Optimization of probabilistic operational characteristics of a neural network algorithm for predictive diagnostics of industrial equipment failures through automated processing of a training sample

The paper considers the optimization of probabilistic operational characteristics of a neural network algorithm for predictive diagnostics of industrial equipment malfunctions. The probability of a false alarm and the probability of detecting a malfunction are accepted as optimized parameters, which are generally accepted metrics for the effectiveness of situation analysis and decision-making systems. The task of optimizing operational characteristics is decomposed to the level of influencing technical parameters of the system and reduced to finding the optimal values of the base calculation time of the derivatives of the measured technological parameters and the time of fault development. Automated processing of the training sample allows you to reduce the time spent on creating a system for predictive diagnostics of industrial equipment failures. The result of optimization of the probabilistic operational characteristics of the neural network algorithm is presented and optimal values of the variable parameters are obtained, as well as a result of training and testing on real telemetry data of the electric power pump of the turbine unit of the CHP, optimal values of the optimized parameters are obtained, according to which conclusions are drawn and further actions are proposed to improve the result. .