Power Forecasting Research Of Large Grid-connected Photovoltaic Station

Energy is an important material basis for economic and social development.With the global economic growing and the world's population increasing,the mode of traditional energy development based on fossil fuels has become unsustainable.The proportion of clean energy.will greatly enhance by the implementation of strong smart-grid and global energy internet in the future,large-scale grid-connected to grid dispatching management has brought enormous challenges,as PV system output has obvious intermittent,random fluctuation.How to make full use of clean energy and to get the most out of become a hot topic of the power system for the premise of meeting the security and stability constraints.So the PV plant designed and their output predicted effectively has important practical significance.This paper analyses the grid-connected photovoltaic power station system principle,system composition and operation mode,and the photovoltaic system key equipment selection,monitoring systems,power prediction and control system was designed.The main influence factor such as weather type index,the solar irradiance,temperature and humidity was identified on the basis on the research and analysis the influence factors and the output characteristics of photovoltaic power station output.The Gaussian Process Regression algorithm was applied to the PV power short-term forecast,and the feasibility of the method is validated,the results show that the prediction effect of this method is better than wavelet neural network and least squares support vector machine.This paper puts forward a classification model based on the thought and principle of similar day combined with particle swarm improvement Gaussian Process Regression forecast method for the photovoltaic power short-term prediction.The method is to set up three kinds of database by weather type index firstly,and apply similar day principle to weather type index,the daily average temperature,daily average humidity as predict when selecting reference index of the similar day,according to different types of forecast,selects the corresponding database,and select 6 days similar day,adopts the combination of covariance function,weighted linear decreasing particle swarm algorithm(LinWPSO)instead of the conjugate gradient algorithm to otimize parameters of super Gaussian Process Regression model,meanwhile,three types of Gaussian Process Regression prediction model was set up Respectively.in the matlab for sense of Shaanxi and Gansu actual grid photovoltaic power station in different areas of the two example analysis,the results show that the proposed classification method and LinW-PSO-GPR combination forecast modeling ideas to improve the photovoltaic power output response time and the prediction precision,high universality and application of the method.