An improved lorentz fitting algorithm for BOTDR using SVM model to capture the main peak of power cumulative average data.

• The proposed method improves SR and measurement accuracy simultaneously. • SVM model is adopted to capture the main peak of power cumulative average data. • Noise is linearly fitted and removed from the main peak data to improve accuracy. • The Brillouin frequency shift of Double-peak Spectrum is precisely determined. • The precision of edge detection in the transition section is improved to 0.1-m. The Brillouin scattering signal is weak and the signal-to-noise ratio (SNR) is low in Brillouin optical time domain reflectometer (BOTDR), these result in a difficulty to improve the spatial resolution (SR) and measurement accuracy at the same time. In this paper, the denoising principle of power cumulative average (PCA) is analyzed theoretically, then an improved Lorentzian-curve fitting (ILCF) algorithm based on PCA data is proposed, in which the support vector machine (SVM) model is adopted to capture the main peak, and the Brillouin frequency shift (BFS) of double-peak spectrum is determined precisely. Furtherly, by adjusting the step of the sliding window, the precision of edge detection in the transition section is improved to 0.1-m. For the fiber under test (FUT) with a length of 12-km, 500 groups of time-domain data are processed under the probe pulse with 50 ns width and 1 GHz sampling rate. The average error of temperature accesses to 0.8 ° C and the SR achieves 6.4-m simultaneously. The algorithm in this paper can reduce BFS standard deviation from 28.21 MHz to 1.27 MHz, improve the SR to the maximum limitation by the probe pulse, determine the BFS of double-peak spectrum and reduce the measurement time. Therefore, the proposed method can improve SR and measurement accuracy at the same time. [ABSTRACT FROM AUTHOR]