Research on electricity price forecasting with Bi-LSTM and ARIMA hybrid models based on Daubechies wavelet

In the electricity market environment, the increase of installed capacity of new energy and the change of market trading mode have brought about the difficulty of electricity price forecasting. This paper proposes a hybrid electricity price forecasting model based on Bi-directional Long Short-Term memory network (Bi-LSTM) and Autoregressive Integrated Moving Average model (ARIMA) under Daubechies N wavelet transform. The data were first decomposed into approximate and detail sub-series by Daubechies N. The Bi-LSTM model optimized by adaptive matrix estimation (Adam) is used to predict the future values of the approximate sub-series of electricity prices, and ARIMA is used to predict the future values of the detailed sub-series of electricity prices. Finally, the sum of the predicted values of the approximate and detailed electricity price sub-series is used as the final electricity price forecasting result. Through example analysis, the MSE, RMSE, MAE and MAPE prediction errors of the proposed method are 3.49, 1.86, 1.52 and 4.99, respectively. Compared with the Bi-LSTM and ARIMA models alone, the model has lowest prediction errors and higher prediction accuracy.