基于隐马可夫模型的SUV车辆侧翻预警研究

发布时间：2018-08-12 12:46

【摘要】：近年来,随着汽车工业和道路交通的快速发展,汽车侧翻事故成为受到人们更多关注的重要安全问题。车辆在高速行驶并进行紧急转向时,在较短的时间内容易发生侧翻,因此车辆的侧翻预警变得尤为重要。本文主要对SUV车辆发生的非绊倒型侧翻进行研究,使用隐形马尔可夫模型(HMM)进行侧翻预警,可以实时的对车辆的运动状态进行监测和预测,并提前发出警示,从而提高了车辆行驶的安全性。首先采用Carsim软件在阶跃转向、斜坡转向、双移线转向、鱼钩转向四种容易发生侧翻的工况下进行仿真,得到HMM模型的可观察序列:侧倾角、侧向加速度。对采集的数据进行预处理,根据车辆的运动状态:直线运动、正常转向、紧急转向、侧翻状态,将四种工况下的数据分类并分割,使用K-means算法确定了运动状态的界限值,作为文中模型训练和辨识的前提。其次建立了HMM的双层运动状态模型,模型的底层为多维的车辆运动参数,模型的高层对应着运动状态的多维高斯隐马可夫模型(MGHMM)。并采用Baum-Welch算法训练模型,对复合工况下采集的数据进行辨识,辨别出车辆当前所处的运动状态。同时运用马尔可夫预测法,对车辆未来3s内将要发生的运动状态进行预测,若发生侧翻则触发预警装置,不发生侧翻则循环进行预测。最后将人工神经网络(ANN)与HMM模型相结合,以HMM当前用于辨识车辆行驶状态的运动参数数值作为ANN模型的输入,并对ANN模型进行训练,选用的BP-神经网络算法对车辆下一时间段的三个运动参数侧倾角、侧向加速度以及方向盘转角的数值进行预测。HMM模型实现了对车辆下一时间段的运动状态进行预测,而ANN实现了对车辆下一时间段的参数数据值进行预测,两者结合,可以使得驾驶员能够更加直观和具体地判定车辆将要发生侧翻的危险程度。
[Abstract]:In recent years, with the rapid development of automotive industry and road traffic, vehicle rollover accidents have become an important safety issue that attracts more and more attention. Vehicles are prone to rollover in a relatively short period of time when they are driving at high speed and making emergency steering. Therefore, vehicle rollover warning becomes particularly important. The tripping rollover is studied. The hidden Markov model (HMM) is used for rollover warning, which can monitor and predict the vehicle's movement state in real time and give warning in advance, so as to improve the vehicle's driving safety. The observable sequence of HMM model is obtained by simulation under the condition of rollover: rollover angle and lateral acceleration. The collected data are pre-processed and classified according to the motion state of vehicle: linear motion, normal steering, emergency steering and rollover. The motion state is determined by K-means algorithm. Secondly, a two-layer motion state model of HMM is established. The bottom layer of the model is multi-dimensional vehicle motion parameters, and the upper layer of the model corresponds to the multi-dimensional Gaussian Hidden Markov Model (MGHMM) of the motion state. At the same time, Markov prediction method is used to predict the motion state of the vehicle in the next three seconds. If rollover occurs, the warning device will be triggered and the cycle will be forecasted. Finally, the artificial neural network (ANN) is combined with the HMM model to identify the vehicle at present. The motion parameters of the vehicle running state are taken as the input of ANN model, and the ANN model is trained. The BP-neural network algorithm is selected to predict the roll angle, lateral acceleration and steering angle of the three motion parameters in the next period of time. The HMM model realizes the prediction of the vehicle moving state in the next period of time. The combination of ANN and ANN can make the driver more intuitive and specific to determine the danger degree of vehicle rollover.
【学位授予单位】：南京林业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U461.6

【参考文献】