Summary Composite generation and transmission system reliability evaluation allows the assessment of the risks of system operation failure, taking into account the uncertainties associated with the availability of equipment. One of the great challenges faced in the use of techniques based on probabilistic assessment during the planning stages is related to the required computational costs. Depending on the reliability levels of the system under study and on the grid size, a large number of operation performance analyzes are necessary. In this sense, this article proposes a new and simple method to efficiently evaluate the composite reliability of electrical power networks. The nonsequential Monte Carlo simulation (MCS) method is combined with unsupervised machine learning (UML) techniques to reduce the computational effort involved in the process of estimating composite reliability indices. The proposed approach allows different unsupervised techniques to be employed, in order to obtain significant reductions in CPU times, without losing the accuracy of the estimated indices. The IEEE‐RTS system, considering the original load and generation and its modified version with the transmission network stressed, in addition to a real large system, is used for evaluating the performance of the proposed method. The results obtained with the use of three different classification techniques (Kohonen self‐organizing map, K‐means, and K‐medoids) are presented and analyzed.

Neste artigo é proposta uma metodologia simples e nova para avaliação da confiabilidade composta de sistemas elétricos de potência. O método de simulação Monte Carlo não sequencial é combinado com técnicas não supervisionadas de aprendizado de máquina com o intuito de reduzir o esforço computacional envolvido no processo de estimativa dos índices de confiabilidade composta. A metodologia permite que diferentes técnicas não supervisionadas sejam empregadas, tendo em vista a obtenção de reduções significativas nos tempos de processamento, sem que haja perda de precisão dos índices de desempenho estimados. Os resultados obtidos com a utilização de três diferentes técnicas de classificação (Kohonen self-organizing map, K-means, and K-medoids) são apresentados e amplamente analisados.