In this paper, in order to evaluate the availability of the boiler combustion optimization of MPC technology for the current method of thermal power generation operation, two cases of experiment were implemented. Firstly, an experiment was performed using simulator linkage and actual process operation information. As a result, it was possible to obtain effects such as reducing overshooting when the target load was reached when linking the simulator. Secondly, an experiment was performed using actual operation information for optimizing combustion-related variables. However. the result of optimization by MPC solution was difficult to apply to actual process control. This is because the mechanical power generation control system is designed for the purpose of following the demand load, ① it is difficult to define a clear causal relationship between the process operation variables and nitrogen oxides and carbon monoxide, and ② data preprocessing techniques for handling disturbances and time delays. It could be confirmed that this was due to the inadequacy of the optimization technique among the multiple conflicting variables. To supplement this, soft-computing approaches including ① data linearization applying Bayesian optimization method, ② generalization of process situation prediction algorithm through application of multi-stage machine learning prediction structure and incremental data utilization, ③ global optimization method through combination of empirical-heuristic optimization algorithm are introduced to optimize the boiler combustion of thermal power plant.