Risk-Averse Optimal Combining Forecasts for Renewable Energy Trading Under CVaR Assessment of Forecast Errors

Abstract

Renewable energy producer is often exposed to huge financial losses in some imbalance hours (meaning that the contracted energy in day-ahead market is not equal to the actual output in real time) caused by extremely large forecast error. To address this challenge, this paper integrates the forecast end-user's risk profile into the development of risk-averse combining forecast approach for renewable energy trading. First, the conditional value-at-risk (CVaR) is applied to evaluate the extreme prediction error of combined forecasts. Then, convex optimization models are formulated with the objective of minimizing the mean square error plus the CVaR of large error. Solving our proposed models determines the optimal weights for individual models participating in the combined forecasts. Finally, the value of risk-averse combined forecasts is verified through examining the financial performance of using risk-averse forecasts as inputs of the bidding strategy in renewable energy trading. Case studies on real-world datasets present that our proposed method not only reduces the mean error but also lowers the extreme error. More importantly, it decreases the imbalance energy and cost in renewable energy trading, thus being less exposed to the risk of large financial losses under extreme prediction errors.