Formation of structural model of objects of power systems on the basis of artificial neural networks

Abstract

The article presents an approach to optimizing the processes of the technical re-equipment and reconstruction (TRR) of power grid facilities using artificial neural networks (ANN). A two-level decision-making structure is proposed, including the tactical level (selection of preferred alternatives for individual facilities) and the strategic level (prioritization for the regional power system or RPS). Mathematical models of private criteria have been developed to evaluate alternatives: economic costs, technical efficiency, reliability, socio-ecological consequences, and equipment unification. The concept of “nomenclature unification criterion” was introduced, which allows clustering of solutions based on the typing of equipment parameters. Training of neural networks was carried out using the algorithm of error back propagation, which provided high accuracy of classification of alternatives (F-measure up to 0.979). Experiments showed the effectiveness of the Levenberg-Marquardt method for training networks with three hidden layers. The integration of ANN into automated decision support systems (ADSS) allows minimizing subjective factors, increasing the adaptability of power systems to changing conditions, and optimizing resource allocation. Results demonstrate the ability to reduce risks and improve infrastructure reliability by processing noisy and incomplete data, as well as real-time analysis. The approach applies to large-scale power systems, balancing short-term and long-term modernization goals.