The Forecasting Approach Of PV Power Based On Weather Conditions Classification

Photovoltaic(PV)power forecasting is of major significance in regards to the grid-connected safety and economic operation of PV plants.However,there are also many problems:The PV output power is closely related to the weather conditions,which are affected by many meteorological parameters.Problems such as numerous weather types,complex weather conditions and limited weather type classification methods make such PV power forecasting a highly challenging endeavor.The PV output power fluctuates more intensely under changing weather conditions and the forecasting error is more obvious.And a single forecasting model is difficult to guarantee forecasting accuracy.At present,PV power forecasting at home and abroad mainly focuses on deterministic point forecasting.Such PV power forecasting can only give a single value within a certain forecasting period and can not describe the reliability of the corresponding forecasting value.Therefore,due to the lack of power error information,the PV point forecasting model is limited in application.To solve above problems,through summarizing the overall development status of PV power forecasting,a novel interval forecasting method for PV power based on generalized weather conditions is proposed in this paper.Firstly,the characteristics of periodicity,intermittentness and volatility of PV output power are analyzed.What's more,the relationship between PV output power and various meteorological factors is analyzed by the path analysis method,and the key meteorological factors of PV power modeling is further determined.Then,the uncertainty of PV output power under different weather conditions is analyzed,and a generalized weather classification method is performed based on solar irradiance reduction index K.Next,a PV power multi-model is established under different generalized weather types and the extreme learning machine(ELM)method is applied in multiple models to minimize the quantity of samples.The confidence interval for forecasting power at a certain confidence level is determined by a kernel density estimation algorithm.Finally,comparative experiments reveal the effectiveness of the proposed forecasting method in terms of model performance,training time and interval accuracy.The proposed method can give the forecasting power and the probability of its occurrence within the forecasting time period,and provide the confidence interval under a certain confidence level,whose information is more comprehensive.