Research On Short Term Load Probability Density Forecasting Method Based On Regression Analysis

With the increasingly serious situation of energy shortage in our country,the pollution caused by fossil energy has more and more impact on people's lives,and many forms of pollution,such as fog and haze,have seriously restricted the development of national economy and the improvement of people's living standard.Accurate load forecasting is an important means of saving energy and reducing emission,improving economic benefits and protecting the ecological environment.With the modernization higher level of the power system and the continuous improvement in the level of intelligent algorithms,how to improve the accuracy of power system load forecasting and prediction,and various load forecasting results,provide reference information for more power staff,has become a major problem currently facing the electric power load forecasting.Therefore,this paper puts forward the research on the short-term load probability density prediction method of power system based on regression analysis,and studies the following content in depth:First of all,the application of traditional forecasting method in load forecasting is analyzed in detail in this paper,and the regression advantage of non parametric regression in load forecasting is analyzed.The main process of load forecasting is studied,which lays a foundation for the probability density prediction of short-term load.Then,a short-term load probability density prediction method based on ACE is proposed.On the one hand,based on the alternating conditional expectation(ACE)theory of nonparametric regression,the historical average temperature,calendar sequence and historical load expectations regression equation is established,and the optimal conversion equation is calculated by regression curve.On the other hand,the history similar days is clustered by fuzzy clustering method and a similar set of days is obtained.The average temperature and calendar sequence under the similar day set are selected,combined with relevant information of forecast date,and optimal conversion equation is used to get the set of regression values of load.Based on the estimation of Gauss kernel density estimation of kernel density estimation,select optimal window width by thumb principle,the probability of short-term load each time the density prediction curve is gotten.Using the actual data of a certain area,the simulation analysis by comparison with other methods proves the validity of the method.Secondly,this paper proposes a short-term load probability density prediction method based on ELM-quantile.Based on the extreme learning machine theory,quantile regression method of nonparametric regression is used to calculate the forecasting load quantile under different quantiles.By serialization of quantile,it is similar to the short-term load probability density prediction method based on ACE.This method also uses Gauss kernel density estimation method and thumb rule to analyze.The probability density prediction curve of each load at each time is obtained by combining the example,and compared with the traditional point prediction method,it is proved that the method is of high accuracy and reliability..Finally,on the basis of the above based on non parametric regression of two kinds of short term load forecasting method of probability density,taking a load as an example,the moment load probability density curves of the same day is predicted,By selecting mode point prediction and comparing the prediction results,this paper analyzes the advantages and disadvantages of the two methods.The validity and advanced nature of the two methods are verified.