Explaining critical clearing time with the rules extracted from a multilayer perceptron artificial neural network

In the past two decades, artificial neural networks (ANN) have been applied to quickly compute the critical clearing time (CCT), a frequently quoted measurement for power systems transient stability. This kind of applications mainly concerns the CCT prediction rather than the explanation because ANN was commonly considered as a black box. This paper will challenge this myth. In this paper, we describe the procedures for explaining CCT by means of a multilayer perceptron (MLP) artificial neural network. The explanation is expressed in terms of “IF antecedent THEN consequent” rules, where the antecedent indicates the power system operating conditions and the consequent refers to whether the CCT is high or low. We can accordingly explain CCT, and in turn we can observe under what circumstances will cause the power system CCT to be high or low. To justify the proposed method, the CCTs of two contingencies in 39-bus power systems are investigated. The results have demonstrated that the CCT can be explained by MLP very well.