Research On Short-term Load Forecasting On Fisher Information And Deep Neural Network

Short-term load forecasting is one of the important components of smart grid energy management system.Meteorological factors have a great impact on the accuracy of short-term load forecasting.Based on the big data environment of smart grid and the correlation analysis between historical load data and meteorological factors,this paper proposes a method of modeling meteorological factors using Fisher information processing,and validates it by combining BP neural network,deep belief network(DBN)and new RBM-Elman network prediction model.For a long time,meteorological factors are one of the influencing factors in the short-term load forecasting.The suitable meteorological factors can improve the final prediction accuracy.Fisher information theory provides a way to monitor system state and state changes by monitoring system variables.Fisher information can be used to solve the problem of introducing meteorological factors into short-term load forecasting.This paper first introduces the treatment methods of single meteorological factors and multi-meteorological factors,and verifies its effectiveness.Secondly,the feature selection based on Fisher information is introduced,which solves the problem of selecting many feature input quantities and further improves the prediction accuracy and speed.In recent years,deep learning algorithms have become more and more widely used in short-term load forecasting.This paper first introduces BP neural network.On this basis,the application of deep belief network and new RBM-Elman network in shortterm load forecasting is discussed.The deep belief network consists of a multi-layer Restricted Boltzmann machine(RBM)and a BP neural network.It overcomes the slow convergence speed and falls into local extremum of traditional networks,and can improve the prediction accuracy.Further,a new type of network structure,RBM-Elman network,is proposed.Unlike static BP network,the Elman network is a dynamic recursive network.The new network uses RBM to initialize the parameters of the Elman network,overcomes the problem of random initialization parameters of the BP network,and further improves the accuracy of short-term load forecasting.Finally,the meteorological factors modeling method based on Fisher information with BP,DBN and RBM-Elman prediction models are used to verify the simulation.The results show that the model prediction results using Fisher information are better than those without Fisher information.Fisher information feature selection model is superior to the model without feature selection in prediction accuracy.DBN model and RBM-Elman model are superior to BP network in prediction accuracy.