| Power distribution systems have been significantly affected by a wide range of fault-causing events; and the current outage restoration procedure may take from tens of minutes to hours. Effective outage cause identification can help to expedite the outage restoration and consequently improve the system reliability. Most current researches are based on system modeling and measurements such as voltage and current; besides, they usually target at a single feeder or a small system due to the difficulty of modeling the large-scale, nonlinear, and time-varying distribution system. In this research, various data mining approaches including statistical methods and artificial intelligence algorithms have been investigated and applied to Duke Energy distribution outage data in order to extract the outage pattern and identify the outage cause; by this means, the additional environmental information recorded in the data can be adopted in the fault diagnosis and the analysis range can be beyond the scope of a single feeder or a small system. Also, the affect of data imperfections such as data noise, data insufficiency, especially the data imbalance issue on the performance of outage cause identification have been investigated.; In this work, logistic regression and artificial neural network are firstly compared on their capability in fault diagnosis; then an existing fuzzy classification algorithm is extended to E-algorithm to alleviate the effect of data imbalance; afterwards, the immune system based Artificial Immune Recognition System (AIRS) algorithm is investigated for its capability in fault diagnosis using real-world data; lastly, a hybrid algorithm based on E-algorithm and AIRS is proposed to embed the rule extraction capability while performing satisfactory fault cause identification. |
