Non-Linear Model Predictive Control For AFR Of SI Engines

Nowadays, environment problems catch more and more attentions by people, and the air pollution is definitely one of the difficulties to solve. Among the sources of air pollution,automobile exhaust is one of the most evilest criminals. With the growth of car ownership,automobile exhaust emissions are increasing year by year, in this situation, the governments have to make a stricter standard to stop it.AFR plays an important role in the control of exhaust emission in the engine parameters.Therefore, the AFR control has been studied by researchers all the world.Currently,the widely used method of engine AFR control is based on the MAP figureand PI feedback,but this method has no universality and flexibility, a huge cost on experiments and low precision in the transient engine operating condition define it will be knocked out in the future. An advanced method is expected.In this paper, a control method with simple mechanism, high precision and strong adaptability is my target.On the base of identification of the SI engine, the non-linear model predictive is used to control the AFR. The main research work as follows:1)Based on the depth research and analysis about the NARX model,using different order of NARX model to do modeling experiment and choose the appropriate order of NARX model.Finally,in this paper adopts third order NARX model modeling the SI engine AFR system.Thus greatly reduces the computational amount of modeling and predictive control.Using fading memory recursive RLS algorithm implementation of NARX model parameters online adaptive update. This algorithm can effectively solve the aging of engine and the changing parameters due to long-term work.So that the model has higher identification accuracy and practical application value.Meanwhile,Based on the depth research and analysis about the RBF neural network,achieving the AFR of SI engine modeling.Training the weight vectors W of output layer by using the fading memory recursive least squares.This can make RBF neural network model suitable for the change of the engine dynamic characteristics and achieve the parameters update adaptive online.And this method has high precision,small amount of calculation,and make the RBF neural network more suitable for the practical application.2)In this paper, a model predictive control method based NARX model for AFR of SIengines was proposed. This method uses NARX model’s structure to predict the system’s outputs and adjust it to separate the linear and nonlinear terms of the fuel mass flow rate in the future, then the least square method can be used to solve the optimal control sequence directly, and the computation of the iterative optimization is effectively reduced. This algorithm has lots of advantages such as high control accuracy, low computation, robustness and adaptability. The simulation proves it is more effective than PID controller for SI engine AFR control.3)To take the advantages of NARX model and RBF neural network model for AFR of SI engines, a united predictive control method was proposed. The method combines he NARX model and the RBF neural network model together, and the RBF neural network model has the advantages of high accuracy and low calculation to predict the output for the SI engine AFR system. At the same time, the NARX model is used to separate the linear and nonlinear parts of the nonlinear dynamic system, so that the least square algorithm can be used to solve the optimal control sequence. Comparing with the independent NARX model for AFR of SI engines, this method has advantages such as less calculation, stronger robustness and higher control accuracy. The simulation proves that the algorithm is effective and accurate.