Temperature extraction for Brillouin optical fiber sensing system based on extreme learning machine

Abstract

The use of extreme learning machine (ELM) network to extract temperature distribution from the measured Brillouin gain spectra (BGSs) along the sensing fiber obtained by Brillouin optical fiber sensors is proposed and demonstrated experimentally. Compared with conventional curve fitting method (CFM), ELM network trained by a set of ideal BGSs can extract temperature information directly from the measured BGSs obtained by Brillouin optical time domain reflectometer (BOTDR) system without the need of determining Brillouin frequency shift (BFS) and converting BFS to temperature. The BGSs linewidth is taken into account to construct the ideal BGSs by using Pseudo-Voigt curve for ELM training. The performance of ELM is analyzed in detail and compared with that of widely-used Lorentzian CFM, and the experiment results show that ELM can provide higher accuracy even at large frequency scanning step and faster processing speed. Therefore, the proposed ELM approach is feasible and effective for temperature extraction in Brillouin optical fiber sensing system.