PV Power Generation Forecast Based On Data Mining Of Haze Parameters

With the global energy crisis and the increasingly serious environmental pollution,development and utilization of new energy sources has become an urgent problem of our times.As a renewable energy,solar energy doesn't need consume any resources of the earth and doesn't cause pollution,so the PV received widespread attention and application.However,the output of photovoltaic power generation has influenced by solar irradiance,humidity and other weather conditions.The output,has a significant variability,intermittent and uncertainty,will result in a great impact on the power grid,and trigger a series of power system security and stability issues.Constraints in meeting the premise of security and stability,how to maximize the use of solar power has become the focus of PV research.The key to solve this problem is the accurately predict of PV power generation,so the study of PV power forecast has a very important value.Traditional PV forecasting methods are mostly based on a single prediction model,without any processing of inputs.That cause the coupling between input variables,and the model exists over fitting.This paper presented a method based on stepwise selection(SS),GMM clustering and RBF neural network to predict the PV output power.First,using stepwise selection method to reduce input factors(hourly meteorological factors),so that the coupling phenomenon between variables can be reduced.Then using GMM method to clustering the samples and establish different RBF forecasting model to avoid over fitting problem in single neural network.This paper based on PV power data from the smart grid key laboratory of Tianjin University from 2012 to 2015 and relevant meteorological data for research.Established PV power forecast model for each season.The simulation results show that the prediction method using forward selection,GMM and RBF has better prediction accuracy and less input factors than general neural network prediction model.