Applications Of Improved Support Vector Machines Methods In Spectroscopy Quantitative Analysis

Spectroscopy quantitative analysis is an indirect measurement technology which developed rapidly in recent years. It has been widely used in scientific research and industry. Support vector machines (SVM) is a new nonlinear modeling method which is suitable for solving small samples and high dimension modeling problems. In this thesis, the basic algorithm of support vector machines is improved from different point of view. The refined algorithm is then applied to build calbration model and detect abnormal samples in spectroscopy quantitative analysis. The main contents of this paper are as follows:1. Introduce the fundamental of spectroscopy quantitative analysis. Review the kernel methods in nonlinear modeling and the main methods for anomaly detection.2. Propose a new algorithm, which is called PLS-SVM, by integrating the existing SVM and Partial Least Square (PLS). Compared with PLS, least sqaure-SVM and knernal PLS in the spectroscopy quantitative analysis, experimental results show that PLS-SVM preserves the prediction precision of SVM while the modeling speed is greatly increased.3. Improve the existing weighed least square SVM (WLS-SVM) algorithm for iteration. This refinement improved the anomaly detection ability and the robustness of calibration model substantially.4. Use PLS-SVM and refined WLS-SVM to build the calibration model for the octane number analysis of gasoline based on near infrared spectroscopy.