Research On Short-term Combination Prediction Of Photovoltaic Output Power Based On FCM And CG-DBN

PV output is so susceptible to external environment that its output power exhibits nonlinear and multi-interference characteristics.Therefore when the photovoltaic system is connected to the grid,the stability of the grid will be affected.Accurate short-term photovoltaic output power prediction method can improve the stability of photovoltaic grid-connected.However,the current short-term photovoltaic output power prediction methods are prone to fall into local optimum solution,long training time and inaccurate prediction.Therefore,it is important to study the short-term prediction of PV output power.This paper conducts detailed research and analysis on the short-term prediction model of photovoltaic output power.The main work of this paper is as follows:(1)According to the relevant theory,the simulation model of photovoltaic cell was established.The characteristics of photovoltaic output power and the relationship between its output characteristics and various elements were determined by historical data and simulation to determine the impact factor.Data is collected and data is preprocessed on the raw data to provide a basis for subsequent classification and prediction.(2)Aiming at the existing problem that the classification model can not grasp the photovoltaic output characteristics and make the combination prediction inaccurate,a combined prediction method based on fuzzy C Means(FCM)and prediction model is proposed.The FCM algorithm is used to cluster the data sets according to the similarity of the data to form different similar day matrices.Then,the corresponding predictive models are established according to the categories,and the data to be tested is put into the corresponding forecasting models for prediction.(3)Aiming at the problems of slow model construction,insufficient feature extraction and inaccurate prediction of short-term prediction model of photovoltaic output power,a deep prediction model based on deep belief network(DBN)is proposed.This method combines the DBN algorithm with the Conjugate Gradient(CG)algorithm.Firstly,the data is unsupervised layer by layer to obtain the initialvalues of each weight of the prediction model,and then the construction of the prediction model is accelerated by the CG algorithm.(4)Simulate and analyze actual historical data of a PV power plant based on the short-term forecasting method of combination of FCM and CG-DBN.Compared with BP neural network prediction model,SVM prediction model and traditional DBN prediction model,the results show that the short-term prediction method proposed in this paper has better prediction effect than other prediction methods.