Data-Driven Based Research On Dynamic Modeling And Optimal Control Of Boiler Combustion System

Nowadays coal-fired thermal power plant boiler still dominates in China’s power generation, meanwhile, the NOx emission of the combustion pollutant becomes the main source of air pollution. Therefore, the research of boiler combustion optimization has significant meaning. With the application and popularization of DCS system, massive field data are preserved, which provide advantageous condition for boiler combustion optimization based on data-driven methods.Boiler combustion system is a complicated nonlinear system, and the influencing factors of the boiler’s thermal efficiency and NOx emission are numerous and intercoupled. In order to eliminate the effects of redundant variables, a multivariate correlation analysis method based on mutual information is introduced. A stepwise feature selection algorithm based on mutual information (MISFSA) is proposed and combustion optimization feature subsets of variables are selected from the original set of variables by using the proposed method.Establishing accurate predictive models of boiler combustion system is the basis of boiler combustion optimization. Since the power plant boiler operation is under dynamic conditions, which means the established model needs to reflect the dynamic characteristics of boiler combustion system. A dynamic modeling method by combining support vector regression machine (SVMR) with nonlinear autoregressive moving average model (NARMAX) is applied to achieve the dynamic model of boiler combustion system based on the selected combustion optimization feature subsets of variables. Particle swarm optimization (PSO) is used to achieve the global optimization of both SVMR and NARMAX parameters. A Grouped PSO algorithm(GPSO), which dividing the whole group into three sub-groups and adjusting the inertia weight and learning factor, is proposed to overcome the problem of local convergence and ensure the global optimization capability. The simulation result shows that the established model has achieved a good performance in following and predicting the dynamic changes of boiler combustion system.To achieve optimal control of boiler combustion system, the proposed GPSO algorithm is used to realize global optimization based on the established models. To achieve hierarchical control of each air throttle, the closed-loop control schemes of boiler combustion system were studied at last.