基于深度学习的分布式驱动轻型电动汽车多状态感知及时滞混杂控制研究

Distributed drive electric vehicle utilizes in-wheel motors to directly drive the wheels such that the torque of each wheel can be controlled independently and fast, which provides unique advantage for vehicle dynamics control. Whereas, dynamic control in distributed drive electric vehicles relies on the real-time coordination of the distributed controller nodes in the vehicle network, and it is necessary to research affect of the network induced time delay for vehicle handling performance and stability. This project is to establish a hybrid time-delay dynamics model of lightweight distributed drive electric vehicles system with complex machine-electric-network coupling, analyze the sensitivity of load parameters in lightweight electric vehicle, and reveal the influence of network induced time delay on vehicle dynamics control. Based on the perspective of cognitive neuroscience, this project is also to construct the vehicle's own multi-state perception system by exploring the information perception mechanism in the human perception system. It uses the deep learning technology to directly learn the motion mechanism of the vehicle and establishes the recurrent neural networks-based internal forward model that can reflect complex nonlinear dynamic characteristics of hybrid time-delay asynchronous system in vehicle dynamics, and then forms vehicle motion perception which can realize high-precision and real-time perception of vehicle states by compensating time delay. This project will further study hybrid delay-based multi-variable robust control method using linear variable-parameter technique and hybrid system theory, and pays attention to the optimal control allocation problem of the underlying multi-actuator for over-actuated electric vehicles. The project results will provide theoretical guidance for the engineering design of the control system in electric vehicles.

分布式驱动电动汽车使用四个轮毂电机直接驱动车轮，为车辆动力学控制提供独特优势，但其动力学控制依赖车载网络对分布式控制器节点的实时协调，车载网络通讯时滞对整车动力学控制有重要影响。本项目拟建立分布式驱动轻型电动汽车系统复杂机-电-网耦合的不确定性混杂时滞动力学模型，分析轻型电动汽车载荷参数的灵敏度，揭示车载网络时滞对车辆动力学控制的影响规律。基于认知神经科学视角探究人类运动感觉系统中的信息感觉机制来构建车辆自身域多状态运动感知系统，采用深度学习技术直接学习车辆的运动规律，建立能反映电动汽车动力学系统本质混杂时滞异步复杂非线性动态特性的递归神经网络内部前向模型，形成能补偿时滞的车辆运动知觉，实现高精度的车辆状态参数实时感知。基于混杂系统理论研究车辆线性变参多变量混杂时滞鲁棒控制方法，关注过驱动轻型电动汽车底层多执行器的最优约束分配问题。项目结果将为电动汽车控制系统的工程化设计提供理论参考。

本项目以分布式驱动轻型电动汽车为研究对象，建立其复杂机-电-网耦合的不确定性混杂时滞动力学模型，发展基于车辆状态空间、模型灵敏度方程的车辆载荷参数轨迹灵敏度的非线性辨识理论模型，定义车辆载荷参数的灵敏度新概念，揭示了车辆载荷参数变化(如乘客或货物加载、惯量增加、质心高度升高等)及时滞对车辆动力学稳定性系统状态估计及控制的作用规律。在融合轮毂转矩等车载多传感器信息的基础上，设计车辆状态及惯性参数的深度学习RNN及非线性卡尔曼滤波等不同种类联合观测系统进行多状态参数的实时感知。针对分布式驱动电动汽车操纵稳定性及平顺性控制系统的中非线性、时滞不确定性及外界干扰问题，阐述了轮毂电动汽车SRM不平衡电磁力的产生机理及其对车辆行驶安全性能的影响，构建了基于动态误差变量的不确定电动汽车运动增广系统及框架，设计车辆线性变参多变量非线性鲁棒控制策略。搭建了硬件在环仿真平台及试验平台，进行了分布式驱动电动汽车相关场地试验验证。本项目的完成推动力智能电动汽车的发展，为保证电动汽车在复杂工况下安全、平顺、可靠性行驶的控制系统工程化设计提供了理论支持与依据。