Anomalous Modes Recognizing Secondary Equipment in Electric Power Industry: Adaptive Neuro Algorithms

Secondary equipment of electric power industry and its information networks require protection from disruptions in information exchange, network hacking and virus attacks. Modern security systems used for this purpose may not be effective enough and require continuous improvement. Detection of anomalies in information flows during the operation of secondary equipment of the power facility (automated control systems and relay protection) is becoming an increasingly important task. The identification of such anomalies can be based on statistical outliers analysis, but the use of data mining (DM) techniques to detect “novelty” in the data structure and system parameters can be a much more effective tool. It is proposed to use the apparatus of artificial neural networks (ANN) as one of the key tools of DM and, specifically, recurrent neural network with LSTM (Long short-term memory) architecture. To detect anomalies in the information support of secondary equipment of the power facility, training of ANN is con-ducted on the maximum possible number of normal modes. At the same time, emergency modes with adequate relay protection operation are also considered as normal. It is shown that the data for ANN training can be obtained on a laboratory bench consisting of a digital network model and a mathematical model of the power unit of the power facility.