Load Forecasting Method Based on CEEMDAN Signal Decomposition and Improved BiLSTM

The accuracy of load forecasting in the power system becomes increasingly important with the step-by-step construction and improvement of new types of power systems. In this article, a method using the CEEMDAN algorithm for modal decomposition of sequences is proposed, where the decomposed sequences are separately analyzed for prediction and then reconstructed to form the complete signal sequence. The main prediction model uses BiLSTM. Compared to traditional LSTM, the ability of bidirectional reading allows BiLSTM to better understand and model the contextual information in the sequence, especially in tasks involving bidirectional dependencies where the effect is significant. Finally, the Love Evolutionary Optimization algorithm was used to improve the hyperparameters of the prediction model, further enhancing the prediction accuracy. By comparing the prediction accuracy results under different methods, further evidence was provided to support the advanced nature of this prediction method.