智能汽车方向与速度综合决策的混合机理与规则建模研究

发布时间：2018-03-29 18:53

本文选题：智能汽车　切入点：复杂交通环境　出处：《吉林大学》2017年博士论文

【摘要】：汽车智能化是当今世界汽车工业的重点发展方向。经过多年的研究推动,针对简单工况的智能辅助驾驶技术日益成熟,诸如车道偏离预警、自适应巡航控制、车道保持辅助等先进驾驶辅助系统已经在工业界得到广泛应用,半自动驾驶和有条件自动驾驶的智能汽车也已经进入测试阶段。各国研究人员正致力于高度自动化智能汽车的攻关,其中,针对复杂交通环境的方向与速度综合决策模型是当前的难点问题之一。驾驶员控制汽车行驶方向与速度的综合决策行为本质上是一个多目标优化问题,通常需要考虑两类因素:一类是可采用连续目标函数进行评价的因素,诸如工效性、易操纵性等;另一类是不连续的二值逻辑因素,诸如安全性、合法性等。现有模型大多基于控制理论和力学原理等机理进行建模,将不连续取值的因素也纳入目标函数,与连续取值的因素一起进行多目标优化,导致难以确定加权系数,求解过程可能收敛到局部解而产生发散,表现为紧急转弯时汽车冲出道路边界等危险情况。针对上述问题,本文提出一种基于机理与规则混合决策的智能汽车驾驶员建模方法。在课题组原有的方向与速度综合控制的驾驶员最优预瞄加速度模型基础上,将安全性与合法性基于规则建模,作为约束条件缩减预瞄加速度的可行域;将工效性、易操纵性等基于机理建模,作为目标函数优化求解最优预瞄加速度。在目前查阅到的相关文献中尚未见到同样做法。本文重点研究了以下几个方面的内容:首先,预瞄加速度可行域缩减方法研究。本文探索了在预瞄纵向加速度和预瞄侧向加速度构成的平面上,依次根据道路可行驶区域、障碍物等约束的行驶安全性,以及法规限速、车道线、交通信号灯与停止线等约束的行驶合法性,逐步缩减预瞄加速度的可行域,进而保证最优预瞄加速度符合汽车行驶安全与交通规则要求。研究了汽车轮廓分别与道路可行驶区域、障碍物、车道线相对位置关系的判定方法,并且提出了一种针对交通信号灯与停止线的合法性分相判定方法。仿真和试验表明有效解决了现有模型难以确定加权系数以及优化求解过程可能发散的问题。其次,预瞄加速度综合评价方法研究。在经过安全性判定与合法性判定缩减之后的可行域内,根据工效性、易操纵性等评价指标对预瞄加速度进行综合评价,优化求解最优预瞄加速度。原有模型主要依靠侧方安全性指标来促使汽车靠近车道中心线行驶。然而,由于侧方安全性指标基本上只在车道线或道路边界附近区域起作用,而在车道中心线附近区域作用很小,导致汽车行驶轨迹的收敛性欠佳,在移线或转弯之后难以稳定地沿车道中心线行驶,呈现出持续的左右摆动。安全性基于规则建模会进一步加剧这个问题。因此,本文提出增加一个横向跟随性评价指标,用于描述驾驶员期望靠近车道中心线行驶的心理特征。仿真和试验表明有效解决了轨迹收敛性欠佳的问题。再次,汽车非线性动力学动态校正方法研究。在经典的最优预瞄加速度模型中,采用被控车辆的低阶等效逆系统模型对汽车动力学进行开环校正,通过精确标定逆模型来保证实际加速度与最优预瞄加速度的一致性。由于汽车动力学除线性区外还有中心区和大滑移区,呈现强非线性,要精确获取其逆模型参数,通常需要在由车速、加速度和路面条件构成的三维空间的大量工作点做标定试验,标定工作量非常大。针对这一问题,本文根据复合校正控制原理,提出以基于逆系统原理的开环校正为主体,代表驾驶员对汽车动力学特性的掌握程度,在此基础上增加一个闭环校正环节,用于描述人的补偿修正能力。仿真和试验表明提高了系统鲁棒性,可以有效减少标定工作量。最后,在Simulink仿真环境下建立了基于机理与规则混合决策的智能汽车驾驶员最优预瞄加速度模型。一方面,以车辆动力学仿真软件Car Sim中的C级样车模型作为控制对象,进行了诸如双移线路段、直角转弯路段、“8”字形路段、3D循环赛道、巡航、超车、交通信号灯路段等多种工况的人-车-路闭环系统仿真;另一方面,以开发型驾驶模拟器作为试验平台,进行了典型ACC工况的试验验证。仿真和试验结果表明:基于机理与规则混合决策的智能汽车驾驶员模型,只需对各评价指标的加权系数进行简单地设定,就能较好地适应上述各种工况,没有出现任何不安全或违反交通规则的现象,有效解决了现有模型优化求解过程可能发散的问题;汽车在移线和通过弯道之后,可以稳定地沿车道中心线行驶,没有出现轨迹持续摆动的现象,有效解决了原有模型汽车轨迹收敛性欠佳的问题;通过对被控车辆的主特性进行快速标定,就可以保证实际加速度与最优预瞄加速度具有较高的一致性,有效解决了标定工作量大的问题。
[Abstract]:The intelligent car is the focus of the development direction of the automotive industry in the world today. After many years of research to promote, in an intelligent driver assistance technology simple conditions increasingly mature, such as lane departure warning, adaptive cruise control, auxiliary and other advanced driver assistance system has been widely used in lane keeping, semi automatic driving and intelligent vehicle condition the autopilot has entered the testing stage. The research, researchers are working on highly automated intelligent vehicle in complex traffic environment, the speed and direction of the comprehensive decision model is one of the difficult problems in current. The driver control decision behavior essence vehicle speed and direction is a multi-objective optimization problem. Usually need to consider two factors: one is the factors that can adopt continuous objective function evaluation, such as efficiency, easy operation etc. ; another kind is the two valued logic factor is not continuous, such as safety, legality and so on. The existing models are based on control theory and mechanics principle of mechanism modeling, the factors of continuous values are included in the objective function of multi-objective optimization with continuous value factors, difficult to determine the weighted coefficient the solving process may converge to a local solution for emergency caused divergence, turning out of the car. The dangerous situation of road boundary according to the above problems, this paper presents an intelligent rule-based decision mechanism and the driver of the car. The driver optimal modeling method based on pre research group of the original direction and velocity integrated control based on tracking acceleration model the safety, validity and rule based modeling, as constraints reduce the feasible domain previewacceleration; the work efficiency, easy handling based on mechanism modeling, as. Standard function optimization of optimal preview acceleration. The same has not been seen in the literature at present refer to. This article focuses on the following aspects: firstly, previewacceleration feasible domain reduction method research. This paper explores in the preview plane longitudinal acceleration and lateral acceleration of the preview, in turn according to the road the driving region, driving safety barriers and regulatory constraints, speed, lane, traffic lights and stop line constraints such as driving legitimacy, gradually reduce the feasible acceleration, so as to ensure the optimal preview acceleration to meet the requirements of safety and traffic rules of the car. The car and road profile respectively. The driving region, obstacles, method to determine the relative position between the lane line, and put forward a set of traffic lights and the method of judging legality phase of the stop line. The simulation and test results show that the effective solution of the existing model is difficult to determine the weighted coefficient and optimization to solve the potential divergence problem. Secondly, research on the comprehensive evaluation method of acceleration preview. After determine the security and legitimacy to determine feasible domain reduction after, according to ergonomic, easy manipulation of the evaluation index of previewacceleration comprehensive the evaluation and optimization of optimal preview acceleration. The original model mainly depends on the lateral safety index to make the car near the center line of the lane. However, due to the lateral safety index basically only in the region near the lane or road boundary effect, and in the center line of the lane near the area of effect is very small, resulting in convergence of the vehicle trajectory the poor, after driving in the lane change or turn to steadily along the center line of the lane, showing a swing continuously. Based on the security rules into modeling This problem is exacerbated in one step. Therefore, this paper proposes to add a transverse follow evaluation index for describing the psychological characteristics of the driver's desired driving lanes near the center line. The simulation and test results show that effectively solves the problem of poor convergence trajectory. Thirdly, vehicle nonlinear dynamics dynamic correction method. The optimal pre classical tracking acceleration in the model, the low order equivalent controlled vehicle inverse system model of vehicle dynamics of open loop correction, the accurate calibration of the inverse model to ensure that the actual acceleration and consistency of optimal preview acceleration. The vehicle dynamics in linear region and the central area and the large slip area, showing the strong nonlinearity, to accurately obtain the inverse model the parameters, usually require a lot of work in the speed, acceleration and three-dimensional space composition of road condition to do the calibration test, the calibration for the workload is very large. This problem, based on the composite correction control principle, put forward to the open-loop inverse system based on the principle of correction as the main representative of the driver master degree of vehicle dynamics, based on a closed loop correction link is used to describe the compensation ability. The simulation and test results show that improve the robustness of the system, can be effective to reduce calibration workload. Finally, established in Simulink simulation environment of intelligent vehicle driver optimal decision rules and mechanism of pre mixed tracking model based on acceleration. On the one hand, with the C prototype vehicle dynamics simulation software Car Sim as a control object, such as the double shift line section, turn Road, "8" cross section, 3D cycle track, cruise, overtaking, traffic lights and other road conditions of driver vehicle road closed-loop simulation; on the other hand, the driving simulator for As the test platform, test of typical ACC conditions. Simulation and experimental results indicate that the intelligent driver model based on hybrid decision mechanism and rules, only weighting coefficient of each index to set up a simple, can better adapt to the various working conditions, without any unsafe or violate traffic regulations the phenomenon, effectively solve the existing optimization model to solve the potential divergence problem; in the car lane and through the corners, can be stably along the center line of the lane road, no track continuous swing phenomenon, effective solution to the original model of vehicle track convergence problem of poor; fast through the calibration on the main characteristics of the controlled vehicle. You can ensure the consistency of the actual acceleration and optimal preview acceleration is high, effectively solve the calibration workload big problem.

【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：U463.6

【相似文献】