Research On Power Forecasting And Economic Operation For Photovoltaic Microgrid

With the development of society and economy,shortage of fossil resources and environmental pollution become more and more prominent.The utilization of renewable energy offers a promising solution for these problems.And microgrid provides distributed renewable energy with a friendly access to the inter-connected power grid,and has aroused widespread academical and industrial interests.In this paper,research on the forecasting techniques for photovoltaic power generation and economic operating of photovoltaic(PV)microgrid is carried out,which mainly includes:1)The characteristics of photovoltaic microgrid are analyzed,including the microgrid structure and its components.The power generation mechanism.and output characteristics of PV power generation system provide the basis for research on PV prediction techniques in the case of a small quantity of training samples.The mathematical model of PV microgrid components provides the basis for research on optimal operation of PV microgrid.2)The prediction of PV power generation is studied.Faced with the deficiency of historical operational data for newly built PV stations,a neural network based PV prediction method with loose requirement on sample size is proposed.According to the decoupling of each factor in the process of PV power generation,an improved two-layer neural network model is used to forecast the output power in the future one day,which realizes the effective training of the neural network based prediction model with small sample size.Based on realistic data,the proposed PV forecasting method is validated.3)The economic operation of PV microgrid is studied.By combining the features of ultra-short-term and short-term prediction of PV power generation,a mixed-time-scale multi-period optimization framework with rolling horizon is proposed for optimal dispatching of microgrid.In the optimization model of day-ahead dispatching layer,the temporal granularity of ultra-short-term and short-term prediction data can be effectively utilized by the proposed mix-time-scale technique,which is able to provide a reliable reference for real-time dispatching layer and hence ensures economic and secure operation.Meanwhile,the proposed mixed-time-scale multi-period optimization method reduces the number of decision variables of optimization model in day-ahead dispatching layer,and hence reduces time consumption for solving this optimization problem.