面向电动车辆智能化控制、考虑传感器采样多速率及延时问题的状态估计研究

Precise and real-time vehicle states are indispensable for intelligent control of electrified vehicles, and some of the states can be estimated by fusing different kinds of sensors. However, the sampling rates and delays are different between exteroceptive sensors (such as vision system and GPS) and proprioceptive sensors (such as inertial sensors). Many conventional methods simply neglect sensor delay and reduce the sampling rate of the estimator to adapt to the slow sensors, but the estimation accuracy is obviously deteriorated. In case of electric vehicles, down-sampling introduces sampling mismatch between the control and the feedback, and may not be able to satisfy the closed-loop bandwidth of the control system. On the other hand, many of the available multi-rate estimation algorithms are too complex to be implemented in real time. Aiming at intelligent control of electric vehicles, this project studies vision-/GPS-based multi-rate estimation methodologies. Firstly, the mapping principle between exteroceptive/proprioceptive sensors and vehicle states such as sideslip angle and vehicle lateral offset is studied. Next, by utilizing frequency analysis, the effects of sampling rate on the control of electric vehicles are studied. Finally, a new estimation algorithm to solve the multi-rate and delay issues is proposed, and its effectiveness is demonstrated through theoretical analysis of convergence/accuracy and experiments. This study will provide an accurate and low computational cost estimation method for electric vehicles and industrial systems with sampling mismatch issues.

电动车辆智能化往往需要融合多种类型的传感器信号以获取实时精确的车辆状态，而外感传感器(如视觉传感器、GPS)和内感传感器(如惯性传感器)有不同的采样速率和延时。传统方法常忽略延时并降低更新速率以适应慢速传感器，但会影响观测精度；对电动车辆而言，降速率还会造成状态反馈和控制速率不匹配，而且可能无法满足控制系统对闭环带宽的要求。另一方面，现有的多速率观测方法存在设计复杂等问题。面向电动车辆智能化控制，本项目开展基于视觉传感器和GPS的多速率状态估计研究，具体包括：探究内/外感传感器组合与质心侧偏角、车路相对位置等车辆状态的映射机理并建模；通过频域分析研究采样速率对电动车辆控制的影响；提出基于采样点间残差补偿的多速率卡尔曼滤波器以实时解决多速率和延时问题，理论证明其在收敛性/准确度上相对传统方法的优势并实车验证。本研究将为电动车辆及存在采样不匹配问题的各种工业应用提供实时精确的状态估计新方法。

电动车辆智能化往往需要融合多种类型的传感器信号以获取实时精确的车辆状态，而视觉等外感传感器和惯性传感器等内感传感器有不同的采样速率和延时。传统方法常忽略延时并降低更新速率以适应慢速传感器，但会影响观测精度。此外，降速率还会造成状态反馈和控制速率不匹配，而且可能无法满足控制系统对闭环带宽的要求。本项目开展了考虑不同传感器采样速率的多速率状态估计研究，具体包括：探究内/外 感传感器组合与质心侧偏角、车路相对位置等车辆状态的映射机理并建模；通过频域分析研究采样速率对车辆控制的影响；提出基于采样点间残差补偿的多速率卡尔曼滤波以实时解决多速率和延时问题并进行了实验验证，实验结果表明，当采样频率高于60Hz时，路径跟踪的精度便不会随采样频率的增加出现显著增加，而当采样频率低于30Hz时，路径跟踪精度出现大幅降低，且控制稳定性遭到破坏。本研究为存在采样不匹配问题的车辆控制提供了参考依据。