基于机器视觉的区域太阳直接辐射动态预测方法研究  

RESEARCH ON REGIONAL SOLAR DIRECT NORMAL IRRADIANCE DYNAMIC FORECASTING BASED ON MACHINE VISION

作　　者：仝勖峰[1] 金嘉祺 袁晓军[3] 陈梦迟 黄文君[3] Tong Xufeng;Jin Jiaqi;Yuan Xiaojun;Chen Mengchi;Huang Wenjun(School of Mechano-Electronic Engineering,Xidian University,Xi’an 710126,China;School of Electronic Engineering,Xidian University,Xi’an 710126,China;College of Control Science and Engineering,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]西安电子科技大学机电工程学院,西安710126 [2]西安电子科技大学电子工程学院,西安710126 [3]浙江大学控制科学与工程学院,杭州310027

出　　处：《太阳能学报》2021年第6期247-255,共9页Acta Energiae Solaris Sinica

基　　金：国家重点研究计划(2018YFB1702200)。

摘　　要：提出一种应用于塔式太阳能热发电站的以全天空动态图像为基础的多维度特征训练融合回归模型处理方法,实现局部区域的太阳直接辐射(DNI)预测。为解决遮挡太阳云团特征难以提取、遮挡情况预测不准确等问题,方案通过阈值分割进行太阳位置检测,使用随机森林检测云团,利用模板匹配测算云速,并从上述信息中提取多维度特征训练融合回归模型实现DNI预测。测试结果表明方案的短期预测误差较小,可准确地预测出DNI的变化趋势,具有较高的工程应用价值。A method of regional solar direct normal irradiance(DNI)dynamic forecasting for concentrated solar power is proposed,which is based on all-sky image processing and fusion regression model trained by multidimensional features. In order to overcome the difficulty in extracting the cloud feature sheltering from the sunshine and inaccurate forecasting,threshold segmentation is used to locate the sun,and the cloud cluster is detected by the random forest. Cloud speed is calculated by the template matching. The fusion regression model is trained to predict DNI whose input feature is the various information extracted from sky images. The experimental results show that the error of short-term DNI forecasting is relatively small,and the trend of DNI change can be predicted accurately.Therefore,the solution is with great engineering application value.

关 键 词：太阳能热发电 太阳辐射 预测 机器视觉 位置检测 特征提取 

分 类 号：TK514[动力工程及工程热物理—热能工程]