分布式光伏电站功率的精准预测对于解决其出力不确定性至关重要。分布式光伏具有数量众多且地理位置分布较为分散的特点，若对每个分布式光伏电站进行功率预测系统配置，将会带来很高的运行成本，为此提出一种基于多源预报动态聚类的分布式光伏集群短期功率预测方法。首先，将预测日的当地公共天气预报信息进行数字编码，并将编码信息与本地区数值天气预报(numerical weather prediction，NWP)数据通过改进自编码器进行特征提取，实现多源预报数据融合；其次，以预测日的多源预报数据融合后的特征作为聚类特征，利用自组织映射(self-organizing mapping，SOM)网络聚类来实现集群的动态划分；最后，通过1维卷积神经网络(1D convolutional neural network，1DCNN)进行集群预测，并将集群预测结果累加实现区域分布式光伏的功率预测。结果表明，所提方法可以得到较为精确的预测精度和更可靠的预测效果。

Accurate power forecasting for distributed photovoltaic (PV) power plants is essential to address output uncertainty. Distributed PV is characterized by a large number and geographical distribution， if a power prediction system is configured for each distributed PV plant， it will bring high operating costs. For this reason， a short-term power prediction method for distributed PV clusters based on dynamic clustering of multi-source forecasts is proposed. Firstly， the local public weather forecast information of the forecast day is digitally encoded， and the encoded information is fused with the numerical weather prediction (NWP) data of the region through an improved self-encoder for feature extraction to achieve the fusion of multi-source forecast data； Secondly， the fused features of the multi-source forecast data of the forecast day are taken as the clustering features， and self-organizing mapping (SOM) network clustering is utilized to realize the dynamic division of the clusters； Finally， the clusters are predicted by the 1D convolutional neural network (1DCNN)， and the cluster prediction results are accumulated to achieve the power prediction of regional distributed photovoltaic. The results show that the proposed method can obtain more accurate and reliable prediction.