Research On Combination Probability Prediction Methods For Photovoltaic Power Generation

As a renewable energy source,solar energy has been rapidly developed due to its cleanliness and huge content.However,since photovoltaic(PV)power generation is greatly affected by various weather factors such as solar radiation,PV power generation has strong intermittent and random nature,and large-scale PV power generation systems are connected to the power grid,which has a great impact on the stability and safety of the power grid.Therefore,accurate and reliable PV power generation prediction can provide reliable data information to staff,which is conducive to flexible management of PV power generation and improvement of grid operation stability and safety.At present,most of the PV power prediction research is about point prediction.It is difficult to express the uncertainty of the prediction result by the determined point prediction value,and the probability prediction can make up for that.Therefore,in this paper,the research on the prediction of PV output is mainly based on the probabilistic prediction research method.Because the PV output is affected by various meteorological factors,there is a certain correlation between them.In order to reduce the computational complexity of the model,this paper uses the feature selection function in the random forest algorithm to select features.The selected features is used as input variables for the predictive model.In different weather types,the PV output differs greatly.In this paper,the fuzzy C-means clustering method is used to divide the data samples according to the weather type to obtain data samples with similar weather types.In this paper,three methods for predicting PV output probability are established respectively.The improved sparse Gaussian process regression method,the neural network quantile regression method and the improved least squares support vector machine error method.In the improved sparse Gaussian process regression model,the introduced sparse Gaussian model greatly reduces the computational complexity of the Gaussian process regression model.The improved gray wolf optimization algorithm is used to obtain the optimal hyperparameters.In the neural network quantile regression model,the PV output at different quantile points is predicted.The kernel density estimation method is introduced to obtain the probability density function of PV power.In the improved least squares support vector machine error model,the improved gray wolf optimization algorithm is used to optimize the least squares support vector machine.The parameter estimation and non-parametric estimation method are used to estimate the probability distribution of the error.The three prediction models are verified by examples.The simulation results show that the three prediction models have different prediction characteristics.In order to give full play to the different prediction advantage of different models,this paper proposes a combined prediction method with time-varying variable weights.The forecast time of one day is divided into four time segments,and each time segment adopts the optimal combination method for three kinds.The predictive models are combined to obtain a combined model with the best predictive performance.The simulation results show that the combined model has the best prediction performance compared with the three prediction models,which can provide more comprehensive prediction information for the grid and improve the stability and safety of the grid operation.