The Study On Application Of Neural Network Technology In Short-term Power Load Forecasting

With the vigorous development of the power industry and the continuous improvement of the people’s living standards,the demand for electrical energy in all walks of life is increasing day by day.For the accurate and reasonable dispatch and distribution of electrical energy to meet the different needs of all kinds of users,short-term power load forecasting has gradually become an important research content in power system dispatching and management.The high accuracy of power load forecasting is beneficial for improving the economic efficiency and competitiveness of power systems,improving the safety and reliability of system operations,and saving energy.Therefore,it is of great significance to study how to improve its accuracy.The prediction of short-term power load has many characteristics,such as many influencing factors,changing randomness and nonlinearity.The performance of the prediction model is influenced by many key factors,such as prediction data,input characteristics,model selection,model parameters,model optimization and so on.In the past,many excellent methods and theories are put forward in the study of power load forecasting.Early mathematical statistics methods are mainly based on data fitting to establish the model.In the era of large data,the prediction effect is not ideal.Modern artificial intelligence method and the most widely used BP neural network have strong computing,self-learning and reasoning ability,However,this method also has uncertainties such as the number of hidden layers,the number of hidden layer nodes,and the selection of initial weight and thresholds,and easy to fall into local optimum problems.This paper focuses on the factors affecting the performance of the predictive model,including the analysis and processing of data,the selection of network model features,the determination of model parameters,and the combination of the prediction model from shallow network to deep network and model optimization to improve the prediction model performance.The main work of this study is to improve the performance of load forecasting model including:(1)Data analysis and processing.Data analysis is the foundation of model establishment,mainly analyzes the relationship between historical load data time series,and finds the periodic relationship between data between year,season,month,week and day.The relationship between load data and meteorological factors is analyzed,and the eight meteorological factors that affect the load change are selected through the calculation of data analysis and entropy weight method.The data is normalized to improve the accuracy of the prediction model.(2)Based on the BP neural network,an optimized BP neural network load forecasting model combined with similar day methods is proposed.For the shortcomings of BP network,a genetic algorithm is used to optimize the initial weights and thresholds of the network,which improve the prediction accuracy and stability of the network.Optimized BP network combined with similar day algorithm,and use the similar-day algorithm to select historical date data with a similarity degree greater than 0.8 on the basis of similarity as the input data of the prediction network for network training and prediction.(3)Use the deep network DBN model to carry out the daily 24-point load forecast.This paper introduces the basic structure of DBN model RBM network and network model derivation and training process,designs a load forecasting framework based on DBN,sets the input characteristics of the network,including date factors,meteorological factors,load values,and designs a 7-day sliding window 24 The hour load value is used as an input feature.The network parameters of the network model are set through extensive experiments and analysis.(4)The model comparison experiment and analysis were completed.Through the actual data of the power grid,an optimized BP neural network load forecasting model combined with similar day methods is proposed for daily average load value prediction experiments,Compared with the experimental results of BP neural network,the accuracy and stability of the optimized model are improved.The daily 24-point load forecasting experiment based on DBN model was completed and compared with the prediction results of ANN model of 1 hidden layer and 2 hidden layers.The experimental results show that the prediction effect of DBN model is better.