Neural Network Based Nonlinear Model Predictive Control for Two-stage Turbocharged Diesel Engine Air-path System

The air-path system of the two-stage turbocharged diesel engine, the characteristics of which include strong nonlinearity, time delay, coupling and constraints, increases the difficulty in engine control. To solve the control problem of the system, a nonlinear model predictive (NMPC) controller based on nonlinear autoregressive model with exogenous input neural network (NARXNN) is developed. At first, a boost pressure predictive model, of which fuel injection quantity is the first input and bypass valve opening is the second input, and the boost pressure is the output, is established based on NARXNN. Through simulation analysis, the absolute error between the output value of the plant model and the predictive model is smaller than 0.05 bar. Then the predictive accuracy of the predictive model when the predictive horizons are different is analyzed, and the Mean Absolute Percentage Error (MAPE) is less than 2% when the predictive horizon is within 30, indicating that the predictive model has good multi-step predictive performance. At last, the NMPC controller based on the NMPC toolbox in MATLAB is established. And the the step response performance and reference-tracking performance of the controller are verified in the co-simulation platform formed by GT-Power and MATLAB/Simulink. It can be concluded from the results that the step response performance of the NMPC controller is better than that of the PID controller, and the relative error of the reference- tracking simulation is smaller than 15%.