DR potential probabilistic forecasting model of load aggregators based on ensemble learning

Day-ahead forecasting on the demand response (DR) potential of the load aggregators could provide important reference information for the quotations and volumes of load aggregators in the electricity market, thus reducing decision-making risks. Aiming at the shortcomings of generalization and reliability of a single-point forecasting model, this paper proposes an online DR potential probabilistic forecasting model of load aggregators based on ensemble learning, which can effectively improve the accuracy and generalization ability of the probabilistic forecasting model. Firstly, the multivariate influencing features of the DR potential of load aggregators are extracted, and the support vector machine-based recursive feature elimination (SVM-RFE) method is used to select features. Secondly, multiple single probabilistic forecasting models are proposed based on the non-parametric kernel density estimation. Finally, a DR potential ensemble probabilistic forecasting model of load aggregators based on "repeated game, dynamic update" is established, which adaptively learns the weights of each base model through the idea of game theory and dynamically update the weights over time. Simulation experiments show that the probabilistic forecasting model proposed in this paper has better accuracy and generalization than a single prediction model.