摘要
典型运行场景提取对制定有效的日前运行策略具有重要意义。微电网中,可再生能源和新型负荷的强不确定性使得微电网的运行场景具有复杂时序特征。传统的场景聚类分析方法缺乏对时序特征的考虑,难以得到有效可信的典型运行场景。为此,文中提出一种基于深度时间聚类的微电网典型运行场景生成方法。首先,基于受路径约束的动态时间规整算法,量度时间序列的形态相似性;其次,设计了一种组合卷积神经网络和双向长短期记忆网络的时序自动编码器结构,提取复杂时序运行场景中的深层次特征并实现数据降维;然后,联合优化时序特征提取与时序聚类,得到有效、可信的典型运行场景;最后,提出考虑时间序列形态相似性的时间轮廓系数以及日内实际场景的运行成本作为聚类有效性评估指标。基于澳大利亚居民微电网的实际算例结果表明,与传统的场景聚类方法相比,所提方法具有更强的复杂时序特征挖掘能力,能够得到更具代表性的典型运行场景。
Typical operation scenario extraction is of great significance for formulating effective day-ahead operation strategies.The strong uncertainty of renewable energies and new loads in microgrids brings complex temporal characteristics to the operation of microgrids.Traditional scenario clustering analysis methods lack the consideration of temporal characteristics,which are insufficient to obtain effective and reliable typical operation scenarios.Therefore,this paper proposes a typical operation scenario generation algorithm for microgrid based on deep temporal clustering.Firstly,the path-constrained dynamic time warping algorithm is used to measure the morphological similarity of time series.Secondly,a temporal auto encoder with combined convolutional neural network and bi-directional long short-term memory network is designed to extract deep features in the complex temporal operation scenarios and realize data dimensionality reduction.Thirdly,the temporal feature extraction and temporal clustering are simultaneously optimized to obtain effective and reliable typical operation scenarios.Finally,the temporal silhouette coefficient considering the morphological similarity of time series and the operation cost of actual intra-day scenarios are proposed as the clustering validity evaluation indices.Experimental results based on an Australian residential microgrid show that the proposed method outperforms other traditional scenario clustering methods with respect to complex temporal feature extraction,and also the typical scenarios are more representative.
作者
庄颖睿
程林
齐宁
陈卫东
吴晓锐
姚知洋
ZHUANG Yingrui;CHENG Lin;QI Ning;CHEN Weidong;WU Xiaorui;YAO Zhiyang(Department of Electrical Engineering,Tsinghua University,Beijing 100084,China;Electric Power Research Institute of Guangxi Power Grid Corporation,Nanning 530023,China)
出处
《电力系统自动化》
EI
CSCD
北大核心
2023年第20期95-103,共9页
Automation of Electric Power Systems
基金
国家自然科学基金资助项目(52037006)
广西电网公司科技项目资助(GXKJXM220222038)
中国博士后科学基金资助项目(2023TQ0169)。
关键词
场景聚类
场景生成
随机优化
微电网
自动编码器
特征提取
scenario clustering
scenario generation
stochastic optimization
microgrid
auto encoder
feature extraction
