Design of an anomaly detection scheme for power distribution

Abstract During the daily power operation, anomaly detection of the faults within the substation is an important work. One of the detection is related to load forecasting which directly affects the economy, safety, power supply quality and distribution planning. It is also one of the significant contents to realize the modernization of power system management. With the prevalence of big data, informatization, and intelligence, how to make better use of these massive data generated during the operation of the distribution network, how to judge the status information of the data, and tailor a suitable operation and maintenance model for the distribution station at the same time need to solve as an urgent problem. Under such a situation, research of electrical state monitoring technology into sharp focus. This dissertation speaking from the needs and problems of the distribution anomaly detection, through a more in-depth analysis of the operation status and SCADA data characteristics of the distribution station, based on the long and short-term memory (LSTM) network model, to carry out research on the load prediction of the substation. Firstly, we analyze the operating principle of the distribution station and the types and causes of failures, determine the SCADA data as the data source for building the model, and summarize its characteristics to lay the foundation for subsequent research methods. Secondly, take the distribution station load data as the research object, and study the distribution station load forecasting method based on long and short-term memory (LSTM) network. Aiming at the correlation of SCADA data, construct a LSTMnetwork model and select appropriate input data to establish an unsupervised model. The significance of this model is to emphasize the relevance of data in the time dimension on the one hand, and to construct the model through unsupervised means on the other hand in order to improve the accuracy of prediction through multiple input parameters. Finally, the model is analyzed and compared with four different models with different indicators to illustrate the correctness and effectiveness of the model. Keywords: anomaly detection, SCADA, LSTM, load forecastingMaster of Science (Communications Engineering