Fault diagnosis model of deep belief networks based on adaptive genetic algorithm optimization

Abstract

Aiming at the problem that the application of deep neural network in fault diagnosis is limited due to its long running time and lack of theoretical basis for initial parameter setting, this paper proposes a fault diagnosis model combining Adaptive Genetic Algorithm (AGA) and Deep Belief Networks (DBN). By defining the error function and initial parameters of DBN as the fitness function and initial chromosome population of AGA, respectively, using the global optimization ability of AGA, iteratively obtains the initial parameters that minimize the DBN error function and brings them into Training in DBN. The AGADBN fault diagnosis model proposed in this paper is verified by the MNIST data set. Compared with DBN and kNN, the convergence speed and recognition accuracy are significantly improved, which is of positive significance for improving the application of deep neural network in fault diagnosis field.