Study Of Regional PV Power Forecast Based On Data Mining

As the rapid growth of photovoltaic(PV)generation capacity,the form of regional PV power integrated by multiple PV plants is becoming more and more common.It is of great significance to control the operation of the power system to grasp the changing law of regional PV power and realize the effective forecast of regional PV power.Because there are many PV plants in the region,the input data involved in the power forecast of regional PV plants is large-scale(including the numerical weather prediction and output power of each PV plants in the region),and there is complex correlation in the input data.Therefore,the key of regional PV power forecast in how to efficiently mine effective information from high-dimensional complex input data to achieve accurate fitting function of high-dimensional input data and regional PV power.At present,the bottom-up method and upscaling method are the two commonly used methods for the regional PV power forecast.The former directly accumulates the forecast results of all PV plants in the region,while the latter uses the benchmark PV plant forecast results to scale up to the regional PV power forecast results.The forecast accuracy of regional PV power still can be improved.What's more,the regional PV power variation has strong uncertainty and is difficult to be accurately forecasted.The bottom-up and upscaling methods are both point forecast methods,which cannot provide power system operators with the uncertainty of regional PV power forecast results.This paper uses the deep learning theory to realize the deep mining of the input data for regional PV power forecast and fully extract the key effective information.This paper starts from two aspects of point forecast and probabilistic forecast,comprehensively carry out the research of regional PV power forecast method based on deep data mining.Firstly,this paper proposes a point forecast method of regional PV power based on convolutional neural network(CNN)algorithm.This method uses the powerful high-dimensional data processing and deep feature mining ability of CNN to carry out deep mining and feature extraction on the input data of regional PV power forecast In addition,this method can effectively improve the accuracy of regional PV power forecast.In order to effectively measure the uncertainty information of the regional PV power forecast results,this paper further combines the CNN algorithm with the nonlinear quantile regression(QR)model,and proposes a probabilistic forecast of regional PV power based on the convolutional neural network quantile regression(CNNQR)model.By improving the structure of CNN,this method improves the ability of feature extraction and data mining,so as to extract the complex features contained in the input data.Combined with the improved CNN,the CNNQR method can deal with high-dimensional and complex input data more directly and effectively,generate the nonlinear QR function and provide quantile forecast results of regional PV power.The forecast of regional PV power in a real power grid in Weifang,Shandong Province is carried out to illustrate the validity of the proposed method in this paper.Test results show that the proposed forecast methods can provide more accurate point and quantile forecast results than the state-of-the-art methods.