地基云图中云团的识别和短时外推方法研究

发布时间：2018-06-04 11:23

本文选题：图像处理 + 阈值分割　；参考：《天津大学》2016年硕士论文

【摘要】：近些年来,随着太阳能并网容量不断地增加,其带来的问题也开始引起人们的关注。其主要问题是光伏输出功率具有间歇性等特点会对电网造成冲击,因此需要在光伏发电功率预测方面进行研究。而随着地基遥感测云仪器研制成功,能很好的对光伏电站上空的天气情况进行监控,结合图像处理技术的发展,使得利用地基云图判断光伏电站是否受到云遮挡而进行光伏功率预测的方式成为可能。同时,经实践证明:基于地基云图的光伏预测方法在短时内具有良好的精度和实用效果。利用云图外推的方式进行光伏功率的预测时,其精度主要依赖于云图中云团识别和外推的准确程度。而目前的研究都主要关注于整个预测模型的建立,而忽视了云图中云团识别和外推的准确程度。因此本文利用全天空成像仪TSI采集的云图数据,首先提取云图的有效区域并对云图进行畸变校正完成云图的预处理阶段。之后利用阈值分割的方法进行云团的识别。最后在识别的基础上利用云图序列进行云团的外推。本文主要的做如下工作:(1)对原始地基云图进行预处理。提取遮光带和镜头支臂影像区域,并对其缺失的云图有效信息进行修复,还原了云图的有效区域,为后续识别和外推的工作打下了坚实的基础。(2)完成云团的畸变校正工作。通过对全天空成像仪TSI采集的云图的畸变特点进行分析,根据其特点完成了云图的畸变校正。在校正后的图像存在缺失信息的区域进行修补,还原真实天空情况。(3)根据云团识别原理对云图进行灰度化,并分析几种传统的阈值分割的方法对云图中云团识别的问题,通过对其改进,提出了基于分块插值的阈值分割方法进行云团的识别方法。(4)利用识别后的图像,结合基于最大互相关法的云团的匹配的方法分块计算云团的运动矢量,并完成云团的短时外推工作。
[Abstract]:In recent years, with the increasing of solar power grid capacity, the problems caused by solar power grid are becoming more and more important. The main problem is that the intermittent characteristics of photovoltaic output power will impact the power grid, so it is necessary to study the power prediction of photovoltaic power generation. With the successful development of ground-based remote sensing cloud measuring instruments, it can monitor the weather conditions over photovoltaic power stations well, and combine with the development of image processing technology. It is possible to predict the photovoltaic power by using the ground-based cloud map to determine whether the photovoltaic power station is blocked by the cloud. At the same time, it is proved by practice that the photovoltaic prediction method based on ground-based cloud map has good accuracy and practical effect in a short period of time. The accuracy of photovoltaic power prediction by extrapolation depends on the accuracy of cloud cluster identification and extrapolation. The present research focuses on the establishment of the whole prediction model, while neglecting the accuracy of cloud cluster identification and extrapolation in cloud images. Therefore, in this paper, the cloud image data collected by the all-sky imager TSI is used to extract the effective region of the cloud image, and the distortion correction of the cloud image is carried out to complete the pre-processing stage of the cloud image. Then the method of threshold segmentation is used for cloud cluster recognition. Finally, the cloud cluster extrapolation is carried out by using cloud image sequence on the basis of recognition. The main work of this paper is as follows: 1) preprocessing the cloud map of the original foundation. Extracting the image region of shading band and lens arm and repairing the missing effective information of cloud image, reducing the effective area of cloud image, laying a solid foundation for the subsequent work of recognition and extrapolation. 2) completing the distortion correction of cloud cluster. Based on the analysis of the distortion characteristics of the cloud image collected by the all-sky imager TSI, the distortion correction of the cloud image has been completed according to its characteristics. In the corrected image there is missing information in the region to repair, restore the real sky. 3) according to the principle of cloud recognition for the gray cloud image, and analysis of several traditional threshold segmentation of cloud image recognition problems, By improving the method, a threshold segmentation method based on block interpolation is proposed for cloud cluster recognition. (4) using the recognized image and the matching method of cloud cluster based on maximum cross-correlation method, the motion vector of cloud cluster is calculated in blocks. And the short-time extrapolation of the cloud cluster is completed.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】