Short-term power load forecasting of CNN-LSTM based on feature selection

Efficient and accurate short-term load forecasting is an important guarantee for the safety, stability, and economic operation of the power system. Aiming at the problem of large prediction errors in peak and valley loads, a prediction model based on feature selection and neural networks is proposed through in-depth analysis of the working principles of feature selection and neural networks. Considering seasonal factors, the load was divided into four seasons: spring, summer, autumn, and winter. Pearson correlation analysis and Spearman correlation analysis were used to comprehensively analyze the climate and electricity price characteristics of the four seasons. A hybrid neural network combining convolutional neural network CNN and long short-term memory network LSTM was used. Using CNN to mine features between data to form new feature vectors, and then inputting this feature vector into LSTM for prediction. Simulate and verify the load forecasting model based on feature selection of CNN-LSTM using a power load dataset, and compare it with the full feature LSTM and feature selection LSTM models. The results show that there is a significant improvement in the accuracy of load forecasting after feature selection and using CNN models.