F2010C028
Design and Experimental Evaluation of Predictive Engine Air-ratio Control Based on Online Neural Network
Vehicle emission is one of the main sources of air pollution in urban area. Reducing the engine emission form automobiles, such as CO, HC, NOx, etc, has become a main concern of governments and automobile manufacturers. Moreover, fuel consumption is also a big concern of customers. Pollution reduction, fuel efficiency and drivability improvement relate mostly to lambda among all of the engine control variables. The best balance between power output and fuel consumption can be obtained by maintaining the air ratio to be the stoichiometric value. This paper presents a nonlinear online predictive control algorithm based on online neural network (ONN) model. Mode based predictive control (MPC), based on predictive model and receding horizon optimization, is an emerging and effective feedback control strategy. For this kind of control strategy, the predictive model is a crucial component because the essence of MPC is to optimize the forecast of process behavior, and the forecast is accomplished with the predictive model. Since the engine design information is usually insufficient and difficult to be identified, the engine model is also highly nonlinear and time varying, using physics laws to determine the engine model is therefore unsuitable. Hence a nonlinear predictive model is proposed to approximate the engine dynamics. Neural network (NN), based on the expectation risk principle, is a powerful machine learning technique which can handle complex and nonlinear function estimation problems. However NN modeling is an offline algorithm, i.e., the engine model built cannot be updated dynamically with the subsequent samples for correction. In order to overcome this weakness, an ONN is proposed so that the estimated engine model can be continuously updated whenever new samples arrive. Thus the accuracy of the engine model can be continuously improved as the latest system samples are available. As a matter of fact, the engine performance changes as the engine wears over time or fair user modifications on the engine, so any changes of engine performance over time can be updated by the proposed online algorithm. The proposed control algorithm is supported and verified by simulation tests with MATLAB SIMULINK. In addition, the control algorithm has also been implemented on a real car for testing. The actual test results in experiments are in good agreement with simulation results. Both simulation and experimental results also show that the proposed control algorithm can regulate the engine air ratio to the stoichiometric value under external disturbance with less than 1% tolerance.
This abstract is supplemented by a PDF, which can be viewed here.
Poster presentation: Test, simulation and calculation methods of vehicles and components