Short‐Term Electricity Price Forecasting Using the Empirical Mode Decomposed Hilbert‐LSTM and Wavelet‐LSTM Models

Abstract

The electricity sector deregulation has led to the formation of short‐term power markets where the consumers can purchase electricity by bidding at the electricity market. The electricity market price is volatile and changes are due to change in electricity demand and the bid price at different span of time during the day. The availability of the electricity price forecast is essential for the electricity market participants to make informed decisions. In this paper, the modified LSTM approach, wavelet‐LSTM, and Hilbert‐LSTM are proposed to predict the electricity price for bidding in the short‐term electricity market. The objective is to improve the precision and adaptability of electricity price predictions by utilizing the temporal dependence identification capability of LSTM and the multiresolution analysis capability of the transforms. The proposed models combine these two effective methods in order to capture both the long‐term trends and short‐term variations present in electricity price time series data. In this approach, the 8‐year dataset is used for training the models, and based on this the day‐ahead price is calculated and compared with the testing data. The proposed techniques show better performance in terms of rank correlation, mean square error, and root mean square error compared to the existing algorithms of LSTM and CNN‐LSTM. The prediction results achieved with wavelet‐LSTM and Hilbert‐LSTM (1‐month dataset of 8 years) are rank correlation 0.9746 and 0.9749, MSE 0.2962 and 0.1363, and RMSE 0.5443 and 0.3692, respectively. The results achieved with the proposed methods are better than the existing forecasting models, and the RMSE for Hilbert‐LSTM and wavelet‐LSTM techniques is improved by 61% and 43%, respectively, compared to the LSTM method. Also, results are calculated for the complete 8 years all 12 months with Hilbert‐LSTM, and the results achieved are rank correlation 0.9645, MSE 0.3876, and RMSE 0.6225. The results achieved with the proposed models are improved in terms of performance parameters compared to the conventional approaches. The proposed models can be used in the day‐ahead electricity price forecasting to bid for electricity accurately in the day‐ahead electricity market.