Optimization Based On Prophet Model And Its Application In Regional Electricity Consumption Forecasting

Electric power is an indispensable energy for social development,and its supply is an important factor that directly affects the development of social economy and science and technology,while regional electric power forecasting is an important basis for improving the unbalanced state of national electric power supply and demand,at present,the regional electricity consumption forecasting method has become a hot topic of scholars at home and abroad.With the rapid development of machine learning technology,many excellent prediction algorithms have emerged,such as time series prediction algorithm,neural network algorithm,etc.,it is widely used to solve the practical problems of regional power consumption forecasting.Prophet is a time series prediction algorithm,which is composed of growth trend,periodic trend,holiday and noise.It performs well in the related fields of business sales forecast and time series forecast.However,each part of Prophet Algorithm focuses on local feature extraction,and it is easy to get into local over-fitting in the process of prediction,which leads to large prediction error.In order to overcome some shortcomings in prediction behavior of Prophet model,this paper proposes a fusion model,which uses the regular term of XGBoost model to solve the over-fitting problem,and introduces a nonlinear error correction mechanism to correct the error of the improved model,in order to achieve the goal of reducing prediction error.The main research contents of this paper are as follows:(1)Aiming at the problem that the prediction of single Prophet model is easy to fall into local over-fitting,an X-Prophet fusion model is proposed in this paper.First of all,according to the characteristics of electricity consumption data affected by special time factors,a Prophet model with mining date features is introduced to extract holiday information from historical electricity consumption data,and the process of adding prediction model training is shown,secondly,in order to avoid the over-fitting problem of the Prophet model on special time nodes such as holidays,a fusion model X-Prophet is proposed and a XGBoost strong classifier is introduced,based on the characteristics of XGBoost regularization which can effectively prevent over-fitting,the feature rules in historical power consumption data are fully mined,thus improving the accuracy of prediction and reducing the mean absolute error(MAPE)and root mean square error(RMSE)of prediction results.The experimental results show that the prediction result of X-prophet is superior to that of single Prophet model,the prediction error MAPE is reduced by about 11.1%,the RMSE is also reduced by about 10.5%,which solves the problem of large prediction error caused by local over-fitting of Prophet.(2)Aiming at the problem of large deviation between the prediction value and the real value of the improved model X-prophet,a nonlinear error correction mechanism is introduced to optimize the X-Prophet model.At the same time,the real error between the reference term and the prediction result term is analyzed and calculated,and the error term is predicted by using the nonlinear error prediction mechanism,the final prediction result is obtained by modifying the prediction value according to the error correction parameters.The experimental results show that the optimized X-Prophet model has smaller prediction error,MAPE reduced by about 1.5%,RMSE reduced by about 1.3%,which proves that the improved algorithm is effective and feasible.(3)According to the research results of this paper,a regional power consumption forecasting system is designed and implemented.The system adopts C/S architecture,including four modules: User Registration/login,data processing,power consumption forecasting and user management,the non-linear Error Correction Model X-Prophet is used in the power consumption analysis and prediction module of the system.