准确的超短期光伏功率预测是可再生能源大规模并网、实现碳达峰和碳中和的关键技术。本文以提高超短期时间尺度下光伏系统的功率预测精度为目标，从光伏系统出力的特性出发，在未来天气模式判断的基础上，分别针对小波动天气和大波动天气建立了结合卫星云图和集群相邻电站多数据源的光伏超短期功率预测方法。论文的主要成果如下：（1）提出了一个基于少量参数和历史功率的光伏电站晴空理想出力计算模型。该模型不需要获取光伏电站详细参数信息，对于已运行电站有着良好的适应性，模型结果可以作为光伏电站功率去除确定性分量进行平稳化的基准值，也可以代替相似日的结果用于其他光伏功率预测的方法中。基于该晴空模型提出了一个针对小波动天气的在线更新的超短期功率预测方法，在已知为小波动天气的前提下，理论上可将第4小时尺度下的光伏预测误差降低到3.78%。（2）提出了基于卫星云图运动矢量分析的光伏电站天气类型判断方法。该方法基于最大类间方差法对云图进行了云检测，基于块匹配法计算了云图中云团区域的位移矢量，进而通过卫星云图与光伏电站之间距离的统计信息，判别了光伏电站在未来4小时是否为晴天。在此基础上，得到了基于出力模式判断光伏功率超短期预测方法，使得小波动天气下的光伏功率预测精度得到了提升。（3）提出了基于卫星云图、数值天气预报和集群相邻电站历史数据的光伏电站超短期预测方法。该方法基于XGBoost模型从数据驱动的角度筛选了与目标电站功率强相关的相邻电站功率数据、数值天气预报数据的特征量，进一步基于晴空辐照度模型对数值天气预报中的短波辐射进行了归一化，去除了季节分量，提高了短波辐射与功率之间的相关性；基于此，训练了LSTM模型建立相邻电站和数值天气预报与目标电站功率的映射关系；最后，通过卫星云图数据得到了目标电站的相关电站，选择与之对应的超短期光伏功率预测模型，有效提高了大波动天气下光伏功率的预测精度。本文深入研究了晴空功率模型，提出了基于卫星云图的晴天功率判断方法，利用卫星云图和时空相关性对大波动天气下光伏超短期功率进行预测，有效地提高了超短期时间尺度下光伏功率预测精度。
Accurate ultra-short-term PV power prediction is a key technology for large-scale grid integration of renewable energy and achieving carbon peaking and carbon neutrality. This paper aims at improving the power prediction accuracy of PV system in ultra-short-term time scale, based on the characteristics of PV system power output and the judgment of future weather patterns, we established PV ultra-short-term power prediction methods combining satellite image and adjacent power plants with multiple data sources for small fluctuation weather and large fluctuation weather respectively. The main results of the paper are as follows.(1) A clear-sky power model for PV power plants based on a small number of parameters and historical power is proposed. The model does not need to obtain detailed parameter information of PV power plants and has good adaptability for already operating power plants. The model can be used as a benchmark value for smoothing the power of PV power plants by removing deterministic components and can also be used in other PV power prediction methods instead of the results of similar days. Based on this clear-sky model, an online updated ultra-short-term power prediction method for small fluctuating weather is proposed, which can reduce the PV prediction error to 3.78% at the 4th hour scale under the premise of known small fluctuating weather.(2) A method for determining the weather type of PV power plants based on the motion vector analysis of satellite images is proposed. OTSU is used for cloud detection, and the displacement vector of the cloud mass region in the satellite images is calculated based on the block matching method, and then the statistical information of the distance between the satellite cloud map and the PV plant is used to discriminate whether the PV plant is sunny in the next 4 hours. Based on this, an ultra-short-term prediction method based on the PV power Weather Classification is obtained, which makes the accuracy of PV power prediction under small fluctuation weather effectively improved.(3) An ultra-short-term prediction method for PV power plants based on satellite images, numerical weather prediction and historical data from adjacent power plants is proposed. From a data-driven perspective, the features of neighboring power plant power data and numerical weather forecast data that are strongly correlated with the power of the target power plant from a data-driven perspective, The correlation between short-wave radiation and power is improved by normalizing the short-wave radiation in numerical weather forecasts based on the clear-sky irradiance model, which removes the seasonal component. based on this, the LSTM model is trained to establish mapping relationship between power from adjacent power plants, numerical weather prediction and the power of the target power station. Finally, the relevant power stations of the target power station are obtained from satellite images, and the corresponding ultra-short-term PV power prediction model is selected, which effectively improves the prediction accuracy of PV power under large fluctuation weather.In this paper, the clear-sky power model is studied, and a clear-sky power judgment method based on satellite images is proposed. Using satellite cloud maps and spatio-temporal correlation to predict PV ultra-short-term power under large fluctuating weather, the prediction accuracy of PV power under ultra-short-term time scale is effectively improved.
