AFR Of SI Engines Nonlinear Predict Control Based On Neural Network Model

With the global car ownership increasing quickly, the vehicle exhaust pollution also becoming seriously, Countries have enacted strict emissions standards. Researchs show that emissions and engine air-fuel ratio(AFR) are closely related. So the research on accurate and effective engine AFR method has become the focus of international scholars study. Currently, the popular AFR control method is based on the MAP with PI method. The weakness the method is that it requires many calibration experiments, moreover, it can not get accurate and effective control of AFR in the transient engine operating conditions. In this paper, based on the model identification of SI engine by using neural network model, a non-linear model predictive control method for AFR of SI engines was proposed. The main research content is as follows:The paper focus on SI engine AFR optimization problem of nonlinear model predictive control, a Reduced Hessian reasonable line search Sequential quadratic programming(SQP) is proposed. Effectively overcome the problem of neural network model’s internal parameters can not be used. Based on the line search SQP optimization algorithm and also introduced the Reduced Hessian and method of improved feasible decreasing direction. In the paper,Solving the sub-problem was converted into solving a series of linear equation which greatly reduced the calculation and ensured the algorithm sufficient decreasing. To avoid the Marotos problems, a higher-order correction algorithm was used to search direction. The convergence and the rate of convergence of algorithm are rigorous theoretical proofed in this paper. Based on the Reduced Hessian reasonable line search SQP optimization algorithm. A non-linear model predictive control method based neural network model for AFR of SI engines was proposed.By using Radial basis function(RBF) and Multi-layer perceptron(MLP) neural network model, the model of SI engine AFR system is obtained. In order to get the whole dynamics of AFR system, we used the random amplitude sequence(RAS) asthe excitation signal of SI engine AFR system. Training the weight vectors of output layer by using fading memory RLS algorithm. This method not only effectively solved the problem due to the wear of engine but also achieved model parameters update adaptive online. The simulations show that the algorithm is efficient for SI engine AFR control.