Studies On Short-term Photovoltaic Power Output Prediction Method

With the serious condition of the lack of fossil energy and environm ental pollution all over the word,as a kind of environmentally friendly renewable energies,photovoltaic generation is one of the most widely used power generation methods in renewable energy industry,which has attracted more and more attention.Due to the volatility,randomness and nonlinearity of photovoltaic radiation and other environmental factors,the output of photovoltaic system changes dynamically with time.Those changes will not only affect the stability of the connected power system,but also increase the investment risk of investors.Therefore,photovoltaic power prediction becomes increasingly crucial.Through analysis of data characteristics of actual photovoltaic plant,this paper is exploring the correlation between different influencing factors and photovoltaic power prediction,and provide theoretical basis for the predictive modeling method,optimize the forecasting model and improve photovoltaic power prediction accuracy further.Firstly,according to photovoltaic plant power model,the main factors that influencing the output of photovoltaic power are studied and analyzed in this paper.And then taking the actual data in 2015 as an object of study,which is selected from the Ashland station in Oregon,USA.The correlation between PV power data and meteorological influence factors are analyzed,which will lay the foundation for the following prediction modeling.Secondly,this paper introduces the basic principles of extreme learning machine and kernel extreme learning machine.Based on that,a short-term photovoltaic power prediction model based on kernel extreme learning machine is established.Then,aiming at the parameter optimization problems of kernel extreme learning machine forecasting model,the influence of kernel parameter ? and penalty coefficient C to kernel extreme learning machine performance are analyzed.A novel short-term photovoltaic power prediction approach based on kernel extreme learning machine optimized by firefly algorithm(FA-KELM)is proposed in this paper.Through the simulation verification of a practical example,comparing the kernel extreme learning machine model,the kernel extreme learning machine model optimized by PSO,and the kernel extreme learning machine model optimized by FA,the results illustrate that firefly algorithm can effectively solve the problem of blind selection of KELM parameters,and more effectively improves the prediction accuracy of kernel extreme learning machine prediction model.Then,according to nonlinear and non-stationary characteristics of photovoltaic power sequence,in order to improve the prediction accuracy of prediction model in the period of power rapid fluctuations,a novel short-term photovoltaic power combined forecasting approach based on variational mode decomposition and sample entropy algorithm is proposed in this paper.Through the simulation verification of a practical example,the results illustrate that variational mode decomposition can effectively reduce the non-stationary of the photovoltaic power sequence,and further improve the prediction accuracy of forecasting model.Sample entropy algorithm reduces training time of forecasting model,and improves prediction efficiency.Finally,compared with various deterministic point prediction models,a short-term PV power interval prediction method is proposed.Through the simulation verification of a practical example,the results illustrate that the interval prediction can ensure that the actual value falls into the prediction interval at a certain confidence level,which can provide decision-makers with more comprehensive information and more conducive to the planning and operation of the power system.As a whole the proposed method can improve the prediction efficiency and accuracy of the forecasting model,which has a good application potential.