Research On Modeling And Optimization For The Coal-fired Power Plants Boiler Combustion System

With the ever increasing demanding of "energy saving and emission reduction", the high efficiency and low pollution of the boiler operation is attracting more and more attention. This thesis studied the data driven based modeling technique and optimization for the thermal power plant boiler combustion system.This thesis gave the overall description of the300MW boiler, field data collection, data pre-processing rules, objective function selection, the process of steady neural networks modeling and model evaluation. Both reducing the normalized coal comsumption and reducing NOx emissions were combined as the objective target for boiler combustion optimization. This thesis compared different training algorithms of neural networks and evaluated the generalization capabilities of the trained models.Furthermore, this thesis discussed data-driven based system identification method in discrete form. It featured dividing the dynamic nonlinear process into linear dynamic part and static (steady) part and then combined these two parts together. Then this thesis gave the definition of static model and dynamic model, the identification techniques, the definitions of gains and methods of computing gains. Through modifying the dynamic gain to match the steady-state gain, thus the dynamical nonlinear process can be approximated locally.Then this thesis discussed the improved neural network based modeling technique considering the existence of the dynamic characteristics in the physical process and derived the training algorithm. This modeling technique has been implemented for modeling of NOx emission and was evaluated. Results showed that the introduced one-order auto-regressive model can capture the dynamics of the formulation of NOx emission and better results can be got compared with the traditional NN.Finally, genetic algorithm was used to get the optimized set points for the MVs to minimize the objective function, which was reduing coal consumption and reducing NOx emissions. The results based on neural network modeling of the two objective variables were got. After the disturbance existed in the NOx modeling data was removed, the new results were got. Then this thesis compared and discussed the two different kinds of optimization results based on different modeling techniques. The gain analysis results for NOx emissions were given and discussed, which can provide guidance for control of NOx emissions.This thesis used the database and dynamic webpage design techniques to make the webpage, which can provides guidance for the operators in the power plant.