Research On Short-Term Load Forecasting Model Based On Deep Learning

With the deepening of power market reform,China's electricity companies began to participate in the power market.The core technology of the power selling company is to accurately predict the power load of users,formulate the power selling scheme and comprehensive power consumption service for customers,and avoid risks.In order to effectively avoid the risk in the process of selling electricity.At present,the common load forecasting model is time series model,but the accuracy of time series model is low.In order to improve the prediction performance of time series model,machine learning model is gradually applied to load forecasting,such as stochastic forest,support vector machine,BP neural network and other models.Machine learning model has stronger stability and prediction performance than time series model,and has improved the accuracy of load forecasting,but the prediction accuracy is still limited.Deep learning improves the prediction performance of the model by building deep neural network.Among them,DNN is a neural network with multiple hidden layers,which has good learning performance.At present,there is little research on DNN in load forecasting.Based on DNN neural network,this paper optimizes DNN and applies it in load forecasting to further improve the accuracy of load forecasting.In this paper,short-term load forecasting is studied.First of all,through the comparative analysis of traditional time series model,commonly used machine learning model and deep learning model,the shortcomings of these models are analyzed.Secondly,particle swarm optimization(PSO)and binary particle swarm optimization(BPSO)are used to optimize the DNN model of deep neural network,and the prediction accuracy of the DNN model is improved by adding the cyclic layer lifting model.Thirdly,in order to verify the performance of this model,this paper conducts empirical research on the model designed in this paper based on the data set of power load forecasting organized by European Intelligent Technology Network(EUNITE).This data set mainly records the electric load data of a certain region in Europe.The goal of this competition is to forecast the daily electric load.Therefore,this paper mainly uses the load data to forecast.In order to test the accuracy of the prediction model,the root mean square error and the mean absolute error are mainly used to measure.In the empirical analysis,pso-dnn and rdnn models are verified,and Holt winter,BP neural network,random forest,support vector machine,gbdt,xgboost model and LSTM model are used for comparative analysis.The analysis structure shows that the prediction performance of PSO optimized DNN model and rdnn model is better in many models,among which rdnn model has the best performance.The robustness of pso-dnn and rdnn models designed in this paper is verified by random sampling time series.The research of this paper has certain reference value for further strengthening the application of Deep Learning in load forecasting and further improving the accuracy of load forecasting.