Optimization Of Nitrogen Oxides Prediction Model Based On Simulated Annealing Algorithm And Particle Swarm Algorithm

By analyzing the relationship between the operating parameters of a heavy-duty gas turbine unit and the NO_x concentration in the flue gas,a NO_x concentration prediction model based on LSSVM was established.The particle swarm algorithm,simulated annealing algorithm and simulated annealing particle swarm algorithm were selected to optimize the model parameters,and the emission characteristics of NO_x in the flue gas of the unit under actual working conditions were studied.The results are as follows:（1）The relationships among ten parameters:exhaust temperature of gas turbine unit turbine,unit load,IGV opening,duty valve opening,diffusion valve opening,and the nitrogen oxide concentration in the flue gas were analyzed.For the collected operating data,the data structure was simplified by removing outliers,eliminating random errors,data standardization and nonlinear PLSR methods,and established a mathematical model for predicting NO_x concentration based on LSSVM.Preliminary prediction results show that the model parametersγandσhave a significant impact on the accuracy of the model,and the trial and error method for model selection has disadvantages of time-consuming and unstable accuracy.（2）The particle swarm algorithm and the simulated annealing algorithm were used to optimize the regularization parameterγand the nuclear parameterσof the nitrogen oxide concentration prediction model based on PLSR-LSSVM.It is found that the particle swarm algorithm has a fast convergence speed when searching for the optimal value,but it is easy to fall into the local area.SA algorithm can accept the imperfect solution with a certain probability,and continuously carry out iterative optimization,so it is easier to find the global optimal value,but the search time is too long.Therefore,the simulated annealing algorithm is used to improve the particle swarm algorithm.The Metropolis criterion is introduced to prevent the particle swarm algorithm from falling into the local optimum and retain the fast convergence characteristics of the particle swarm algorithm.The average RMSE of the NO_xprediction model optimized by the SA algorithm is 0.0174,the average running time is 9449 s,and the variance is 1.07×10^-5;the average RMSE of the NO_x prediction model optimized by the PSO algorithm is 0.090,the average running time is 47 s,and the variance is 5.01×10^-3;The PSO algorithm which was improved by the SA algorithm was used to optimize the NO_x concentration prediction model of LSSVM.It is found that the average of RMSE of the model is 0.0174,the average calculation time is 120 s,and the variance is 1.29×10^-6.The PSO-LSSVM model improved by the simulated annealing algorithm could simultaneously meet the higher generalization ability and accuracy of the SA-LSSVM model and the fast convergence speed of the PSO-LSSVM model when predicting the concentration of nitrogen oxides.（3）Based on the LSSVM model,the PSO-LSSVM model,and the SA/PSO-LSSVM model,the nitrogen oxide emission characteristics were predicted,when the flue gas was under the steady load condition and the load increase condition of the gas turbine.The results show that:under the steady load condition and the load increase condition,the nitrogen oxide concentration predicted by the SA/PSO-LSSVM model is in good agreement with the measured data.In the above models,the absolute error is the smallest,the prediction effect is the most stable,and it satisfies the accurate prediction of nitrogen.