Wang Bohui

Data-driven modelling and fuzzy multiple-model predictive control of oxygen content in coal-fired power plant

In the combustion system of a boiler, oxygen content in the flue gas is a significant economic parameter for combustion efficiency. As a combustion system is highly complex and there are many constraints in a real process, traditional control cannot achieve satisfying performance in the practical oxygen content tracking control problem. In this paper, we build a combustion process model with a data-driven method and present a multiple-model-based fuzzy predictive control algorithm for the oxygen content tracking control. The combustion process model is presented as a multiple-model form, which can represent the real process more accurately. A data-driven method with fuzzy c-means clustering and subspace identification is used to identify the model parameters. Then, model predictive control integrated with a fuzzy multiple-model is used to control the oxygen content tracking problem. As the coal manipulated variable is decided by the load demand in the real process, a real-time measured value is applied to the process. All data used to obtain the process model is historical real-time data generated from a 300-MW power plant in Gui Zhou Province, China. Real-time simulation results on the 300-MW power plant show the effectiveness of the modelling and control algorithms proposed in this paper.

A hybrid model for furnace exit gas temperature monitoring based on CM-LSSVM-PLS

Monitoring system of furnace ash fouling is the foundation of the soot-blowing operation on furnace area. For furnace exit gas temperature (FEGT) is the key parameter in monitoring system, a new CM-LSSVM-PLS method is proposed to predict FEGT. In the process of CM-LSSVM-PLS method, considering the characteristics of operational data, c-means (CM) cluster algorithm is used to partition the training data into several different subsets. Submodels are subsequently developed in the individual subsets based on least squares support vector machine (LSSVM). Finally, partial least squares (PLS) algorithm is employed as the combination strategy. The single LSSVM is established to make a comparison with CM-LSSVM-PLS method. The proposed model is verified through operation data of a 300MW generating unit. The comparison result shows that the new CM-LSSVM-PLS method can predict FEGT accurately while the time consumed in modeling decrease drastically.