Method Research On The Prediction Of Solar Power Output From Photovoltaic System

Fossil fuels,such as coal,oil and gas,are becoming more and more deficient.Nowadays photovoltaic(PV)technology has been rapidly developed and explored,due to the fact that the solar energy is abundant,environmental-friendly and renewable.With the development and advancement in PV technology,in addition to the wide establishment in remote areas,PV systems are also becoming popular in grid-connected applications.There are many factors that could affect the power output of a PV system,Due to the variability of solar irradiance and other environmental factors,the power output of a PV system dynamically changes with time.The variability of power output not only adversely affects the stability of the electrical system being connected but also adds more risk to the profit of PV system owners.For this reason,there is an increasing need for more accurate prediction of power output.In this paper,first of all,correlation analysis between varieties of environmental factors and solar power output is conducted to explore the valuable predictors for generating power.And a large number of methods have been proposed for forecasting the power output of photovoltaic systems,including statistical methods such as time series analysis,spatialtemporal series analysis,linear regression,and some machine learning algorithms like artificial neural network,support vector machine,classification and regression tree,random forest and gradient boosting decision tree.Every model has its instinctive advantages and disadvantages in real applications,this paper suggests a fairly simple nonlinear regression model known as multivariate adaptive regression splines(MARS),as an alternative to forecasting of solar power output.Two data sets from Macau in China and Oklahoma Mesonet in America are used for fitting and forecasting based on different models,and MARS model is proved to be more reliable and stable according to model's accuracy,intelligibility,efficiency and feasibility.Finally,an operational-friendly prediction system for solar power output is developed with four modules,including data loading,statistic analyzing,data modeling and data visualizing.Successfully it helps users have a comprehensive understanding of real data and choose the suitable method for data modeling.