关于出租车载客地点序列推荐技术的研究

在大城市中,出租车已成为实现智能交通运输系统不可或缺的一环。然而,由于一些出租车司机的驾驶经验,和对城市活动的熟悉程度的不足,使得其在寻找乘客时会采取毫无目的的随机漫游策略。这就导致了出租车司机的收益不高,同时也造成了能源的消耗以及环境的污染。针对此问题,将提出出租车载客地点的推荐模型,使得模型给出的推荐地点序列能获得较高的期望收益。具体来说,将基于出租车GPS轨迹数据建立出租车载客地点的马尔科夫决策过程模型,并给出求解该模型的2种算法。仿真实验结果显示,与典型的TopK方法相比,给出的推荐结果能更好地提高单位时间内出租车司机的收益。     

Using Non-cooperative Game Theory for Taxi-Sharing Recommendation Systems

This paper presents a recommendation mechanism for taxi-sharing. The first aim of our model is to respectively recommend taxis and passengers for picking up passengers quickly and finding taxis easily. The second purpose is providing taxi-sharing service for passengers who want to save the payment. In our method, we analyze the historical global positioning system trajectories generated by 10,357 taxis during 110 days and present the service region with time-dependent R-Tree. We formulate the problem of choosing the paths among the taxis in the same region by using non-cooperative game theory, and find out the solution of this game which is known as Nash equilibrium. The simulation of SUMO, MOVE, and TraCI are adopted to fit our model to verify the proposed recommendation mechanism. The results show that our method can find taxis and passengers efficiently. In addition, applying our method can reduce the payment of passengers and increase the taxi revenue by taxi-sharing.     

基于图结构的城市出租车交通流量可视化

为了更好地分析出租车在城市区域中的运行规律,提出一种基于图结构的城市交通流量可视化分析方法。通过对路网进行聚类,将路网连接路段划分为区域结构,并用点-线连接形式表示城市路网,同时以区域车流量为权重,结合图中心性概念对区域重要性进行了分析。以北京市四环为例,对一亿多条出租车GPS数据进行可视化分析,实验结果表明,该方法可以直观有效地展现不同区域的出租车流量随时间变化规律和不同区域的重要程度。     

基于趋势预测的合乘收益研究

首先提出一种基于车载自组网的城市出租车合乘的方案；其次针对北京市6499辆出租车的GPS信息进行了较为全面的数据分析，获取81785个乘客上车时间及位置，从整体和局部两个方面定量地讨论了出租车合乘带来的益处，运用灰色预测和曲线拟合两种不同的方法建立数学模型预测上车点个数的变化趋势；最终得出在北京市六环内乘客采用合乘方式后，可减少1/3以上的上车点以及约20～30%的行驶里程，对缓解交通压力及构建绿色城市具有十分重要的意义.     

基于稀疏轨迹数据的出租车载客区域推荐

基于短期出租车轨迹数据的载客区域推荐能极大减少系统开销,提高推荐效率,但常伴随着数据稀疏性的问题.针对该问题,本文提出了一种融合地理信息的隐语义模型-GeoLFM.该模型通过将出租车司机所处的客观地理环境信息,融合到司机-载客区域矩阵分解的过程中,从而弥补数据稀疏性带来的不足.同时,根据出租车实时的空间位置信息,为身处不同地点的出租车推荐不同的载客区域.实验证明,本文提出的方法与常用方法相比,推荐结果与真实的出租车司机载客情况间的平均绝对误差和均方根误差都得到大幅降低,较好的提升了推荐效果.     

Understanding the relationship between complicated crossings and frequently visited locations – a case study with boro taxis in Manhattan

ABSTRACT

Urban mobility has complex patterns and principles. Data of moving entities on the underlying transportation infrastructure can help understanding those complex patterns and principles. Therefore, we need static infrastructural information and knowledge on spatio-temporal movement patterns of public transport services and of various vehicle fleets. We focus on inspecting data partitions of individual taxi movement acquisitions in New York City (NYC), together with OpenStreetMap (OSM) data extracts, for gaining more knowledge about the complex daily mobility patterns in NYC. We select trip information of tracked boro taxi drivers, who are restricted to pick up customers at the airports and the southern part of Manhattan. By computing with taxi customer drop-off positions, we define drop-off clusters as the customer destination hotspots of selected Saturdays in June 2015. These hotspots are then related to the OSM road network, in particular to its derivatives: complicated crossings. By comparing with a previous assumption of detecting ‘fast leaving’ behaviour within the restricted zone, we receive characteristic matching results: only few destination hotspots appear at complicated crossings. Nearly all the matching intersections have nearby situated pedestrian zones and many are associated with previous construction measures. Finally, we reason on the usefulness of the proposed method.     

Autonomic Navigation System Based on Predicted Traffic and VANETs

To avoid an expected traffic jam, drivers make detours based on limited information; however, the majority following the alike routes may result in an unexpected congestion. Conventional navigation approaches are unable to respond to the unexpected congestion because these approaches do not consider the routes taken by other vehicles. Navigation systems that utilize global traffic information can improve gas consumption, CO₂ emissions and travel time. Therefore, in this paper, the authors propose an autonomic navigation system (ANS) operating over vehicular ad-hoc networks (VANETs). The proposed ANS adopts a hierarchical algorithm to plan vehicle routes. The proposed ANS imitates the human nervous system when managing the navigation system, in which vehicles monitor traffic via VANETs. Moreover, this paper proposes a time-dependent routing algorithm that uses a novel traffic prediction method based on the routes of vehicles. This paper adopts EstiNet as simulator tool that dominates hundreds or thousands of VANET-based vehicles routing in two maps, Manhattan area, and Taipei city. The results show that the proposed ANS improves the average speed by 60.02 % when compared with the shortest path first (SPF) algorithm and by 15.49 % when compared with the distributed method of a traffic simulation in the Manhattan area. The proposed ANS also improves the average speed by 30.5 % when compared with the SPF algorithm and by 15.8 % when compared with the distributed method of a traffic simulation in the Taipei area. Furthermore, to emulate real environments, there is a scenario in which only a portion of the vehicles complies with the proposed ANS.     

基于分时MDP的出租车载客预测推荐技术研究

针对出租车盲目寻客导致空载率高的问题,提出了一种出租车载客热点推荐策略,以最大程度优化匹配乘客过程,提高寻客效率。基于出租车历史轨迹数据,结合热点乘客信息的时间序列特性,提出基于循环神经网络的分段预测(SPBR)算法,以及基于分时马尔可夫决策过程(TMDP)的载客推荐模型。实验表明,SPBR算法预测结果的RMSE比SVR、CART和BPNN等算法分别降低了67.6%、71.1%和64.5%;TMDP模型出租车期望回报比历史期望提升了35.9%。     

Spatial experience based route finding using ontologies

Spatial experiences in route finding, such as the ability of finding low‐traffic routes, exert a significant influence on travel time in big cities; therefore, the spatial experiences of seasoned individuals such as taxi drivers in route finding can be useful for improving route‐finding algorithms and preventing using routes having considerable traffic. In this regard, a spatial experience‐based route‐finding algorithm is introduced through ontology in this paper. To this end, different methods of modeling experiences are investigated. Then, a modeling method is chosen for modeling the experiences of drivers for route finding depending on the advantages of ontology, and an ontology based on the taxi drivers’ experiences is proposed. This ontology is employed to create an ontology‐based route‐finding algorithm. The results are compared with those of Google maps in terms of route length and travel time at peak traffic time. According to the results, although the route lengths of route‐finding method based on the ontology of drivers’ experiences in three cases (from nine cases) are greater than that based on Google maps, the travel times are shorter in most cases, and in some routes, the difference in travel time reaches only 10 minutes.     

基于蚁群优化算法的出租车调度方法研究

城市交通工具的合理调度能够有效缓解日益严峻的交通压力,出租车作为公共出行的交通工具满足了大量的出行需求。蚁群算法(ACO)作为仿生算法的代表,根据蚂蚁个体产生的信息素,通过不同策略和信息素更新等操作,逐步接近最优解,适合解决城市交通资源路径规划问题。文章给出一种改进的蚁群算法进行出租车调度,在不同时间段内,对非热点区域向热点区域以及热点区域向非热点区域转移进行研究,根据信息素差异化特征,首先建立了时间区域优化算法和区域调度模型,通过对数据样本的训练得到不同情况下的转移概率和行驶里程,从而确定最优的抑制因子和调节参数,提高出租车转移概率并减少空载行驶距离,实现对出租车资源的合理分配。     

VANET-based secure taxi service

Taxi service is an important type of point-to-point transportation in many cities. One of the major concerns is the safety of both passengers and taxi drivers. In this paper, based on the framework of VANET (Vehicular Ad-hoc NETwork), we propose a novel secure taxi service scheme to tackle the problem. First, a driver needs to authenticate himself using a password before he can provide any service. Second, after picking up a passenger, the taxi’s OBU (on-board unit) needs to continuously send its current location to the trusted server. This location is then double-verified with RSU (road-side unit) signatures. We also design a cooperative tracking mechanism to locate a taxi in case of emergency. In some occasions, taxi passengers as well as taxi drivers are concerned with their privacy (such as travelling route). We incorporate privacy-preserving measures into our scheme. The taxi’s OBU uses pseudo identities instead of real identity for all ongoing transmissions so that a passenger’s travelling route (or the driver’s travelling route when he/she is not carrying any passengers) cannot be traced by a third party easily. We show by simulation that the total delay (from 1 to 9 s) introduced by all security measures is small while the total number of messages required to be transmitted in the network is reasonable.     

Faster than a speeding bullet train

China is throttling up a 430-km/h magnetically levitated train to link Shanghai and its airport. Built in China by a trio of German companies and the Shanghai Maglev Transportation Development Co., it reaches 430 km/h (268 mi/h)-130 km/h faster than Japan's famous bullet train. And even as it goes faster than any commercial vehicle without wings, the Chinese train is smoother and quieter than Amtrak's wheel-on-rail Acela-the state of the art in the United States-which pokes along when it can at a maximum 240 km/h. After many false starts and the completion of full-scale experimental maglev systems in Japan and Germany in the 1980s, maglev in China will finally start shuffling passengers in October in a reasonably large-scale, commercial system. The trains will run from downtown Shanghai's financial district to Pudong International Airport, making a 9-minute run that will shave about 40 minutes off the typical trip time in a taxi. With three five-car trains, each carrying as many as 574 passengers, and trains leaving every 10 minutes, the US $1.2 billion system could carry more than 10 million passengers a year.     

Neural Technologies for Precise Timing in Electric Power Systems with a Single-Frequency GPS Receiver

The global positioning system (GPS), with its ability to provide time synchronization with accuracy to 200 ns over the wide area, provides an ideal tool for performing time tagging in electric power systems. The observed GPS pseudo-range varies from the true range because of range measurement errors. GPS errors sources are ionospheric delays, atmospheric delays, troposphericelays, multi-path effects and dilution of precision etc., affecting the GPS signals as they travel from satellite to user on Earth. In this paper neural models has been presented are for more accurate GPS timing in electric power systems with a single-frequency GPS receiver. The proposed methods use back-propagation (BP), extended Kalman filter (EKF) and particle swarm optimization (PSO) training algorithms, which achieves the optimal training criterion. We use actual data to evaluate the performance of the proposed methods. An experimental test setup is designed and implemented for this purpose. Results using the three methods are discussed. The experimental results obtained from a coarse acquisition (C/A)-code single-frequency GPS receiver are provided to confirm the efficacy of the approaches to give high accurate timing. The GPS timing RMS error using neural network based on PSO learning algorithm reduces to less than 105 and 36 ns, before and after SA, respectively.     

Ring of fresh water [London Water Ring Main]

Thames Water is rejuvenating London's water supply, creating a forward-looking network, with 50% more capacity, aimed at the next century. It will be based around the £250 million London Water Ring Main (LWRM)-part of a decade-long, £4 billion capital expenditure programming. Deeper than the lowest tube lines, longer than the Channel Tunnel and wide enough to take a London taxi, the 2.5 m-diameter, leakfree completed ring will be 80 km long carrying 50% of London's water 40 m below the surface. The ring will pump water to local distribution systems, via strategically placed vertical shafts, at a maximum rate of 1295 million litres per day. The vital role of the control system is to match supply with demand at all times. The control and monitoring system has four levels which are briefly described, as is its operation     

Detecting Hot Road Mobility of Vehicular Ad Hoc Networks

Vehicular Ad Hoc Networks (VANETs) can provide value-added services to both drivers and passengers with on-board vehicular communication systems. Node mobility and volatile wireless connection in VANETs affect inter-contact time (T _I ) between mobile nodes, which greatly degrades the performance of vehicular applications. Nevertheless, the node spatial distribution in VANETs is another important factor especially in real applications. It positively affects the inter-contact time of vehicular nodes. By leveraging it, we can significantly improve the performance of data transmissions and inter-vehicle communication. To this end, we investigate the data collected from around 4,000 taxisin Shanghai and propose in this paper an efficient hot road mobility model. We find that most taxis distribute on some hot roads, which makes the node spatial distribution follow the power law. Based on this observation, we propose the concepts of indirect contact and heterogeneous inter-contact time (T _H ) to reveal how hot roads can change the distribution of inter-contact time. We find that the tail distribution of T _H also appears the power law, and both node spatial distribution and T _H distribution decay at least as the power law. We further propose a model for detecting vehicle mobility in hot roads, which can generates synthetic traces that captures both spatial and temporal features of nodes in VANETs.     

Thermoelectric Performance Optimization in p-Type Ce y Fe3CoSb12 Skutterudites

In this study, we investigated the impact of the Ce filling fraction on the thermoelectric properties of p-type filled skutterudites Ce _y Fe₃CoSb₁₂ (y = 0.6 to 1.0). The electrical conductivity decreases gradually with increasing y, while the Seebeck coefficient displays an opposite variation tendency, consistent with the expected electron donor role of the Ce filler in this compound. The overall power factors are invariable among all the samples. Alteration of the Ce filling fraction exerts little influence on the phonon transport, but the total thermal conductivity markedly declined with increasing y due to the reduced contribution to heat transfer from carriers. As a consequence, the maximum thermoelectric figure of merit ZT reaches ～0.8 for the sample with y = 0.9, comparable to that of pure Fe-based skutterudite CeFe₄Sb₁₂; more importantly, the former possesses a much larger average ZT between 300 K and 800 K than the latter, showing superior potential for use in intermediate-temperature thermoelectric power generation applications. Further enhancement of ZT in p-type Fe₃CoSb₁₂-based skutterudites could be realized via nanostructuring or a multiple-filling approach.     

The discrete multidimensional MPUM

Given a finite set of polynomial-exponential, multivariate, and vector-valued sequences, we show how the smallest linear shift-invariant set containing the data trajectories can be written as the solution set of a system of linear difference equations with constant coefficients. The resulting representation is known as the most powerful unfalsified model (MPUM) in behavioral systems theory. We address the case where the components of the given sequences take their values in a field, as well as the case where these values belong to a finite ring of the form ${{\mathbb{Z}}/m{\mathbb{Z}}}$ for an integer m > 1.     

A 7.9 μA multi-step phase-domain ADC for GFSK demodulators

The phase-domain analog-to-digital converter (Ph-ADC) is proved to be more power efficient than traditional amplitude ADCs in wireless receivers . A low power multi-step Ph-ADC for zero intermediate frequency (IF) GFSK receivers as defined in Bluetooth low energy protocol is proposed in this paper. With dedicatedly designed binary code scheme and multi-step operation, the Ph-ADC requires only 52 current elements and one comparator, in contrast to the design in literature using 260 current elements and 8 comparators. Non-idealities due to transconductance errors and offset errors are theoretically analyzed, followed by a design strategy to minimize trip point errors. Simulation results show that the digital intensive Ph-ADC consumes only 7.9 μA current from a 1.8 V supply when implemented in a 180 nm CMOS process. Monte-Carlo simulations show that the maximum trip point error is only 2.3°, which is less than 1/8 least significant bit. When the Ph-ADC is used in a GFSK demodulator, the required IF E_b/N₀ is 13.5 dB to achieve a bit error rates of 0.1%.     

Reading DRAM cells using two properly designed cascaded inverters

During the reading of one-transistor one-capacitor dynamic random-access memory (1T–1C DRAM) cells, the need arises to amplify a small voltage difference (in the order of 30 to 100 mV) by a suitable sense amplifier. The net result is that the higher voltage will rise to V _DD while the lower one will decrease to 0 V. Simulation results for the 0.13 μm CMOS technology reveals that approximately 40 % of the read access time is associated with the sense-amplifier operation in addition to the area required by each sense amplifier for each column in the memory array due to the need to raise or lower the relatively large bitline parasitic capacitance. Also, this process consumes relatively large amount of power. In this paper, a novel readout technique for reading DRAM cells using two cascaded inverters is proposed. Simulation using the 0.13 μm CMOS technology with V _DD =1.2 V reveals that 37 % of the read access time using the conventional sense-amplifier reading scheme is saved for stored “1” which has a longer access time compared to stored “0”.     

Influence of Li-B-Si Additions on the Sintering and Microwave Dielectric Properties of Ba-Nd-Ti Ceramics

Li₂O-B₂O₃-SiO₂ (LBS) synthesized via a solid-state reaction process was chosen as a novel sintering aid for tungsten-bronze-type Ba₄Nd_9.3Ti₁₈O₅₄ (BNT) ceramic. The effects of LBS additions on the sintering behaviors, microstructures, and microwave dielectric properties of the BNT ceramic have been investigated, indicating that LBS addition obviously lowered the sintering temperature of the BNT ceramic without damaging its microwave dielectric properties. BNT ceramic doped with 3 wt.% and 4 wt.% LBS addition could be well sintered at 975°C and 950°C for 3 h and had excellent properties: ε _r = 65.99, Q × f = 4943 GHz (f = 4.4 GHz), τ _f = 19 ppm/°C, and ε _r = 64.56, Q × f = 4929 GHz (f = 4.3 GHz), τ _f = 11 ppm/°C, respectively.