A power electronic circuit fault diagnosis method based on an optimized deep belief network

Abstract

The invention discloses a power electronic circuit fault diagnosis method based on an optimized deep belief network. The method comprises the following steps: (1) using an RT-LAB semi-physical simulation platform to set a fault expierment, and acquiring direct current side bus output voltage signals under different fault modes to serve as original fault characteristic quantities; (2) extracting anintrinsic mode function component and an envelope spectrum thereof of the output voltage signal by utilizing empirical mode decomposition, calculating a plurality of statistical characteristics, andconstructing an original fault characteristic set; (3) removing redundancy and interference features in the original fault feature set based on a feature selection method of an extreme learning machine, and performing normalization processing to serve as a fault sensitive feature set; (4) dividing the fault sensitive feature set into a training sample and a test sample, and preliminarily determining the structure of the deep belief network; (5) adopting a doodle search algorithm to optimize the deep belief network, and setting the number of hidden neurons of the network; And (5) obtaining a fault diagnosis result. According to the invention, the fault feature data size and the fault identification accuracy are improved.

Description

technical field [0001] The invention belongs to the field of fault diagnosis of power electronic circuits, and in particular relates to a fault diagnosis method of power electronic circuits based on an optimized deep belief network. Background technique [0002] The reliability of power electronic converter conversion system is of great significance in energy, transportation, communication, industrial manufacturing, aerospace, environmental protection and national defense military applications. [0003] The fault diagnosis of power electronic converter conversion system is mainly to monitor and diagnose the power electronic devices in its main circuit. At present, power electronics fault diagnosis methods are mainly divided into three categories: analytical model fault diagnosis method, signal recognition method and fault diagnosis method based on knowledge fusion. The analytical model fault diagnosis method needs to accurately establish the fault model of the circuit to be...

AI Technical Summary

Problems solved by technology

The analytical model fault diagnosis method needs to accurately establish the fault model of the circuit to be diagnosed; the commonly used processing methods in the signal recognition method include Fourier transform method, Park transform method and wavelet transform method, but these methods may lead to effective signal processing during signal processing. The loss of fault information, when there are many fault types, the fault feature quantity selected after transformation cannot effectively distinguish different fault types; the fault diagnosis method based on knowledge fusion, such as the artificial neural network method, but the usual BP neural network recognition method is used in training. It is easy to fall into the local optimal solution, and has great blindness in the selection of initial connection weights and thresholds, and it is difficult to adjust parameters and select fault feature vectors

Images