Research On Power Forecasting Of Distributed Photovoltaic Power Based On Trajectory Characteristics

Due to the small installed capacity,wide spatial distribution,and the limit of cost and technology,a large proportion of small-capacity distributed photovoltaic power plants are not equipped with a complete power plant monitoring system.And it is difficult to obtain accurate numerical weather prediction for such plants.Therefore,unlike large-capacity photovoltaic power plants,such plants cannot use the forecasting model based on weather prediction information directly.In order to realize the power prediction of distrributed photovoltaic power plant as much as possible based on existing conditions,or to provide a new idea for ultra-short-term power prediction of distributed photovoltaic plant,this paper proposes a kind of ultra-short-term forecasting method of distributed photovoltaic power based on spatial correlation by making use of big-data analysis.Then the target photovoltaic power plant without prediction condition is mapped to the reference photovoltaic power plant with abundant forecasting condition,so as to complete the ultra-short-term prediction of target photovoltaic power plant.This paper mainly carried out the following specific work.(1)Matching the trajectory characteristics of distributed photovoltaic power based on hierarchical clustering algorithm.Firstly,based on the research on trajectory characteristics of distributed photovoltaic power,the related concepts and determinations of spatial correlation are proposed,and the differences and relationships between timing correlation matching and spatial correlation matching are analyzed.Then,the program of spatial correlation matching and timing correlation matching using hierarchical clustering algorithm is proposed,and its feasibility is verified by simulation.(2)Master-slave prediction method based on spatial correlation of distributed photovoltaic power.This paper studies the spatial correlation characteristics of the historical power trajectories and establishes a mathematical model to characterize the spatial correlation relationship between target photovoltaic plant and reference photovoltaic plant.The ultra-short-term power prediction of reference photovoltaic plant is taken as input of the model,then output the prediction result of target photovoltaic plant.In this paper,reference photovoltaic plant with abundant forecasting conditions can obtain the artificial neural network prediction model,which was established by training the historical sample data(including the meteorological information as input and the historical power as output.The prediction results of reference photovoltaic plant can be obtained by taking the numerical weather prediction(NWP)into the BP-ANN model.And then the prediction curve of the target photovoltaic power plant is calculated according to the correlation relationship.(3)Key factors affecting the prediction accuracy of master-slave prediction methods.The influence of Four key factors on spatial correlation matching and master-slave prediction method are studied,including dynamic spatial correlation matching and static spatial correlation matching,different reference photovoltaic power plants,clustering thresholds and matching days.It gives the basis to determine the reasonable range of key factors further.Finally,the feasibility of distributed photovoltaic power ultra-short-term prediction method based on spatial correlation is verified by several simulation examples.The prediction effect of this method is better than the prediction method based on timing correlation.It solves the problem that distributed photovoltaic power cannot use the prediction method based on meteorological information,which provides the basis for the effective regulation and control of grid-connected power of distributed photovoltaic power and has important application value.