In this paper, hydrogen sensitive film of tungsten trioxide/Platinum (WO3/Pt) was prepared by sol-gel method, and the hydrogen concentration was detected based on transmission optical fiber hydrogen sensing system. Due to the hydrogen sensitive film is easily affected by temperature when it reacts with hydrogen, and water will be generated in the process and hinder the reaction. Therefore, the relationship between hydrogen concentration and the transmitted optical intensity signal received by the photodetector is nonlinear. In order to improve the accuracy of detection system, in this paper, Support Vector Regression (SVR), Back Propagation (BP) and Radial Basis Function (RBF) neural network are innovatively used to optimize the data. The effect of the three algorithms are compared by average relative error rate and coefficient of determination (R2). Compared with BP and RBF neural networks, SVR algorithm has a better prediction effect on hydrogen concentration with average relative error of 3.8% and determination coefficient (R2) of 0.9998.