MINIMIZAÇÃO DE MULTAS PAGAS POR CONCESSIONÁRIAS DE DISTRIBUIÇÃO ATRAVÉS DA OTIMIZAÇÃO DOS INTERVALOS DE MANUTENÇÃO

Abstract

The National Agency of Electric Energy (ANEEL) supervises distribution utilities by counting individual and collective indices associated with the continuity of electricity supply. In this context, the utilities are penalized when there are violations in the targets associated with the indices. An alternative to mitigate these penalties is to schedule maintenance activities for electrical network equipment to maximize the reliability indexes. This maximization must be carried out considering the minimization of maintenance costs to avoid wasting resources and loss of revenue. In this way, the objective of this dissertation is to optimize the maintenance frequency associated with tree trimming in overhead distribution networks by concurrently minimizing the penalties due to violations in the SAIDI (System Average Interruption Duration Index) target and the costs of tree trimming. The impact of maintenance on the monthly penalty values was estimated by combining the continuous Markovian model in transient regime of the feeder sections with three probabilistic techniques: Monte Carlo Simulation (MCS) and two Surrogate Models based on the lognormal and Pearson distributions. In addition, the penalties were estimated considering load constraints (voltage drops and circuit loading) during load transfers. The multiobjective problem associated with determination of the tree trimming intervals was solved using PSO (“Particle Swarm Optimization”) based on Pareto Theory. The proposed methodology for tree trimming optimization was tested in the RBTS (“Roy Billinton Test System”) for reliability studies in distribution networks. The tests results demonstrated that the proposed methodology obtained less losses of revenue for the distribution utility than the conventional paradigm of minimizing only the expected value of the SAIDI index. In addition, the Surrogate Models are accurate with respect to SMC, but have a considerably lower computational cost.