Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back The Statistical Characteristics Evaluation of Protection Operation Based on a Feed Forward Neural Network S.O. IVANOV, M.V. NIKANDROV, L.A. SLAVUTSKII The article is devoted to neural network modeling of relay and cybernetic protection. An algorithm based on the simplest feed forward neural network is used, which, after the ANN training, can be entered into industrial microprocessor equipment with standard characteristics. Using the example of the overcurrent protection with time delay for three-phase grids, errors and the quality of neural network training and testing are analyzed. It is shown that the maximum errors in the recognition of test data by an artificial neural network occur near the threshold of protection activation. At the same time, the characteristics of the protection operation depend significantly on the statistical uniformity of the test data. The results obtained can be used not only for relay protection, but also for cybernetic protection of industrial facilities, since they can serve for real-time "surveillance" and anomalies detection in information flows. Keywords: relay and cybernetic protection, operation characteristics, feed forward neural network. DOI: 10.25791/pribor.9.2023.1439 Pp. 26-31.