A Method Of Probabilistic Distribution Estimation Of PV Generation Based On Data Mining

As the energy crisis and environment pollution become increasingly serious,countries are committed to developing new energy and expanding renewable energy applications.Compared with fossile energy,solar energy has the advantage of clean,widely distributed and abundant,and increasingly attract attention of people.The technology of photovoltaic power generation get rapid development.Due to the strong randomness and intermittent,photovoltaic power generation is easy to cause impact to power grid.The reliable forecast of photovoltaic power is especially important to ensure the safety and stability of the power system.Now many prediction methods use all the attributes of the historical data as the input variables of model,and each different input variables are treated equally,which cause poorer effect of similar data filtering.Existing prediction methods focus on point prediction to obtain determined value which doesn't provide any risk information.Considering the above questions,this paper proposes a method based on Adaptive weights algorithm(AW),k-Nearest Neighbor(KNN),K-means algorithm,quantile regression neural network(QRNN)and kernel density estimator(KDE)to estimate probabilistic distribution of PV generation.Firstly,adaptive weights algorithm based on particle swarm optimization(PSO)is used to dynamically determine the weight of different input variables.Then K-means is employed to divide historical data according to weather types based on appropriate similarity metrics.Finally,the model of quantile regression neural network is established for different weather types.The results of experiments show that,probability prediction model in this paper,which supports dynamic changes in the numbers of input variables and the corresponding weights,improves the prediction accuracy and provides more information on the risk of prediction results for the power grid planning.