A new framework for electricity price forecasting via multi-head self-attention and CNN-based techniques in the competitive electricity market

Due to recent technical improvements, the smart grid has become a feasible platform for electricity market participants to successfully regulate their bidding process based on demand-side management (DSM) perspectives. At this level, practical design, implementation, and assessment of numerous demand response mechanisms and robust short-term price forecasting development in day-ahead transactions are all critical. The accuracy and effectiveness of the day-ahead price forecasting process are crucial concerns in a deregulated market. In this market, the reason for low accuracy is the limitation of electricity generation compared to the electricity demand variations. Hence, this study proposes a suitable technique for forecasting electricity prices using a multi-head self-attention and Convolutional Neural networks (CNN) based approach. Further, this study develops a feature selection technique using mutual information (MI) and neural networks (NN) to choose suitable input variable subsets significantly affecting electricity price predictions simultaneously. The combination of MI and NN reduces the number of input features used in the model, thereby decreasing the computational complexity of the NN. The actual data sets from the Ontario electricity market in 2020 are acquired to verify the simulation results. Finally, the simulation results proved the efficiency of the proposed method by demonstrating increased accuracy by attaining the lowest average value for MAPE and RMSE with a value of 1.75% and 0.0085, respectively, and compared to results obtained by recent computational intelligence approaches. By attaining accurate electricity price results, the significance of this study can be summed up as aiding the electricity industry's operators in administering effective energy management, efficient resource allocation, and informed decision-making.