Regularized Probabilistic Forecasting of Electricity Wholesale Price and Demand

Abstract

Electricity price forecasting is a fundamental step for power producers in competitive electricity markets, although it is a challenging task. Participation of renewable power plants in the electricity supply chain has increased uncertainty of electricity supply, demand, and price. Probabilistic forecasting approaches are the proper tools to take into account this uncertainty. In this paper, we use the double exponential smoothing (DES) as well as triple exponential smoothing (TES) methods to forecast electricity price volatility. Regularized forecasts for volatility have been studied using the elastic net regularization method. Sample sign correlation of standardized electricity prices (standardized by volatility forecasts) is used to identify the conditional distribution of electricity price time series. Validation of the regularized volatility forecasts is demonstrated using the publicly available hourly electricity price data of Ontario. Our data analysis results show that TES forecasts of volatility outperforms DES. In addition, elastic net regularization decreases the mean square error of TES volatility forecasting from 72.95 to 59.16. The regularized probabilistic forecast of electricity demand is used to implement a decision analysis approach and to model the scheduling of power generation units in the electricity market.