Load Forecast Model Development

Transcription

1 Load Forecast Model Development Load Analysis Subcommittee June 20, 2018

2 Disclaimer This is the current status of ongoing model development. The results are not final and are only indicative of what potential impacts on the forecast might be. 2

3 Potential Process First Model Estimated Base Load Day-type Economics Model Estimation Forecasted Base Load End-Use Characteristics Load Forecast Distribution Weather Simulation Model Estimation Base Load Weather End-Use Characteristics Second Model 3

4 Commercial Load Previously used a static weight to transform employment trends to square footage trends. Stakeholder concern that this would miss the changing relationship over time. Looked at the relationship of U.S. square footage to U.S. employment Commercial Equipment indexes are expressed in intensity per square foot (not per customer). Getting a proxy for square foot per customer, can establish intensity per customer. 4

5 1.6 U.S. Building Space and Employment Index, 1998 = Commercial Square Footage Employment Sources: Itron, Moody s Analytics 5

6 1.3 Ratio of Building Space to Employment Sources: Itron, Moody s Analytics 6

7 Coincident Peak Model Spreadsheet Base Contains base history and forecast values assumed in the model Coefficients Model parameters from the Second Model Residuals Model Estimates and Residuals 7

8 PJM RTO Summer Forecast 180, , , , , , , Unrestricted Peak Weather Normalized Peak 2018 Forecast Test Model 8

9 Summer 2021 Forecast Distributions 1.2 Per-Unitized Peak % 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Peak Percentile 2018 Forecast Test Model 9

10 6% Summer Zonal CP Differences Test Model vs 2018 Forecast For Delivery Year 2021/22 4% 2% 0% -2% -4% -6% 10

11 Summer Forecast Model Accuracy Forecast Model Solved vs Top Ten Summer Days per Year Mean Absolute Percent Error 3.0% 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Forecast Horizon Current Model Test Model 11

12 Summer Error Zero Year Forecast Horizon Percent Error 8% 6% 4% 2% 0% -2% -4% -6% -8% Forecast Vintage Pct Err on Rank Day 2-10 Pct Err on Rank Day 1 Current Model (Abs Pct Err) Test Model (Abs Pct Err) 12

13 Summer Forecast Observations Test Model has very similar growth pattern ( % growth) to 2018 forecast, though starts out about 0.7% lower Summer forecast distribution is a little bit more spread out, with generally slightly higher per-unitized values above the median. Zonal changes generally range between +/- 2%. Several zones in New Jersey have positive changes outside this range, while several Pennsylvania zones have negative changes outside this range. 13

14 PJM RTO Winter Forecast 160, , , , , , , Unrestricted Peak Weather Normalized Peak 2018 Forecast Test Model 14

15 Winter 2021/22 Forecast Distributions 1.2 Per-Unitized Peak % 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Peak Percentile 2018 Forecast Test Model 15

16 Winter Zonal CP Differences Test Model vs 2018 Forecast For Delivery Year 2021/22 10% 8% 6% 4% 2% 0% -2% -4% -6% -8% -10% 16

17 4.0% Winter Forecast Model Accuracy Forecast Model Solved vs Top Ten Winter Days per Year Mean Absolute Percent Error 3.5% 3.0% 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Forecast Horizon Current Model Test Model 17

18 8% Winter Error Zero Year Forecast Horizon 6% 4% Pct Error 2% 0% -2% -4% -6% -8% Forecast Vintage Pct Error on Rank Day 2-10 Pct Err on Rank Day 1 Current Model (Abs Pct Err) Test Model (Abs Pct Err) 18

19 Winter Forecast Observations While starting higher, Test Model has slower growth relative to 2018 forecast (0.2% vs 0.4%). More so than Summer, Winter forecast distribution is also spread out, with higher per-unitized values above the median. Zonal changes generally range between +/- 2%. PENLC, DQE, and RECO stood out as zones with negative changes outside this range. EKPC, PEPCO, and VEPCO stood out as zones with positive changes outside this range. 19

20 Next Steps Stakeholder Feedback Investigate additional refinements Develop Energy, Non-Coincident, and LDA CP models 20