SCIENTIFIC ARTICLE COMPLEX FUZZY MODELS AND INFERENCE MECHANISMS IN LOCAL EXPERT SYSTEMS FOR ELECTRIC GRID MAINTENANCE

Abstract

The transition of power grids to service on the actual technical condition instead of rigid repair intervals is complicated by stochasticity of defects and many influencing factors. Classic analytical models work poorly in such conditions, so hybrid intellectual approaches combining the processing of expert knowledge and numerical data are in demand. (Research purpose) The research purpose is performing a synthesis of mathematical models and specialized algorithms of logical inference, which allow transforming data on current damage to structural elements of territorial power grids into reasonable recommendations for choosing repair strategies. (Materials and methods) The apparatus of fuzzy logic based on the maximinal algebra of L. Zadeh was used to formalize production rules in the form of disjunctive normal forms. It was shown that input variables are technological defects of subsystems of supports, conductors and substation equipment, their current values and short-term forecast are obtained through artificial neural networks with subsequent normalization by a segment [0; 1]. The output mechanism according to the modified Sugeno scheme of the first order, a single-phase circuit for homogeneous subsystems and a two-phase one for mixed ones were implemented, which takes into account the possibility of replacing a set of local repairs with one capital one. (Results and discussion) A two-phase logical inference algorithm was developed, presented in the form of a BPMN diagram; in the first phase, the need for overhaul is estimated according to a separate rule: if the threshold (δ = 0.7) is exceeded, the system gives a directive recommendation; otherwise, the second phase is launched, where the vector of relevance of local repairs is determined, the integral effect of accumulated defects is sorted and calculated. It was revealed that with a critical set of minor damages, the final indicator reaches one, which makes local repairs inexpedient. We set the threshold for selecting local events empirically. (Conclusions) We proposed a mathematical apparatus that translates high-quality rules of grid maintenance into formalized computational algorithms. It was determined that the combination of neural network forecast and fuzzy logic increases the adaptability of estimates; the two-phase mechanism allows to objectively determine the boundary beyond which the accumulation of small defects requires not local, but major repairs, which provides cost optimization.