基于分段学习模型的自动驾驶行为决策算法研究

在具有车道线的特定自动驾驶场景中,针对目前端到端的行为决策算法直接输入原始图像进行决策导致的网络模型迁移性差、预测精度欠佳、泛化能力不足等问题,提出一种基于分段学习模型的车辆自动驾驶行为决策算法。首先,基于GoogLeNet建立一种端到端的车道线检测网络模型,并引入车道中心线作为决策的重要线索提高算法的迁移能力,同时利用YOLOv3目标检测模型对本车道内前方最近障碍物进行位置检测;而后,经几何测量模型将两者检测结果转换成环境状态信息向量为决策做支撑;最后,构建基于长短期记忆(LSTM)网络的驾驶行为决策模型,根据编码的历史状态信息刻画出动态环境中车辆的运动模式,并结合当前时刻的状态推理得到驾驶行为参量。使用建立的真实驾驶场景数据集对模型分别进行训练、验证与测试,离线测试结果显示车道线检测模型的检测位置误差小于1.3%,车道内前方障碍物检测模型的检测精度达98%以上,驾驶行为决策网络模型表征预测优度的决定系数 大于0.7。为进一步验证算法的有效性,搭建了Simulink/PreScan联合仿真平台,多种工况下的仿真验证试验中多个评价指标均达到工程精度要求,实车测试的试验结果也表明该算法可实现复杂驾驶场景下平稳、准确无偏航的预测效果并满足实时性要求,且与传统端到端模式的算法相比,具有更好的迁移性和泛化能力。     

A decision support approach to the computerisation of some operational decision problems in community transport

Community Transport (CT) in the UK operates a diverse range of services, and organisations are computerising management and operational functions. This paper describes the approach which has been taken to computerising four operational decision making functions.

The paper considers models of human decision making and problem solving, with particular reference to an information processing view of cognitive activity and to perception and memory. The design of decision support systems is also discussed.

Four decision problems are considered. For each, the paper considers how people tackle the problem, how computers can be used to tackle it and the approach which has been adopted.

For allocating trips to vehicles using a diary, the approach has been to provide a representation on screen of a manual diary. For vehicle brokerage, vehicles are presented to the operator allocating a booking in an order based on the Sequence Number, an index of how ‘difficult to book’ a vehicle is, and the distance of the vehicle's base from the start point of the trip. For the sorting of passenger pick‐ups into an efficient tour, traditional solutions to the travelling salesperson problem have been rejected in favour of a solution using spacefilling curves. Finally, for allocating dial‐a‐ride passenger trips to vehicle shifts an approach has been chosen which presents the operator with appropriate information rather than attempting to automate the scheduling.

The paper concludes that the approach to the diary has been successful and accepted by operators, although the similar approach to the dial‐a‐ride scheduling has not, as the system has not yet been able to replace manual scheduling aids. The facility to order passenger pick‐ups is little used by operators. Finally, it is suggested that the vehicle brokerage problem may be an appropriate use of fuzzy logic.     

Future electricity demand from battery‐powered vehicles

An important decision faced by airline schedulers is how to adapt the flight schedule and aircraft assignment to unforeseen perturbations in an established schedule. In the face of unforeseen aircraft delays, schedulers have to decide which flights to delay, and when delays become excessive, which to cancel. Current scheduling models deal with simple decision problems of delay or cancellation, but not with both simultaneously. But in practice the optimal decision may involve results from the integration of both flight cancellations and delays. In Part I of this paper, a quadratic programming model for the integration decision problem is given. The model can formulate the integration of flight cancellations and delays as well as some special cases, such as the ferrying of surplus aircraft and the possibility of swapping different types of aircraft. In this paper, based on the special structure of the model, an effective algorithm is presented, sufficient computational experiments are conducted and some results are reported. These show that we can expect to obtain a sufficiently good solution in terms of reasonable CPU time.     

邻近建筑物条件下青岛地铁盾构施工风险评价

以往构建的邻近建筑条件下地铁盾构施工风险指标没有考虑指标间的相关性,且忽略专家的犹豫程度,首先从方法创新的角度利用决策实验室法（DEMATEL）确定准则层间综合影响关系,作为控制层结构关系构建网络层次结构模型,求出DEMATEL和ANP相结合的主观混合权重; 再以熵权法求解客观权重,采用主客观权重偏差平方和最小为目标函数求解组合权重; 最后考虑了专家犹豫程度,运用vague集法进行施工风险评价。通过计算得出青岛地铁1号线安安区段风险等级与工程实际相符合,进而验证了该评价方法的实用性。     

路段人行横道二次过街感应式信号控制算法设计

在现有感应控制二次过街系统中,道路两边以及安全岛按钮控制多为分开控制,没有相互协调,行人在安全岛等待时间过长易导致违章过街等问题。在一次过街感应控制算法基础上,结合二次过街相关理论、机动车与行人延误机理以及行人过街信号配时理论,提出"人行绿波"协调思路,对路段二次过街感应控制算法进行设计,并应用Vissim软件进行仿真验证,结果显示设计感应控制算法能有效地降低行人和机动车延误。     

快速公交预感应信号优先协调控制策略

提出了一种考虑交叉口协调控制的预感应公交信号优先策略,包括两个部分:信号配时优化和协调控制.假设公交车行驶时间已知的情况下,该策略通过按绿信比分配理想时间差和基于公交车站点时空距离转换的改进数解法来实现交叉口的配时优化和协调控制.为了获得该策略的实施效果,以常州市通江南路两相邻交叉口的单个方向为例,设计了四种信号控制情景(无优先、传统优先、预感应优先和预感应协调优先),并利用微观仿真软件VISSIM进行仿真分析.结果表明,四种情景中预感应协调优先策略大大减小了公交在交叉口的延误,提升了公交服务的可靠性,并且对社会车流的干扰最小.     

呼和浩特火车站交叉口交通组织优化分析研究

针对呼和浩特火车站交叉口的交通组织方式和信号配时进行实地调查,并对当前组织方式下的交通运行状况进行分析,以车辆安全畅通运行及延误时间短为目标,提出适合该路口交通条件的提前放行左转的保护加许可型信号配时优化方案,并应用synchro进行仿真,仿真结果表明该优化方案有效提高了交叉口通行能力且降低了延误时间。     

CJK6150H数控车床主轴变频改造

本文通过对我公司使用的CJK6150H数控车床,针对日常使用中出现的较为频繁的故障,应用目前技术领先的变频技术,对机床进行了变频改造,大大降低了故障率,使月维修费用从5000余元降低到目前不到200元,减少了维修费用,降低了员工的劳动强度,提高了生产效率,为企业创造了不少的经济效益。     

多径衰落信道下OFDM系统定时同步优化

在多径环境下,正交频分复用（OFDM）系统对第1径的寻找是解决定时同步的关键,多数算法将最强径当作第1径,使得定时同步向后延迟,造成严重的ISI.本文在PN训练定时同步的基础上,通过寻找滑动求和值的第1个下降沿来确定第1径的位置.仿真结果证明了本算法的优越性,同时在第1径不为最强径的情况下算法也能得到较好效果.     

应急救援物流中物资分发点多目标排队选址优化

基于模糊数学、排队理论与选址理论,建立了应急救援物流中物资分发点的多目标模糊排队选址优化模型,其中：受灾点的需求率和物资分发点的服务率为模糊数,物资分发点建立成本最小和救援反应时间最短为目标函数．设计了模型的求解算法,确定了物资分发点的选址以及各个分发点所覆盖的受灾点区域范围．结合具体算例,验证了模型的实用性,并分析了物资分发点应急服务能力对救援效果的影响．