Prediction Of PV Power Output Based On Ensemble Neural Network And Bayesian Model

Due to environmental pollutions and climate change,the application of renewable energy has been rapidly growing in recent years and photovoltaic(PV)is one of the most widespread renewable technology.PV output is characterized by significant intermittency,which arises from a variety of meteorological factors.Therefore,the increasing penetration of PV creates severe challenges to the stability of the power grid.Therefore,it is very important to conduct research on the prediction of PV power output.Prediction of PV output has been a popular topic of research.A variety of methods,ranging from physical models to artificial neural networks,have been applied to this topic and every method has its own advantage.In order to reduce the impacts of the uncertainty on the prediction of PV output,this paper performed predictions of the solar radiation intensity and PV module temperature.On this basis,prediction of PV power output and evaluation of its probability distribution were implemented,including three parts:(1)An integrated neural network method was developed on the basis of Elman neural network model(EANN),extreme learning machine(ELM)and Echo state model(ESN),for the prediction of solar radiation.The results show that the integrated neural network method achieves a higher prediction accuracy than an individual neural network model does and the prediction resulting from the integrated neural network method appear more concentrated distribution.Therefore,the prediction quality of the integrated method is higher than an individual one.(2)A heat transfer analysis was carried out to establish the nonlinear relationship between the temperature of PV modules and solar radiation,environmental temperature and wind speed.Next,the least square method was employed to fit the module temperature,which was validated using the measured data obtained from a terrestrial PV experimental platform.Finally,a grey model was used to optimize the obtained function,where the fixed influence coefficient was converted into a variable influence coefficient,so as to improve its accuracy.(3)A model based on the dynamic Bayesian network theory was established to evaluate,the probability distribution of PV outputs probability.Key parameters,e.g.solar radiation,module temperature and historic outputs,were included as nodes in the dynamic Bayesian network for short-term prediction.The model was solved in MATLAB.Compared with traditional determined point prediction,probability prediction provided more comprehensive information on the uncertainty of PV outputs,which is valuable for power dispatching and grid controlling.