Temperature prediction of Bragg grating sensing based on a one-dimensional convolutional neural network

Abstract

In this paper, we propose a new (to us) way of demodulating the grating sensing spectrum using a one-dimensional convolutional neural network (1DCNN) to overcome the limitation of the traditional fitting method of temperature demodulation for subway tunnel fires. This method constructs a one-dimensional convolutional neural network model and combines it with the experimental device of a fiber Bragg grating (FBG) temperature measurement. One thousand eight hundred spectra of experimental data are selected as sample data for training. Adam’s random optimization algorithm is used in training to predict the temperature of multiple periods, with an accuracy of 99.95% and a root-mean-square deviation (RMSE) of 0.0832°C. The experiment shows that the algorithm in this paper is better than the GRU and LSTM algorithms, as traditional maximum peak methods, and can effectively improve the measurement accuracy. This article aims to provide a high-speed demodulation solution for FBG-based sensing systems to meet the practical needs of large-scale real-time monitoring.