The Research Of Boiler Flue Gas Oxygen Content Soft Sensor Based On Improved LSSVM

Thermal power is the main method of power generation in our country,it is the pioneer of the national economy.So that,improving the efficiency of boiler combustion and lowing the consuming of coal are urgent problems in power filed.The excess air coefficient is an important parameter to reflect the combustion efficiency and thermal efficiency of the boiler.But it can not be measured directly during the boiler operation and usually obtained by flue gas oxygen content.So the high-accuracy measurement of oxygen content is important to direct production operation.Now Zirconia oxygen sensor is used in power plant widely,but it has higher measurement error and complex maintenance for some time.So it is significant to take the research of flue gas oxygen content soft measurement.This paper based on boiler combustion system of Weihe power plant in Shaanxi province,and conduct deep study in technological process of boiler combustion system,primary and final selection of instrumental variable,data collection and preprocess,building and calibrating the soft measurement model,realization of soft sensor of flue gas oxygen content,the detail is following:(1)Primary and final selection of secondary variablesAfter analyzing the technological process of boiler burning system and the factors of flue gas oxygen content,choosing the following variables as auxiliary variables,such as the boiler load,temperature of main steaming,etc.Then the instrumental variables are determined by the maximum amount of information and correlation of flue gas oxygen content,and the instrumental variables that used in soft sensor model are selected based on the principals of component analysis and grey correlation degree.In this paper,the nine instrumental variables are used: air volume,flow rate of main steaming,current of air supply,wind current,fuel consuming,boiler load,temperature of main steaming,pressure of main steaming and the quantity of water.(2)Data collection and preprocess300 sets of data are selected from the database of DCS system.The maximum-minimum principle is used to process the collected data firstly,then the data are processed based on 3? and five-spot triple smoothing algorithm,this process can eliminate defect error and random error.In the 275 groups of remaining data,The 192 sets of data are selected to establish the system model,the remaining 83 sets of data are used to verify the model's accuracy of prediction and generalization ability.(3)Building and calibrating the soft measurement model of flue gas oxygen contentUse LSSVM,PSO-LSSVM and improved PSO-LSSVM algorithms to build soft senor model of flue gas oxygen content respectively.After simulation and comparison,the conclusion can be drawn.The regularization and kernel parameters of traditional LSSVM are determined by cut-and-try,the model has low accuracy and it is a random process.For the PSO-LSSVM model,the regularization and kernel parameters of LSSVM are optimized by PSO algorithm,which improve the accuracy of model,but the result trap to local optimization easily.For the improved PSO-LSSVM algorithm,parameters of LSSVM are optimized by CPSO with random inertia weight,the improved PSO-LSSVM has higher accuracy of prediction and reliability,and more suitable for field measure of flue gas oxygen content.Finally,the slide window to calibrate model is built.(4)Realization of flue gas oxygen content sensorIn order not to affect the normal operation of the existing boiler combustion DCS system,the measured values of auxiliary variables are directly obtained from the upper computer of the existing boiler burning DCS system.The communication between MATLAB and WinCC are realized by OPC technology,online measurement of flue gas oxygen content is realized.In this paper,the improved PSO algorithm is used to optimize the regularized and kernel parameters of LSSVM,which improve the prediction precision,generalization ability and reliability of the soft measurement model of flue gas.The project of measurement of flue gas oxygen content of boiler combustion system by OPC and industrial communication technologies arerealized.In this way,the gas oxygen content can be detected with high accuracy.