Electricity Demand Forecasting at ISO-NE: Review of Peak Load Models prepared for Stakeholder Meeting Installed Capacity (“IC”) Methodology Review July 14 th , 2006 Westborough, MA by Benchmark Forecasts Consulting Douglas R. Hale, [email protected]Frederick L. Joutz, [email protected]
Electricity Demand Forecasting at ISO-NE: Review of Peak Load Models prepared for Stakeholder Meeting Installed Capacity (“IC”) Methodology Review July 14 th , 2006 Westborough, MA by Benchmark Forecasts Consulting Douglas R. Hale, [email protected] Frederick L. Joutz, [email protected]. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Electricity Demand Forecasting at ISO-NE: Review of Peak Load Models
- Provide an independent evaluation of ISO-NE’s energy and peak demand forecasting models
- Compare ISO-NE’s methodology with industry
- Recommendations for improving models and forecasts
• Milestones
– Briefing and preliminary report June 8th.
– Draft report and briefing late June / early July
– Final report early / mid July
• Long Run and Medium Term Forecasting Models
• Short Run and Peak Forecasting Models
3
Preliminary Findings on ISO-NE’s Energy Models and Methodology
• The ISO-NE forecasting methodology• The short term energy model forecasts demand two years out• The long term energy model forecasts annual demand ten years out• We have examined the peak demand forecasts
4
Replication of Seasonal Peak Models
• Peak Models for Resource System Planning 2006 (RSP06)– Winter– May– June– July– August– September
• Sample Weekdays January 1992 through August 2005
5
Replication of Seasonal Peak Models• Winter Peak Model Specification is Standard and includes
– a Base Load trend– Heating Degree Day Measure – Monthly Dummy Variables– Monday and Friday Effects
• Separate Summer Peak Model Specification is Standard and includes– A Base Load Trend– 3-Day Weighted Temperature Humidity Index (SWTHI)– Heat Wave Variable– Monday and Friday Effects
The ADL model in levels or natural logarithms demonstrates a significantincrease in explanatory power.
12
Cooling Load Indexes
• Cooling Load Index (CLI)
• Base-Cooling Load Index (BCLI)
• Estimated from Daily Peak Models
• Peak Load(t) = b0 + b1 CDD(t)
• CLI(t) = b1(t) / b1(1992)
• BCLI(t) = b0(t) / b0(1992)
• Able to Replicate these models and the Normalizations with Trends
• Tested whether adjusting for July August vs. May September made a difference. None found.
13
• Coefficient on WTHI a measure of peak load sensitivity to temperature/humidity– Indexed to 1992 value– Estimate a trend– Strong growth over historical period Doubling 1992-2005– Increasing penetration of Air Conditioning
• Constant a measure of peak load dependent on more general economic conditions– Indexed to 1992 value– Estimate a trend– Slower growth over historical period 16% increase 1992-2005
• Heating Index and Heating Base Load Index for heating season (Jan-Apr,Oct-Dec)
14
1.0
1.2
1.4
1.6
1.8
2.0
92 93 94 95 96 97 98 99 00 01 02 03 04 05
CLICLIJJA
TCLITCLIJA
Comparison of Cooling Load Indexes Summer vs. June-August
15
1.00
1.04
1.08
1.12
1.16
92 93 94 95 96 97 98 99 00 01 02 03 04 05
BCLITBCLI
BCLIJJATBCLIJJA
Comparison of Base Cooling Load Indexes Summer vs. June-August
16
14000
16000
18000
20000
22000
24000
26000
28000
80 82 84 86 88 90 92 94 96 98 00 02 04
PKA PKWN
Annual Peak (Actual and Weather Normal)
17
9000
10000
11000
12000
13000
14000
15000
16000
80 82 84 86 88 90 92 94 96 98 00 02 04
NELA/8.76 NELWN/8.76
Average Annual Hourly Load (Actual and Weather Normal)
18
Load Factor Construction
Load Factor is Ratio of Annual “Adjusted” Hourly Load to Annual “Adjusted” Peak Loads
DSM adjusted load factor =
(Energy + DSM on Energy) / ( 8.76 * (Peak + DSM on Peak) )
The Change to DSM-adjusted Load Factor shows the impact over time of carrying forward the change in the short-run forecast and the load factor decrement
Long-run reference summer peak forecast based on long-run energy forecast and a declining load factor.
Driven by increasing relative growth in Peak Demand
But part of this may be do to inclusion of DSM factors in calculation.
It is not clear that this offers a clear picture
The DSM factors or adjustments are not directly measured and taken as given by ISO-NE.
There can “political” or institutional factors driving these which the ISO-NE rightly chooses not to get involved in.
22
Specification of the Estimated Models
• Diagnostic Testing of Estimated Models– Model Fit– Standard Error – Adj. R squared– Constant Variance– Correlated Errors– Model Stability– Elasticities (Price, Income, Weather)
• Time Series Properties of the Data– Short-run Dynamics– Long-run Relations– Integration / Cointegration
23
Replication of the Forecasts
• Forecast Simulation
– Performed simple simulation conditioned on actual explanatory variables
– Long-run State Models are reasonably close
• Forecast Replication is close. Still need to compare notes with ISO-NE staff
• Forecasting Process and Theory
– Recent Developments in Econometric and Forecasting Techniques
– Archiving and Documentation Procedures
24
Our Replication Experience
• Good news-we did it, all the data are extant, programs worked as advertised, etc.
• Not so good news-couldn’t have done it without lots of help• Learned a lot about models that was not obvious from reported
statistics
25
Recommendations• Improve Documentation• Data and Model Archiving• Seasonal Peak Models (Specification Issues)
– Choice of Included Variables– Levels or Natural Logarithms– ADL Model Dynamics (Serial Correlation Correction)
• Cooling Load Indexes• Load Factor
– DSM Issue– Choice of Projection (Level, Continued Trend, and Smoothing)
26
Recommendations• Switch from Current Annual Aggregate ISO-NE Model to a Quarterly
or Monthly Model.• Level of Detail
– Total Load – avoid data problems in the data– Sectoral - advantage might be a gain for peak use by residential
and commercial sectors
• Consider the MIT Center for Energy Policy Research Center for Evaluating Actual Decision Making process regarding Capacity Expansion– New Director specializes in the techniques– Real Options approach
27
Are the Energy Forecasting Models Clearly Described in the Documentation?
• The equations for the short and long term models are exhibited• Data sources are identified and some data series are included• The estimation results, some diagnostics and some forecast errors
are reported• The general approach to merging the short and long term models is
described
28
Assessment of ISO-NE Energy Model Documentation
• In our experience ISO-NE has done more than most forecasters to document their methods and make them accessible
• The content (models, data, estimation, error experience, etc.) is good
• Certain transformations to splice the short and long term forecasts are not fully explained
29
Replication: Why is Replication Important?
• Important component of scientific process.• Provide confidence in methodology.• Serve as double-check on models and data.• Starting point for further analysis and diagnostic tests.• Verification of Documentation and Archives
30
Replication: Why is Replication Important?
• Examination of the Historical Data• Specification of the Estimated Models• Economic Theory• Statistical Theory • Time Series Properties of the Data• Diagnostic Testing of Estimated Models• Forecast Simulation• Forecasting Process and Theory