Abstract This study describes the deployment of machine learning algorithm called generalized linear model (GLM) to improve the temperature prediction performance in Brillouin optical time domain analysis (BOTDA) fiber sensor for distributed temperature sensing application. In GLM, the temperature prediction is made from the Brillouin gain spectrum (BGS) and the link function, without the need to determine the Brillouin frequency shift (BFS). In this proof-of-concept experiment, the performance of GLM was investigated by collecting the BGS and comparing it to the conventional Lorentzian curve fitting (LCF) method. From the experimental results, we have found that the GLM method produced a more consistent temperature prediction than the conventional LCF method. Furthermore, the proposed GLM method could still retain an accurate temperature measurement regardless of low signal-to-noise ratio (SNR) and large frequency scanning step while collecting BGS, which is difficult to be achieved by the conventional LCF method at certain level. In addition to that, the prediction obtained is 655 times faster than the conventional LCF method. The small and negligible deterioration to the temperature resolution confirmed the robustness of GLM in performing fast and accurate temperature measurement for BOTDA.

中文翻译：

---

提高布里渊光时域分析光纤传感器测温性能的广义线性模型

摘要 本研究描述了称为广义线性模型 (GLM) 的机器学习算法的部署，以提高用于分布式温度传感应用的布里渊光时域分析 (BOTDA) 光纤传感器的温度预测性能。在 GLM 中，温度预测是根据布里渊增益谱 (BGS) 和链接函数进行的，无需确定布里渊频移 (BFS)。在这个概念验证实验中，通过收集 BGS 并将其与传统的洛伦兹曲线拟合 (LCF) 方法进行比较来研究 GLM 的性能。从实验结果来看，我们发现 GLM 方法比传统 LCF 方法产生了更一致的温度预测。此外，所提出的 GLM 方法在收集 BGS 时，无论信噪比 (SNR) 低、频率扫描步长大，仍能保持准确的温度测量，这是传统 LCF 方法在一定程度上难以实现的。除此之外，获得的预测比传统的 LCF 方法快 655 倍。温度分辨率的微小且可忽略不计的恶化证实了 GLM 在为 BOTDA 执行快速准确的温度测量方面的稳健性。