针对传统车辆控制策略辨识过程费时、费力,需大量实车试验的问题,提出了一类混联式混合动力车辆(power-split hybrid electric vehicles,PS-HEV)的伪谱最优控制构型辨识方法。该方法基于车辆功率分配器的纵向动力学模型与积分型性能指标...针对传统车辆控制策略辨识过程费时、费力,需大量实车试验的问题,提出了一类混联式混合动力车辆(power-split hybrid electric vehicles,PS-HEV)的伪谱最优控制构型辨识方法。该方法基于车辆功率分配器的纵向动力学模型与积分型性能指标,构建了一类Lagrange型最优控制问题。由于该类问题呈非线性不连续的特性,传统方法难以求解,本文中利用Legendre伪谱法将原问题高精度转化为非线性规划问题(nonlinear programming,NLP),并调用成熟算法对NLP进行求解。本研究所提出的构型辨识方法可用于巡航、加速和排放等方面的控制策略辨识,是PS-HEV最优控制策略辨识的通用框架。最后以车辆加速-滑行巡航策略下的算例说明了所提辨识框架的有效性。展开更多
Abnormal movement states for a mobile robot were identified by four multi-layer perceptron. In the presence ot abnormality, avoidance strategies were designed to guarantee the safety of the robot. Firstly, the kinemat...Abnormal movement states for a mobile robot were identified by four multi-layer perceptron. In the presence ot abnormality, avoidance strategies were designed to guarantee the safety of the robot. Firstly, the kinematics of the normal and abnormal movement states were exploited, 8 kinds of features were extracted. Secondly, 4 multi-layer pereeptrons were employed to classify the features for four 4-driving wheels into 4 kinds of states, i.e. normal, blocked, deadly blocked, and slipping. Finally, avoidance strategies were designed based on this. Experiment results show that the methods can identify most abnormal movement states and avoid the abnormality correctly and timely.展开更多
文摘针对传统车辆控制策略辨识过程费时、费力,需大量实车试验的问题,提出了一类混联式混合动力车辆(power-split hybrid electric vehicles,PS-HEV)的伪谱最优控制构型辨识方法。该方法基于车辆功率分配器的纵向动力学模型与积分型性能指标,构建了一类Lagrange型最优控制问题。由于该类问题呈非线性不连续的特性,传统方法难以求解,本文中利用Legendre伪谱法将原问题高精度转化为非线性规划问题(nonlinear programming,NLP),并调用成熟算法对NLP进行求解。本研究所提出的构型辨识方法可用于巡航、加速和排放等方面的控制策略辨识,是PS-HEV最优控制策略辨识的通用框架。最后以车辆加速-滑行巡航策略下的算例说明了所提辨识框架的有效性。
基金Project (60234030) supported by the National Natural Science Foundation of China
文摘Abnormal movement states for a mobile robot were identified by four multi-layer perceptron. In the presence ot abnormality, avoidance strategies were designed to guarantee the safety of the robot. Firstly, the kinematics of the normal and abnormal movement states were exploited, 8 kinds of features were extracted. Secondly, 4 multi-layer pereeptrons were employed to classify the features for four 4-driving wheels into 4 kinds of states, i.e. normal, blocked, deadly blocked, and slipping. Finally, avoidance strategies were designed based on this. Experiment results show that the methods can identify most abnormal movement states and avoid the abnormality correctly and timely.
