Analysis And Simulation Of Gas Turbine Flue Gas Emission Characteristics

In the context of global energy saving and emission reduction,the application of gas-steam combined cycle power generation technology is increasingly widespread.The pollutants in the flue gas of gas-fired power plants are mainly NOx.It is of practical significance to study its emission characteristics and establish emission characteristics models using machine learning methods.This article takes Siemens V94.3A gas turbine as the research object,the influence of operating parameters on the NOx production was analyzed based on the historical operation data.The LSSVM toolbox of MATLAB was used to establish the characteristic model of NOx emission.The specific research content is as follows:(1)The NOx emission characteristics of gas-steam combined cycle units are analyzed.That a comprehensive analysis of the combustion process of the gas turbine and the NOx generation process determines the main factors Which affect the formation of NOx.Meanwhile,the part analyzes that the specific impact of operating parameters on NOx with historical data.Finally,10 operating parameters Which affect the generation of NOx was determined: duty valve position,diffusion valve position,IGV opening,turbine exhaust temperature,oxygen content of flue gas,gas turbine load,exhaust temperature,atmospheric temperature,atmospheric pressure,atmospheric humidity.(2)308 sets of data were collected from the DCS system,and the collected data is preprocessed.The method of min-max is used to normalize the data and eliminate the impact of the dimension.The method of normal distribution test is used to remove data outliers,which can improve the data quality.The method of five-point three-time smoothing is used to smooth the data,which can eliminate the random errors.The method of partial least squares is used to reduce the dimensionality of the data,which can overcome the multiple correlations between independent variables.The principal component matrix that extracted by the PLS method is used as the input variable of LSSVM.After preprocessing,296 sets of data were obtained.222 sets of data were used as training data to build the model,and the remaining 74 sets of data were used as test data to verify the accuracy of the model.(3)LSSVM,PSO-LSSVM,and improve PSO-LSSVM were used to establish the characteristic model of NOx emission.Simulation results show that LSSVM can obtain better regularization parameters and kernel parameters by trial and error method,however,its optimization process was time consuming,and its training results were contingency and blindness,moreover the prediction accuracy was not high.The PSO algorithm can effectively optimize the regularization parameters and kernel parameters of the LSSVM,its prediction accuracy was higher,but the PSO algorithm was easy to premature and fall into the local optimal.The improved PSO-LSSVM algorithm has higher optimization performance,which improves the learning factor and inertia weight of PSO algorithm to enhance the global search ability and local search ability.The improved PSO-LSSVM has higher prediction accuracy and reliability,which can effectively realize the prediction of NOx concentration.Finally,the NOx emission characteristics when the load is stable,sequentially increased,and decreased were analyzed respectively,then the improved PSO-LSSVM algorithm was used to establish the characteristic model of NOx emission.The results show that: on the above three cases,the number of principal components extracted by the PLS method was not the same,which indicates that the influence of each parameter on the production of NOx in these three states is not completely consistent.However,using the improved PSO-LSSVM for NOx emission modeling has achieved good prediction results.