Research on Neural Network Model and Algorithm for Short-term Load Prediction

With the promotion of China's smart grid construction, the short-term load forecasting of power system is regarded as an important part of energy management system. At present, the amount of data in the whole grid has reached the scale of big data, especially in the user side, the uncertainty of user behavior and the scattered distribution of user data are particularly complex, and the traditional load forecasting methods have gradually withdrawn from the historical stage. In addition, due to the different economic development levels in different regions and the different power grid price competition mechanisms, different requirements are put forward for power load forecasting.This paper proposes a LSTM Prediction Model Combining Long-term with Short-term Prediction for power load, specifically for the load data with strong periodicity. The method can not only learn the characteristic information within each cycle, but also learn the characteristic information between each cycle. Therefore, it can learn data characteristics from multiple perspectives.