Research On Soft-sensor For Thermal Parameter And Optimization Of Combustion In Power Plant

With the popularization of DCS system in the power industry,power plant are mostly accumulated a lot of production data.The data often contains abundant knowledge behind it.Through the analysis of historical data,we can obtain the knowledge or means of optimization of power system operation,which provides decision basis for the power plant operation,maintenance and accident treatment.Based on the data analysis of thermal parameter,soft measurement and combustion optimization is one of the important content.According to the problems of great investment and low accuracy in measuring the oxygen content of flue gas,basing on the flue gas oxygen content theory,selecting reasonable secondary variables,the complicated relation model of secondary variables and oxygen content in flue gas was establish by support vector machine(SVM)method.Genetic algorithm(GA)was used to optimize the penalty coefficient C and kernel function parameter g,and then the GA-SVM soft measurement model of oxygen content in flue gas was built by using the optimal value.The accuracy and generalization of the model were be tested,which showed that the error of GA-SVM model less than 0.2% and relative error less than 4%,and met the requirements of different loads and different time prediction.Compared with the parameter optimization results of particle swarm optimization algorithm and grid method,genetic algorithm was easy to find the global optimal solution,which was more accurate for the measurement of oxygen content in flue gas.Oxygen and second air distribution are important parameters that affect the boiler efficiency of coal-fired units and NOx emissions.Aiming at the combustion optimization of a 600 MW unit,this paper established the model of association rule mining algorithm based on clustering partition,quoted density parameters and then chose the point of big density areas as the initial clustering centers for improving the stability of clustering.Meanwhile put the accurately data of flue gas oxygen content based on the soft sensor model to the database of combustion optimization.Based on the history data,the optimal oxygen content and the second air distribution for efficiency of boiler and NOx emission was determined by data mining.The result shows that optimal oxygen content lower than running average oxygen content,auxiliary air gatage and surrounding air gatage present waist type distribution.Ensuring the boiler thermal efficiency not lower than 92.7%,mainly load periods’ NOx emissions decreases from 220~420 mg/m3 to 220~320 mg/m3.