Power Load Forecasting Based On LSTM Deep-Network

In recent years,with changes in energy production and consumption patterns,it has brought more uncertainties to the power grid.Power load forecasting is an essential basic work in the development of the power industry.A reasonable and accurate power load forecast is Provide important reference for power grid dispatching,planning,construction and operation.Therefore,adopting a more flexible forecasting method to improve the accuracy of short-term power load forecasting is still a work that we urgently need to solve,which is of great significance to the development of the power grid.Power load forecasting will be affected by many factors.The analysis of influencing factors is the basis of power load forecasting and can provide a reference for the determination of the input of the forecasting model.In this paper,the Pearson correlation coefficient is used to analyze the correlation between different influencing factors and load.The analysis shows that historical load,weather type,date type and other factors have a strong correlation with power load.Artificial neural network has been used in the field of electric load forecasting due to its powerful self-learning function,associative storage function,and ability to find optimized solutions at high speed,and has achieved good results.With the in-depth study of artificial neural networks by scholars,a new research direction of deep learning has emerged.Based on the LSTM neural network prediction model in deep learning,this paper proposes an LSTM multivariate time series model incorporating multiple influencing factors.By adding a multi-feature fusion layer on the basis of the LSTM model,it effectively compensates for the defect that the LSTM model cannot extract multiple features.It has been verified that the LSTM multivariate time series model incorporating multiple influencing factors can effectively improve the prediction accuracy of short-and medium-termpower load forecasting.In order to make up for the defect that the multi-influence factor LSTM multivariate time series model cannot extract multi-dimensional features from the data,this paper further uses a dilute autoencoder to optimize the model and establish the SAE-LSTM prediction model.This model can extract high-level features of influencing factors and mine their deep-level feature factors,effectively making up for the shortcomings of the LSTM multivariate time series model incorporating multiple influencing factors.Based on the load data of Shanxi Province and the data of related influencing factors,this paper uses the LSTM model,the LSTM multivariate time series model incorporating multiple influencing factors,and the SAE-LSTM model to make short-and medium-term power load predictions.The prediction results prove that both the LSTM multivariate time series model and SAE-LSTM model incorporating multiple influencing factors can improve the accuracy of short-term power load prediction,and the prediction performance of the SAE-LSTM model is more superior.