SVM Modeling On Slowly Time Varying System And Online Correction Studying

The time varying widely exists in many fields such as process control, avigation and spaceflight, and fault diagnosing. As the extensiveness and complexity of time varying systems, the modeling on the time varying system is always a hotspot. Traditional modeling algorithms base on the hypothesis of time invariant system or stationary process. But in fact systems always present the characteristics of fluid processes and time varying, especially the slowly time varying parameter system, so there are limitations in traditional algorithms. Therefore for eliminating above limitations, the modeling method aiming at slowly time varying parameter system is intensively studied.Firstly, the characteristics of the time varying systems and the traditional modeling algorithm are introduced berifly. Basing on the above algorithms and their shortages, the intelligent modeling algorithm of support vector machine is intensively studied, and denoising quality of the SVM is vertified. Secondly, two incremental SVM algorithms are researched, which are ZD-SVM and AO-SVM. Mater the advantages and disadvantages in both ZD-SVM and AO-SVM. An improved AO-SVM modeling algorithm is proposed. The algorithm in this paper solves the three problems in AO-SVM:(1) matrix calculations are ulitized to eliminate the problem that support vector set is empty; (2) alternative FIFO is used to substitute the traditional data deletion strategy; (3) mistaken data is recalculated to aim at the problem of unstable algorithm. Thirdly, there exist some problems in the general continuous online modeling strategy, such as wasting resource and erroneous learning. Online correction is applied to solve these problems. Monitor the process characteristics by PCA monitoring method and then model by online SVM. At last wavelet denoising is used to to solve the lost of the process monitoring ablity which is caused by PCA offline modeling on noisy datas. And finally a new online modeling method:PCA-SVM algorithm which models on noisy slowly time varying systems is conduct. And the algorithm is proved effective by simulation.