Application of SVM algorithm based on thulium doped fiber ring system in ammonia quantitative analysis

Abstract

According to the difficulty in selecting the parameter of SVM when modeling on the gas quantitative analysis, and existing methods need a long time, SVM optimized by improved grid search method is proposed to build an NH3 gas quantitative analysis model of ring cavity system based on thulium doped fiber laser. Firstly, the initial spectral data of NH3 are preprocessed and dimensionality reduced by principal component analysis (PCA). Then, the optimized cost factor C and RBF kernel function parameter g are obtained by the improved grid search method, which is substituted into a support vector machine for concentration regression analysis. Experiments show that the optimal SVM parameters obtained by the improved grid search method are C = 1.6245 and g = 2.2191. And it greatly reduces the optimization time compared with the traditional grid search method. The mean square error of the prediction results is within 10%, R2 is 0.984, which basically meets the requirements of gas concentration prediction.