Research On Ultra-Short-Term Power Prediction Of Photovoltaic Power Generation

Affected by solar radiation intensity and meteorological conditions,photovoltaic power generation is highly random,volatile and intermittent,which brings severe challenges to large-scale photovoltaic power grid-connected.Therefore,accurate prediction of photovoltaic power generation has important practical significance.This thesis summarizes the research progress in recent years in related fields at home and abroad,and analyses and studies the ultra-short-term power prediction of photovoltaic power generation.In this thesis,the photovoltaic power station of DKASC in Australia is taken as the research area.The data are from its official website(http://dkasolarcentre.com.au/locations/alice-springs).Firstly,the factors affecting photovoltaic power,such as solar radiation,temperature,humidity and wind speed,are analyzed.In order to improve the forecasting efficiency,the historical data affecting photovoltaic power generation are normalized.The value of K in K-means algorithm is selected by hierarchical clustering method,and the original data is clustered by K-means algorithm.Appropriate similar days are selected as training samples for sunny,cloudy and rainy days respectively.Secondly,aiming at the problem that BP neural network is easy to fall into local optimum,this thesis establishes the BP neural network(PSO-BP)prediction model based on particle swarm optimization and the least squares support vector machine(LSSVM)prediction model.The simulation results show that the prediction effect is good.Finally,in order to solve the problem that the prediction accuracy of PSO-BP model and LSSVM model is not high in cloudy and rainy days,the ultra-short-term prediction model of photovoltaic power generation is established by using RandomForest algorithm to improve the prediction accuracy.On the basis of the RandomForest model,based on the existing LSSVM model of PSO-BP model,the correlation matrix of the two single models is determined by using Grey Relational Analysis,and it is regarded as a human being.The training objective of ANN is to obtain the optimal weight matrix of a single model and establish two combined forecasting models.Mean absolute percentage error and root mean square error are used to test the prediction effect of the model.The simulation results and error analysis show that the combined forecasting model based on randomforest and LSSVM has higher prediction accuracy.