Research Of Non-linear Model Predictive Control Theory And Method For AFR Of SI Engines

With the vehicle exhaust emission standards becoming strictly and theincreasingly people’s requirements for the fuel economy and the dynamicperformance of engines, the accurate control of engine air-fuel ratio (AFR) hasbecome one of the hot research issues of scholars. Currently, the widely usedautomobile engine AFR control methods are based on the MAP method. Thesemethods not only require a lot of the calibration experiments, and in the transientengine operating conditions, the accurate control of AFR can not be guaranteed. Inthis paper, as the target is researching about the SI engine AFR methods which issuitable for engineering applications, high control accuracy, robustness and adaptiveability, as well as based on the model identification of SI engine, and by usingnon-linear model predictive control method, in depth research about the theories andthe methods of SI engine AFR control. The main research work as follows:1). By using Radial basis function (RBF) neural network model achieved themodeling of SI engine AFR system. Set the random amplitude sequemce (RAS) signalas the excitation signal of SI engine AFR system, so that will make the training data toreflect the whole nonlinear dynamics of AFR system. And by using fading memoryRLS algorithm achieved the SI engine AFR system modeling with RBF neuralnetwork model and the model parameters update adaptive online. This effectivelysolved the system parameter variations problems due to the wear of engine and themanufacturing tolerances which will affect the accuracy of the model prediction. Sothat model has the advantages of less amount of the training computing, highmodeling accuracy and the parameters adaptive online.2). Based on the depth research and analysis about the traditional second-orderVolterra model, for the problem of the serious difference between the model inputvariables, the traditional second-order Volterra model was improved, and proposed avariable sampling periods Modified Volterra model of SI engine AFR system.Compare with the traditional second-order Volterra model, this model effectively reduced the dimension of the model while ensure the modeling accuracy, so thatgreatly reduced the calculation of the modeling and the prediction. Moreover, byusing fading memory RLS algorithm achieved the parameters of the ModifiedVolterra model update adaptive online, and make it more suitable for the practicalapplication.3). In this paper, for the SI engine AFR nonlinear model predictive controloptimization problems, a non-monotonic trust region SQP with the inequalityconstraints optimization algorithm was proposed. This method based on the trustregion SQP optimization algorithm, for the inequality constraints problem, with theaddition of the trust region and the Non-monotone line search method, greatlyreduced the calculation of the method and improved the convergence rate. Moreover,it effectively avoided the Marotos problems. And the feasibility and the convergenceof the algorithm had rigorous theoretical proofed in this paper.4). In this paper, by using the a non-monotonic trust region SQP with theinequality constraints optimization algorithm, a non-linear model predictive controlmethod based RBF neural network model for AFR of SI engines was proposed. In thismethod, the optimization algorithm just used the predict output sequence of the model,independent from the structure and the internal parameters of the prediction model,thus effectively solved the problem of that the internal parameters of RBF neuralnetwork model can not be used. So that it makes the model predictive control methodwith the advantages of high control accurate, robustness and adaptive ability. Thesimulation shows that the algorithm is efficient for SI engine AFR control.5). In this paper, a model predictive control method based Modified Volterramodel for AFR of SI engines was proposed in this paper. This method uses ModifiedVolterra model’s structure to separate the linear terms from the model, and using leastsquare method to solve the optimal control sequence directly. Thus this algorithm hasthe advantages of high control accuracy, low computation, robustness and adaptiveability. The simulation shows that the algorithm is efficient for SI engine AFR control.6). In this paper, a model predictive control method based Modified Volterramodel combined with RBF neural network model for AFR of SI engines was proposed in this paper. In this method, Modified Volterra model and RBF neuralnetwork model were combined together. The method not only has the advantages ofthe low calculation, high prediction accuracy which are from RBF neural networkmodel, but also using least square method to solve the optimal control sequencedirectly which by using the structure of Modified Volterra model. So that comparewith the other two methods proposed in this paper, this method has the advantages ofless calculation, stronger robustness and higher control accuracy. The simulationshows that the algorithm is efficient for SI engine AFR control.