Performance Comparison Between Different Methods Of Photovoltaic Power Forecasting

As the increasing demand for global energy,photovoltaic(PV)power generation has been widely applied.The output of PV power generation systems is influenced by many random factors,which make the output power uncertain and intermittent.Therefore,accurate power forecasting of the PV power generation systems is of great significance for the construction planning,dispatching management and operation control of the power system.In Section ?,the current research status of PV power generation and the development of power prediction are analyzed.And the applications of back propagation(BP)neural network,support vector machine(SVM)and Bayesian(Bayes)algorithms in PV power generation are summarized.In Section ?,the three prediction models(BP-ANN?SVM?Bayes)are discussed in detail from the aspects of algorithm principle,main parameters,model training and testing,and their advantages and disadvantages are also briefly introduced.In Section ?,through the description of the empirical test platform,the source of the experimental data is clear,and through the selection of the historical operating data in 2016,the characteristics of PV output power are analyzed and the input variables of the model are selected according to Pearson correlation coefficient.In Section IV.according to the above three methods,six groups of time are designed from the aspect of data sample size to compare the performance of different models.The performance comparison is mainly analyzed from the training time of the model,the precision of the model,and the memory of the equipment needed for the training model.The three models are trained according to the data of the best size,and the prediction accuracy is compared,and the best modeling method is made clear.In Section V,the content of this paper is summarized and the future work is prospected.This paper mainly explores the adaptability and advantages and disadvantages of three main statistical modeling power prediction methods on different training sample sizes,which has a higher reference value for the selection of each model and the application scenario.General has the following conclusions:For small sample data,the Bayes method has less training time,occupies less memory and high forecasting accuracy.For medium sample,the result of SVM model is more stable and the overall effect is best under the condition of less training time and less computer memory occupation.For large sample data,SVM and Bayes methods are caused by training time and computer memory problems separately,are not suitable for PV power forecasting,but BP neural network has high accuracy and high performance.There are no fixed selection criteria for the parameters of machine learning algorithm model,which is closely related not only to the sample data,but also to the scientific research ability of researchers.Although the above three methods are compared in detail in photovoltaic power generation forecasting,there are still some deficiencies due to the limitation of-personal scientific research capacity and time.Hope to be able to further explore this topic on this basis to provide more useful reference for photovoltaic power prediction in the selection of methods and model building.