Web-based Multicriteria Decision Analysis and ...1119331/FULLTEXT01.pdf · their decisions. Some software that deal with Business Intelligence (BI) and budget de-cisions are in use

IN DEGREE PROJECT CIVIL ENGINEERING AND URBAN MANAGEMENT,SECOND CYCLE, 30 CREDITS

, STOCKHOLM SWEDEN 2017

Web-based Multicriteria Decision Analysis and Visualization for Reinvestments in Power Networks

NATALIE EKROTH

JOSEFIN LENNARTSSON

KTH ROYAL INSTITUTE OF TECHNOLOGYSCHOOL OF ARCHITECTURE AND THE BUILT ENVIRONMENT

Abstract

It can be a hard and time consuming task for a decision maker to decide which parts of anetwork to reinvest in. There are a lot of parameters to take into consideration regardingreinvestments, for example age, number of outages, number of inspection remarks and thedegree of inspection remarks. Without any visualization, it is difficult to detect patternsin the data. Therefor, the decision maker is required to really know the network he/she isworking with and to have a gut feeling of where to reinvest.

The purpose of this thesis is to show that the decision making process can be muchsimpler and better supported when using GIS tools for analysis and visualization. Thisis done by designing a prototype of a web application that can produce multicriteriadecision analysis on the parameters of interest for reinvestments in a power network.Traditionally, heavy desktop clients are for expert users while web-based clients are betterfor layman users. One of the greatest advantages of a web-based client over a desktopclient is that it can be reached externally from any device that has access to internet.Because of this, the prototype is developed as a web-based client. Customer data can besensitive information, this means that the data needs to be secure and directly accessiblefor the users of the application. Therefor, a 3-tier architecture with client, server anddatabase is used.

The result is visualized in a map, which makes it easy for anyone to interpret theresult. Since the prototype is developed to be used by none GIS experts, the weightedlinear combination method is used for the analysis. The prototype is not fully automatedand does not deliver an absolute decision, the goal is rather for it to function as an aid forthe decision maker when deciding on the final reinvestment area.

The prototype is evaluated by the prospective users of the application through a ques-tionnaire and the results show that a tool like this would be very useful for reinvestmentsdecisions. Since the prototype does not rely on topology or network structure, it can beadapted to other spatial decision problems than just reinvestments in power networks.

Keywords: GIS, web-GIS, MCDA, Visualization, Reinvestment

AcknowledgementsDaniel Sedell, (Digpro AB), co-supervisor. For providing help, ideas and support andfor keeping us positive and focused throughout this thesis.

Associate Professor Gyozo Gidofalvi, (KTH Geoinformatics), co-supervisor. For pro-viding feeback, ideas and research material.

Professor Yifang Ban, (KTH Geoinformatics), examiner. For exmaniation of this thesis.

Fredrik Hilding, (Sweco Position AB). For providing help with the start-up phase ofthe prototype development.

Ella Syk, (Digpro AB). For providing ideas and help during the development of theprototype.

Bjorn Persson, (Digpro AB), senior supervisor and Jonas Jacobsson, (Digpro AB).For feedback, inputs and sharing extensive knowledge throughout the work of this thesis.

Susanne Christoffersson (Vaxjo Energi), Peter Karlsson (Vaxjo Energi) and OrjanKvist (Vaxjo Energi). For participating in interview and evaluation and for providing uswith real world data.

Cathrin Backstrand (Jonkoping Energi), Mats Javebrink (Jonkoping Energi) andAshfaq Taimor (Kraftringen). For participating in interviews and sharing importantknowledge and experiences.

Finally we want to address a great thanks to everyone that attended our session at Dig-pro’s customer meeting for listening to the presentation and answering the questionnaire.

I

Declaration of Individual ContributionThe work of this thesis has been divided equally between Natalie Ekroth and JosefinLennartsson. The literature study about related work and common technology was mainlycarried out by Josefin whereas the literature study about web GIS was mostly conductedby Natalie. The interviews and questionnaires were designed, held and summarized byboth of the authors. The methodology, result and discussion were conducted together andit is impossible to divide the individual contribution on these parts of the thesis.

The functionality of the prototype was mostly developed by Natalie and the designof the prototype was mainly implemented by Josefin. However, both authors have con-tributed with ideas and suggestions of functionality and design of the prototype.

II

ContentsAcknowledgments I

Declaration of Individual Contribution II

List of Figures V

List of Tables V

Terms and Abbreviations VII

1 Introduction 11.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 Objectives and Research Issues . . . . . . . . . . . . . . . . . . . . . . . 11.4 Limitations and Delimitations . . . . . . . . . . . . . . . . . . . . . . . 21.5 Disposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Related Work 22.1 Common Concepts and Methodologies . . . . . . . . . . . . . . . . . . . 2

2.1.1 Analysis with GIS . . . . . . . . . . . . . . . . . . . . . . . . . 22.1.2 Spatial Decision Support Systems (SDSS) . . . . . . . . . . . . . 32.1.3 Web GIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.3 Related Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.1 Digpro Products . . . . . . . . . . . . . . . . . . . . . . . . . . 152.3.2 Web GIS Architecture . . . . . . . . . . . . . . . . . . . . . . . 15

3 Research Methodology 163.1 Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1.1 Interviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173.1.2 Data Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . 173.1.3 Basemap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 Prototype Development . . . . . . . . . . . . . . . . . . . . . . . . . . . 183.2.1 Front-End Development . . . . . . . . . . . . . . . . . . . . . . 183.2.2 Back-End Development . . . . . . . . . . . . . . . . . . . . . . 183.2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193.2.4 Operational Layers . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.3 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Results and Analysis 204.1 Interviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204.2 Prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214.3 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

III

5 Discussion 275.1 Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.1.1 Interviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275.1.2 Data Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5.2 Prototype Development . . . . . . . . . . . . . . . . . . . . . . . . . . . 285.2.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285.2.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5.3 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

6 Conclusions and Future Work 306.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

References 32

Appendix 34

IV

List of Figures1 Different representations of the same layer on different background maps 112 Flowchart showing the process of the research . . . . . . . . . . . . . . . 163 Image showing example of WLC for power networks . . . . . . . . . . . 194 The startpage of the prototype . . . . . . . . . . . . . . . . . . . . . . . 215 Meny button clicked . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 Light gray basemap and age layer . . . . . . . . . . . . . . . . . . . . . 227 Dark basemap and age layer . . . . . . . . . . . . . . . . . . . . . . . . 228 Parameter form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 Tooltip shown on hover . . . . . . . . . . . . . . . . . . . . . . . . . . . 2210 Weight form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2311 Alert box with warning message . . . . . . . . . . . . . . . . . . . . . . 2312 Result from analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2313 Click and pan to polygon . . . . . . . . . . . . . . . . . . . . . . . . . . 2314 Result from questions 1 to 6 . . . . . . . . . . . . . . . . . . . . . . . . 2515 Result from questions 7 to 12 . . . . . . . . . . . . . . . . . . . . . . . . 2616 Result from question number 13 . . . . . . . . . . . . . . . . . . . . . . 27

V

List of Tables1 Scale for Pairwise Comparison . . . . . . . . . . . . . . . . . . . . . . . 62 Pairwise Comparison of the Evaluation Criteria . . . . . . . . . . . . . . 6

VI

Terms and AbbreviationsBI Business IntelligenceCSS Cascading Style SheetsCTA Call to ActionDM Decision MakerGIS Geographic Information SystemGUI Graphical User InterfaceHCI Human Computer InteractionHTML HyperText Markup LanguageHTTP Hypertext Transfer ProtocolJSON JavaScript Object NotationMCDA Multicriteria Decision AnalysisMCE Multicriteria EvaluationSA Sensitivity AnalysisSDSS Spatial Decision Support SystemS-MCDA Spatial Multicriteria Decision AnalysisUI User InterfaceURL Uniform Resource LocatorUX User ExperienceUXD User Experience DesignWLC Weighted Linear Combination

VII

1 Introduction

1.1 BackgroundMaking decisions regarding reinvestments in power networks can be a hard and timeconsuming task. There are a lot of parameters that need to be considered and evaluated inorder to reach a conclusion. For example, for a pipe these parameters can be age, numberof outages, previous inspections and more. How to weigh these different parametersagainst each other is not trivial. Today, many electric companies go through these factorsmanually and try to get an overview of where an reinvestment should be made in thenetwork (Sedell, 2016). For these companies, an interactive web-based tool would be ofgreat use and a good way to support and back up their decisions. The tool will be a webapplication where, from case to case, the user can choose which parameters they wantto prioritize and how to weigh these parameters of interest against each other. This willresult in an output map that highlights the areas suggested for future reinvestments in thepower network.

1.2 Problem DefinitionToday a lot of the decisions for reinvestments in the power network are done on “gutfeeling”, which requires expertise and previous knowledge of the network in question.Decisions are highly dependent on who makes them and their knowledge of the network.The people in charge of reinvestment decisions are often people with little or no knowl-edge and experience of working with GIS systems. There is no advanced analysis behindtheir decisions. Some software that deal with Business Intelligence (BI) and budget de-cisions are in use but few of them uses a map component and or an advanced analysis.

1.3 Objectives and Research IssuesThe objective of this thesis is to develop a prototype of a web application that will aid indecision making for reinvestments in power networks. The purpose of the prototype isto:

1. perform multicriteria decision analysis,

2. list areas in need of reinvestments and

3. be visually helpful in decision making by showing these areas on a map.

The main question this research aims at answering is: What type of analysis is ap-propriate for layman users when making reinvestment decisions and how can this infor-mation be efficiently communicated to the decision maker? To reach the answer to thisquestion, goals are set up to achieve along the way. These goals consist of finding outhow companies make these decisions today, if they see a need for a tool that will aid indecision making and what type of analysis that is appropriate for this.

1

1.4 Limitations and DelimitationsThe prototype can be extended with more features that would be of great interest, but dueto time limits only a small portion of features are implemented.

The developed prototype can be used for all types of reinvestment decisions. How-ever, since power network data is used in the present thesis, the prototype is appliedto this. For other types of data, other parameters and aggregation methods need to beconsidered, as described in 3.1.2 and discussed in 5.1.2.

The prototype will only be evaluated by a small subset of users. Since these arethe prospective users of the developed prototype they are assumed to be representative.However, a greater subset of users with different kinds of backgrounds would probablyhave been an even more accurate representation of reality. Due to time limits, this wasnot considered in the present thesis.

1.5 DispositionSection 2 presents related work and other research done on the subjects presented in thisthesis as well as related technologies such as Digpro products and web GIS architecture.This is followed by Section 3, where the methodology for this research is presentedin detail. Further, in Section 4 the results of the research are presented together withscreen shots of the developed prototype. The results are thereafter discussed in Section5. Finally, conclusions and recommended future work are presented in Section 6.

2 Related Work

2.1 Common Concepts and MethodologiesThe main focus of this research can be seen as a combination of decision making withthe help of GIS and the presentation in the form of maps on the web. To provide thereader with the relevant background information, common concepts and methodologiesare summarized in the following subsections.

2.1.1 Analysis with GIS

The abbreviation GIS stands for Geographical Information System (Heywood et al.,2011). There are several viewpoints as to how the term should be defined, one beingthat GIS is a computer system consisting of three principal components; the hardware,software and the appropriate procedures. GIS uses spatially referenced and geographicaldata to carry out various management and analysis tasks, hence it is said that the maingoal of GIS is to assist when making spatial decisions (Malczewski, 1999).

2

2.1.2 Spatial Decision Support Systems (SDSS)

Spatial decision support systems (SDSS) are systems that help the user solve complexproblems that include a spatial component, such as site selection, routing or urban de-velopment (Sugumaran et al., 2011). For spatial problems it is unusual to have exactlythe right kind of or enough data, thus both hard and soft information are almost alwaysconsidered in spatial decisions. Hard information is verifiable data and knowledge whilesoft information includes feelings, opinions and preferences. Both hard and soft informa-tion involve some uncertainty, therefore spatial problems cannot be solved with absolutecertainty. Spatial decision support system is used for the analysis in this research.

2.1.2.1 Multicriteria Decision Analysis A Multicriteria Decision Analysis (MCDA)problem can be described as a problem where the possibilities of actions are based onincomparable and contradictory criteria (Malczewski, 1999). Throughout this thesis theterm MCDA is used although another commonly known name for this kind of analysisis Multicriteria Evaluation (MCE). When the problem also involves the spatial compo-nent i.e., when the result is based on the alteration and union of geographical data, it iscalled spatial MCDA (S-MCDA). The values of a group of evaluation criteria in com-bination with the preferences of the DM are considered to be the vital characteristics ofspatial MCDA. This means that the final result will consist of the relation between thegeographical components in combination with value judgments. Since spatial problemscan be very complex, with their many components and the relationship among them, theymight be of great difficulty to solve for a decision maker. GIS and MCDA can help theDM in reaching better effectiveness and efficiency. S-MCDA is a well known method forsolving spatial problems with contradictory criteria, therefor this approach is used in thepresent thesis.

Malczewski (1999) suggests that there are seven steps in the process of decision mak-ing with MCDA. It starts with identifying the problem and ends with (a) recommenda-tion. Among these there are three main components that need to be thoroughly consid-ered; value scaling, criterion weighting and combination rules (Malczewski and Rinner,2015). These three are described more in depth below.

Value ScalingAfter defining the problem and deciding on the criteria to be included, the next step is toscale the values (Malczewski, 1999). To be able to compare attributes, they have to be onthe same scale. There are several methods with which attributes can be made commensu-rate. The methods are usually divided into deterministic, probabilistic or fuzzy. Amongthe deterministic approaches, the linear scale transformation is considered to be the mostcommonly used one. With the linear scale transformation the data is stretched linearly.There are two linear scale transformation methods mentioned in the literature; maximumscore and score range, where the score range is the most frequently used method for stan-dardizing evaluation criteria. When applying the linear scale transformations one has tofirst decide if the criterion is considered to be a benefit or cost criterion. Benefit criteria

3

are criteria where high values are preferable, for example the further away from waterthe better the location. Cost criteria are thus the opposite and low values are preferable.There are two mayor differences between these two methods. By standardizing with thescore range procedure the worst possible scenario always gets the value of 0 and the bestthe value of 1, and all other values are ranged in between. For the max score method onthe other hand, the highest value is not necessarily 1 and the lowest is not always equalto 0. The max score keeps the proportional changes whereas the score range does not.

x′

ij =xij − xmin

j

xmaxj − xmin

j

(1)

x′

ij =xmaxj − xij

xmaxj − xmin

j

(2)

Equations 1 and 2 display the calculations of the score range procedure. Equation 1 isapplied to benefit criteria, whereas Equation 2 is applied to cost criteria.

x′

ij =xij

xmaxj

(3)

x′

ij = 1− xij

xmaxj

(4)

Equations 3 and 4 display the calculations of the max score procedure. It standardizesthe the data by dividing the raw data with the maximum value of each criterion. Equation3 is applied to benefit criteria and Equation 4 is applied to cost criteria. For Equations1 - 4, x

′ij is the standardized score for the i:th object and the j:th attribute. xij is the raw

score, xmaxj is the maximum score for the j:th attribute and xmin

j the minimum score forthe j:th attribute.

As mentioned above, there are multiple ways in which the attribute values can bescaled (Malczewski, 1999). For the deterministic methods there are something calledvalue / utility function approaches. The shape of the value function is determined by thedecision maker’s preferences and in practise it is often approximated by performing amid-point value method (Malczewski and Rinner, 2015). This means that the decisionmaker first decides on the end points, i.e assigns the values of 0 and 1. After that thedecision maker chooses what value would be the mid-point of these two and thus assignsit 0.5. After this a value function can be derived or more mid-points can be assigned tofind the most fitting function.

Besides the deterministic ways to standardize maps other methods are based on prob-ability theory and fuzzy logic. Scaling with fuzzy logic is done by first specifying afuzzy set membership function and then assigning a value to a decision alternative basedon its membership (Malczewski and Rinner, 2015). The common fuzzy set membership

4

functions are sigmoidal, J-shaped, linear or user-defined. The probabilistic way of deriv-ing commensurate maps is based on probability theory (Malczewski, 1999). This has todo with the likelihood of a value occurring. Numbers between 0 and 1 can be used torepresent the relative frequencies with which the different possible outcomes occur. 0 isassigned if it is impossible for the event to occur and 1 is assigned it it is certain that anevent will occur.

In the case of this thesis the authors’ unawareness of the distribution of the parame-ters precludes any option but a linear standardization, as discussed in Section 5.

Criterion WeightingWhen all criteria have been made commensurate it is time to weigh them against eachother. The method that is considered to be the simplest one is called ranking (Malczewski,1999). With ranking methods the criteria are ranked in ordered and then normalized. Forexample, with straight ranking the most important criterion is set to 1, the second mostimportant to 2 and so on. After ranking the criteria there are different methods to derivenormalized weights. Another way of assigning weights is by something called ratingmethods, where one of the simplest approaches is called the point allocation. It is amethod in which the DM distributes 100 points between the criteria of interest. Thismeans that if one criterion is given 100 points it is the only one that will influence theoutput and if a criterion is given 0 points it will be ignored. As an example, if the DMis considering three criteria, the points can be allocated to be [33,33,33] if they are con-sidered to be of equal importance or [50,25,25] if one is more important but the othertwo of equal importance, and so on. In comparison to ranking methods, with rating sev-eral criteria can be considered as equally important and the 5th criterion does not haveto be 5 times worse than first one. These two methods can be criticized for their lack oftheoretical foundations. The attributes need to be clearly defined for these methods tobe of any use. On the other hand, if the person distributing the weights is not familiarwith advanced weighting techniques, the rating and ranking methods can be a simple wayto assign weights. Since the prototype developed in this particular thesis addresses GISlaymen users, the weights will be assigned by rating. A more complex method, pairwisecomparison, will be briefly described below since it is often mentioned and used. SeeSection 5 for the discussion of choice of weighting method.

The pairwise comparison, was originally developed as a part of the Analytic Hierar-chy Process (AHP) (Malczewski, 1999). The weights are derived from a matrix wherethe DM compare the relative importance of the criteria against each other. See Table 2for an example of a pairwise comparison matrix and Table 1 for the comparison scale.Once the pairwise comparison matrix is developed, a process of developing normalizedweights can begin.

5

Table 1: Scale for Pairwise Comparison

(Malczewski, 1999, p.183)Intensity of Importance Definition

1 Equal importance2 Equal to moderate importance3 Moderate importance4 Moderate to strong importance5 Strong importance6 Strong to very strong importance7 Very strong importance8 Very to extremely strong importance9 Extreme importance

Table 2: Pairwise Comparison of the Evaluation Criteria

(Malczewski, 1999, p.183)Criterion Price Slope View

Price 1 4 7Slope 1

41 5

View 17

15

1

Combination RulesAt the end of MCDA you want to know which alternatives are the best and / or the worst(Malczewski, 1999). To be able to order the alternatives you need some kind of decision/ combination rule. Among the many decision rules to chose from, additive rules arebest known and most widely used. When handling spatial multiattribute decision makingthe most commonly used techniques are simple additive weighing, or Weighted LinearCombination (WLC) which are based on weighted average. The acronym WLC refersto the process of the combination rule; the criterion is firstly scaled linearly, followedby being assigned a weight and then all of these are summed together. The criteria aredirectly given weights according to their relative importance as perceived by the DM. SeeEquation 5 for the mathematical expression of WLC.

Ai =∑j

wjxij (5)

In Equation 5 xij is the score of the i:th alternative with respect to the j:th attribute, wj

is a normalized weight. Ai is the sum of all xij multiplied by their respective weight.

Ai =∑j

wjxij = 0.1 ∗ 0.5+ 0.2 ∗ 0.25+ 0.3 ∗ 0.25 = 0.05+ 0.05+ 0.075 = 0.25 (6)

6

Equation 6 shows an example where WLC is used. Three criteria are firstly scaled tohave values of 0.1, 0.2 and 0.3 and are then assigned the weights of 0.5, 0.25 and 0.25respectively. According to Equation 5 the final sum for this attribute is 0.25.

If paying attention one can note that there is a trade-off to be aware of when usingWLC (Drobne and Anka, 2009). If a parameter with a high original value gets weighedlow (value of 0.3 * weight of 0.25=0.075 ) it might still add to the final value more than aparameter with a low original value that gets weighed high (value of 0.1 * weight of 0.5 =0.05). WLC can be said to offer full trade-off, being that a low value can be compensatedby a high given weight.

For this thesis the WLC combination rule is used since it lets the user set the weightsin one straightforward step and leaves a lot of control to the decision maker. Section 5 willdiscuss the benefits and challenges with the method. To be able to discuss the differenceof methods further, two other methods that are often mentioned regarding combinationrules are also described, Analytic Hierarchy Process (AHP) and Ordered Weighted Aver-aging (OWA). The AHP was developed by Thomas Saaty in 1980 (Malczewski, 1999).It is used with pairwise comparison and consists of three principles; decomposition, com-parative judgment and synthesis of priorities. This means that the decision problem firsthas to be decomposed into a hierarchy. The comparative judgment refers to pairwisecomparison of the elements within a level of the hierarchy. The final step is about con-structing an overall priority rating. In general one can say that AHP uses the pairwisecomparison firstly when comparing the criteria and then again for comparing the alter-natives to finally sum it all together. OWA is a weighted sum with ordered evaluationcriteria. This means that both criterion weights and ordered weights are applied. Withthe order weights the level of trade-off between criteria can be chosen. In OWA the firststep is to create ranked layers. The ranked layers are created by going through each pixelone by one and for the highest ranked layer only selecting the best pixel from all criterionlayers. This means that the rank 1 layer will include only the pixels with the highestvalues from all criterion layers, rank 2 will have the second best pixels and so on. Eachpixel in the ranked layers are then multiplied by its respective criterion weight. If takenfrom layer x, multiply by criterion weight a, if taken from layer y multiply by criterionweight b. Finally the ranked layers (with their criterion weights already multiplied in) aremultiplied by their respective order weight. WLC can be seen as a version of OWA wherethe ordered weights are set to be equal (Drobne and Anka, 2009). When setting equalorder weights the step of ranking and multiplying by order weights is thus unnecessary.

2.1.2.2 Sensitivity Analysis Sensitivity Analysis (SA) is a part of MCDA and refersto how the errors in the input data affect the error in the final output (Malczewski, 1999).The aim of multicriteria spatial error analysis is basically to evaluate the effect of errorsthat the criterion maps and set weights have on the decision outcomes. To perform sen-sitivity analysis an analysis of uncertainties must first be done (Malczewski and Rinner,2015). The main sources of uncertainty are the values and weight of the criteria. At thispoint there is no method for choosing the optimal MCDA model. If applying differentmulticriteria decision rules to the same decision problem the results will be inconsis-

7

tent. When selecting a MCDA model one should consider the nature of the problem,data requirements, consistency of results and computational complexity. To summarize,a sensitivity analysis is done to conclude the robustness of the recommended solution.The goal of this thesis is not to find the most accurate result from the analysis, but tosee if a web-based analysis tool like the one presented in this thesis would be useful fordecision makers in power network business. For this reason a sensitivity analysis in thissense is not preformed in this thesis although it is further discussed in Section 5.

2.1.3 Web GIS

Prior to the development of Web GIS most of the GIS applications were designed forspecialists working on desktop computers (Fu and Sun, 2011). Web-based applicationsdo not require a locally installed software and have therefor changed the use of GIS bymaking it more accessible to the public. There are a lot of benefits with Web GIS insteadof desktop GIS (ESRI, 2017). One of the advantages with Web GIS is that the applicationis easily available via a Uniform Resource Locator (URL). The application can be easilyaccessed via any device that has access to the Internet. This means that the applicationcan be reached by multiple users at the same time, in contrast to the desktop applicationwhich can only be used by one user at a time. This also brings a challenge to the WebGIS - it needs higher performance and scalability than the desktop GIS. Due to the factthat GIS applications are not only used by GIS experts but also by users with no GISbackground, one of the challenges is to design for simplicity and comfort (ESRI, 2017).This means that the User Interface (UI) and the User Experience (UX) need to be takeninto consideration in order to create a successful GIS. In this section, these terms aredeeper investigated.

2.1.3.1 Human Computer Interaction Before the later 1970s, the ones who inter-acted with computers were mainly information technology specialists (Carrol, 2017).When computers became more available for layman users the need for research in Hu-man Computer Interaction (HCI) increased rapidly. HCI refers to design and use ofcomputer technology with focus on the interfaces between people and computers. Onepurpose of HCI is to increase the usability of computer technology. Within HCI the termusability is continually reconstructed but it often includes properties like fun, flow andaesthetic tension to mention a few. Usability is a concept with no end and it can not bereduced to a static checklist.

As the focus of HCI is moving towards User Experience (UX), more research is doneon what creates a good experience for the user (Vermeeren et al., 2016). Therefor, furtherstudies are done on emotional aspects rather than only functional aspects. Design forexperience needs to affect the users feelings in a positive way, thus the role of design hasbecome more and more important for HCI. This is further looked into in Section 2.1.3.1and Section 2.1.3.1.

HCI needs to overcome many obstacles, one of them is that different users have differ-ent cognitive style, i.e different ways of learning and keeping knowledge (Rouse, 2017).

8

Furthermore, the user interface technology is a field that is changing rapidly, causing newinteraction possibilities that previous research cannot be applied to. This means that therewill always be a need for new research in the field of HCI.

User InterfaceThe User Interface (UI) is the way a user interacts with a computer or other electronicdevices. The UI consist of both software interface, i.e. menus, buttons and other controlsin a web page, and hardware interface, i.e. keyboard and mouse, to control the software(Stopper, 2012).

The first user interface for GIS was command-line interface, where the user interactedwith the computer by typing commands and the system responded with text in the com-mand prompt (Egenhofer and Kuhn, 2010). These GIS were mainly accessed and usedby experts. In the second half of 1980 GIS was developed with a Graphical User Inter-face (GUI) which included windows, menus, buttons, icons and other controls. The GUIallowed the user to interact with the software mainly through hardware such as keyboardand mouse (Stopper, 2012). With GUI, GIS became more accessible since the user nolonger needed to remember and understand all commands, they were integrated with theGUI (Egenhofer and Kuhn, 2010).

GIS has long been developed by need of functionality and the user interface has notbeen prioritized (Egenhofer and Kuhn, 2010). Recently, GIS developers have started tounderstand the importance of a good GUI as well, not just good functionality. Although,since GIS often includes a lot of data and functionality it is a hard task to design a goodUI that all users, despite their GIS knowledge, can understand and appreciate (Andersen,2015).

User ExperienceUser Experience (UX) is all the experiences that a person has when interacting with adigital tool. These experiences are for example physical, emotional and mental (Stokes,2017). It refers to the overall satisfaction a user gets from interaction. When designingfor the user, the following questions are good to think about:

• Who is the user?

• What are the user’s wants and needs from your platform?

• What are the user’s capabilities, web skills and available technology?

One of the major things of UX is usability. Usability is about making the interactioneasy and intuitive (Stokes, 2017). The users should not have to think, they should justdo. The most important part of usability is sticking to standards, for example to havenavigation menus at the top or left of the web page. Another thing to think about is todesign the components in the same way (Andersen, 2015). For example, if the closebutton is in the top-right corner in one window, it should be located in the same placein other windows as well. Furthermore, the mouse click should behave in the same waythroughout the application. One example of this is that a right click on one object in the

9

application should generate the same response as a right click on another object in theapplication.

Another major thing in UX is simplicity, the simpler the better (Stokes, 2017). Onepart of simplicity is to have a lot of empty spaces since it makes it easier for the user tonavigate in the web page. Further, when the user needs to make a lot of choices there isa physiological part that kicks in. The user might worry about making the right choiceand similar, therefor fewer options makes the experience better for the user. Additionally,clear, simple and plain language is important when designing a good UX. If there are alot of functions in the application, it is a good idea to group similar functions together toachieve a cleaner interface (Andersen, 2015). This will make the interface feel lighter forthe user. It is also good to keep the main functionality easily accessible for the user.

Besides usability and simplicity, credibility is of great importance in UX (Stokes,2017). If the web page looks professional and trustworthy the experience will be betterfor the user.

User Experience DesignThe success of a digital product, a web page for example, depends on how the user in-terprets it (Stokes, 2017). A good User Experience Design (UXD) can please a customerand generate more customers whereas a bad UXD can lead to less customers. A greatUXD is reliable, functional, convenient and more importantly it is enjoyable and givesan experience worth sharing.

When developing a web page with focus on UXD, it is good to start with creating thebasic structure of the page (Stokes, 2017). Most web pages have a hierarchical structurewith broad important pages on the top and narrow less important pages in the bottom ofthe structure. The second step is to analyze the content of the web page i.e what contentis needed and how and when it should be created. This is done by analyzing what thesite should achieve, what the user wants and needs and the tone and language of the site,to mention a few. The next step is to create a sitemap of the web page. A sitemap is astructured plan for how the pages of the web page will be organized. Further, the layoutof the web page needs to be visualized. The layout of a web page is designed basedon the type of page but it typically consists of four elements; header, footer, side barsand central content. The main navigation menu, search tools and login features shouldbe placed in the header. The footer is used for important but not leading features, suchas legal information and additional navigation elements. Secondary content and toolsshould be presented in the side bar. Finally, the central content area is used to present themain content of the web page. The final step is to fit together all the other elements ofthe web page such as Call To Actions (CTA) and forms.

It is important to understand that one user experience design cannot and will not workfor everyone in every situation, since all human beings are different (Stokes, 2017). Thus,the user experience design from one web page cannot be directly copied to another webpage. The design needs to be adapted to the goals, values and product for every individualweb page.

10

2.1.3.2 Web MappingIt is difficult to detect patterns in raw data, for example several rows of number or text(Murray, 2013). However, when data is visualized in a bar chart, a pie chart or othersimilar visualization techniques, even small children can detect patterns in the data. Thus,visualization is a good way of communicating data.

A map is a graphic representation that shows spatial relationships (Tyner, 2010).Maps are often generalized in the sense that only the interesting information are ex-tracted, in contrast to a photography where all available information is present. Whencreating a map, the mapmaker needs to know the purpose of the map and extract the rightkind of information. For example, a sea map does not need to have road information.Since the mapmaker decides what to show on the map, a map is always biased in onesense. The map has moved from paper to digital, thus the maps are no longer designedfor map readers but to map users (Muehlenhaus, 2013). Web maps are expected to beinteractive, responsive and possible to manipulate to fit the user’s need.

Figure 1: Different representations of the same layer on different background maps

According to the Swedish Standards Institute (SIS), web map services should offermap layers that include few and associated object types (SIS, 2015). This means that theuser should be able to choose the object types that are relevant for the application and beable to turn on and off object types in the map. Additionally, the web map service shouldoffer different styling of the background map to be able to emphasize the object types inthe map, this can be seen in Figure 1. The level of details in the map should partly bedependent on the zoom level; zoomed in views should show more details than zoomedout views. SIS states that this principle is often used for background maps, but is not ascommonly used for other map layers. The guidelines from SIS is taken in considerationfor the prototype development in the present thesis.

2.2 Previous WorkThe field of GIS can be seen as fairly wide and several studies can be found that coversits evolution. The following section gives a brief summary of some of the things thathave been researched regarding SDSS, MCDA and GIS on the Web. This provides anoverview of what has been done and what experts in the field think regarding differentapproaches, which is of great importance when making a decision regarding the method-ology for this thesis.

11

Jankowski et al. (2001) presented a map-centred exploratory approach to multiplecriteria decision making. The authors state that maps are sometimes used to visualizeand / or evaluate the results however they are usually not used as decision support tools.With this, they saw the relevance of presenting new spatial decision support tools wheremaps play a more important role in the structure of multiple criteria spatial decision prob-lems. With this tool the user is for instance allowed to order decision options and assignpriorities to decision criteria. They ask the question “What are the effective means of us-ing maps in order to support decision problem exploration and structuring?”. They thenpresent four multicriteria cases, with a different decision support tool for each. Finallythey conclude that a high level of interaction between maps and attribute data graphs aidsthe decision maker in an understanding of the problem structure. Just as in this article,the current thesis presents interaction between maps and attribute data. The result of theanalysis is presented in a map as well as listed in a table.

Karnatak et al. (2007) developed a multicriteria decision analysis tool for spatial de-cision making in the Web GIS environment. The authors emphasize how the traditionalGIS can only serve devoted users with desktop software and how the web enabled GISgives users convenient and efficient access to the system. Although multicriteria tech-niques such as AHP help selecting the optimal alternative, they discovered that expertknowledge is of great importance when assigning weights. The present thesis also usesa web-GIS environment and the experts are allowed to select the weights, thus keepingthe importance of the experts’ knowledge. The current thesis uses the combination ruleWLC and not AHP as in this article.

Rinner (2007) suggested in this article the principles of combining MCE methodswith Geographic visualization (GeoViz) to support spatial decision making. The ap-proach is tested on the Urban Quality of Life in Toronto, Canada. The method chosenfor measuring the quality of life is the Analytic Hierarchy Process (AHP) which accord-ing to the author gives the user a way of interacting with decision making strategies andshowing spatial patterns in the evaluation results. The method was evaluated on util-ity and usability by interviewing three users. Rinner (2007) suggests that since the useof GeoViz tools is accelerating, there is also a requirement of correctly evaluating theirusefulness. When looking at design principles from studies within HCI, researchers havefound a difficulty in measuring the success of GeoViz tools. In the case study presented inthe article one of the difficulties lies in the fact that there is no agreed upon definition andmeasure of QoL. The challenges faced when evaluating the QoL include the definition ofthe neighbourhoods, the choice of parameters that affect QoL and the processing of theseparameters. By MCE the QoL indicators can be weighted individually. In the case studytwo models are compared, one modern and one contemporary. The difference betweenthe models is the indicators involved in deciding the QoL. The two models are then com-pared by weights that are set by sliders. By putting one model at 100 percent weight andthe other at 0, the areas displayed are significantly different. For the evaluation, domainexperts were interviewed and they answered questions while an investigator operated the

12

tool. It was concluded that when an analyst is allowed to manipulate the MCE settingsthey can observe changes in the result and compare it to their previous knowledge. Re-sults from the case study showed that even though the interviewee was already awareof some spatial patterns regarding the QoL in the area, the visual analytic approach waswell received. Future research should look at the possibility of decision makers to ma-nipulate the original attribute values and their weights. Just as in this article, the presentthesis evaluates the prototype by letting the potential user see the prototype in action, askquestion and fill out a questionnaire. Weights are set by using sliders. HCI and UX istaken into consideration when developing the user interface. The difference between thisarticle and the present thesis is the choice of combination rule. The article uses AHPwhereas this research uses WLC.

Jankowski et al. (2008) presented a concept of a Web-based spatial multiple criteriaevaluation tool for individual and group called Choice Modeler (CM). The objective ofthe article was to present a prototype that used current web-based technologies and withthis contribute to the process of developing MCE as either a part of SDSS or a stand alonemethodology. The purpose of the CM was to be used as a tool for evaluation of decisionvariants, which would help to reduce the complexity of the decision having to be madeby the decision maker regarding the multiple decision options, evaluation criteria andcriterion weights. The authors mention previous research done in the area of web-basedspatial decision support systems where they believe that a lot of it has covered applicationspecific models for what-if scenarios and visualization of such, although not much hasbeen done on the study of tools that amplify the human judgment about the componentsof the decision situation. The article questions whether MCE tools should be moved fromdesktop and towards web services. For the Choice modeler the three-tier architecture, aversion of a distributed architecture, is used as system implementation. The three-tierarchitecture uses a client tier, middle-ware tier and the data storage tier. In this researchthe CM server retrieves data from the database and executes MCE functions selected bythe user. The CM represented the beginning of providing MCE functions in the shape ofdistributed web services. The authors suggest that the next step in research should be todevelop MCE functions as open source standardized Web services. In accordance withthis article, the present thesis also presents GIS trough web services and with a three tierarchitecture.

Ligmann-Zielinska and Jankowski (2008) presented a framework for performing sen-sitivity analysis in spatial multiple criteria evaluation. The framework is organized as aguide in selecting the SA technique most appropriate for the problem . The authorsstress the lack of research regarding spatial sensitivity analysis. Since the spatial MCEproblems involve the spatial component there should also be a way of evaluating if themethods and weights should be the same for an entire layer throughout space. Maybeone criteria should be looked at differently in different places, for example urban andrural. The goal of the framework is to help when making a decision about what kindof SA best suits the problem at hand. This framework was looked at to help figure out

13

what kind of evaluation should be done for the current research. No spatial SA is donein the current thesis due to constraints such as time and underdeveloped methods but isdefinitely something that could be considered in the future when more research has beendone.

Drobne and Anka (2009) did a study on merging GIS and MCDA as well as com-paring the two MCE methods WLC and OWA. The authors claim that about 80 percentof data used by managers or decision makers include a geographical component. Mapshave been used as support in decision making for a very long time. It is important tonote that no GIS by themselves make decisions, they are simply there to aid the decisionmaker. According to the authors, the limitations of WLC include the trade-off amongcriteria as well as the scaling. As the name suggest the scaling of WLC is linear, but insome cases a non-linear scaling might make more sense. One advantage with WLC is theability to give relative weights to each of the factors. WLC can be seen as one variant ofOWA, where the order weights are given equal importance. With order weights one morestep is included where the decision makers themselves can decide how much trade-offis desired. In using OWA there are three groups that the criteria should be divided into;hard constraints, factors that can trade-off and factors that should not trade-off. WhenOWA method is used in the case study, the factors are first divided into two groups; en-vironmental concerns and development costs, since they do not have the same level oftrade-off. For the cost factors a full trade-off and average risk was selected and thus theWLC method was used. For the environmental factors order weights that gave both lessrisk and less trade-off was selected. When finally combining the two layers an OWA ap-proach with low risk and no trade-off was chosen (order weights of 1 and 0). Accordingto the authors the main purpose of the application was not to find a suitable place in thearea of the case study, but rather to describe and test the WLC and OWA methods ofMCE. The practise and research of GIS-based multicriteria decision making is rapidlygrowing, and the tools provided today give decision makers of spatial decision problemsadvantages. Still, there are some topics that need to be further investigated and devel-oped. The authors list them as: selection of attributes, weights, scale and methods foraggregation, error assessment and the inclusion of database and decision rule uncertaintyand sensitivity analysis. The pros and cons of WLC and OWA in this article are lookedat and for the current thesis the WLC is chosen.

According to Silva et al. (2014) there is a need to investigate how to integrate GIS,MCDA, the Internet, modelling and databases with the goal of creating Web Multicri-teria Spatial Decision support systems (Web MC-SDSSs). Therefor this article presentsa fully integrated system for combining GIS and the MCDA method called ELECTRETRI with the help of ArcGIS software. This specific article applies this method on acase study where the sustainability of dairy farms is analyzed. The authors claim that thegreatest benefit of using Web services is the fact that there are no limitations in terms oftime, data and communication. Through the web, the services can be accessed conve-niently and effectively. Silva et al. (2014) list the three major advantages of integration

14

of GIS and MCDA as; enhancing the evolution of GIS, improving the desired level ofusability and enriching approaches to problem solving. Within the MCDA-GIS integra-tion several decision rules have been suggested, among them the most common ones areWeight Summation/Boolean Overlay, Ideal Reference Point, Analytic Hierarchy Processand outranking methods (being the one mentioned in this article). In the concludingpart the authors mention that a limitation with this approach is the requirement of Inter-net connection. The current thesis also uses web-services to create a prototype, but theprototype stands alone and not on top of ArcGIS.

2.3 Related TechnologyIn this section related technologies such as Digpro products and web GIS architecture isdescribed.

2.3.1 Digpro Products

Digpro is a company that offers different kinds of GIS solutions. Their application, dp-Power, can be used for representation of power networks. Different modules can be addedfor further functionality, for example the module Operator includes functions for troublecall handling and outage management. The module Maintainer is as the name impliesused for planning, execution and follow-up of inspections. Data from dpPower, Operatorand Maintainer will be used for this thesis.

dpWebmap is a web-based GIS solution that is available as complement to all Dig-pro’s utilities. In dpWebmap, the network and background data can be visualized withany web browser. This allows the customers of Digpro to make their data available bothexternally and internally. The base of dpWebmap is used as a guideline for the prototypedevelopment in this thesis.

2.3.2 Web GIS Architecture

A three-tier architecture is often used when developing a web application. This meansthat the user-interface (presentation tier), the data access (logical tier) and the data storage(data tier) are developed as three different modules (AL-Mukhtar and Hadi, 2012).

In a web GIS the data is stored in a GIS database, the data access is done by a GISserver and the user interface is a client (Fu and Sun, 2011). The client can be a webbrowser, desktop application or a mobile application. Upon interaction from the user, theclient sends a request to the GIS server which sends the request to the GIS database. TheGIS database sends back the requested data to the GIS server which processes the dataand then sends the result back to the client. Finally the client presents the results to theuser.

15

The functions and the performance of the web GIS server are the most important partsof a web GIS application (Fu and Sun, 2011). A GIS database stores and manages thedata for a GIS (Fu and Sun, 2011). This kind of database can handle both spatial data(points, lines and polygons) and non-spatial data. Some GIS databases store collectionsof features while others store the data model that defines spatial relationships, for exampletopologies or networks. The database can range from small, single-user database up tolarge multiple-user database where the latter allows simultaneously accessing and editing.The GIS application is never better than the quality of the data stored in the database.Therefor, it is important to think about the purpose of the GIS application when the datais collected - a professional application needs good, current geographic information. Theclient can either be a web browser, a desktop application or a mobile client (Fu andSun, 2011). Web browsers are the most commonly used clients for web GIS. Earlier,the browser clients where static and tedious. Nowadays, with technologies like AJAX(Asynchronous JavaScript and XML) and additional APIs, one can create a dynamic,interactive and user-friendly interface which can perform many types of GIS operations.

3 Research Methodology

Figure 2: Flowchart showing the process of the research

The methodology can be divided into three main parts consisting of five steps as seenin Figure 2. The preparation phase, followed by the development phase and finally theevaluation phase. After the first evaluation the prototype might be adjusted according torecommendation before the final evaluation.

3.1 PreparationBefore the development can start some preparation is necessary. The following sectionsdescribe this process.

16

3.1.1 Interviews

Visualization can be very different for someone who is used to looking at maps com-pared to someone who normally work with lists and tables. At the same time creatinga tool when you are not sure what the important features are can be really hard with-out asking someone who is familiar with the subject at hand. Therefore, to understandwhat the customer actually wants and demands from a tool like the one handled in thisresearch, interviews are carried out with three power company customers of Digpro. Allof the interviews are conducted with people involved in the decision making process ofreinvestments.

The purpose of the interviews is to understand how the companies make their de-cisions today and if they see a need for a tool that will help them provide means fordecisions. Apart from general questions like the subjects role in the decision makingprocess and their GIS experience, open-ended questions are asked in order to allow forfree-form answers. The main questions are listed below:

• How is the decision process carried out today?

• Which are the important parameters when looking for reinvestments?

• How should the result be aggregated and visualized?

3.1.2 Data Handling

The data for this thesis is extracted from a real customer of Digpro and consist of bothspatial and non-spatial data that is stored in an Oracle database. The spatial data containsline objects (for example cables) and node objects (for example substations). The cus-tomer data includes a lot of sensitive information, therefor the data needs to be anonymizedbefore it can be used in the web application. Furthermore, the database that the data isextracted from consists of a lot of data that is not of interest in this thesis, therefor onlya few schemas are extracted. The extracted schemas include base data (ID, installationyear and geometries to mention a few), outage data (for example length of outage andnumber of affected customers) and inspection data (for example number of inspectionremarks and the degree of the remark).

The outage data gathered for this thesis is grouped at bay level and this data cannotbe extracted for individual objects. To be able to perform a consistent analysis, the rest ofthe data needs to be aggregated at bay level as well. A bay can be defined as the physicalbox in a power station along with the outgoing line that includes all the objects that areconnected under the bay, for example cables and delivery points. Every object belongsto one and only one bay and in the data the bay is stored as an attribute for each object.

17

3.1.3 Basemap

The basemap provides the geographic context of the application. Since the applicationin this thesis is developed for power networks, a basemap that includes main roads andbuildings is chosen. Since too many details in the basemap can disturb the visualizationof the network, a light gray-scale with a low level of detail is used for the basemap. Tobe able to enhance the colors of some of the operational layers, a dark basemap is alsoavailable.

3.2 Prototype DevelopmentIn order to see if the chosen analysis and visualization are conceivable, a prototype isdeveloped. The development is described in detail in the following sections.

3.2.1 Front-End Development

The front-end of the web is what the user can see and interact with i.e. the UI. The UI ofthis prototype is created with HTML5, CSS and JavaScript. The biggest challenge of thefront-end development is to design the web page so that the user is faced with relevantand useful information. In the ideal web application, the user does not have to think tomuch about the functionality, they should understand by the interface how it works. Inorder to try to achieve this and to create a good UX, the keywords usability, simplicityand similarity are taken into consideration. Since most of the intended users are familiarwith Digpro’s software, the GUI of dpWebmap is used as inspiration. This can mainly beseen in the side navigation bar of the prototype. Furthermore, the concealed functionalityto click on an entry in the resulting table and zoom into the corresponding feature in themap is also a behaviour that is included in dpWebmap.

The open source JavaScript library OpenLayers 3 is used to create the map and thevector layers. OpenLayers 3 also includes a lot of functions that allow for interaction andmanipulation of spatial data that is included in the prototype, for example convex hull.

3.2.2 Back-End Development

The back-end of the web consists of the parts that the user cannot see or interact with, i.e.server and database. In the web application developed in this thesis the communicationbetween the database and the client is done by a Python server. This server was createdby Ella Syk and Fredrik Hilding in a master thesis conducted 2016, Syk and Hilding(2016).

After the relevant data is extracted from the original database, the Oracle geometryobjects need to be converted into GeoJSON format. This is done directly in the databasewith a in Oracle 12.2 built in function. The original data consists of base data, outagedata and inspection data and is stored in separate tables. In order to get all of the relevantdata in one single table, these different tables are joined together. This makes it easier towrite the queries on the server side and also improves the response time for the prototype.

18

Due to the fact that data sent between the server and the browser have to be plain text,the Python server has to query the database for JSON-data. JSON is a text that easilycan be converted into JavaScript objects and vice versa. In Oracle 12.2 there are built infunctions to aggregate data into JSON-data, these functions are used in the queries on theserver in order to get data that is easy to handle in JavaScript.

3.2.3 Analysis

The combination rule for this research is WLC. It is done with a linear score range scal-ing. The scaling will give all parameters a value between 0 and 1, where 1 refers to anelement with poor values and thus in need of reinvestment. The weighing is done throughthe rating method of point allocation.

Figure 3: Image showing example of WLC for power networks

Figure 3 shows an example of how WLC is performed on power network data thathas been aggregated to bay level (as mentioned in Section 3.1.2). In Figure 3 the bays arerepresented as blue polygons. The first step of the WLC is the scaling of the parameters,which is done as a function in the JavaScript. The weights are set by the user through anHTML form presented in the interface and the final summation is done with a functionin the JavaScript after the parameters and weights are chosen.

For the presentation of the result, a table is displayed. In this resulting table, the baysare sorted in accordance to their final sum. The bay with the highest sum is ranked asnumber 1 and displayed at the top of this resulting table. This is the bay in greatest needof reinvestment. Besides the rank, this resulting table also shows information about theID of the bays and a colored dot that indicates the final score of the bay. The coloringof the dots in this resulting table is red for values between 1 and 0.8, yellow for valuesbetween 0.8 and 0.4 and green for values from 0.4 and below. An example of how thisresulting table looks for the prototype developed in this thesis can be seen in Figure 12in Section 4.2.

19

3.2.4 Operational Layers

The operational layers are layers that the user work directly with or layers that are aresult of an algorithm in the application. The operational layers are placed on top of thebasemap. In this thesis one of the operational layers includes the network which consistsof all arcs and points. This layer is styled similar to how the network is styled today indpPower in order to achieve similarity.

According to a new law, the power companies are not allowed to charge for networkparts that are over 40 years old. Due to this, a layer that includes arcs and points that areover 35 years old is created. This layer makes it easy for the customer to quickly see theold network parts in the map.

Another operational layer is created from the result of the analysis in the application.The features of this layer (the bays) are created by a convex hull around the arc and pointfeatures. Each arc and point belongs to a specific bay. After several joins in the databasethe associated bay is stored as an attribute for each arc and point feature. This attribute isthen used to create the convex hull for each bay.

3.3 EvaluationEvaluation of the research is done by sending out a questionnaire to the customers. Thequestionnaire is first sent out to the interviewed customers. After this, the prototype isadapted to the opinions that are expressed by the customers. The questionnaire is alsoanswered by the customers that are attending the presentation of this thesis at Digproscustomer meeting. However, the prototype is not adapted after this questionnaire dueto lack of time. The questionnaire concerns questions about functionality, design andvisualization techniques. The questionnaire can be found in Appendix 6.

4 Results and AnalysisThis part starts with presenting the result from the interviews in Section 4.1. Followedby several screen shots of the developed prototype in Section 4.2. Lastly, a summary ofthe evaluations is presented in Section 4.3.

4.1 InterviewsAfter conducting three telephone interviews with customers of Digpro the relevant back-ground information was gathered to be able to proceed with the prototype. The mainquestions asked were how the decision process is carried out today, if they think it worksperfectly or if they see a need for a new tool and also what they consider to be the impor-tant parameters when deciding on their reinvestments.

The current decision making process is dependent on expertise and it is important thatthe decision maker knows the network. The companies have the information, for exampleage of the network parts, in an excel document that they take into consideration when they

20

are looking for reinvestments in the network. None of the interviewed companies useany analysis or algorithm to support the reinvestment decisions today. After describingessential functionalities of the proposed prototype they all thought that a web applicationlike the one presented in this thesis could be helpful.

In the interviews, the companies presented which parameters that they often lookedat when doing reinvestments. These parameters are listed below:

• Age of the network

• The outages in the network

• The result of inspections in the network

• What type of network part (free or isolated)

4.2 Prototype

Figure 4: The startpage of the prototype Figure 5: Meny button clicked

In Figure 4 the startpage of the prototype is shown. The menu button is placed in the topleft corner. In Figure 5 the menu button has been clicked and the left side navigation barhas been extended. In this menu the user can select which base map to show and whichoperational layers to place over the base map.

21

Figure 6: Light gray basemap and age layer Figure 7: Dark basemap and age layer

In Figure 6, the age layer and the light background map has been chosen. The darkbasemap overlayed with the age layer can be seen in Figure 7. In the age layer, thenetwork parts between 35 and 40 years old are styled orange and network parts over 40years old are styled red. This information is shown in the legend displayed in the bottomright corner of the map.

Figure 8: Parameter form Figure 9: Tooltip shown on hover

When the ”Run analyze” button is clicked, a form with selectable parameters isopened. This form is showed in Figure 8. As seen in Figure 9, the user can hover overthe parameters to get a short explanation of the respective parameter.

22

Figure 10: Weight form Figure 11: Alert box with warning message

When the wanted parameters are selected, a form where the user can select weightsfor each parameter opens. This form can be seen in Figure 10. If the total weight doesnot add up to 100% an alert box is displayed to the user, this is shown in Figure 11.

Figure 12: Result from analysis Figure 13: Click and pan to polygon

The result from the analysis is presented with a table and polygons in the map. Thered polygons are automatically visualized in the map after the analysis, this can be seenin Figure 12. The user can click on the rows in the resulting table to show and pan tothe corresponding polygon in the map, this behaviour is shown in Figure 13. Figure 13also shows that it is possible to click on the polygons in the map to get a popup withinformation about that polygon (bay).

23

4.3 EvaluationFor the questionnaire a total of 30 answers were gathered. The questions in the question-naire are listed below. Thereafter, the results are summarized and presented in stackedcharts.

1. Are the selectable parameters appropriate for reinvestment analysis?

2. Is it appropriate to aggregate the data on bay level?

3. Is any important operational layer missing?

4. Does the pop-up show the right information about the object?

5. Does the pop-up show the right information about the bay?

6. Is the right information shown in the resulting table?

7. Is the placement of the main menu good?

8. Is the placement of the resulting table good?

9. Is the background map showing the right level of detail?

10. Is the coloring of the age layer easy to interpret?

11. Is the coloring of the result of the analysis easy to interpret?

12. Do you think a tool like this would be useful for you?

13. Overall impression of the design

24

q1 q2 q3 q4 q5 q6

10

20

30

#par

ticip

ants

Yes No Don’t know

Figure 14: Result from questions 1 to 6

In Figure 14, the result from the questions regarding functionality is presented. Thestack for q1 shows that most of the subjects thought that the selectable parameters wereappropriate for reinvestment analysis in power networks. The subjects who said nowanted to be able to select more parameters such as land use and cost. There were afew comments about being able to select which kinds of objects the analysis is done on,for example stations or cables. The stack for q2 shows that most of the subjects thoughtthat it was appropriate to aggregate the data on bay level. Some of the subjects wantedto aggregate on specific types of objects, for example stations and cables. Another sug-gestion was to do the analysis on every single object with no level of aggregation. Whenlooking at the result regarding q3 it is clear that many subjects were missing some oper-ational layers. The suggestions were to also include operational layers for land use, cost,airborne cables and outages. Many of the subjects had comments regarding the questionsabout the information in the popups of arc/node objects and bay objects and the informa-tion in the resulting table as seen in the stacks for q4, q5 and q6. Most of them thoughtthat the chosen id should be replaced with another id that is more informative for the user.Some of the subjects desired that the type of object, for example feeder cable, should bestated in the popup. The subjects also thought it would be desirable to show the totalsuitability score from the analysis in the resulting table.

25

q7 q8 q9 q10 q11 q12

10

20

30

#par

ticip

ants

Yes No Don’t know

Figure 15: Result from questions 7 to 12

The questions regarding design and visualization is presented in the stacks for q7 toq11 in Figure 15. All of the subjects thought that the placement of the menu and theresulting table was good, this can be seen in stack for q7 and q8. Some of the sub-jects did not answer the questions regarding the visualization but all of the subjects whodid thought the visualization was good. This is shown in the stacks for q10 and q11.There was a comment about the need to take color blindness into account for the coloringthroughout the prototype. Some of the subjects did not answer the question regardingusefulness of the prototype but all of the subjects who did thought that an application likethe one developed in this thesis would be useful for reinvestment decisions, this can beseen in the stack for q12.

26

7 %3

50 %

4

43 %

5

Figure 16: Result from question number 13

In Figure 16 the result from the question regarding the overall impression of thedesign is summarized. The overall impression was measured on a scale from 1-5, 1meaning that the overall impression was not at all good and 5 meaning a very goodimpression. Most of the subjects rated it as 4 or 5, only a few rated it as 3, as seen in thefigure.

5 Discussion

5.1 PreparationBefore getting started on the actual development of a prototype, relevant informationneeded to be gathered. It was chosen to do so through literature study for the GIS analysisand visualization parts. For knowledge about the current decision making process interms of reinvestments of power networks and the demand for a new process, performinginterviews was considered to be the best way to go.

5.1.1 Interviews

The interviews were conducted at an early stage of the development process as a wayof gaining more insight in the workings of power networks. At this early stage it is adifficult task to know what questions to ask, if not familiar with the subject at hand.Thus, presenting several companies with a questionnaire to get quantitative informationdid not seem appropriate for this thesis. Instead, to gain a general understanding andinsight it was chosen to conduct fewer, qualitative interviews where the customers wereasked open questions in the hopes of describing the process and needs in more than a yesor a no. Furthermore, it is easier to orally describe the concept of the proposed prototypethan in writing.

27

5.1.2 Data Handling

For the sake of keeping the real topology and the connections between components ofthe network, the use of real data was preferred over generating random data. It was alsoassumed that the customers would find it easier to evaluate the prototype if it remindedthem as much as possible of reality. Gathering and working with real data meant ex-tensive work since anonymization (of for example customer data) was required, but thebenefits of using real data were considered to override the time it would take and thechallenges that were to be faced.

The aggregation step was a big concern in this thesis. After conducting the inter-views and looking through the gathered data, the bay aggregation was considered themost appropriate. Since the objects in a bay are topologically related and connected itis more reasonable to use this as aggregation than for example a clustering algorithm.Another choice could have been to not aggregate at all, but to perform the analysis onsingle objects. As mentioned earlier in this thesis, the gathered data consisted of data thatwas already summarized for bays, which made it impossible to decompose this to singleobjects without generalizations and guesses. The evaluation showed that some of thefuture users thought it would be good to aggregate on geographical area or administra-tive boundaries. This would not consider the connections between the objects and wouldtherefore be difficult to apply on a network. For other kinds of reinvestments analyses,a geographical aggregation would probably work and might even be better. Since manyof the users want different kinds of aggregation methods, one solution could be to imple-ment an aggregation choice in the application. This would mean that the user him/herselfchoose the appropriate aggregation method before the analysis is run.

5.2 Prototype DevelopmentIn this section the choice of architecture and analysis is discussed.

5.2.1 Architecture

For the presentation tier of the prototype in the current thesis, a web-based client wasimmediately chosen over a desktop client. A web-based prototype means access forseveral people at the same time and from anywhere. Historically, desktop clients areheavier and might be needed for experts that perform analyses on every day basis. In thecase of this thesis, the users are laymen and will not perform the analysis more than everynow and then. A web client is usually straightforward and reminds the user of everydayuse in its similarity to other web pages, whereas a desktop client requires introductionand help to get through it and all its applications.

A 3 tier architecture was mainly chosen due to the fact that it is simple to maintain.Since it includes three different tiers (client, server and database), they can be workedwith and developed separately and independently of each other. This means that if theserver needs to be updated, you do not have to update every single client. The server alsomakes it possible for several users to use the application at the same time. Additionally,

28

security is another important factor why a 3 tier architecture was chosen. Since the clientdoes not interact directly with the database, the risk of unauthorized data is decreased.This also makes the database structure hidden for the user. The server improves the dataintegrity as well since it validates the correctness of the data. This is very important whenworking with sensitive customer data, which this application is supposed to do. However,it needs to be taken into consideration that the 3 tier architecture is quite complex andharder to develop than for example a 2 tier architecture. This makes the 2-tier architectureless expensive.

5.2.2 Analysis

The combination rule chosen for this thesis was Weighted Linear Combination (WLC).It was chosen because it is a simple and commonly used technique that lets the user setthe weights in one straightforward step and leaves a lot of control to the decision maker.When using WLC in an analysis it is important to be aware of the trade-off, where aweight can compensate for an attribute value and vice versa. Other combination rulesthat were looked into were AHP and OWA. Both of these two methods could have beenapplied for this analysis but after some research they were considered to be a bit toocomplicated and involve too many steps for none GIS experts. In accordance with thesimplicity step of HCI one simple step of setting weights seemed best suited for thisprototype.

With the unfamiliarity of the parameters in power networks, it is a difficult task to tryand guess their distribution and thus linear scaling seemed like the most appropriate wayto go. There are other scaling methods that might be a better fit for this kind of problemif the parameters are researched deeper. For this thesis a scaling of 0 to 1 was chosen.Another scale, for example 0-255 could also work and is commonly used within GIS andrasters, but since the prospective users of an application like the one developed in thisthesis are non GIS experts, the scale of 0-1 was considered to be easier to overview andinterpret.

The prototype will function as an aid for the decision maker when deciding on thefinal reinvestment area. This means that the result delivered by the analysis does notnecessarily have to be the optimal one. The decision maker will still have to interpret theresult and decide if it is suitable for reinvestments in accordance with other factors likemoney constraints, laws and similar.

5.3 EvaluationSince the authors of this thesis are not that familiar with the power network business,the evaluation of the the prototype was considered to get most accurate results by askingactual potential users. The questionnaire was therefore sent out to the prospective usersof the application during Digpro’s customer meeting. One reason why a questionnairewas chosen for evaluation was because it allows for quick response from a large audience.It would take long time to conduct for example individual interviews with everyone after

29

the presentation. Furthermore, the result from a questionnaire can be easily analyzed andvisualized in charts. One issue with a questionnaire though, is that they cannot capturefeelings and emotions in the same way as a face-to-face interview. This means that usefuldata can be missed. One more issue noticed in this thesis is that some subjects did notanswer all of the questions. In this case, unanswered questions were chosen to be markedas “Don’t know”. Another option could have been to leave out these answers from theresults.

The main error source of the used analysis is the subjectivity in the critical parametersand their corresponding weights. If the roles of the parameters are perceived incorrectlythe result from the analysis will be incorrect. Sensitivity analysis is a good way to getan overview of the robustness of the used analysis. However, the true purpose of theevaluation conducted in this thesis was to see if this kind of prototype is at all requestedby the market and thus worth developing. To be able to find this out a prototype wasdemonstrated to the customers. This way they got to see an application in action andcould evaluate if it would be useful or not. No sensitivity analysis was performed for thisprototype since the actual correctness of the analysis was not considered to add any valuein such an early development stage.

6 Conclusions and Future Work

6.1 ConclusionsThe purpose of this thesis was to show that the decision making process can be muchsimpler and better supported when using GIS tools for analysis and visualization. Fromconducted interviews it was concluded that none of the companies used any technicalanalysis for the reinvestment decisions. It is mainly gut feeling and knowledge about thenetwork that is important when making these decisions today. The interviewed compa-nies thought that a web application for analysis and visualization of potential reinvest-ment areas could be helpful. The prototype was developed with Javascript, HTML andCSS and the goal was not to have a perfect prototype to sell to the customers but rather tosee if there was any use in developing a real product that would handle analysis and vi-sualization and how it can be done. The user interface of the developed prototype is verysimple and straight forward, the user does not have to search for hidden menus or similar.This seemed to be appreciated by the prospective users since the overall impression of theinterface was highly rated in the evaluation. The evaluation also showed that a tool likethe one developed in this thesis could be very useful for reinvestments. However, moredevelopment is needed before the application could be used on real problems. Due to thefact that the network and topology is not a big part of the prototype, the prototype can beused for reinvestments in any area, not just power networks. To conclude, a straightfor-ward, simple analysis with visualization in a map as decision support, is an appropriateway to communicate reinvestment suggestions to laymen users of GIS.

30

6.2 Future WorkSince this is the first stage in developing a product that handles decision making for rein-vestments there are a few things that can be implemented as improvements. For thisthesis five parameters were chosen after interviewing experts. These five parameters canvary between companies and utilities. A big set of general parameters where each com-pany can select which are applicable for them in a first step, before the step of choosingwhich parameters to use for the actual analysis, might be a way to expand the prototype.

As mentioned in section 4.3 there was a desire to include more operational layers. Forexample, it would be interesting to be able to visualize other factors such as constructionscosts and power consumption. Following, land cover data could be implemented to seewhere a lot of airborne cables are placed surrounded by trees. There might not be currentoutage data to look at, but it might be worth implementing some kind of probability ofmeters of airborne cables in forest since they are at high risk of outages.

Having every user choose parameters and weights could be troublesome for the com-panies. For example, if one decision maker thinks that the number of outages is veryimportant and another decision maker thinks that the number of inspection remarks is themost important factor, the results from these two analyses will be different. A suggestionis that once a company has found what they consider to be the best weight combination,it could be a good idea for them to “freeze” these weights and thus get the same valuesindependent on who is running the analysis.

The current prototype only shows the bay that is in need of reinvestment. An im-provement would be to be able to click on the bay and get the statistics of each object inthe bay to see exactly where the outages and inspection remarks are located. Maybe allthe faults of the bay are because of one (1) really bad object and for this the whole baymay not need reinvestment, only this one object needs to be replaced.

Due to time constraints, no evaluation of how well the areas showing up in the webapplication match the actual investment areas has been done in the work of this thesis. Away to implement this already today could be to let the people in charge of reinvestmentsin the same area as the borrowed data came from, check how well the areas suggestedby the web application developed in this thesis match the areas that are actually aimed atreinvesting in for the next few years. Further on, the analysis method could be changed tosee if the results then match better or worse and thus eventually find an optimal analysismethod.

There is room for improvements in the user interface to make the user experienceeven better. One thing is to make it more responsive, for example adding a loadingsymbol when the analysis is running. As it is today, it is not possible for the user to see ifthe analysis is running or if the application has frozen. Due to the time limit the prototypeis only adapted for Chrome. It would probably be a good idea to develop the prototypeto work just as well in other web browsers such as Firefox or Safari.

In the current prototype there is no possibility of saving the result from the analysisother than a screen shot. In the future, it would be interesting to look into a better way tostore the result from previous runs.

31

ReferencesAL-Mukhtar, M. and Hadi, S. (2012). Developing a three-tier web data management

application for higher education admission environment.

Andersen, M. (2015). Evaluating and improving the user-interface of exist-ing gis-applications by the use of hci evaluation techniques. http://kth.diva-portal.org/smash/get/diva2:824772/FULLTEXT01.pdf. Col-lected: 2017-04-19.

Carrol, J. M. (2017). Human computer interaction. http://www.qlik.com/us/products/qlik-geoanalytics. Collected: 2017-04-04.

Drobne, S. and Anka, L. (2009). Multi-attribute decision analysis in gis: Weighted linearcombination and ordered weighted averaging. Informatica, 33:459–474.

Egenhofer, M. J. and Kuhn, W. (2010). Interacting with gis. http://www.geos.ed.ac.uk/˜gisteac/gis_book_abridged/files/ch28.pdf. Collected:2017-04-17.

ESRI (2017). About web gis. http://server.arcgis.com/en/server/latest/create-web-apps/windows/about-web-gis.htm. Collected:2017-04-25.

Fu, P. and Sun, J. (2011). Web GIS Principles and Applications. Esri Press, Redlands,California.

Heywood, I., Cornelius, S., and Carver, S. (2011). An Introduction to GeographicalInformation Systems. Pearson Education Limited, Essex, England.

Jankowski, P., Andrienko, N., and Gennady, A. (2001). Map-centred exploratory ap-proach to multiple criteria spatial decision making. International journal of Geo-graphic Information Science, 15(2):101–127.

Jankowski, P., Swobodzinski, M., and Zielinska, A.-L. (2008). Choice modeler: A web-based spatial multiple criteria evaluation tool. Transactions in GIS, 12(4):541–561.

Karnatak, H. C., Saran, S., Bhatia, K., and Roy, P. (2007). Multicriteria spatial decisionanalysis in web gis environment. Geoinformatica, 11:407–429.

Ligmann-Zielinska, A. and Jankowski, P. (2008). A framework for sensistivity analysis inspatial multiple criteria evaluation. Geographic Information Science, pages 217–233.

Malczewski, J. (1999). GIS and Multicriteria Decision Analysis. John Wiley Sons, Inc,New York, New York.

Malczewski, J. and Rinner, C. (2015). Multicriteria Decision Analysis in GeographicInformation Science. Springer, Berlin, Germany.

32

Muehlenhaus, I. (2013). Web Cartography: Map Design for Interactive and MobileDevices. Taylor & Francis.

Murray, S. (2013). nteractive Data Visualization for the Web. OReilly Media.

Rinner, C. (2007). A geographic visualization approach to multi-criteria evaluation ofurban quality of life. International Journal of Geographical Information Science,21(8):907–919.

Rouse, M. (2017). Hci (human-computer interaction). http://searchsoftwarequality.techtarget.com/definition/HCI-human-computer-interaction. Collected: 2017-04-11.

Silva, S., Lus, A.-A., and Dias, L. C. (2014). Development of a web-based multi-criteriaspatial decision support system for the assessment of environmental sustainability ofdairy farms. Computers and Electronics in Agriculture, 108:46–57.

SIS (2015). Geografisk information webbkartografi webbkarttjanster med av pro-ducenten definierade kartskikt. http://www.sis.se/PageFiles/12181/Rapport%200.9%20Webbkartografi.pdf. Collected 2017-01-12.

Stokes, R. (2017). emarketing - the essential guide to marketing in a digitalworld. https://www.redandyellow.co.za/wp-content/uploads/5-User-Experience-Design_Quirk-Textbook-5.pdf. Collected:2017-04-11.

Stopper, R. (2012). Graphical user interface - layout and design. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.527.6895&rep=rep1&type=pdf. Collected: 2017-04-18.

Sugumaran, R., Degroote, J., and Sugumaran, V. (2011). Spatial decision support sys-tems.

Syk, E. and Hilding, F. (2016). User-centric web-based system for visualization of nis-data forlayman users.

Tyner, J. (2010). Principles of Map Design. Guilford Press.

Vermeeren, A. P., Roto, V., and Vnnen, K. (2016). Design-inclusive ux research: designas a part of doing user experience research. Behaviour & Information Technology,35(1):21–37.

33

Appendix A

34

35

36

37

38

39

40

41

42

43

44

45

Reports in Geodesy and Geographic Information Technology

The TRITA-GIT Series - ISSN 1653-5227

17-001 Priscilla Adjei-Darko. Remote Sensing and Geographic Information Systems for Flood Risk Mapping and Near Real-time Flooding Extent Assessment in the Greater Accra Metropolitan Area. Master of Science thesis in Geoinformatics. Supervisor: Osama Yousif. March 2017.

17-002 Ines Otosaka. A Historic Record of Sea Ice Extents from Scatterometer Data. Master of Science thesis in Geoinformatics. Supervisors Supervisors: Maria Belmonte Rivas and Ad Stoffelen, Royal Netherlands Meteorological Institute (KNMI), the Netherlands and Yifang Ban, KTH. April 2017.

17-003 Shérazade Gadhoumi. Platforms for Real-time Moving Object Location Stream Processing. Master of Science thesis in Geoinformatics. Supervisors: Andreas Degwerth, Airbus Defense and Space and Gyözö Gidofalvi, KTH. April 2017.

17-004 Jonas Bengtsson and Mikael Grönkvist. Performing Geographic Information System Analyses on Building Information Management Models. Master of Science thesis in Geodesy No. 3146. Supervisors: Carine Hals (Agima) and Milan Horemuz (KTH). June 2017

17-005 Peng Zhang. 3D Building Models, Production and Application. Master of Science thesis in Geodesy No. 3147. Supervisor: Milan Horemuz, June 2017.

17-006 Violeta De Lama. Precision Analysis of Photogrammetric Data Collection Using UAV. Master of Science thesis in Geodesy No. 3148. Supervisor: Milan Horemuz, June 2017.

17-007 Josefin Lennartsson and Natalie Ekroth. Web-based Multicriteria Decision Analysis and Visualization for Reinvestments in Power Networks. Master of Science thesis in Geoinformatics. Supervisor: Daniel Sedell, Digpro and Gyözö Gidofalvi, KTH. June 2017.

TRITA-GIT EX 17-007

ISSN 1653-5227

ISRN KTH/GIT/EX--17/007-SE

TRITA GIT EX 17-007

www.kth.se