【摘要】 针对省级电网新能源发电概率预测中存在的场站对象多、特征变量维数高及区域气象因素影响显著等难点，提出一种全省区域新能源功率非参数概率预测方法。首先，构建基于气象相似区聚类划分的风电和光伏集群，将各集群气象信息加权聚合为全省区域预测模型的可用气象因子；其次，利用斯皮尔曼及最大信息系数多准则筛选出与新能源出力最相关的气象因子，作为模型输入；最后，构建基于时间卷积神经网络（TCN）-分位数回归（QR）-核密度估计（KDE）的非参数概率预测模型。以西南某省区域风电和光伏总出力概率预测为应用实例，对比验证所提方法的有效性。更多还原

【Abstract】 The probability prediction problem of regional new energy power generation in the province faces difficulties such as multiple power station objects, high dimensional characteristic variables, and influence of regional meteorological factors. Therefore, this paper proposes a non-parametric probabilistic prediction method for regional new energy power in the province. Firstly, the historical meteorological factors of each station are used as input data, and the improved clustering method is used to construct wind power and photovoltaic clusters based on meteorologically similar areas, and then the meteorological information of each cluster is weighted and aggregated into available meteorological factors for the provincial regional prediction model. Based on this, the meteorological factors most related to new energy output are screened using Spearman and maximum information coefficient as the final input factors of the model to improve the model training efficiency. Finally, a nonparametric probabilistic prediction model based on time convolutional neural network(TCN)-quantile regression(QR)-kernel density estimation(KDE) is constructed. Taking all the wind power and photovoltaic plants in a southwestern province as application examples, the effectiveness of the proposed method is verified.更多还原