Electricity price forecasting with high penetration of renewable energy using attention-based LSTM network trained by crisscross optimization

Abstract

Accurate electricity price forecasts is the common concern of market participants. With the integration of high penetration of wind and solar energy resources into the power system, the renewable energy sources will have a great impact on the electricity price volatility undoubtedly. In this regard, a novel attention mechanism (AM) based electricity price forecasting model for electricity market with high proportion of renewable energy is proposed in this paper. In order to investigate the effect of renewable energy on the electricity price prediction, the wind power generation, solar power generation, predicted load and the historical price series are simultaneously taken as the input features. In the data preprocessing stage, the empirical wavelet transform (EWT) is applied to decompose each of the input features into multiple components to avoid learning the autocorrelation of the original sequence. In the model training stage, a hybrid AM-based long short-term memory network (LSTM) is proposed as the forecasting model, aiming to make full use of the AM to dynamically evaluate the importance of different input feature. Furthermore, the crisscross optimization algorithm (CSO) is adopted to retrain the parameters of fully-connected layer so as to further enhance the generalization ability. The proposed method is validated on the datasets of Danish electricity market with a high proportion of renewable energy, and the experimental results show that the proposed model is superior to other hybrid models involved in this study.