Research On Load Forecasting Model Based On Improved Gaussian Process Regression

With the rapid development of the power grid,load forecasting plays an increasingly significant role in the operation of the power system.Precisely predicting the load of a certain place can not only improve the stability of the power grid operation to a certain extent,but also has a positive effect on the economy.With the continuous innovation of science and technology,the methods of load forecasting have gradually diversified,and the forecasting accuracy has also been continuously improved.However,there is still no forecasting method with absolute superiority and adaptability.Therefore,the improvement of load forecasting method has become a problem to be solved in the new power grid environment.Short-term load forecasting is the key to power dispatching and provides basic data for power generation planning and system security analysis.In this paper,short-term load forecasting is studied.Due to the volatility and randomness of short-term load data,after summarizing and analyzing the existing short-term load forecasting methods,Gaussian Process Regression,which is good at dealing with complex problems such as small sample,high dimension and nonlinear,is selected as the basic prediction algorithm.As a classical machine learning algorithm,Gaussian Process Regression has been applied in the field of load prediction.Compared with Neural Network and Support Vector Machine,Gaussian Process Regression has the advantages of fewer tunable parameters,strong generalization ability and easy implementation.However,the algorithm also has the problems that the parameters are difficult to determine and the training data selection is unreasonable.Therefore,this paper proposes a load forecasting model research based on the improved Gaussian Process Regression,and mainly discusses the following contents:Firstly,because the accurate use of the hyperparameters in the Gaussian Process Regression model directly affects the prediction accuracy,this paper proposes an improved Gaussian Process Regression model by combining Antlion Optimizer to solve the problem that the conjugate gradient method is unable to accurately obtain the hyperparameters.Firstly,the information sharing mechanism is introduced to solve the problem that the traditional Antlion Optimizer is easy to fall into local optimum.By reassigning the value of poor antlions,the optimization optimizer can jump out of local extremum and improve the performance of optimization.Information sharing Antlion Optimizer is used to obtain the optimal hyperparameters,which can effectively improve the prediction accuracy of GPR.Secondly,a hybrid prediction algorithm based on Density Peak Clustering and improved Gaussian Process Regression is proposed according to the strong similarity of short-term load series and the prediction characteristics of Gaussian Process Regression in handling small samples.Before the prediction,a new Density Peak Clustering algorithm is used to cluster the historical data and find the similar days of the prediction date,so as to build a more reasonable and effective data sample set.After training and learning the sample set,the prediction accuracy can be guaranteed and the prediction time can be greatly reduced.Finally,according to the above two improvement methods for Gaussian Process Regression,taking the power grid load in a certain place as an example,the simulation test is carried out in the exact same environment,and compared with the currently commonly used prediction methods.After the results are obtained through simulation and the error is calculated,it is concluded that the mixture model based on Density Peak Clustering and improved Gaussian Process Regression can better reflect the changing trend of load and further improve the accuracy of short-term load forecasting,which proves the feasibility of this forecasting method.It provides a new idea for short-term load forecasting theory.