Research On Modeling For Combustion System Of Large Capacity Utility Boilers Based On Support Vector Machines

In the long term,the main force of the development of China's power industry is still thermal power generation.Therefore,the thermal power unit is the key object of energy saving and emission reduction.In the case of the continuous improvement of the capacity and parameters of the unit and the continuous complexity of the system's ontology characteristics,it's difficult to build an accurate mathematical model through the analysis of the mechanism.Compared with the traditional modeling method,the modeling method of boiler combustion system based on intelligent algorithm has a great advantage.Therefore,in-depth research on the modeling method of large-scale boiler combustion system based on intelligent algorithm has a great significance to achieve the goal of high efficiency and low emission for thermal power generation units.In this paper,the support vector machine modeling method of large-scale boiler combustion system was studied,and the emission factors of a 1000 MW ultra supercritical boiler was analyzed.A prediction model of boiler emission characteristics was established.The support vector machine(SVM)algorithm has the advantages of global optimization,avoiding "over learning" and "dimensionality disaster".In order to overcome the shortcomings of support vector machine constrained by small sample modeling,a core vector machine(CVM)method based on large-scale data was adopted.Meanwhile,aiming at the problem of parameter selection of support vector machine and core vector machine prediction model,an improved particle swarm optimization algorithm was proposed to optimize the parameters of model kernel to improve its prediction performance.In this paper,the SVM prediction model and CVM prediction model for the emission characteristics of large boilers were established and the predictive performance of the two were compared.The results showed that the modeling method of support vector machine and core vector machine were feasible for large boiler emission characteristics,and CVM had faster convergence speed and better generalization ability.Meanwhile,with the increase of modeling data,CVM prediction model had greater advantage in prediction accuracy and modeling time compared with SVM prediction model.