Online Modeling Of Thermal Process Based On Least Squares Support Vector Machines

Energy is an important basis for human survival and social development. Its consumption, accompanied by prominent environmental problems, increases dramatically in today’s China. Occupying the subject in China’s energy consumption structure, thermal power, although its share is gradually reduced, will still exist as the main power in the future. Along with it are the task of energy saving and emission reduction. With coal consumption indicator further compressed, requirements of desulfurization and denitrification are becoming stricter, optimization of the thermal power plants will rely more on optimization theory than personal experience. So, it is important to establish a precise mathematical model of thermal system. The thermal process of power plant is a nonlinear, slow time-varying and large time-delay process, which is complex in its internal environment while vulnerable to outside environmental changes. Obviously, the extreme operating data could not be easily obtained because of the need to ensure the safe and stable operation. Least squares support vector machines (LS-SVM) is a method that suitable for thermal modeling, because it has the advantage of small sample learning ability. Since LS-SVM have a high requiment for the stability of training samples, the steady-state detection method, which is suitable for online detection and based on least-squares curve fitting, is specifically studied. And further, its steady-state index formula is improved. Above-combined are applied to the steady-state detection for the total fuel quantity, the primary air damper position, the economizer outlet flue gas oxygen content of the combustion system of a600MW power plant, individually and multivariate integratedly, and the specific step for online calculation is designed. On the basis of the steady-state data sample extracted using the above method, and by means of LS-SVM, the parameter model of the combustion system, represented by the emission concentration of nitrogen oxides (NOx), is established. The model is used to simulate the new data obtained online, thus verifying the feasibility of LS-SVM to model thermal process online combining online steady-state detection and online data updating.