Data Centric Load Forecasting and Allocationstaff.utia.cas.cz/kulhavy/powergen00s.pdf · Data-centric model ! database requested data statistics monitoring data 5th Jan 2000, 7:15

1

Zdeněk Schindler, Rudolf KulhavýZdenZdeněěk Schindler, Rudolf Kulhavýk Schindler, Rudolf Kulhavý

Data Centric Load Forecasting

and Allocation

Data Centric Data Centric Load ForecastingLoad Forecasting

and Allocationand Allocation

Honeywell Technology Center Europehttp://nero.htc.honeywell.cz/

June 2000

Honeywell Technology Center EuropeHoneywell Technology Center Europehttp://nero.htc.honeywell.cz/http://nero.htc.honeywell.cz/

June 2000June 2000

2

Problem DefinitionProblem Definition

Combined heat Combined heat and powerand powerproduction production

and distributionand distributionsystemsystem

Goal:Goal:Realization of decision support system for the load Realization of decision support system for the load dispatching center with these functions:dispatching center with these functions:

1. Heat demand forecasting2. Hot water distribution enhancement3. Heat & power generation enhancement

3

MotivationMotivation

• to avoid unnecessary boiler start-ups and shutdowns

• to improve cogeneration contracting

• to enhance the efficiency of heat production and distribution

4

red lines steamblue lines hot water

1 km

Combined cycleCHP

Heating plantMunicipal incinerator

CHP

CHP

• Five heating plants • 96 km of steam pipeline network• 74 km of primary hot water pipelines (five networks )

District Heating System - Pilot Project District Heating System - Pilot Project

5

Heat Demand ForecastingHeat Demand Forecasting

Quantities to be forecasted:Quantities to be forecasted:• total supply of steam• total supply of heat• total supply of electricity• heat supply in individual hot water pipelines - sepa rate and

cumulative values• heat supply in hot water• heat supply in steam• total gas consumption• total steam production• total heat production• total heat production in steam

Total Total -- 17 quantities17 quantities

6

Heat Demand ForecastingHeat Demand Forecasting

Mode of operation: Mode of operation: 24 x 724 x 7

Forecast time ranges:Forecast time ranges:• 15’ averages one-day ahead … short-term forecast• 1 h averages one week ahead … medium term forecast• daily averages one month ahead … long term forecast

Computer workload:Computer workload:• 1,632 forecasts (of a single value) every 15 minute s• 2,856 forecasts (of a single value) every hour• 1,581 forecasts (of a single value) every midnight

Extra requirement:Extra requirement:• to implement a process history database with nearly

500 variables monitored in 5’ period

7

District Heating System ModelDistrict Heating System Model

consumers

distribution

production

Complex stochastic system

First-principle model(s) ?

weatherconditions

time of day

day of week

holidayheat demand

steam demand

hot waterdemand

8

District Heating System ModelDistrict Heating System Model

input output

BLACK BOXBLACK BOXBLACK BOXweather conditions

time of dayday of week

holiday

heat demandsteam,

hot water

Data-centric model !

databaserequesteddata

statisticsmonitoring

data

5th Jan 2000, 7:15 h-5°C

...pipeline P2:

150 000 kg/hour,...

9

Data-Centric = Memory-Based ForecastingData-Centric = Memory-Based Forecasting

Alldata

Estimation

Globalmodel

Complex, non-linear/non-Gaussian behavior

fit globally witha single model

Adaptation

Recentdata

Localmodel

Complex, non-linear/non-Gaussian behavior fit locally in time with a

simple model

Querydriven

retrieval

Relevantdata

Localmodel

Complex, non-linear/non-Gaussian behavior fit locally in data cubewith a simple model

10

Predictedtarget variable(heat supply,

steam production)

ALGORITHM1. Assess the conditions

at the query point (weather conditions)

2. Search database for similar conditions

3. Extrapolate from the past values(heat production)

4. Estimate precision of the forecast

Data-Centric Forecasting Algorithm Data-Centric Forecasting Algorithm

Situation (weather conditions,

time of day, type of day)Query point

Neighborhood

11

Output dataOutput data� process variables forecasts� enhanced process variables

Operational Operational datadata

DataData--centric centric forecasting and forecasting and

optimizationoptimization

ActionsActions

Process history databaseProcess history database

Implementation - Honeywell DSSImplementation - Honeywell DSS

Input dataInput data� external weather forecast � process monitoring data

12

DSS - Data Selection ScreenDSS - Data Selection Screen

13

DSS - Forecaster Basic ChartDSS - Forecaster Basic Chart

Column = forecast

Line = actual values

14

DSS - Forecaster Composite ChartDSS - Forecaster Composite Chart

15

Hot Water Distribution EnhancementHot Water Distribution Enhancement

Problem:Problem:Define the “best”- pressure difference (set-point for circulating pumps controls)- supply temperature (set-point for heat exchanger controls)for hot water pipelines

Optimal control of complex dynamic system!!!Optimal control of complex dynamic system!!!

Complexity consists in:Complexity consists in:• consumers - stochastic behavior,

- different (unknown) control strategies• network - variable, state dependent time delays• environment - concurrent environment influence

is reflected in load gradually • unknown pipeline characteristics• etc.

16

Con

trol

var

iabl

e(p

lant

op.

poi

nts)

ALGORITHM1. Assess the current

conditions (weather conditions)

2. Search database for similar conditions (red points)

3. Estimate profit for retrieved controls under the current conditions

4. Conduct a “fine-trim” search by adjusting the process controls (blue points)

Data-Centric Optimization AlgorithmData-Centric Optimization Algorithm

Situation (weather conditions,

time of day, type of day)Query point

Objective function(profit)

Neighborhood

17

Hot Water Distribution EnhancementHot Water Distribution Enhancement

Optimality Criterion:Optimality Criterion:Maximize efficiency of heat distribution

Method:Method:Comparing• available heat supplied to consumers,• unused heat returning back to a heating plant.

The applied algorithm attempts to satisfy the necessary optimality conditions

Motivation:Motivation:Circulating energy (excepting intentional accumulat ion) results in higher heat losses.

18

Hot Water Distribution EnhancementHot Water Distribution Enhancement

Solution:Solution:Combination of • data-centric optimization procedures

• simplified model-based objective function

Conditions descriptors:Conditions descriptors:• outdoor temperature• current electricity production• time of day• type of day (day of week, holiday)• age of data

Closed loop safeguard:Closed loop safeguard:• Low level controllers keep critical variables in technology limits autonomously• Operator can increase or decrease the supplied heat

temporarily

19

Heat & Power Generation EnhancementHeat & Power Generation Enhancement

Problem:Problem:

Find optimal outputs of boilers that minimizeproduction costs

Solution method:Solution method:Data-centric optimization

Knowledge:Knowledge:operating conditions past actions and costs

implicitheat load line

Boiler 1Boiler 2

Productioncosts

Localcosts Resource Allocation Problem

20

DSS Network SchemeDSS Network Scheme

POWERFAULT DATA ALARM

Win NT/98

Modem

Win NT/98

Oracle8i ClientDSS Client

IDC

PHD Server

Win NT 4.0 Webbrowser

Webbrowser

Webbrowser

DSS Thin clients

IDC

Win NT 4.0

Oracle8i ServerOracle Developer Server

Oracle Application ServerDSS Server

21

Typical Tangible BenefitsTypical Tangible Benefits

• lower production costs (e.g., elimination of unnece ssary

boiler start-ups, increased efficiency)

• lower distribution costs (reduced heat losses)

• improved product quality (reduced variation of

cogenerated power)

• improved asset management (less boiler start-ups)

• reduced safety margins (owing to lower uncertainty)

• reduced labor costs (less operators required)

• increased productivity (automation of some function s)

22

Typical Intangible BenefitsTypical Intangible Benefits

• improved data reliability due to built-in data vali dation

• transparent access to multiple operational data

• schedule-driven 24x7 delivery of information

• access to the analytic results via both clients and Web

• faster more informed decisions

• better customer service

• improved process knowledge

23

DSS - General AdvantagesDSS - General Advantages

DSS allows the plant operators, managers, and execu tives toDSS allows the plant operators, managers, and execu tives to

• anticipate changes in the environment;

• respond faster to changing conditions;

• anticipate abnormal situations well ahead;

• detect operating conditions leading to off-spec pro duction;

• operate closer to the optimum or limits;

• reduce time to optimal operating conditions;

• better understand the processes;

• learn continually from the manufacturing experience .

24

Data-Centric Method PrinciplesData-Centric Method Principles

Not confined to heat production and distributionNot confined to heat production and distribution

Basic application principle:Basic application principle:• The software mimics the real system behavior

Application policy:Application policy:• Describe the problem carefully• Ask correct questions

Application prerequisites:Application prerequisites:• Enough relevant data for required processing• Correctly specified input-output relations• Well chosen approximating functions