中文摘要：

为全面挖掘扰动事件的内在关系、分析影响扰动源关联特性的外界因素，从扰动源的特征提取、时空关联分析、关联异常辨识方面开展研究，通过数据分析的方法验证了基本扰动事件之间的时空关联性。为克服传统数据清洗方法重复检出率低、计算成本高的缺点，提出自适应近邻排序数据清洗方法，并且实现了扰动源特征信息的快速提取。在此基础上，利用随机矩阵理论从宏观角度分析扰动源的时空关联特性。针对可能存在的局部扰动源关联异常，提出一种基于相关矩阵信息增益的动态辨识方法。最后，依据实际电能质量监测系统记录的暂态事件数据，验证了该方法的可行性和有效性。研究结果表明，所提出的数据清洗方法适用于大数据分析框架，扰动源之间的关联程度与电网的运行方式和天气状况有关。信息增益方法利用移动窗口和动态阈值技术有效跟踪电网运行状态变化，相较于PCA方法具备更高的异常事件识别率。

英文摘要：

In the complex power system, the temporal-spatial correlation of disturbance sources may be detected by power quality monitoring system. To explore the inherent relationship among disturbance events and analyze outside influential factors, focusing on disturbance feature extraction, temporal-spatial correlation, correlation anomaly detection, we veri- fied the temporal-spatial correlation of power disturbance events based on data analysis methods. In order to overcome shortcomings of traditional data cleaning methods, e.g., low detection rate and high computational cost, we proposed an adaptive sorted neighborhood data cleaning method, thus the feature of disturbance event was extracted quickly. Thereby, the random matrix theory was used to analyze temporal-spatial correlation of disturbance sources in a long term. A dy- namic detection method based on the information increment of correlation matrix was proposed for the possible disturbance source correlation anomaly in a short term. Finally, the field test data were taken as an example to verify the feasibility and effectiveness of the proposed methods. The results of this research illustrate that the proposed data cleaning method is applicable for big data analysis framework. The degree of correlation between disturbance sources is related to network operation mode and weather conditions. The moving-window based and dynamic-threshold based techniques make the information increment method track the variation of power grid operation state and improve the abnormal event recognition rate, compared with the PCA method.