基于神经网络的车用发动机广义预测控制

为了改善具有非线性特性的发动机燃油控制效果,以达到高效率、低污染的要求。利用一种前向神经网络作为非线性系统的模型,并将其分为线性部分和非线性部分。其中非线性部分用单隐层的BP神经网络对其建模,采用学习速度较快的Davidon最小二乘法在线调整网络权值;线性部分采用受控自回归积分滑动平均(CARIMA)模型作为其数学模型,用递推最小二乘法(RLS)作为其参数辨识的方法。每步将所得非线性系统的网络模型线性展开,得到线性回归模型,并以非线性前馈增益方式补偿建模误差,建立了一种适合非线性系统的自校正广义预测控制器。仿真结果表明该算法收敛速度快,控制动作平稳,控制效果理想。

Generalized Predictive Control of Vehicle Engine Based on Neural Network

In order to achieve the requirements of high efficiency and low pollution,this paper aimed at the improvement in the fuel control effect of engine with nonlinear characteristics.A feedforward neural network was adopted as nonlinear system model,and was divided into linear and nonlinear components.For the nonlinear part,the modeling used BP neural network with a single hidden layer,and realized online adjustment of network weights by using Davidon least squares with faster learning;For the linear part,controlled auto-regression integral moving average(CARIMA) model was adopted as the mathematical model,and recursive least squares(RLS) as its parameter identification method.At each step,the network mode of nonlinear system was expanded linearly to get linear regression model,and modeling error was compensated in the nonlinear feedforward gain way.A new generalized predictive self-tuning controller suitable for nonlinear system was presented.The simulation results show that the algorithm had several advantages such as fast convergence,stable control actions,an ideal control effects.