On Data-driven Predictive Control Of Power Plant Boiler Combustion Process

With increasing of energy shortage and environmental pollution, in order to meetthe requirements of social development, the operation of thermal power plant withhigh energy consumption and high pollution emission must be improved. The pollutantemission lies in the combustion process, it has great practical significance to researchthe theory and methods for operation optimization and economic combustion. Thecombustion is a nonlinear time-varying process with strong disturbances and stronglycoupled multivariate. Because of complex physical and chemical changes contained inthe combustion, it is difficult to establish accurate mathematical model byconventional methods. So the optimization control of the boiler can’t meet the newrequirements. On the other hand, a large amount of data was sampled and storedduring the process running. The data contains much system information, the modelingand control method based on data-driven provides an effective way to the unit’soperation optimization.On the basis of the analysis of the existing theory and methods, a sampleselection and update policy was presented to solve the problem of time and memorycost during support vector regression (SVR) training on big data. The time seriescharacteristics lie in industrial process data, the sampling mechanism of largehistorical data is proposed according to the data’s timeliness. Then a memory modesupport vector regression algorithm (MM-SVR) was constructed. After this, for theonline modeling of the key parameters of combustion process, the comprehensiveutilization of the history and online data is studied, and an online sample updatingstrategy is presented, then the memory mode online support vector regressionalgorithm (MM-OSVR) is constructed. The simulations carried out on severalbenchmark data show that MM-OSVR can improve the training speed withoutdecreasing the accuracy. NO_xanalysis on the actual operation data of a boiler verifiesthe effectiveness of the method. On the basis of the modeling, the predictive control method of boiler combustionprocess based on data-driven model is studied. Taking NO_xas the control indicator,the prediction model is established using MM-SVR, and the combustion controlsystem based on predictive control is designed to further optimize the combustionprocess. Simulations on the data with different loads and load varying show that thecontrol system has good dynamic performance.