Human-like autonomous car-following model with deep reinforcement learning

This study proposes a framework for human-like autonomous car-following planning based on deep reinforcement learning (deep RL). Historical driving data are fed into a simulation environment where an RL agent learns from trial and error interactions based on a reward function that signals how much the agent deviates from the empirical data. Through these interactions, an optimal policy, or car-following model that maps in a human-like way from speed, relative speed between a lead and following vehicle, and inter-vehicle spacing to acceleration of a following vehicle is finally obtained. The model can be continuously updated when more data are fed in. Two thousand car-following periods extracted from the 2015 Shanghai Naturalistic Driving Study were used to train the model and compare its performance with that of traditional and recent data-driven car-following models. As shown by this study’s results, a deep deterministic policy gradient car-following model that uses disparity between simulated and observed speed as the reward function and considers a reaction delay of 1 s, denoted as DDPGvRT, can reproduce human-like car-following behavior with higher accuracy than traditional and recent data-driven car-following models. Specifically, the DDPGvRT model has a spacing validation error of 18% and speed validation error of 5%, which are less than those of other models, including the intelligent driver model, models based on locally weighted regression, and conventional neural network-based models. Moreover, the DDPGvRT demonstrates good capability of generalization to various driving situations and can adapt to different drivers by continuously learning. This study demonstrates that reinforcement learning methodology can offer insight into driver behavior and can contribute to the development of human-like autonomous driving algorithms and traffic-flow models.     

Dynamic demand estimation and prediction for traffic urban networks adopting new data sources

Nowadays, new mobility information can be derived from advanced traffic surveillance systems that collect updated traffic measurements, both in fixed locations and over specific corridors or paths. Such recent technological developments point to challenging and promising opportunities that academics and practitioners have only partially explored so far.The paper looks at some of these opportunities within the Dynamic Demand Estimation problem (DDEP). At first, data heterogeneity, accounting for different sets of data providing a wide spatial coverage, has been investigated for the benefit of off-line demand estimation. In an attempt to mimic the current urban networks monitoring, examples of complex real case applications are being reported where route travel times and route choice probabilities from probe vehicles are exploited together with common link traffic measurements.Subsequently, on-line detection of non-recurrent conditions is being recorded, adopting a sequential approach based on an extension of the Kalman Filter theory called Local Ensemble Transformed Kalman Filter (LETKF).Both the off-line and the on-line investigations adopt a simulation approach capable of capturing the highly nonlinear dependence between the travel demand and the traffic measurements through the use of dynamic traffic assignment models. Consequently, the possibility of using collected traffic information is enhanced, thus overcoming most of the limitations of current DDEP approaches found in the literature.     

Subjective-utility travel time budget modeling in the stochastic traffic network assignment

ABSTRACT

In this article, we propose a new model called subjective-utility travel time budget (SU-TTB) model to capture travelers' risk-averse route choices. In the travel time budget (TTB) and mean-excess travel time (METT) model, a predefined confidence level is needed to capture the risk-aversion in route choice. Due to the day-to-day route travel time variations, the exact confidence level is hard to be predicted. With the SU-TTB model, we assume travelers' confidence level belongs to an interval that they may comply with in the route choice. The two main components of SU-TTB are the utility function and the TTB model. We can show that the SU-TTB can be reduced to the TTB and METT model with proper utility function for the confidence levels. We can also prove its equivalence with our recently proposed nonlinear-expectation route travel time (NERTT) model in some cases and give some new interpretation on the NERTT with this equivalence. Finally, we formulate the SU-TTB model as a variational inequality (VI) problem to model the risk-averse user equilibrium (RAUE), termed as generalized RAUE (GRAUE). The GRAUE is solved via a heuristic gradient projection algorithm, and the model and solution algorithm are demonstrated with the Braess's traffic network and the Nguyen and Dupuis's traffic network.     

盾构下坡掘进对推进阻力的影响研究

以神华新街6°煤矿斜井盾构工程为背景,针对盾构下坡掘进工况,通过建立盾构上作用的水土压力载荷与坡度的数学关系模型,推导出推进阻力与坡度的数学公式。结合工程实例,计算水平掘进与6°下坡掘进各推进阻力值,并利用MATLAB软件绘制下坡掘进时推进阻力随坡度变化的曲线。结果表明:1)6°下坡掘进的总推进阻力相对于水平掘进仅减小了约8%,坡度小于13°时,可通过水平掘进时的推进阻力减去重力沿掘进轴线上的分量来近似求得,且偏差小于10%;2)总推进阻力会随坡角的不断增大而减小;3)当坡度大于50.4°后,盾构有自动向下滑移的趋势,刀盘将自动压紧开挖面,导致刀具自动嵌入开挖面,增加了启动扭矩与换刀的难度。     

State estimation in roll dynamics for commercial vehicles

Accurate lateral load transfer estimation plays an important role in improving the performance of the active rollover prevention system equipped in commercial vehicles. This estimation depends on the accurate prediction of roll angles for both the sprung and the unsprung subsystems. This paper proposes a novel computational method for roll-angle estimation in commercial vehicles employing sensors which are already used in a vehicle dynamic control system without additional expensive measurement units. The estimation strategy integrates two blocks. The first block contains a sliding-mode observer which is responsible for calculating the lateral tyre forces, while in the second block, the Kalman filter estimates the roll angles of the sprung mass and those of axles in the truck. The validation is conducted through MATLAB/TruckSim co-simulation. Based on the comparison between the estimated results and the simulation results from TruckSim, it can be concluded that the proposed estimation method has a promising tracking performance with low computational cost and high convergence speed. This approach enables a low-cost solution for the rollover prevention in commercial vehicles.     

Design of tyre force excitation for tyre–road friction estimation

Knowledge of the current tyre–road friction coefficient is essential for future autonomous vehicles. The environmental conditions, and the tyre–road friction in particular, determine both the braking distance and the maximum cornering velocity and thus set the boundaries for the vehicle. Tyre–road friction is difficult to estimate during normal driving due to low levels of tyre force excitation. This problem can be solved by using active tyre force excitation. A torque is added to one or several wheels in the purpose of estimating the tyre–road friction coefficient. Active tyre force excitation provides the opportunity to design the tyre force excitation freely. This study investigates how the tyre force should be applied to minimise the error of the tyre–road friction estimate. The performance of different excitation strategies was found to be dependent on both tyre model choice and noise level. Furthermore, the advantage with using tyre models with more parameters decreased when noise was added to the force and slip ratio.     

Power network parameter estimation method based on data mining technology

The parameter values which actually change with the circumstances, weather and load level etc.produce great effect to the result of state estimation. A new parameter estimation method based on data mining technology was proposed. The clustering method was used to classify the historical data in supervisory control and data acquisition (SCADA) database as several types. The data processing technology was impliedto treat the isolated point, missing data and yawp data in samples for classified groups. The measurement data which belong to each classification were introduced to the linear regression equation in order to gain the regression coefficient and actual parameters by the least square method. A practical system demonstrates the high correctness, reliability and strong practicability of the proposed method.     

基于层次分析法的区间估计的车站能力查定误差因子分析研究

本文归纳总结了车站能力查定计算过程中存在的多种影响因素,深入分析了其产生原因和不可避免性。然后利用层次分析法的区间估计方法计算得出各种误差因子对计算结果的影响权重大小,并对比不同误差因子的权重区间,筛选出强误差因子,剔除弱误差因子,为精确计算车站能力值提供依据。     

基于频域搜索的梳状谱信号检测方法

由于梳状谱信号多普勒容限较小,利用常规的匹配/相关检测方法进行检测时,需要较大数量的拷贝信号与计算量。针对这一问题,本文提出一种基于频域搜索的梳状谱信号检测算法。通过利用梳状谱信号频域多峰结构的特点,根据信号的先验信息计算出不同径向速度下回波信号的频点分布,对接收的信号在频域进行搜索累计的方法同时实现信号的检测与速度估计。仿真分析证明了本文算法的有效性。     

基于EMD的矢量水声器方位估计算法

对经验模态分解算法(EMD算法)进行深入研究,对其基本原理、基本性质以及存在的问题进行分析。为改善EMD算法在方位估计的精度,减少高频噪声、间断干扰的影响,本文采用Hilbert-Huang变换方法来消除模态混叠现象。实验结果表明,本文提出的基于EMD的矢量水声器方位估计算法,能够有效提高方位估计的精度。