基于无人机遥感成像及分布式数据协作的光伏发电预测理论研究

Improving the forecast accuracy of renewable power generation is one important way to decrease renewable energy curtailments and increase the integration of renewable energy into power systems. Cloud moving and shedding has the greatest influence on the forecast accuracy of solar power generation, leading to its output’s randomness. Under very strict regulation to solar power forecast accuracy from the government, the penalty to solar power stations has increased quite a lot in recent years, which has great impacts on the development of renewable energy. Traditional forecasting tools of solar power generation mainly rely on historic measurements and numeric weather predictions, which did not consider the impacts from cloud moving and shedding. So, such approaches can not accurately predict the sudden change of solar power output. To solve this challenge, this research first introduces the Unmanned Aerial Vehicle (UAV) remote sensing image system. This imaging system will achieve the real-time monitor to the cloud moving and shedding, and then help us analyze whether clouds would have impacts on solar power output. Compared with the traditional total sky imager, UAV remote sensing image system has great advantages of high flexibility, fast deployment and low price. Main research contents of this project include: 1) very short-term forecasting method for solar power generation using UAV remote sensing image systems, 2) very short-term forecasting method for solar power generation considering the collaboration of distributed data from multiple stations, and 3) regional forecasting method of aggregated solar power output considering spatio- temporal correlations. This research would boost the integration of disciplines between UAV remote sensing image and energy forecasting technologies. It would also provide supports to increase the integration of renewable energy into power systems.

提高新能源预测精度是减少新能源弃量、提高电网消纳能力最有效的手段之一。光伏发电具有随机性，尤以云团遮挡对其预测精度影响最大。这导致在政府部门日益严格的预测考核要求下光伏电站被罚严重，也不利于进一步提高新能源消纳。传统的光伏发电预测方法主要借助历史数据和数值天气预报，并未考虑云团遮挡这一重要影响因素，因此无法对发电功率瞬时波动做出准确预测。为解决上述难题，本项目将创造性引入无人机遥感成像系统，对云团遮挡进行实时监控，判断光伏发电是否受到云团遮挡影响。与传统的地基测云仪器相比，无人机遥感成像系统具有机动、快速、经济等优势。本项目的主要研究内容包括：基于无人机遥感影像技术的光伏发电功率超短期预测理论、考虑多站点分布式数据协作的光伏发电功率超短期预测理论、计及分布式光伏时空相关性的区域光伏发电功率预测理论。本项目将促进无人机遥感成像技术与电力预测理论的学科交叉融合，为提高新能源消纳提供技术支撑。

近年来随着整县光伏政策的逐步推进，光伏发电装机规模迅速增长，传统方法的预测精度已经无法满足电网安全稳定和经济运行要求，其中云层阴影及其移动对光伏发电功率预测影响最大。为了突破光伏发电预测精度的技术瓶颈，本项目引入无人机遥感成像系统，针对新型云层图像采集技术、数字化图像处理技术、云层移动建模分析、多站点分布式数据协作等技术难题，开展了持续三年的深入研究。. 项目执行期内开展的主要研究内容包括：第一、基于Boosting集成框架的新能源发电功率异常值检测方法；第二、基于无人机遥感成像的光伏发电超短期预测方法；第三、基于随机梯度提升树XGBoost的短期光伏发电确定性预测方法；第四、基于重尾分布和深度神经网络的光伏发电概率预测方法；第五、基于分位数插值和深度自回归循环网络的超短期光伏出力概率预测；第六、基于二次样条分位数函数的光伏发电概率预测方法。. 项目执行期内取得的重要研究结果包括：第一、研制了基于无人机遥感影像技术的光伏发电功率超短期预测系统；第二、系统性研究了光伏预测问题，实现了超短期预测和短期预测、确定性预测和概率性预测；第三、通过参加国内权威预测竞赛（例如国网AI创新大赛-赛道3新能源发电功率预测），在全国多个实际光伏电站上验证了本项目所提方法的有效性。. 截止2023年1月，依托本项目取得的研究成果包括：获得陕西省自然科学二等奖1项，发表期刊论文8篇（其中SCI论文4篇，EI论文4篇），申请国家发明专利4项（其中已授权2项），获得软件著作权2项，培养博士研究生1名、硕士研究生2名，作特邀学术报告4次。. 本项目的主要科学意义总结如下：第一、获取了云团遮挡造成的阴影图像信息，分析了连续图像上的阴影移动情况，有效预报了太阳辐射短时波动；第二、提高了光伏发电预测精度，为提高光伏发电经济效益提供了有效保障，对保证系统的安全稳定、促进电网的优化运行具有重要意义；第三、促进无人机遥感成像技术与电力预测理论的学科交叉融合，为提高新能源消纳提供技术支撑。