Study on Magneto-Hydro-Dynamics Disturbance Signal Feature Classification Using Improved S-Transform Algorithm and Radial Basis Function Neural Network

• Main Authors: Nan YU, Jiarong LUO
• Format: Article
• Language: English
• Published: IFSA Publishing, S.L. 2014-09-01
• Series: Sensors & Transducers
• Subjects: S-Transform algorithm; Magneto-hydro-dynamics (MHD); Radial basis function neural network; Short time Fourier transform; Wavelet transform.
• Online Access: http://www.sensorsportal.com/HTML/DIGEST/september_2014/Vol_178/P_2381.pdf
• ISSN: 2306-8515; 1726-5479

The interference signal in magneto-hydro-dynamics (MHD) may be the disturbance from the power supply, the equipment itself, or the electromagnetic radiation. Interference signal mixed in normal signal, brings difficulties for signal analysis and processing. Recently proposed S-Transform algorithm combines advantages of short time Fourier transform and wavelet transform. It uses Fourier kernel and wavelet like Gauss window whose width is inversely proportional to the frequency. Therefore, S-Transform algorithm not only preserves the phase information of the signals but also has variable resolution like wavelet transform. This paper proposes a new method to establish a MHD signal classifier using S-transform algorithm and radial basis function neural network (RBFNN). Because RBFNN centers ascertained by k-means clustering algorithm probably are the local optimum, this paper analyzes the characteristics of k-means clustering algorithm and proposes an improved k-means clustering algorithm called GCW (Group-cluster-weight) k-means clustering algorithm to improve the centers distribution. The experiment results show that the improvement greatly enhances the RBFNN performance.