Research On Short-term Pv Power Prediction Method Based On Machine Learning

Energy crisis and environmental problems together promote the change of the energy structure of the whole world,and solar energy,as a typical representative of clean energy,with its huge potential of cleanliness,efficiency and so on,has gradually become one of the leading roles in this change.However,photovoltaic power generation has strong uncertainty and randomness,and its grid-connection brings great security risks to the power system,such as impact,volatility,damage to power quality,etc.Therefore,the accurate prediction research of photovoltaic power generation is of great significance to the construction of a strong and intelligent power system with new energy as the main body.In this context,this paper is based on machine learning algorithm to carry out in-depth research on photovoltaic power generation power prediction.(1)Taking the data from the Australian Solar Research and Development Center as the sample,the main factors affecting photovoltaic power generation were screened,and the dimensionality of the model input was reduced.Moreover,in order to weaken the influence of different weather types on the proposed model,the similar day algorithm is combined to convert different weather types into the photovoltaic power of their corresponding similar days,and the data set of photovoltaic power of similar days is composed.Compared with the data set without adding similar day photovoltaic power,a comparative experiment was carried out under the same model.The experimental results show that adding photovoltaic power of similar days to the data set can effectively extract data features and significantly improve the prediction accuracy of the prediction model.(2)Based on the two sets of data sets with and without similar daily photovoltaic power,the prediction model of error back propagation neural network(BPNN),the prediction model of support vector regression(SVR)and the prediction model of ensemble learning in machine learning were established respectively.The experimental simulation showed that: Under the above two data sets,the ensemble learning prediction model has a good prediction performance,and the prediction accuracy is significantly improved compared with the unoptimized single BPNN prediction model and the SVR prediction model.(3)Aiming at the problem that the prediction model is easy to fall into local optimum,particle swarm optimization(PSO)and genetic algorithm(GA)were adopted to optimize the parameters of BPNN model,GA was used to optimize the parameters of SVR prediction model,and three combined prediction models were established by adding the data set of photovoltaic power generation of similar days.Experimental comparison and analysis were conducted with the original prediction model respectively.Compared with the original model,the prediction accuracy of the three combined prediction models was significantly improved,among which,regardless of the time cost,the GA-SVR prediction model had the best prediction ability.