Electricity market clearing price prediction based on mode decomposition and improved transformer

To enhance the accuracy of settlement price forecasting in the electricity market, this paper proposes an innovative forecasting model, MCPformer, based on the Transformer framework, aimed at overcoming the limitations of existing models in capturing the complex seasonal, trend, and non-stationary characteristics of electricity market settlement price. Utilizing the ICEEMDAN algorithm, the model decomposes the original sequence and reorganizes the data based on sample entropy, while a multi-head causal attention mechanism is introduced to explore the temporal dependencies of long sequences. Furthermore, a module combining TCN and multi-head causal attention mechanisms is designed to extract and amplify the seasonal and trend features of the series. Finally, a multi-head attention summation layer integrates these features, achieving precise real-time forecasting of electricity market settlement price. Experimental evaluations confirm the superiority and universality of MCPformer, demonstrating varying degrees of improvement in accuracy compared to benchmark models.