The Day-Ahead Electricity Price Forecasting Based on Stacked CNN and LSTM

Abstract

In the competitive electricity market, achieving accurate electricity price forecasting is important to the participants. To improve the electricity price forecasting accuracy, the stacked CNN and LSTM model is proposed in this paper. Periodic patterns exist in the electricity price time series, i.e., dependency between different timestamps exists, and the features are selected based on the patterns to forecast day-ahead electricity price. Then, the CNN model is designed and the original time series is transformed into image-like samples based on the periodic patterns, which will help CNN to learn the data more effectively. Next, LSTM model is designed based on the selected features. Last, the stacking method, which is an ensemble learning strategy, is adopted to achieve better accuracy by fusing the forecasted values of CNN and LSTM models. The proposed model is validated on the Pennsylvania - New Jersey - Maryland market data, and the results show that the proposed model can indeed improve the forecasting accuracy.