Multi-feature Short-Term Power Load Prediction Method Based on Bidirectional LSTM Network

Power system load prediction is an important task of power enterprises, and the analysis method and prediction accuracy of regional grid load characteristics are a key factor in building smart grids and improving the consumption capacity of distributed power generation. Aiming at the problem of reducing the accuracy of the prediction model due to the large number of factors affecting the load forecast, the degree of influence is different, and the strong correlation between multiple influencing factors is caused. In this paper, a multi-feature short-term load prediction method based on deep learning is proposed. The core is to organically combine the improved bidirectional long-short-term memory (BiLSTM) model with the multi-feature data mining method, extract the high-dimensional features of the input vector by using deep convolution through the Inception structure, optimize the weight distribution of the output vector based on the Attention attention mechanism, and construct the Inception-BiLSTM-Attention model. Based on Inception-BiLSTM-Attention, a multi-chain fusion model is constructed, and the feature learning of multiple time period dimensions is carried out, and the short-term power load prediction with high accuracy is realized. This study can provide a reference for regional power system optimization decisions.