服务于区域光伏预测的天空图像K-means云空辨识模型

Cloud Identification Model Based on K-means for Regional Photovoltaic Forecasting

地基天空图像的云空辨识及云团预测是研究区域电网光伏发电功率分布与变化的前提,对支撑调度提高光伏发电消纳比例具有重要意义。首先分别采用较高的红蓝分量比值和较低的红蓝分量比值作为固定阈值分割地基天空图像,依次提取辨识结果中的天空像素点和云像素点的位置信息并获取原图像中对应位置的天空像素点和云像素点的RGB值;其次对获得的天空像素和云像素求均值并将各自均值中的红蓝分量相除获取初始聚类中心;然后使用K-means算法,利用加权欧式距离计算每一个聚类样本与聚类中心之间的距离,通过数次迭代得到聚类结果,进而将聚类结果还原成矩阵得到地基天空图像的云空辨识结果图;最后利用云南某光伏电站全天空成像仪TSI-VIS-J1006采集的天空图像进行仿真,结果表明该方法较固定阈值法的收敛速度更快、聚类精度更高,能够有效实现地基天空图像的云空辨识。

Cloud space identification of ground sky image and cloud prediction are the premises of studying the distribution and change of photovoltaic power in regional power grid. They are important for supporting scheduling and increasing the proportion of consumptive PV power generation. Firstly,the high and the low ratio of red and blue components were adopted respectively as fixed thresholds to segment ground sky images. We sequentially extracted the location of the sky and the cloud pixels in the identification results and got the RGB value of the corresponding cloud and the sky pixels in the original image. And then we calculate the average of the sky and cloud pixels that were obtained and make mean value of the red and blue components divided to get the initial cluster center. Then basing on the Kmeans algorithm and Weighted Euclidean distance,we calculate the distance between each cluster sample and cluster center. We get the clustering results by several iterations. Then the clustering results are reverted to a matrix so as to obtain the cloud space identification results of ground sky images. Finally,we use the sky images collected by the total sky imager TSI-VIS-J1006 of a PV power station in Yunnan to carry out simulation. And the results show that comparing with the fixed threshold method,this method is faster and its clustering accuracy is higher,and it can effectively realize the cloud space identification of the ground sky image.