Short-term Probabilistic Prediction Of Distributed Photovoltaics Considering Temporal And Spatial Characteristics

Solar power generation is a renewable and pollution-free method,which has developed rapidly in recent years.But photovoltaic energy is typically intermittent,its output power largely depends on weather factors,with strong random volatility.Therefore,the theory of probability and method should be used to take full account of the uncertainty of photovoltaic production.The topic of this work is forecasting of short-term photovoltaic power generation.It intends to use ARIMA model,LSTM model and GeoMAN model based on different analysis structures to analyze and forecast the photovoltaic power output data of a certain region.The greatest advantage of the ARIMA model is its ability to accurately predict the seasonal cycle data indicators,while revealing the nonlinear characteristics of the explained variables.LSTM has "forget" and "update" functions,which can solve the problem of long time series dependence and significantly improve the prediction speed;In addition,Grey correlation analysis,Pearson coefficient and Spearman coefficient were used for spatial correlation analysis,and then GeoMAN model was improved by stepwise regression method.Finally,the photovoltaic power plant in Jiangsu,China,is used for practical analysis.The results show that compared to the traditional ARIMA model,the short-term probability prediction of photovoltaic power generation obtained by GeoMAN model has higher accuracy.