Predicting the Reliability of Industrial Enterprise Power Supply System Using Artificial Intelligence

Abstract

Efficient management of the industrial electric power supply system operational reliability requires timely forecasting of the restoration time (MTTR, Тv) and the failure flow parameter (ω) for different voltage classes and equipment types. These values determine the damage from underproduction and make it possible to form the budget for capital and current repairs of electrical grid and substation equipment. Traditional statistical methods (exponential smoothing, moving average, linear regression) do not always capture the nonlinear relationships between failures, weather conditions and substation load. This paper proposes an approach based on a neural network for predicting MTTR and ω for the next year. A multilayer perceptron with eight input features and two output features, namely MTTR and ω, was used to ensure stable operation over short time series. A sliding window method was used to evaluate the forecast accuracy. The initial data include emergency outage logs for 2008 2020, information on equipment quantity and substation capacity for the considered node (Central Power Station – Steam Blowing Power Station –Substation No. 95). For each voltage class (3, 6, 10, 35, 110, 220 kV) and equipment type (transformer, cable line, overhead line, circuit breaker), restoration time distribution laws are constructed and their evolution is assessed. The developed algorithm also calculates the undelivered electrical energy for the given power node. The results are visualized in multi-page reports (PDF, Excel) containing graphs of MTTR and ω dynamics, distribution law tables and the comparison of the predicted values with the actual data. The proposed method can be used by dispatch and maintenance services for planning preventive maintenance and assessing the probable damage from emergency outages.