塔式光聚热发电地基云图超短期辐照功率预测系统研究

发布时间：2019-06-22 14:49

【摘要】：在光聚热发电中,云层作为影响辐照量的根本原因之一,其生消与移动都会对电力输出稳定性造成影响。因此,对镜场范围内云层运动趋势的预测是实现辐照功率预测的关键。对于中长期的辐照量变化,已可通过蓄能装置进行相应调节。而辐照量的突变却会给系统带来较大干扰,为消除这一影响,光聚热发电系统由吸热器温度调控系统通过调节入口低温流体流速对吸热器温度损失进行补偿。但是此调控系统存在分钟级纯滞后,当云层发生对太阳遮挡行为时,需要一定的反应时间,而这段时间内吸热器出口高温流体温度稳定性无法得到保证。因此,本次研究的预测系统可为吸热器温度调控系统提供一个前馈信号,由超短期预测结果克服调控系统存在的纯滞后(分钟级),当遮挡发生时经过一个小的波动后,吸热器出口流体温度可快速恢复稳定。不同于基于历史气象数据与卫星云图的中长期预测,本次研究是基于地基云图。通过跟踪拍摄系统完成以太阳为中心的目标图像的获取,在此基础上结合计算机视觉技术对云层进行观测与分析。具有实时性好、准确率高的特点,能够满足分钟级超短期辐照功率预测的要求。本文分别对系统初始化、镜头畸变校正、云层检测、云层匹配、云层预测五个方面进行了系统的研究与验证。本文具体的研究工作包括:(1)利用跟踪支架、CCD相机、广角镜头构成跟踪拍摄系统,获取以太阳为中心的云图,由镜头中心遮光片对太阳进行遮挡。由RS485通讯协议实现了跟踪支架与PC端的数据交互,通过调节相机曝光时间与信号增益实现曝光自动调节;(2)确定了一种桶形畸变的校正模型,由多项式地址修正结合同心圆模板的校正方法,完成了广角镜头桶形失真校正,取得了明显效果;(3)提出一种先聚类再分类的云检测思想,以此思想提出颜色特征结合K-Means聚类的云层检测算法,并对检测结果进行了指标评价。相比于灰度阈值分割方法,其在检测效果上有极大的提升;(4)利用SIFT算法与误匹配消除方法获得了准确的匹配结果,在此基础上提出一种针对于间隔1分钟以上的大时间跨度匹配方法,可获得数量更多的匹配点。由云层分片跟踪方法对所有面积有效云层分别统计匹配点与轮廓点信息,充分体现云层运动状态的独特性;(5)根据分片跟踪结果提出一种分级式预测模型结合粒子滤波的云预测方法,可对4分钟以内的云层移动进行准确预测。通过所预测到的云量信息,来对应太阳辐照量以及辐照功率的变化。
[Abstract]:In optical thermal power generation, cloud layer is one of the fundamental reasons that affect the irradiance, and its generation, elimination and movement will have an impact on the stability of power output. Therefore, the prediction of cloud motion trend in the range of mirror field is the key to realize the prediction of irradiation power. For medium and long term irradiance variation, it can be adjusted by energy storage device. However, the sudden change of irradiance will bring great interference to the system. In order to eliminate this influence, the temperature loss of the endothermic power generation system is compensated by adjusting the inlet low temperature fluid flow rate by the endothermic temperature control system. However, there is a minute stage pure lag in this control system, and a certain reaction time is needed when the clouds block the sun, and the temperature stability of the high temperature fluid at the outlet of the absorber can not be guaranteed during this period. Therefore, the prediction system can provide a feedforward signal for the temperature control system of the absorber, and overcome the pure lag (minute stage) of the control system from the ultra-short term prediction results. After a small fluctuation when the occlusion occurs, the fluid temperature at the outlet of the absorber can be quickly restored to stability. Different from the medium and long term prediction based on historical meteorological data and satellite cloud images, this study is based on foundation cloud images. The sun-centered target image is obtained by tracking and shooting system, and the cloud layer is observed and analyzed by computer vision technology. It has the characteristics of good real-time and high accuracy, and can meet the requirements of minute level ultra-short-term irradiation power prediction. In this paper, five aspects of system initialization, lens distortion correction, cloud layer detection, cloud layer matching and cloud layer prediction are systematically studied and verified. The specific research work of this paper includes: (1) using tracking bracket, CCD camera and wide-angle lens to form a tracking shooting system, the sun-centered cloud image is obtained, and the sun is obscured by the lens center shade. The data interaction between the tracking bracket and the PC end is realized by RS485 communication protocol, and the exposure automatic adjustment is realized by adjusting the exposure time and signal gain of the camera. (2) A correction model of barrel distortion is determined, and the correction method of bucket distortion of wide angle lens is completed by the correction method of multinomial address correction combined with concentric circle template, and obvious results are obtained. (3) an idea of cloud detection based on clustering and then classification is proposed, and a cloud layer detection algorithm based on color feature and K-Means clustering is proposed, and the detection results are evaluated. Compared with the gray threshold segmentation method, its detection effect is greatly improved. (4) the accurate matching results are obtained by using SIFT algorithm and error matching elimination method. On this basis, a large number of matching points can be obtained for the long time span matching method with an interval of more than one minute. According to the cloud slice tracking method, the matching point and outline point information of all areas of effective cloud layer are counted respectively, which fully reflects the uniqueness of cloud motion state. (5) according to the piecewise tracking results, a hierarchical prediction model combined with particle filter cloud prediction method is proposed, which can accurately predict the cloud movement within 4 minutes. The predicted cloud amount information is used to correspond to the variation of solar irradiance and irradiation power.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM615

【参考文献】