基于智能体——神经网络的轮胎侧偏特性

在轮胎接地压力试验台上测量轮胎在充气气压、垂直载荷、轮速等变化的18种工况下获得的轮胎侧偏特性,并定性地分析侧偏特性与侧偏角、轮速、垂直载荷和充气压力等众多因素关系,揭示出它们之间的函数关系表现出复杂的非线性。据此,将558个样本数据点作为网络特征参数,训练和建立自适应神经网络模型,进行网络逼近。由于存在网络层数和节点的增加带来的计算量大和众多非线性传递函数相互叠加造成的网络不确定性以及误差大的问题,应用智能体技术,对神经网络的训练过程进行优化。仿真结果表明,效果较好,该曲线的平均误差小于1.5%。通过与魔术公式轮胎模型比较分析,得到所采用的模型更加逼近轮胎试验曲线,表现出明显的优越性,可为汽车理论研究提供一定参考。

Tire Lateral-slip Characteristics Based on Agent-neural Net

Under the combined action of inflation pressure,vertical load and wheel velocity,the lateral forces of 18 working conditions are measured by the analytic system of tire contact pressure on ground.And the complex relations between them are analyzed qualitatively,which they are the complicated nonlinear implicit function.For samples,558 test dates are selected and used as the network characteristic parameters,an agent-neural network for kind identification is trained and constructed.Because the increasing of network nodes and layers and the superposition principle of nonlinear transfer function,the identification rate is low and node positions are uncertainty in networks.In order to solve these problems,the agent is introduced to optimize the training process of the artificial neural networks.Simulation results showed that the mean error achieves 1.5%.According to the analysis of simulation results compared with magic formula,adaptive model based on agent has the advantage of numerical approximation the for test curve.This method can provide support for car safety design.