Single-channel distributed Raman temperature sensing based on a 1-dimensional convolutional neural network

Abstract

We present a simple and cost-efficient single-channel Raman distributed temperature sensing (DTS) system based on temperature prediction by a 1-dimensional convolutional neural network (1D-CNN) from the Raman anti-Stokes backscatter trace. The proposed Raman DTS system is based on incoherent optical frequency domain reflectometry with homodyne down-conversion with excitation of spontaneous Raman backscattering by an L-band laser diode and detection of the Raman anti-Stokes in the optical C-band. A 1D-CNN is employed to predict the spatially resolved temperature profile along the fiber from the obtained anti-Stokes backscatter trace only and thus, solves the problem of temperature referencing for single-channel Raman DTS systems. The network was trained on three different scenarios, consisting of uniform and non-uniform temperature profiles along the fiber in a temperature range from 0 °C to 60 °C. The obtained results show that the measurement and signal processing pipeline presented here is capable of predicting the temperature distribution to an accuracy of approximately 1 K in the tested scenarios.