生成式对抗网络估算车辆参数及路面不平度

Estimation of Vehicle Parameters and Road Roughness by Generative Adversarial Network

主流的通过逆向分析得到路面不平度的方法很少从车辆的认知不确定性来考虑所识别的路面不平度是具有概率分布性。为了应对这一挑战,提出使用测试得到的车辆加速度响应的方法。基于车辆运动系统约束的生成对抗网络(GAN)进行车辆参数的动态校准和估算路面不平度。通过引入生成模型来描述估计的路面不平度和车辆参数的概率分布。在生成对抗网络训练过程中通过分别最小化其车辆运动系统约束构建物理约束,达到使生成对抗网络的最后输出符合工程问题的强物理性。然后通过使用数值模拟测得的车辆加速度响应训练来学习生成模型,进行了数值仿真实验。使用标准规范的路面不平度和经典车辆模型来证明所提出的方法的可行性。结果表明,所提出的方法可以成功地捕获车辆认识的不确定性。

The mainstream method of obtaining pavement unevenness through reverse analysis rarely considers the identified pavement unevenness to have a probability distribution from the cognitive uncertainty of the vehicle.To address the challenge,the method of using the vehicle acceleration response obtained from the test is proposed.Based on the generative adversarial network(GAN)constrained by the vehicle motion system,the dynamic calibration of vehicle parameters and the estimation of pavement unevenness are carried out.A generative model is introduced to describe the estimated probability distribution of pavement unevenness and vehicle parameters.In the training process of the generative adversarial network,the physical constraints are constructed by minimizing the constraints of the vehicle motion system respectively,so as to make the final output of the generative adversarial network conform to the strong physicality of the engineering problem.Then,the generative model is learned by training the vehicle acceleration response measured by numerical simulation,and numerical simulation experiments are carried out.The proposed method is demonstrated using standard gauge pavement unevenness and classic vehicle models.The results show that the proposed method can successfully capture the uncertainty of vehicle cognition.