| Support vector machine (SVM) has been one hotspot of machine learning. SVM is based on statistical learning theory. Because of having advantages such as global optimization and good generalization, Support vector machine has achieved a serial of successes in pattern recognition, regression, etc.In this thesis, improvements on SVM and applications are studied. The main content is as follows:(1) To overcome the effects of noise and outliers, the weighted proximal SVM (WPSVM) is presented. Experiments prove that the WPSVM has good performance.(2) The online weighted proximal SVM is proposed for function regression, time sequence prediction and dynamic system identification. The results of experiments show that the online weighted proximal SVM can reach success.(3) The DNA genetic algorithm with varible population size is used to optimize the parameters of SVM and WPSVM. The experiment results of system identification and inverse model control show this method is effective. |
