Depth wavelet extreme learning machine-based multi-level inverter fault diagnosis method

Abstract

The invention discloses a depth wavelet extreme learning machine-based multi-level inverter fault diagnosis method. The method comprises steps: the inverter AC-side voltage fault signals under variousworking states are acquired, acquired three-phase AC voltage is intercepted and combined to a piece of data, a total sample data set is obtained, the total sample data set is normalized, a training set, a verification set and a test set are generated, a deep feature extraction and fault diagnosis model is constructed the test set is used as to-be-tested samples, and the deep feature extraction and fault diagnosis model is used to diagnose a modular multi-level inverter fault. A wavelet extreme learning machine self-encoder formed by a wavelet base, an extreme learning machine and a self-encoder has the advantages of good feature extraction effects and quick training speed, and a sparse coefficient is added to the wavelet extreme learning machine self-encoder realization process, and the depth wavelet extreme learning machine has good anti-noise performance.

Description

technical field [0001] The invention belongs to the technical field of modular multilevel converters, and in particular relates to a multilevel inverter fault diagnosis method based on a deep wavelet extreme learning machine. Background technique [0002] Modular Multilevel Converter (MMC) has the characteristics of high voltage and large capacity, low switching current stress, and low harmonic distortion rate. MMC has become the preferred converter topology in flexible direct current transmission systems. However, the MMC structure usually includes multiple sub-modules, many switching devices, and is in a fast switching state for a long time. Compared with the two-level inverter, the failure rate is significantly increased. After the MMC switch device has an open circuit fault, the output three-phase voltage is unbalanced, which will seriously affect the system. There is a problem of high similarity between faults of different modules of MMC, and the degree of discriminati...

AI Technical Summary

Problems solved by technology

At present, the method of feature extraction mainly adopts signal processing (such as FFT, wavelet) and statistical analysis (such as PCA, ICA). These methods have the disadvantages of large manual intervention and low degree of automation.

In 2006, Professor Hilton proposed deep learning, which can adaptively extract deep feature information from massive data by simulating the learning process of the human brain, which overcomes the shortcomings of traditional feature extraction methods of manual intervention, but deep learning has training speed compared with general methods slow problem

Images