HMTF Understanding PLF August 31, 2015 Kevin Harris, ColumbiaGrid TEPPC\Hydro Modeling Work Group - Chair.

HMTF Understanding PLFAugust 31, 2015

Kevin Harris, ColumbiaGridTEPPC\Hydro Modeling Work Group - Chair

2

Outline

• Background• Is Hydro generation proportional to load?• Change in Hydro Operations on Columbia River 2011+• Understanding PLF K Factor• Developing K Values for PLF• Hydro Dispatch Against Load – Wind• Summary of Findings• Hydro Thermal Co-Optimization (HTC)• Proposed GridView Improvements

3

Background

4

Objective of Current Hydro Modeling Review

• Review the modeling of Core Columbia River projects:– What does it mean when you change a modeling

coefficients for PLF or HTC?

– Determine if existing modeling represents current operations?

– Make recommendations to correct any operational issues

– Develop tools/method to determine appropriate Hydro modeling parameters/coefficients for in GridView

5

Hydro Background• The Northwest represents 74% of the Hydro generation in WECC

US System

• The Core Columbia River projects represents (Coulee – Bonneville):

– 19,810 MW of capacity (8,905 aMW of generation 2001-2012)

– The Core is 61% of the Northwest Hydro generation

– The Core is 45% of the Hydro generation in WECC US System

• WECC current Hydro modeling assumptions were created for use in Promod with calendar year 2010 data for the PNW

Focus first round of Hydro review on Core Columbia River

6

The Core Columbia

• The Core Columbia River is used to evaluate Hydro modeling in GridView

• Current Modeling:– Fixed Hourly Shape, 5 Projects

which represents 23% of the Capacity

– PLF/HTC, 6 Projects which represents 77% of the Capacity

Upper Columbia: Coulee - Priest RapidsLower Columbia: McNary - Bonneville

7

Compare Operations

• Created spreadsheet to compare operation on the Columbia River by project or by aggregated sets (~65 MB)

• Contains hourly Hydro and load for: 2001, 2002 and 2005-2013• It capable of comparing (all tables and/or charts):

– Compare monthly operation: min, avg min, avg, avg max, and max– Compare average weekday peak hour for both load and generation– Compare average weekday operations (hour ending 1-24)– Calculate K Factor based on hourly data by month– Calculate K Factor based on monthly stats– Calculate resulting Hydro generation based on K Factor– Compare daily allocation generation vs. load and (Load-Wind)– Calc Polynomial for daily min and max generation– Compare daily operation by year: min, average and max gen

8

Is Hydro Generation Proportional to Load?

9

Example Historic Operations• Example operation for January 2010 through 2013

Per unit of generation the operating range of

Upper Columbia is greater than the Lower

Columbia

Load Used:= 100% of BPA

+ 100% of MidC + 6% of CAISO

Aggregated and some individual projects Hydro

generation are proportional to load

10

Change in Hydro Operations on Columbia River 2011+

11

Operational Change Starting in 2011• Starting in 2011 the annual average

daily operating range decrease by 2,224 MW (38%)– During the spring run-off (Apr-Jul) an

average reduction of 3,619 MW (65%)– Balance of year (Jan-Mar & Aug-Dec) an

average reduction of 1,527 MW (25%)• Any forecast run should reflect this

reduction in operational flexibility

12

Operational Change Starting in 2011

Note the polynomial for 2011, 2012, and 2013 for operational

Min and Max rating over-lap

Date PointsSpring Run-Off: 122/yrBalance of year: 243/yr

13

Understanding PLFK Factor

14

Proportional Load Following (PLF)• How PLF works

– The reference frame for PLF is the average monthly load and Hydro generation– Hourly Hydro generation is equal to the hourly percent change in load, from

average load, multiplied by K Factor and applied to average monthly Hydro generation

– Min and Max rating is enforced on Calc Hydro generation– K=0 results in a flat monthly shape equal to the average monthly Hydro generation

15

Measuring Success

What’s the Criteria to measure reasonable Hydro operations?

Used Criteria• Confirm that Hydro generation is proportional to load• Average Weekday Operating Range: Focus on keeping the average

weekday operating range within reasonable historic limits (Avg WKD Max-Avg WKD Min)– Weekday Max set to the average of the maximum 3-4 hours– Weekday Min set to the average of minimum 3-4 hours

• Error Check: Compare calculated average weekday generation shape with actual operations

16

K Factor - Hourly Shape (PLF)

• Positive K Factor result in Hydro generation proportional to load– K Factor 0> and < 1, results in a contraction of the daily operating range

in Hydro generation relative to the load shape– K Factor > 1, results in expanding the daily operating range in Hydro

generation relative to the load shape• Negative K Factor result in

Hydro generation inversely proportional to load

Note: Daily average is equal to monthly average

17

K Factor - Daily Allocation (PLF)

• High K Factors leads to greater operational flexibility that historic operations

• Lower K Factor reduce the Hydro generation daily volatility• Modeling Objective: Lower K

Factor reduce the Hydro generation daily volatility

• K Factor for Avg 2011-13– Grand Coulee: 2.44– Core Columbia: 1.44

• Develop K Factor for aggregated projects versus individual project

• Desired new feature: A means to set daily upper and lower limits to bound daily allocation by month

18

Calculating K Factor

• The reference frame for PLF is the average monthly load and Hydro generation

• K Factor:= Slope of– Y= Hydro Gen(Hr i)/Avg Mo Hydro Gen– X:= Load(Hr i)/Avg Mo Load– Example K:= 2.4111

Is their a simpler way to calculate K?

19

Alternative Method for Calc K Factor

• You only need two data points to calculate a slope:– Lower Point: The average weekday minimum Hydro and load– Higher Point: The average weekday maximum Hydro and load

• The dark red line uses the weekday daily min and max• Observation: There can be

significant volatility in min generation

• Consider using a multi hour average for min and max (The Green Line use the avg min and max 3 hour)

The avg WKD min/max do not set operational min/max rating

20

Two Point K Factor

If Slope is all we need:= Delta(Y Axis)/Delta(X Axis)• Delta(Y Axis) and Delta(X Axis) can be independently calculated• K Factor can be split into two components:

– Hydro K’ [Delta(Y Axis)]: Is tailored to the desired average weekday operating range

– Load K’ [Delta(X Axis)]: Is based on the load Hydro is to be dispatched to

• Formula:

• Feedback loop: Calculate resulting hourly Hydro generation based on K to determine if its hourly shape mimic desired historic behavior

𝑲 (𝑺𝒍𝒐𝒑𝒆)=

𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏(𝒂𝒗𝒈𝒅𝒂𝒊𝒍𝒚 𝒎𝒂𝒙)𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏(𝒂𝒗𝒈)

−𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏(𝒂𝒗𝒈𝒅𝒂𝒊𝒍𝒚𝒎𝒊𝒏)

𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏(𝒂𝒗𝒈)𝑳𝒐𝒂𝒅(𝒂𝒗𝒈𝒅𝒂𝒊𝒍𝒚 𝒎𝒂𝒙)

𝑳𝒐𝒂𝒅 (𝒂𝒗𝒈)−𝑳𝒐𝒂𝒅(𝒂𝒗𝒈𝒅𝒂𝒊𝒍𝒚 𝒎𝒊𝒏)

𝑳𝒐𝒂𝒅 (𝒂𝒗𝒈)

=𝑫𝒆𝒍𝒕𝒂 (𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏)

𝑫𝒆𝒍𝒕𝒂(𝑳𝒐𝒂𝒅)=𝑯𝒚𝒅𝒓𝒐𝑲 ′

𝑳𝒐𝒂𝒅𝑲 ′

𝑳𝒐𝒂𝒅𝑲 ′=𝑳𝒐𝒂𝒅𝑹𝒂𝒏𝒈𝒆¿¿𝑯𝒚𝒅𝒓𝒐𝑲 ′=𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏¿ ¿

21

Need Spin?

• Maximum operational may be lower than physical capability– Typically the available energy runs out serving load/obligation prior to

meeting maximum physical capability• If the desired maximum operating rating limits a projects ability

to provide spin, consider modeling a spinning reserve unit:– Model a second unit representing

this project ability to provide spinning reserves:• Max Rating: Spin capability• Min Rating: Zero• Cost: High >= $50,000/MWh

– For simplicity set spin reserve capability to zero on original Hydro project

22

Developing K Values for PLF

23

Develop Modeling Hydro Coefficient

• Split Columbia River Projects into two groups:– Upper Columbia: Coulee through Priest Rapids (7 projects)– Lower Columbia: McNary through Bonneville (4 projects)– Note: The operating range per unit of generation for Upper Columbia is

greater than Lower Columbia– Projects that are not proportional to load are modeled as a flat monthly

shape (Previously modeled as hourly shapes)

• Split the calendar year into two seasons: – Spring Run-Off: Apr-Jul– Balance of year: Jan-Mar, and Aug-Dec

• Based Hydro operation on average operating from 2011-13:– Calculate coefficient for each year then average resulting values

24

Summary of Modeling Change

• Split the 11 projects into upper and lower Columbia River to calculate PLF coefficients and compare to expected operations– Upper Columbia: Coulee through Priest Rapids– Lower Columbia: McNary through Bonneville

• Use a flat monthly generation shape for previously hourly shapes K:=0

ObjectiveTake a desired

operating year and have it’s monthly

operation conform to operation from 2011-

13

25

Min/Max Polynomial and Hydro K’

• Polynomial used to determine min and max ratings assume:– Daily max is based on the average of the maximum 3 to 4 hours– Daily min is based on the average of the minimum 3 to 4 hours– Weekdays only (Weekend data is excluded from the poly)

• Calc Hydro K’: Use target average monthly generation with the appropriate seasonal curve to calculate min and max rating used in calculating Hydro K’

𝑯𝒚𝒅𝒓𝒐𝑲 ′=𝑷𝒐𝒍𝒚𝐌𝐚𝐱 [𝑻𝒂𝒓𝒈𝒆𝒕 𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏 (𝒂𝑴𝑾 )]−𝑷𝒐𝒍𝒚 𝑴𝒊𝒏[𝑻𝒂𝒓𝒈𝒆𝒕𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏(𝒂𝑴𝑾 )]

𝑻𝒂𝒓𝒈𝒆𝒕𝑯𝒚𝒅𝒓𝒐𝑮𝒆𝒏(𝒂𝑴𝑾 )

26

Calculate Operational Min/Max Ratings

• Calc operational min and max rating based on historic operations– Base min rating on min gen curve at a 15% probability and max on the max gen

curve at 85% probability– Use the target average monthly generation for the average– The StDev is based on:

• Backcast use actual StDev• Forecast use calc StDev based on 2011-13 operations

27

Adjustments

• Calc min rating is greater than or equal to 0• Calc min rating is less than or equal to physical max• Using polynomial: Adjustment are made to Outlier data points

(outside the operating range of the polynomial)– This is driven by high spring run-off during 2011-13– The adjustments are:– Operating range was locked at the limiting bounds (upper or lower

operating range of the polynomial)– The relative location of operating range is applied to the target

operating• Min Rating:= (Target Gen aMW)*bound[(Avg – Min)/(Max – Min)

28

Calculate K

• Calc Load K’ based on modeled load:

• Calculate:

𝑳𝒐𝒂𝒅𝑲 ′=𝑳𝒐𝒂𝒅𝑹𝒂𝒏𝒈𝒆¿¿𝑲=

𝑯𝒚𝒅𝒓𝒐𝑲 ′𝑳𝒐𝒂𝒅𝑲 ′

Example K ValuesJan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

For 2010 BackcastUpper Columbia 5.08 4.85 5.09 5.04 3.98 2.59 3.07 4.70 4.33 4.81 3.53 4.29Lower Columbia 2.24 2.08 2.35 1.94 2.26 0.98 1.65 1.15 2.37 2.61 3.15 3.49

For 2010 ForecastUpper Columbia 4.04 3.99 4.35 2.98 2.43 1.12 1.77 3.13 4.76 4.22 3.43 3.55Lower Columbia 2.74 2.73 2.63 3.63 2.52 0.69 2.37 0.80 2.12 2.60 2.74 2.69

29

Error Check 2013 BackcastCompare calculated hourly weekday generation shape with historic operation for accuracy in duplicating hourly shape

(hour ending 1-24)

2013 backcast matches actual operations

30

Error Check 2010 BackcastCompare calculated hourly weekday generation shape with historic operation for accuracy in duplicating hourly shape

(hour ending 1-24)

2010 backcast matches actual operations

31

Error Check 2010 ForecastCompare calculated hourly weekday generation shape with historic operation for accuracy in duplicating hourly shape

(hour ending 1-24)

Reduced operating range can be seen in the above

chart

32

Hydro Dispatch Against Load – Wind

33

Hydro Dispatch Load - Wind• Relative to load, wind generation serves up to 64% of BPA daily load in 2014

or 20% of annual load• Changing from “Load” to “Load – Wind” increases the deviation in daily a

factor of 2.7• The expanded daily StDev directly impacts the daily allocation of Hydro

generation. This impact can be amplified when K is > 1BPA 2014 BA Load and Wind

Avg StDev StDev/Avg

Load 6,282 646 10%Wind 1,271 1,084 85%Load-Wind 5,011 1,351 27%

Installed Wind Capacity in BPA for 2014 4,515

MW

34

Hydro Dispatch Load - Wind• The top right chart BPA load vs Core

Columbia gen (R^2=0.616)• The bottom right chart: BPA Load –

Wind vs Core Columbia gen (R^2:= 0.252)

• Bottom chart show daily Hydro generation tracking BPA load

Op Range (X Axis)BPA 0.9 to 1.1BPA-Wind 0.5 to 1.5

StDevAct HY 0.081HY (Load) 0.063HY (Load-Wind) 0.041

35

Hydro Dispatch Load - Wind

• Compare backcasting April 2013 Hydro operation:– Against Load: A better match against actual Hydro

operation– Against Load-Wind: When wind comes on and off for a

couple of days the result Hydro diverges from actual Hydro generation

36

Hydro Dispatch Load – Wind

• What we want: – Daily Hydro allocation to reflect historical operations

• There is no one clean answer when determining monthly Hydro allocation– Intra day Hydro generation to take into account generation from non-

dispatchable supply (Wind and Solar) when dispatching• Competing issues:

– K values based on Load has a tendency to be higher, resulting in increase volatility in daily Hydro allocation

– K values based on Load –Wind tend to be lower while the volatility in load increases: resulting mixed results, i.e. volatility daily Hydro allocation may increase or may decrease

• K based on Load is more predictable while Load – Wind increase volatility in monthly allocation

37

Summary of Findings

38

Summary of Findings

• Current modeling does not reflect operational changes on the Columbia River which starting in 2011– Base Hydro operating on year 2011-2013 for any forecast year

• Individual project Hydro generation are not always proportional to load but Net Columbia, Upper Columbia and Lower Columbia are proportional to load

• Individual project that are not proportional to load are modeled as a flat monthly profile (Base Load)

• Calculating coefficients for two aggregated systems: – Upper Columbia: Coulee to Priest Rapids– Lower Columbia: McNary to Bonneville

39

Summary of Findings

• The use of high K values result in inappropriate intra monthly allocation of daily Hydro generation– Develop K factors in a manor to minimize its value (Consider develop K

Factors on an aggregate basis instead of individual)– The use of HTC can lower K value: For example: Splitting 50% of the

dispatch range to HTC reduces K value by 50%

• Dispatching Hydro against “Load – Wind – Solar”– K based on “Load” is more predictable while “Load –Wind” increase

volatility in monthly allocation

40

Hydro-Thermal Co-optimization (HTC)

41

• HTC reshapes a share of PLF Hydro generation based on LMP• The original equation for HTC

– Where: • C – Plant Capacity• A – Range of plant generation (Plant Operating Range)• This assumes a 50/50 split between PLF/HTC hence ½*A

• Re-writing where– HTC(Share) + PLF(Share) = 1 (Share of operating range)– A:= Range of Plant Gen:= C*OpRange(%)– ½:= HTC(Share)

4pC p

4*HTC(Share) p p

p:=

42

Proposed GridView Improvements

43

Proposed GridView Improvements• Load – Wind – Solar: Add two new dimensions that control

how Load – Wind – Solar is set:– The ability to set Load – Wind – Solar by area/region– The ability to set a percentage (0% to 100%) of wind and solar that is

subtracted from the load– Example:

• BPA: 20% Wind and 100% Solar• CAISO: 100% Wind and 100% Solar

• A means to limit daily allocation of Hydro generation when K Factors is high– Example: Weekday limit between 110% and 90% of average daily

generation. Weekend would allow a lower limit?

44

Kevin Harris [email protected](503) 943-4932