Electricity Price Forecasting Combined with Wavelet Packet Decomposition and a Hybrid Deep Neural Network in Spot Market

Abstract

Accurate electricity spot price forecasting is significant for market players to make decisions on bidding strategies. However, electricity spot prices are extremely volatile to forecast due to the influences of various factures. This paper develops an electricity price forecasting framework in spot market combined with wavelet packet decomposition (WPD) algorithm and a hybrid deep neural network. The WPD algorithm has higher decomposition accuracy and it can identify fluctuating trends and occasional noise in the data. The hybrid deep neural network is embedded with temporal convolutional neural (TCN) network, long and short-term memory (LSTM) neural network. The new hybrid framework is designed for improving the ability of feature extraction via TCN model and enhancing the efficiency of price forecasting. Case studies on the electricity market in UK confirm that the proposed model outperforms alternatives on the forecasting accuracy. Comparing to mean errors of other techniques, the average mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) of the proposed method are reduced by 27.3%, 66.9% and 22.8% respectively. Meanwhile, case studies on different denoising methods and datasets demonstrate that the proposed prediction model can better analyze the fluctuations in time series data and has certain generalization ability and robustness.