基于动态时间规整分层凝聚的光伏场景缩减方法研究

Research on the photoelectric field reduction method based on dynamic time regularized layer condensation

随着我国光伏并网比例不断增大,亟须探索能有效挖掘有典型统计特征的光伏序列的场景缩减方法。针对传统场景缩减方法存在着缩减场景特征不明显、异常值过高、时序样本度量效果不佳等问题,提出一种基于动态时间规整的光伏场景缩减方法。首先,对光伏数据集进行预处理,并引入动态时间规整对光伏样本时间序列进行相似性度量;其次,以样本凝聚分层树状图枝节垂直跨度最大为准则进行最佳连通策略的选择;然后,采用凝聚分层聚类的方法对光伏场景数据集进行场景缩减;最后,利用澳大利亚沙漠太阳能中心(DKASC)屋顶太阳能阵列公开数据集,通过算例验证了所提方法的评价指标优于传统缩减方法,而且具有更优的尾部事件提取性能。此外,进行了敏感度分析,由此选取了最佳的连通策略和时间窗口参数。

With the increasing proportion of photovoltaic power connection,it is urgent to explore scene reduction methods that can effectively mine photovoltaic sequences with typical statistical features.To address the problems of traditional scenario reduction methods such as unclear reduction of scenario characteristics,high abnormal values,and poor performance of time series samples a photovoltaic scenario reduction method based on dynamic time warping that balances the complexity and coverage of scenario reduction is proposed.Firstly,the photovoltaic dataset is preprocessed,and then the dynamic time warping method is introduced to measure the similarity between the time series of photovoltaic samples.Furthermore,the best connectivity strategy is selected based on the criterion of maximum vertical span of dendrogram branches,and the photovoltaic scenario dataset is reduced using an agglomerative hierarchical clustering method.Finally,the proposed method is verified by case studies using the DKASC rooftop solar array public dataset,which not only shows better performance evaluation indicators than traditional reduction methods,but also exhibits superior tail event extraction performance.Additionally,sensitivity analysis is conducted to select the optimal connectivity strategy and time window parameters.