Capacitor Parameter Estimation Based on Wavelet Transform and Convolution Neural Network

doi:10.1109/TPEL.2024.3409534

Capacitor Parameter Estimation Based on Wavelet Transform and Convolution Neural Network

2024-11-01

2

PRE

AI

DOI：10.1109/TPEL.2024.3409534

Original

Original request for help

Save

Abstract

En 中文

This article proposes a capacitor parameter estimation method based on wavelet transform and convolution neural network (CNN). By fully utilizing wavelet transforms and the inherently nonideal properties of bandpass filters, the low-frequency and midfrequency band features contained in capacitor voltages are extracted with high resolution. Leveraging these features, a subsequent CNN network simultaneously estimates two crucial aging indicators of capacitors, i.e., capacitance and equivalent series resistance (ESR). While most existing methods can only identify either capacitance or ESR, the proposed method stands out by addressing both. The integration of two different frequency features enables the proposed method to exhibit broader applicability across different modulation schemes and control strategies, and is less sensitive to load conditions and sampling frequency. Experiment results based on a modular multilevel converter case study prove the effectiveness of the proposed method.

Keywords:

Capacitors

Time-frequency analysis

Feature extraction

Frequency modulation

Wavelet transforms

Transforms

Estimation

Capacitor

convolution neural network (CNN)

feature fusion

parameter estimation

pattern recognition

Capacitor Parameter Estimation Based on Wavelet Transform and Convolution Neural Network

Abstract

Journal

Organization