A rollover warning methodology for the sport utility vehicle (SUV) based on Multidimensional Gaussian Hidden Markov Model (MGHMM) and BP Artificial Neural Network (ANN) is proposed. The roll angle and lateral acceleration are used as the observed sequence of HMM and the motion status are taken as the state sequence of HMM. The Baum-Welch algorithm is adopted to train the HMM and the Markov prediction algorithm is applied to forecast the motion status in the future 3s of the vehicle. The unnecessary ANN training is reduced while the training efficiency and prediction accuracy are improved by using the predicted vehicle movement status as a guideline to make the ANN learn purposefully. The simulation result show that the established rollover warning method not only can predict the vehicle movement status, but also can forecast specific movement parameters, which can be used by the driver to judge the rollover quantitatively as well as providing data for the anti-rollover electronic control system with less parameters and high efficiency.%提出一种基于多维高斯隐马尔可夫模型(MGHMM)和BP人工神经网络(ANN)的SUV车辆侧翻预警方法,采用侧倾角和侧向加速度作为隐马尔可夫(HMM)的可观测序列,车辆行驶运动状态作为HMM的状态序列,采用Baum-Welch算法对模型进行训练,运用马尔可夫预测算法对未来3s内车辆的行驶运动状态进行预测,用预测出的车辆运动状态作为指引,使ANN有目的学习,减少不必要的ANN训练,提高训练效率和预测精度.仿真结果表明,建立的侧翻预警方法所需参数少,效率高,不仅能预测车辆行驶运动状态而且能预测具体的运动参数,可使驾驶员量化判断侧翻,也可为抗侧翻电子控制系统提供数据.
展开▼
