A distributed ensemble bpnn used for status assessment of circuit breakers based on coil current characteristics

Abstract

The switching coil of circuit breaker operating mechanism has many current characteristics and it can reflect the potential failure state. By analyzing these characteristics, the state of circuit breaker can be evaluated reasonably. However, unbalanced class data, large data volume, and poor classification efficacy are the problems that conventional classification methods must face to. This paper presents a distributed ensemble back propagation neural network (DE-BPNN) method to evaluate circuit breaker state based on coil current characteristics. Firstly, the coil current data are de-noised by wavelet packet with db5 and 9 characteristic parameters are extracted. In order to deal with imbalanced class data, the SMOTE method is adopted to ensure that training samples are the same volume. And it also provides a differential sample extraction method to segment training samples. Then, by building multiple differentiated subset BPNN, the voting strategy is used to generate the final results. Finally, DE-BPNN algorithm is deployed on the distributed large data computing platform Spark to reduce the computation time. Two classification algorithms such as K-means and standalone BPNN are used to compare classification accuracy and efficiency with DE-BPNN. The experimental results show that DE-BPNN has high classification accuracy with the unbalanced training data volume. In addition, the computation time of DE-BPNN keeps stable with the increase of data quantity.