【摘要】 针对传统光缆故障分析与识别过程中，OTDR信号受噪声干扰和人工经验依赖导致事件识别准确率低、耗时长的问题，提出一种基于ICEEMDAN-SVD与IBES-TWSVM相结合的OTDR事件识别算法。首先，利用ICEEMDAN将OTDR信号分解为多个IMF，并通过样本熵区分IMFs中的有用分量和噪声分量；接着，保留有用分量，对噪声分量使用SVD分解以提取其中残留的有用信号；然后，重构两次处理后的有用信号以获得降噪后的OTDR信号；最后，采用改进的BES算法对TWSVM惩罚因子和核函数参数进行优化，利用特征数据集进行训练和测试，最终实现OTDR事件识别。实验结果表明，当训练样本为730条，测试样本为312条时，该算法对光缆线路中的事件识别率约为95.52%，识别时间为7.12 s，所提算法在事件识别准确率与识别速度方面均优于传统的OTDR事件识别算法。更多还原

【Abstract】 In response to the low event recognition accuracy rate and lengthy processing time in the traditional optical cable fault analysis and identification due to the facts that the OTDR(optical time domain reflectometer) signals are interfered by noise and dependent on artificial experience, an OTDR event recognition algorithm based on the combination of ICEEMDAN-SVD and IBES-TWSVM is proposed. Firstly, ICEEMDAN(improved complete ensemble empirical mode decomposition with adaptive noise) is utilized to decompose the OTDR signals into multiple IMFs(intrinsic mode functions), and the sample entropy is employed to distinguish between useful and noise components within the IMFs. Next, the useful components are retained, while the noise components are subjected to SVD(singular value decomposition) to extract any residual useful signals. Subsequently, the useful signals from both processings are reconstructed to obtain denoised OTDR signals. Finally, an improved BES(bald eagle search) algorithm is used to optimize the penalty factor and kernel function parameters of TWSVM(twin support vector machine). The optimized TWSVM is then trained and tested by a feature dataset to achieve OTDR event recognition. Experimental results demonstrate that when the number of the training samples is 730 and that of the test samples is 312, the algorithm achieves an event recognition rate of approximately 95.52% in optical cable lines, with a recognition duration of 7.12 seconds. The proposed algorithm outperforms the traditional OTDR event recognition algorithms in terms of both event recognition accuracy rate and recognition speed.更多还原