针对隧道工程监测现场应用布里渊光时域反射仪（BOTDR）系统存在的问题，提出基于双树复小波变换（DT-CWT）和改进的LMS算法的组合降噪模型，用于对BOTDR分布式光纤监测信号进行降噪处理。首先，基于双树复小波变换算法分解原始信号； 然后，使用样本熵作为目标函数自动选择最优小波分解层数的模型，并基于优化双曲余弦函数改进LMS算法的收敛速度和收敛性； 最后，为验证所提出算法的有效性，进行BOTDR温度信号降噪试验。试验结果表明： 1）DT-CWT-LMS算法的降噪效果明显优于传统的小波阈值降噪方法，在6个温度梯度上的平均SNR值比WDD、EMT、EMD分别高出43.98%、17.5%、8.4%，平均RMSE值分别降低33.18%、17.14%、9.23%，平均SE值分别降低29.04%、21.17%、20.67%。为验证所提算法在工程现场的有效性，依托北京地铁隧道监测项目，使用DT-CWT-LMS算法对光纤监测信号进行降噪研究，降噪后信号的样本熵平均降低幅为64.03%； 2）和常规WDD、EMT、EMD 3种方法相比，DT-CWT-LMS算法在3条光纤上的表现均优于其他3种算法，在1号、2号、3号光纤上的SNR指标比其他3种算法的平均值分别高出22%、38%、27%，说明该算法可作为隧道工程光纤监测信号的一种有效降噪方法。

Herein, a combined denoising model based on the dualtree complex wavelet transform (DTCWT) and an improved least mean square(LMS) algorithm is proposed to denoise monitoring signals collected by distributed fiber optic of Brillouin optical timedomain reflectometer(BOTDR) systems in tunnel monitoring sites. First, the original signals are decomposed using the DTCWT algorithm. Second, the sample entropy(SE) is used as the objective function to automatically select the model with the optimal wavelet decomposition level. The LMS algorithm’s convergence and convergence speed are improved based on the optimized hyperbolic cosine function. Finally, BOTDR temperature signal denoising experiments are conducted to validate the effectiveness of the proposed algorithm. The results show the following: (1) The denoising effect of the DTCWTLMS algorithm is significantly superior to that of the traditional wavelet threshold denoising method. The average values of signaltonoise ratio(SNR) for the six temperature gradients is 43.98%, 17.5%, and 8.4% higher; the average root mean square error is 33.18%, 17.14%, and 9.23% lower; and the average SE value is 29.04%, 21.17%, and 20.67% lower than those for wavelet domain denoising(WDD), empirical wavelet transform(EMT), and empirical mode decomposition(EMD), respectively. The DTCWTLMS algorithm is applied to the denoising of fiber optic monitoring signals in a metro tunnel monitoring project in Beijing, China, and the average decrease in the SE of the denoised signal is 64.03%, validating the feasibility of the proposed algorithm. (2) Compared with the conventional WDD, EMT, and EMD methods, the DTCWTLMS algorithm performs better on the three fibers. The SNR indices on fibers Nos. 1, 2, and 3 are 22%, 38%, and 27% higher than the average values of the other three algorithms, respectively, demonstrating that the proposed algorithm is an effective denoising method for fiber optic monitoring signals in tunnel engineering.