Application Research Of Short-term Power Load Forecasting Based On Neural Network

China’s huge population and industrial output lead to a huge demand for the electric power.However,due to the complex operation mode of the power system and the uncertainty of user’s electricity consumption,as well as the influence of the power equipment,the studies of power load forecasting have a very high value in both academia and industry.Accurate prediction of short-term power load can help eclectic companies adjust load equipment in time,reduce resource wasting,improve performance and stability of power network.For the short-term power load forecasting,there are not only studies on peak load forecasting,but also studies on daily load at specific time points.With the in-depth studying of artificial intelligence,domestic and foreign experts used the neural network which with fast computing ability and self-learning ability in the studying of power load forecasting.However,the training model of neural network algorithm is affected by parameters,its training speed and convergence results are uncertain,and it is easy to fall into local optimal results.In view of the problems in the studies of short-term power load forecasting,combining the short-term power load data and the characteristics of application demand,this thesis puts forward two models of 7-day peak load forecasting and 7-day 24 points load forecasting based on neural network.This thesis puts eyes on the optimization and performance improvement of short-term power load forecasting model from the aspects of data processing,data analysis,model determination,model parameter determination,model optimization and improvement.The main work of this thesis includes:(1)Completing the data analysis and processing.According to the power load data set used in this thesis,the general characteristics of power load data are analyzed,and the periodicity characteristics of the year,month,week and day are emphasized.According to the characteristics of it’s periodic change,the missing data complementation,abnormal data processing and data standardization are realized.(2)A 7-day peak load forecasting model based on optimized RBF is proposed in this thesis.Genetic algorithm is used to optimize the center vector,center point’s width,weight between the hidden and output layer in RBF network,which can weaken the influence of parameter selection on forecasting results and improve the generalization ability and prediction accuracy of the model.Through the comparative experiment,the feasibility of the daily peak forecasting model is proved,and it’s prediction accuracy improved..(3)This thesis used a 24-point daily load forecasting model based on optimized LSTM.In order to avoid the local optimization caused by the artificial input parameters of LSTM,particle swarm optimization algorithm is used to train the model parameters.By this way to improve the accuracy of LSTM network.and the comparison experiment shows that the prediction accuracy of the 7-day 24-point prediction model is improved.