AESO Load Forecast Application for Demand Side Participation. Eligibility Working Group September 26, 2017

Transcription

1 AESO Load Forecast Application for Demand Side Participation Eligibility Working Group September 26, 2017

2 Load forecasting for the Capacity Market Demand Considerations Provide further information on forecasting process including how flexible loads are reflected in current load forecasting methods and whether any conclusions can be drawn regarding whether coincident peak would be expected to coincide with system stress periods in forecast models Develop and provide a high-level assessment of the administrative implications of demand participation in the capacity market for each alternative (Supply and Demand side of the market) AESO Public, without prejudice 1

3 New AESO load forecasting tool Background Recognizing evolving business needs, the AESO kicked off a project in 2016 to find and implement a new load forecasting tool Reliable substation-level hourly load forecasts Flexibility to accommodate capacity market requirements including alignment of capacity market load forecast with transmission planning forecast SAS Energy Forecasting was chosen through a competitive process SAS LTLF (Long-term load forecast) tool completed implementation in May 2017 AESO still assessing and refining substation and area-level models and results 2

4 New AESO load forecasting tool SAS LTLF Capabilities Substation, planning area, planning region, and AIL-level hourly load forecasts, all reconcilable Probabilistic (e.g. P10/P90) or deterministic forecasting Inputs include: historical load data, weather variables, calendar variables, economic data Economic scenario modelling Energy Efficiency can be modelled (using an efficiency trend variable) or assumptions can be applied post-forecast Post modelling adjustments (e.g. for known retirements) Significantly less time to run means more up-to-date information included in forecast Future Capabilities Photovoltaic and Electric Vehicle adjustments Initial results At AIL level, results are similar to 2017 LTO Reference Case load growth (preenergy efficiency adjustment) 3

5 Response to group questions 1 Question: Provide further information on forecasting process including how flexible loads are reflected in current load forecasting methods and whether any conclusions can be drawn regarding whether coincident peak would be expected to coincide with system stress periods in forecast models. Historical performance of flexible loads can be captured through historical substation load data Past load patterns can be assessed to determine correlation with price, economic signals, system peaks, and other variables with the goal of estimating coincidence with system stress periods 4

6 Response to group questions 2 Develop and provide a high-level assessment of the administrative implications of demand participation in the capacity market for each alternative (Supply and Demand side of the market) Once all capacity market design elements are fully determined the AESO will ensure that demand response is captured in the supply or demand curve. Expected approaches: Supply side participation: The supply curve will be adjusted to reflect demand as offered into the supply curve Demand side participation: Baseline demand curve will be determined, the demand curve will be adjusted (moved to the left) with demand side offer price/ quantity pairs. Demand response eligibility in the capacity market a condition precedent for all these considerations 5

7 Thank You

8 Appendix 7

9 New AESO load forecasting tool - Additional details SAS LTLF Uses a generalized linear model framework The new tool utilizes a exhaustive diagnose process 4 different model structures are tested while running: Many combinations of calendar and temperature effects are iterated through and chosen based on certain criteria (e.g. reduce hourly MAPE) Model(s) accuracy assed during defined in-sample hold-out period to ensure model stability (comparing forecasted values to actuals) Winning model structure identified and used for forecast 8

10 New AESO load forecasting tool Additional details Capable of probabilistic forecasts based on historic profiles Currently working to integrate 30 years of historic temperature profiles, yielding a distribution of 30 observations per hour of the forecast (distribution defines the likelihood of forecast values) Multiple economic scenarios and variables can be tested in tandem Allows for easy economic uncertainty testing and comparisons Hourly modelling allows the AESO to use the most recent data while achieving high levels of statistical significance Flexibility to eliminate outliers based on user defined criteria Data quality checks ensure that historic data is representative 9