Time Series (TS) analysis is a hot topic in Data Mining community. Currently, the detection of the operation state in energy system often relies only on human judgment or even on-site inspection. But TS analysis can help automatizing this task and has become attractive in the energy field. In this paper, we propose a method for detecting and recognizing system operating pattern based on change points and complex network features. We first explain how the change point detection method can be applied in different pipeline operation scenarios. The results obtained by this method can help TS to segment subsequences and extract shapelet. Then, the shapelet are transformed into the form of visibility graph. The structural features of such TS graphs corresponding to different operating patterns of the system are extracted. Finally, we validate the graph feature-based representation method on datasets from an oil pipeline system in China. We compare it with statistical features-based representation baseline method for classification tasks. The results show the interpretive and accuracy of our proposed method. The method could be a basis for intelligent detection and recognition in the field of energy systems.
Keywords time series, change point detection, visibility graph, structural features.