Short-term load forecasting based on MB-LSTM neural network

Short term load forecasting is the basic work for the safe and stable operation of power system. With the development of power market, the variability and randomness of short-term load increase the complexity of power load. Higher requirements are put forward for the accuracy of load forecasting. In this paper, a short-term load forecasting framework based on multi-layer stacked bidirectional long short-term memory (MB-LSTM) is proposed. The goal of the model is to solve the problem of error accumulation caused by traditional LSTM and improve the forecasting accuracy. A feedback mechanism is designed to improve the unidirectionality of traditional LSTM network prediction. In order to help the convergence to the optimal solution more quickly, the correlation between the historical load data is mined by the deep learning training method of multi-layer superposition. Finally, the experimental results show that the method can extract data features on the basis of the prediction results, and the effectiveness of the improved method is proved by comparing and analyzing the traditional prediction models.