Fault Identification Based on Stochastic Fuzzy Broad Learning System

Abstract

Compared with traditional models, Fuzzy Broad Learning System (FBLS) shows better performance in multi-classification applications. Stochastic Collocation Networks (SCNs) are suitable for problems of large-scale data because SCNs can automatically update model parameters and construct universal approximator under inequality constraint supervision mechanism. In order to overcome the shortcomings of FBLS and SCNs, fully develop their advantages and enhance the accuracy of the model, a method of algorithm fusion is proposed. We use fuzzy system to improve the stability of SCNs model and enhance its accuracy in multi-classification applications; At the same time, the stochastic algorithm can automatically adjust the model parameters to a certain extent, so that the FBLS can adapt to large-scale input. Aiming at the problem of wind turbine fault detection, the Stochastic Fuzzy Broad Learning System (SF-BLS) is used to classify and identify the working state of wind turbine planetary gearbox. Experiments show that the proposed SF-BLS model has certain advantages over FBLS and traditional SCNs model in both recognition accuracy and operation efficiency.