基于IMPSO-BPNN 的天然气掺氢发动机性能预测

To further improve the performance of traditional particle swarm optimization back-propagation neural network (PSO-BPNN) model, based on the influence mechanisms of inertia weight and acceleration factor on particle swarm optimization, an improved particle swarm optimization back-propagation neural network (IMPSO-BPNN) method adopting non-linear decreasing inertia weight and non-linear acceleration factor was proposed. The IMPSO-BPNN method was applied to the regression analysis and prediction of performance parameters such as torque, equivalent brake specific fuel consumption, and brake specific NOx emission of a hydrogen enriched compressed natural gas (HCNG) engine. It was also compared with other neural network methods in terms of prediction accuracy, generalization ability, and convergence speed, including PSO-BPNN, genetic algorithm optimized back-propagation neural network (GA-BPNN), and back-propagation neural network (BPNN) methods. Research results show that the fuel-air ratio and spark advance angle can significantly affect the torque, equivalent brake specific fuel consumption, and brake specific NOx emissions of the HCNG engine. With torque as the predictive variable, the average absolute percentage error of the optimal IMPSO-BPNN model is 5.85%, 12.62%, and 17.96% smaller than those of PSO-BPNN, GA-BPNN, and BPNN models, respectively, and the correlation coefficient of the optimal IMPSO-BPNN model is 0. 999 86, also the highest among these models, which indicates that the prediction performance and generalization ability of the model established by the IMPSO-BPNN method are generally superior to those established by other methods. With brake specific NO* emission as the predictive variable, the CPU running times reduce by 95% in both the optimal PSO-BPNN model and the optimal IMPSO-BPNN model compared with the optimal GA-BPNN model, which demonstrates the superiority of PSO-BPNN and IMPSO-BPNN methods to the GA-BPNN method in terms of time dimension. Therefore, compared with PSO-BPNN and GA-BPNN methods, the proposed IMPSO-BPNN method has significant advantages in prediction performance and generalization ability, and ensures high computing efficiency. 2 tabs, 16 figs, 31 refs.