Probabilistic training procedure for neural network based power system diagnostic software

This paper outlines a procedure which uses a power system description language to train a neural network to generate conditional probabilities for device failures. The proposed method minimizes the need for the use of explicit conditional statements.