SVM Theory And Its Application On Coordinated Intelligent Control Of Ship Boiler-Turbine

Ship steam power equipment is one of the most important equipments of ship, when it was running, the loads varied frequently, and the varying range is big. So it required a high response performance for the system. Furthermore, the boiler-turbine control is one difficulty of the steam power equipment control. Because the characters of nonlinear, strong coupling, time-vary and big delay, routine control strategy can't satisfy the requirements of the system. Based on the analysis of the dynamic characters of the boiler-turbine system, this dissertation researched the mathematic modeling and control of the coordinated system by using support vector machine (SVM) algorithm arisen in recent years.This dissertation introduced SVM linear and nonlinear regression theories. Then the comparing research of SVM and neural network was made. The simulation results showed that SVM has better performance than neural network for identification of the nonlinear systems. The realization of SVM algorithm was studied. And the sequential minimization optimization (SMO) algorithm of solving regression problem was deduced with the help of the SMO algorithm that was put forward for classification.During the course of dynamic system on-line identification, the model needed to renew for the reason of training set data increasing along with each sampling. So the implement of SVM regression on-line training algorithm was researched. The increment SVM on-line training algorithm based on sliding time window was proposed. This method decreased the training time of SVM greatly and improved the identification precision.The nonlinear system was identified by SVM using different kinds of kernel function, and then the basic method of selecting kernel function was given here. The parameter selection of the kernel function represented by RBF function was researched as an emphasis. Based on analyzing the influence of each parameter in SVM regression algorithm, the search space of genetic algorithm was defined. With the help of the global search ability of genetic algorithm, the parameters of SVM can be chosen automatically. This method offered an effective approach to optimizing SVM parameter automatically.Moreover, the dynamic model of ship boiler-turbine coordinated system was established by using a simplified mechanism analysis method. Then its dynamic performance was analyzed. Based on analyzing two basic schemes——boiler followed turbine and turbine followed boiler in the traditional boiler-turbine control system, this dissertation expatiated the principle of coordinated control system and given its structure diagram.The problem of inverse control based on SVM and its realization algorithm were researched. Aiming at the greater variation of the dynamic characteristic, when the operating condition changed or the parameter time varying, a compound intelligent control combined SVM and fuzzy control was put forward. This control strategy took full advantage of system modeling ability of SVM, and utilized each advantage of inverse control and fuzzy control. The simulation results showed that the SVM compound intelligent control proposed in this dissertation can overcome the limitation caused by imprecision of the inverse model identification, and has good control performance.This dissertation made deeply research on training algorithm, parameter selection and control algorithm for SVM. And some beneficial results were obtained. These results promoted, enriched and deepened the research on the theory of SVM regression algorithm, especially application of SVM in automation control domain.