Short-term Load Forecasting Based On DBN-RVM

Power load forecasting is of great significance to the safety,reliability and economic operation of power grid.There are many factors affecting the change of power load,which can be roughly divided into natural factors and social factors.Due to the coupling and changeable of various factors,the mechanism of load impact is complex,so the change of power load presents certain randomness.Any power load forecasting model will have some errors.How to improve the accuracy and reduce the forecasting errors has always been a research hotspot of scientific researchers.Aiming at the problems of low forecasting accuracy,poor robustness,strong noise influence,easy to fall into local optimum,insufficient generalization ability,and unable to consider many load factors in the models established by previous forecasting methods,combined with the characteristics of power load variation,this paper applies deep learning and relevant vector machine to short-term power load forecasting,and carries out in-depth research on the forecasting model.On this basis,a single forecasting model is improved to further improve the accuracy of load forecasting,and satisfactory forecasting results are obtained.In this paper,the main factors affecting the variation of power load are analyzed,and the periodic characteristics of daily average load are studied.Firstly,time series method and BP neural network are used to roughly predict the daily average temperature in the future.Then taking the main influencing factors of power load as the input of the forecasting model and the average daily load as the output of the model,the short-term power load forecasting model based on DBN and the short-term power load forecasting model based on RVM are established.The short-term power load forecasting is carried out by the above two models with and without temperature.In order to overcome the shortcomings of previous short-term load forecasting models and the disadvantage that the accuracy of single algorithm model is greatly affected by the accuracy of temperature forecasting,a new short-term load forecasting model based on DBN-RVM fusion algorithm is proposed.Firstly,the fusion algorithm model uses the method of relevant vector machine to build a general model of power load cycle change.Then the forecasting error compensation model of the general model is established by using the Deep belief network,and the forecasting error of the general model is compensated by the forecasting error compensation model so as to realize the high-precision forecasting of power load.The forecasting results of the single algorithm model and the fusion algorithm model are compared,considering the temperature and without considering the temperature,the average relative errors of forecasting results based on DBN model are 4.22% and 5.23% respectively,and those based on RVM model are 3.43% and 1.76% respectively.The average relative errors of forecasting results based on DBN-RVM fusion algorithm are 1.63% and 1.55% respectively.The results show that the fusion algorithm model based on DBN-RVM greatly improves the forecasting accuracy,and can achieve high-precision forecasting of short-term power load without considering temperature and temperature.The fusion algorithm model reduces the dependence on the accuracy of temperature forecasting,weakens the influence of random noise in load factors on power load forecasting,improves the robustness,applicability and reliability of the model and is a new and practical method for short-term load forecasting.Figure [31] table [15] reference [49]...