基于DT⁃CWT⁃LMS算法的煤矿井筒光纤监测数据自适应降噪研究

Research on Adaptive Denoising of Coal Mine Shafts Fiber Optic Monitoring Data Based on DT-CWT-LMS Algorithm

摘要:

为了解决在煤矿立井井筒监测现场应用中布里渊光时域反射仪（BOTDR）系统信噪比低的问题，提出了一种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，设计了基于双树复小波变换分解原始信号，并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，随后，使用LMS算法计算原始信号的自适应降噪阈值，并通过优化双曲余弦函数来改进LMS算法的收敛速度和收敛性。为了验证所提出算法的有效性，进行了BOTDR温度信号降噪实验，最后，依托山东省郭屯煤矿井筒监测项目，使用DT-CWT-LMS算法对光纤监测信号进行了降噪研究。实验结果表明，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，平均SNR指标提高了32.03%，平均RMSE降低了33.2%；现场研究结果表明，经过降噪后信号的样本熵平均降低幅度为64.75%，与光纤光栅传感器的数据对比相差在5%以内， 说明信号中的背景噪声得到了有效的抑制。该研究为BOTDR技术在煤矿立井监测中应用提供了一种有效的信号降噪方法。

Abstract:

To address the issue of low signal-to-noise ratio (SNR) in the Brillouin Optical Time-Domain Reflectometer (BOTDR) system used for on-site monitoring of coal mine shafts, a combined denoising model based on the Dual-Tree Complex Wavelet Transform (DT-CWT) and an improved LMS algorithm is proposed for denoising the BOTDR distributed fiber optic monitoring signals. Firstly, a model based on DT-CWT was designed to decompose the original signal, with sample entropy used as the objective function to automatically select the optimal wavelet decomposition level. Subsequently, the LMS algorithm was used to calculate the adaptive denoising threshold of the original signal, and the convergence speed and performance of the LMS algorithm were improved by optimizing the hyperbolic cosine function. To verify the effectiveness of the proposed algorithm, a denoising experiment on BOTDR temperature signals was conducted. Finally, based on the monitoring project at the Guotun Coal Mine shaft in Shandong Province, the DT-CWT-LMS algorithm was used to study the fiber optic monitoring signals. The experimental results showed that the denoising effect of the DT CWT-LMS algorithm is significantly better than that of the traditional wavelet threshold denoising methods, with an average SNR improvement of 32.03% and an average RMSE reduction of 33.2%. The on-site research results indicated that after denoising, the average reduction in sample entropy was 64.75%, with the data difference compared to fiber optic grating sensors being within 5%, confirming that the background noise in the signal had been effectively suppressed. This study provides an effective signal denoising method for the application of BOTDR technology in coal mine shaft monitoring.