Support vector machine based distributed optical fiber leakage detection system and experimental analysis

Leakage experiments were performed on the distributed optical fiber pipeline leakage detection system. The system was based on a mixed interference principle of Mach-Zehnder and Sagnac. 20 groups of leakage point samples were gained between 1km to 9km from the leakage experiments. Support vector machines(SVMs) were used to build the sample of data model, the related parameters of SVM were optimized , and the sensitivity of the system was analyzed. The results show that the processed samples are significantly closer to the actual leakage points than the experimental leakage points through regression analysis. The mean absolute error of experimental leakage points is 118.85m while that of the regression samples is 92.01m. The accuracy rate of positioning improves to 99.85%. The sensitivity of the detection system is lower when the leakage points are farther away from the Faraday rotation mirror. The sensitivity of the system is less than 0.5 Hz / m when the leakage points are over 10 km.