Combustion Optimization for Utility Boilers Based on Sharing LSSVM Model

The boiler combustion system is a complex multi-input & multi-output system. To achieve the objectives of high efficiency and low pollutant emission for the system in one model, some improvements were carried out on the standard regression algorithm ofleast squares support vector machine(LSSVM). Based on the field combustion adjustment data of a 1 000 MW ultra supercritical boiler, a sharing LSSVM model was established by taking both the boiler thermal efficiency and mass concentration of NO_x emission as the output variables, with which an improved particle swarm algorithm was selected to optimize the operating conditions of the boiler. Results show that with the sharing LSSVM model, the mean prediction error for boiler thermal efficiency and mass concentration of NO_x emission can reach 0.028% and 2.16% respectively, the searched parameter combination of high thermal efficiency and low NO_x emission may serve as a reference for optimization of boiler operation.