Research On Application Of Support Vector Machine In Thermal System Modeling

Electricity is the main form of energy in modern society, with the rapid development of national economy, mankind depends on increasing electricity supply. The government attaches great importance to the development of electric power industry, and great-capacity and high-parameter thermal power generating unit becomes the direction of development of thermal power generating unit. With the improvement of the capacity and parameters of modern thermal power generating unit and the increasing complex of the system feature, higher requirement for the control system of thermal process in electric power industry production is proposed. The establishment of mathematical model for thermal system is very important to achieve precise control of production process in power plant. However, it’s very difficult to establish mathematical model by Mechanism Analysis because of the complex of the features of thermal system, the common method is to estimate mathematical model by input and output data. As a new machine learning algorithm, Support Vector Machine (SVM) has many advantages such as requiring small training set, global optimization, to avoid "over learning", to avoid "the curse of dimensionality". It has achieved remarkable success in solution of many kinds of problem such as pattern recognition, regression estimation and so on. As a special form of SVM, Least Squares Support Vector Machine (LS-SVM) not only inherits many advantages of SVM, but also obtains more wide application for the advantage of more quickly training. Owing to the problem of thermal system modeling itself is a problem of regression estimation, there has very considerable potential of application of SVM and LS-SVM in solution of this problem. In this paper, firstly the performance of LS-SVM including sparsity and robust was studied, and two algorithm based on clustering thinking was proposed to improve the performance of LS-SVM. Then, LS-SVM was applied in thermal system modeling, and three soft sensor models about unburned carbon content, exhaust gas temperature and NOx emission were established, so as to prove that the application of SVM in thermal system modeling is feasible.