Showing papers by "Zhiyuan Liu published in 2019"

DeepPF: A deep learning based architecture for metro passenger flow prediction

Abstract: This study aims to combine the modeling skills of deep learning and the domain knowledge in transportation into prediction of metro passenger flow. We present an end-to-end deep learning architecture, termed as Deep Passenger Flow (DeepPF), to forecast the metro inbound/outbound passenger flow. The architecture of the model is highly flexible and extendable; thus, enabling the integration and modeling of external environmental factors, temporal dependencies, spatial characteristics, and metro operational properties in short-term metro passenger flow prediction. Furthermore, the proposed framework achieves a high prediction accuracy due to the ease of integrating multi-source data. Numerical experiments demonstrate that the proposed DeepPF model can be extended to general conditions to fit the diverse constraints that exist in the transportation domain.

An effective spatial-temporal attention based neural network for traffic flow prediction

Abstract: Due to its importance in Intelligent Transport Systems (ITS), traffic flow prediction has been the focus of many studies in the last few decades. Existing traffic flow prediction models mainly extract static spatial-temporal correlations, although these correlations are known to be dynamic in traffic networks. Attention-based models have emerged in recent years, mostly in the field of natural language processing, and have resulted in major progresses in terms of both accuracy and interpretability. This inspires us to introduce the application of attentions for traffic flow prediction. In this study, a deep learning based traffic flow predictor with spatial and temporal attentions (STANN) is proposed. The spatial and temporal attentions are used to exploit the spatial dependencies between road segments and temporal dependencies between time steps respectively. Experiment results with a real-world traffic dataset demonstrate the superior performance of the proposed model. The results also show that the utilization of multiple data resolutions could help improve prediction accuracy. Furthermore, the proposed model is demonstrated to have potential for improving the understanding of spatial-temporal correlations in a traffic network.

A tailored machine learning approach for urban transport network flow estimation

Abstract: This study deals with urban transport network flow estimation based on Cellphone Location (CL) and License Plate Recognition (LPR) data. We first propose two methods to filter CL data and extract the spatio-temporal traffic features for a specific road segment. A tailored machine learning approach is developed, including two components: a tangible multi-grained scanning ensemble learning model and a novel two-stage zero-shot learner. The former aims to estimate traffic flow on a single link with both filtered CL data, extracted spatio-temporal traffic features, and LPR data by incorporating the unique merits thereof. The latter is capable of estimating traffic flow on those links with only CL data by considering the spatial features of these links and relevant land-use information. Finally, case studies are analysed to demonstrate the impressive performance of the tailored machine learning approach.

Surrogate-based simulation optimization approach for day-to-day dynamics model calibration with real data

Abstract: This paper investigates the day-to-day dynamics model from the perspective of travelers’ actual route choice behaviors, and calibrates and validates the route-based day-to-day dynamics model with the real-world license plate recognition (LPR) data. Due to the highly nonlinear and multi-modal response function in the calibration of the optimization problem, traditional gradient-based nonlinear regression algorithms or other analytical optimization approaches are inapplicable to deal with the calibration work. In this paper, a surrogate-based simulation optimization approach is proposed to deal with the expensive-to-evaluate response function in the day-to-day dynamics calibration work. More specifically, the kriging metamodel is adopted to surrogate the optimization function of the calibration process. With this meta-modeling approach, a sound solution can be achieved with only a few sampling points in a comfortably afforded computation burden, thus giving a valid estimation of the parameters in the day-to-day dynamics model. Finally, a case study based on the real-world LPR data is conducted to validate the proposed model and calibration method.

A Two-Stage Destination Prediction Framework of Shared Bicycles Based on Geographical Position Recommendation

Abstract: In this study, we investigate the destination prediction problem for the trips of shared bicycles, which can assist authorities in overcoming the existing challenges like indiscriminate parking, bicycle rebalancing, etc. An integral part of E-business, the recommendation system can predict the buyers' needs and recommend personalized contents for each buyer. Likewise, prediction of travelers' destinations is essentially a geographical position recommendation. In this paper, we present a two-stage destination prediction framework inspired by the recommender system. It is an innovative framework for the destination prediction problem in transportation studies. The architecture is highly flexible and extendable while it utilizes the user's historical data such as departure time, geographic position, and traveling behavior. The experiments demonstrate that the proposed method performs well in the real situation.

A cell-based dynamic congestion pricing scheme considering travel distance and time delay

Abstract: This study introduces the dynamic congestion pricing (DCP) problem with the consideration of the actual travel distance and time delay (i.e. a joint distance and time-delay toll, JDTDT) in a dynami...

Prediction of City-Scale Dynamic Taxi Origin-Destination Flows Using a Hybrid Deep Neural Network Combined With Travel Time

Abstract: Predicting city-scale taxi origin-destination (OD) flows takes an important role in understanding passengers' travel demands as well as managing taxi operation and scheduling. But the complex spatial dependencies and temporal dynamics make this problem challenging. In this paper, a hybrid deep neural network prediction model based on convolutional LSTM (ConvLSTM) is proposed. For improving the prediction accuracy, the implicit correlation between travel time and OD flow is explored and they are combined as inputs of the prediction model. Moreover, in order to realize OD flows prediction at the road network level, and solve the problem that grid-based representation method cannot distinguish traffic flow at different heights, such as in multi-layer overpass areas, this paper presents a grid and road nested method to represent ODs. With the time of day partitioned into time slots, OD flows are extracted and predicted in both spatial and temporal domain. In the experiment, real taxi data are used to verify the proposed model and prediction method fully. And the experimental results show that the proposed model can effectively predict city-wide taxi OD flow, and outperforms the typical time sequence models and existing deep neural network models.

Traffic Volume Estimation in Multimodal Urban Networks Using Cell Phone Location Data

Abstract: This paper proposes a method to estimate multimodal traffic volume on urban road networks using cellphone location data. The study considers the fact that there can be more than one phone user in each driving vehicle. Firstly, a temporal and spatial method is used to distinguish whether the cellphone signals are sent by cellphones in running vehicles on urban roads. Secondly, a minimum spanning tree clustering method is proposed to calculate the number of commuters in each vehicle. Based on the proposed method, the various travel modes, including drive alone, carpooling, and bus, can be identified while their hourly traffic volume can be estimated. Finally, the predicted traffic volumes are compared with the actual values obtained from License Plate Recognition system. The experiment result shows that the proposed method can accurately estimate the hourly traffic volumes of different travel modes and the estimation errors of this method are within a reasonable range.

An Analytical Gate Delay Model in Near/Subthreshold Domain Considering Process Variation

Abstract: Voltage scaling technique is widely employed in state-of-the-art low power circuits with excellent power reduction. However, voltage scaling to sub-threshold (STV) and near-threshold (NTV) domain introduces performance degradation and high process variation sensitivity. Accurate modeling of the statistical characteristics especially the probability distribution function (PDF) and the cumulative distribution function (CDF) is urgently required with process variation consideration. In this paper, a novel analytical model is derived based on log-skew-normal (LSN) distribution to precisely evaluate the gate delay variation. The multi-variate threshold variation in stacked gates are modeled with a linear approximation method in delay distribution derivation. By applying the CDF of the proposed model, the maximum and minimum delay indicated by ±3σ percentile point can be calculated essentially different from the common method with much higher accuracy. Experimental results show the proposed model is highly fitted with Monte Carlo (MC) results for stochastic delay modeling of generic logic gates in near/subthreshold regime with less than 8% and 6% error in delay variability and ±3σ delay prediction, showing maximum accuracy improvement about 40 times compared to preproposal models.

Long-Term Tai Chi Experience Promotes Emotional Stability and Slows Gray Matter Atrophy for Elders.

Abstract: Brain adverse structural changes, especially the atrophy of gray matter, are inevitable in aging. Fortunately, the human brain is plastic throughout its entire life. The current cross-section study aimed to investigate whether long-term Tai Chi exercise could slow gray matter atrophy and explore the possible links among gray matter volume (GMV), long-term Tai Chi experience and emotional stability in a sequential risk-taking task by using voxel-based morphometry. Elders with long-term Tai Chi experience and controls, who were matched to Tai Chi group in age, gender, physical activity level, participated in the study. A T1-weighted multiplanar reconstruction sequence was acquired for each participant. Behaviorally, the Tai Chi group showed higher meditation level, stronger emotional stability and less risk-taking tendency in the sequential risk-taking compared to the control group. Moreover, the results revealed that the GMV of the thalamus and hippocampus were larger in the Tai Chi group compared with the control group. Notably, the GMV of the thalamus was positively correlated with both meditation level and emotional stability. The current study suggested the protective role of long-term Tai Chi exercise at slowing gray matter atrophy, improving the emotional stability and achieving successful aging for elders.

Practical taxi sharing schemes at large transport terminals

Abstract: This paper addresses taxi sharing at large transport terminals, where passengers are queuing for taxis. It aims to develop an allocation method for the optimal taxi sharing plan. Two key questions ...

Analysis of the Influencing Factors of Carpooling Schemes

Abstract: Carpooling, as an emerging travel mode, has not yet been paid sufficient attention despite its merits of convenience, cost-efficiency, and sustainability. The studies of carpooling, especially on the factors influencing carpooling, are inadequate. To tackle this problem, what and how the factors influence travelers' mode choice are studied with a practical trial. Based on a qualitative survey, we found that the travel time, travel cost, safety, punctuality, comfort, social contact, and identity promotion are potential influencing factors. Then a stated preference (SP) survey is conducted to address the main factors influencing carpooling. Finally, three key factors (travel time, travel cost, and safety) are singled out and the interaction of travel time and travel cost are explored by a binary logit regression analysis, respectively. Following the findings of the main factors and their interaction, related suggestions are proposed to promote the future application of carpooling.

The emotional stability of elders with tai chi experience in the sequential risk-taking task.

Abstract: The present study investigated the difference of emotional stability between elders with long-term tai chi experience and elders in a control group during the sequential risk-taking task. Twenty-eight tai chi practitioners (age = 67.36 ± 5.91 years, exercise years = 9.23 ± 4.19) and 28 control participants (age = 65.21 ± 3.55 years, other physical exercise without meditation component) participated in the experiment. Participants were asked to open a series of boxes consecutively and decided when to stop. Each box contained a reward, except for one, which contained a "devil." This box would eliminate the participant's score in the trial. Once participants stopped, both collected gains and missed chances were revealed. Then, participants rated how they felt on a 9-point scale from extreme regret to extreme relief. Additionally, participants filled out the Chinese version of the Beck Depression Inventory, the Barratt Impulsiveness Scale Version 11, the NEO Five-Factor Inventory, and the Five Facet Mindfulness Questionnaire. The tai chi group showed a higher meditation level, stronger emotional stability, and less risk propensity (all ps < .05) than the control group. In the tai chi group, meditation level was positively correlated with emotional stability (r = -.47, p < .01). Moreover, subjective emotion was negatively correlated with subsequent behavior (tai chi: β = -.44, p < .001; control: β = -.27, p < .001). Subjective emotion rating entirely mediated the relationship between objective outcome and subsequent behavioral change in the control group, whereas it partially mediated such relationship in the tai chi group. The current cross-sectional study found a difference of behavior and feeling between elders with long-term tai chi experience and elders in the control group, which suggests that long-term tai chi experience could enhance elders' emotional stability and reduce their risk propensity during the sequential risk-taking task.

Analysis of the acceptance of park-and-ride by users: A cumulative logistic regression approach

Abstract: Park-and-ride (P&R) schemes are an important way of increasing the public transport mode share, which relieves the negative impact caused by excessive automobile usage. Several existing studies have been conducted in the past to explore the factors that can influence the acceptance of P&R by travelers. However, quantitative analyses of the pertinent factors and rates of traveler choice are quite rare. In this paper, the data collected from a survey in Melbourne, Australia, is used to analyze the acceptance of P&R by travelers going to the central business district (CBD). In particular, we explore the influence that specific factors have on the choice of travel by those who are currently using P&R. The results indicate that the parking fee in the CBD area, travel time on public transport, and P&R transfer time affect traveler use of P&R. A quantitative assessment of the impact of these three factors is conducted by using a cumulative logistic regression model. Results reveal that the P&R transfer time has the highest sensitivity while public transport travel time has the least. To maximize the use of P&R facilities and public transport, insights into setting parking fees and designing P&R stations are presented.

Station-Level Hourly Bike Demand Prediction for Dynamic Repositioning in Bike Sharing Systems

Abstract: Regarding the issue of unbalanced bike distribution in the bike sharing system (BSS), the accurate prediction of bike demand is of importance to dynamic repositioning. This research focuses on the prediction accuracy of the hourly bike number change on station level. Three frequently used machine learning models, including Random Forest (RF), Gradient Boosting Regression Tree (GBRT) and Neural Network (NN) are trained on the same preprocessed dataset. Meanwhile, two training methods, training a check-in prediction model as well as a check-out prediction model, respectively, and training the model with processed bike number change data directly, are both applied to these three models in order to improve the prediction accuracy. The results show that the method of training on the bike number change data is more effective and the GBRT model trained in this way outperforms other models.

Building Effective Short Video Recommendation

Abstract: How to build an effective personalized recommendation system is a challenging but highly valuable problem in social media services. This paper focuses on constructing a universal framework for short video recommendation by predicting the probability of finishing watching the entire video and pressing the 'like' button. Four novel techniques are proposed to improve the prediction accuracy. Firstly, we present an Incremental Multi-Window Scanning approach to extract the features pertaining to the users' behaviors. Also, a User Interaction Behavior Hierarchy is designed to capture a larger quantity of information and reduce the computing time. Additionally, the model transfer is capable of transferring the knowledge learned by the model on other datasets to the final model. Lastly, a rank-based ensemble approach which is suitable for tasks based on the evaluation metric of AUC is proposed. Our method long ranked first in the final stage of ICME Short Video Understanding Challenge (Track1) before the revision of competition rule.

Cost-Effectiveness Analysis of Plug-In Hybrid Electric Vehicles using Vehicle Usage Data Collected in Shanghai, China:

Abstract: This paper investigates the economic viability of plug-in hybrid electric vehicles (PHEV) in Shanghai, China based on a real-world in-use PHEV dataset. To quantify PHEV drivers’ gross profit compar...

Building Effective Large-Scale Traffic State Prediction System: Traffic4cast Challenge Solution.

Abstract: How to build an effective large-scale traffic state prediction system is a challenging but highly valuable problem. This study focuses on the construction of an effective solution designed for spatio-temporal data to predict large-scale traffic state. Considering the large data size in Traffic4cast Challenge and our limited computational resources, we emphasize model design to achieve a relatively high prediction performance within acceptable running time. We adopt a structure similar to U-net and use a mask instead of spatial attention to address the data sparsity. Then, combined with the experience of time series prediction problem, we design a number of features, which are input into the model as different channels. Region cropping is used to decrease the difference between the size of the receptive field and the study area, and the models can be specially optimized for each sub-region. The fusion of interdisciplinary knowledge and experience is an emerging demand in classical traffic research. Several interdisciplinary studies we have been studying are also discussed in the Complementary Challenges. The source codes are available in this https URL.

Analytical Gate Delay Variation Model with Temperature Effects in Near-Threshold Region Based on Log-Skew-Normal Distribution

Abstract: The near-threshold design is widely employed in the energy-efficient circuits, but it suffers from a high sensitivity to process variation, which leads to 2X delay variation due to temperature effects. Hence, it is not negligible. In this paper, we propose an analytical model for gate delay variation considering temperature effects in the near-threshold region. The delay variation model is constructed based on the log-skew-normal distribution by moment matching. Moreover, to deal with complex gates, a multi-variate threshold voltage approximation approach of stacked transistors is proposed. Also, three delay metrics (delay variability, ± 3 σ percentile points) are quantified and have a comparison with other known works. Experimental results show that the maximum of delay variability is 5% compared with Monte Carlo simulation and improves 5X in stacked gates compared with lognormal distribution. Additionally, it is worth mentioning that, the proposed model exhibits excellent advantages on − 3 σ and stacked gates, which improves 5X–10X in accuracy compared with other works.

Estimation Models for the Safety Level of Indoor Space Pedestrian Flows

Abstract: This paper presents a method to safely organize the indoor space pedestrian flows. To quantitatively describe the indexes related to pedestrian safety, a capacity estimation model is proposed for the number of participants, and then the characteristic of pedestrians of different age is considered in analysing pedestrian speed by using step frequency data. Furthermore, we develop a pedestrian safety estimation model and obtain the probability of safe crossing. Based on the estimation results of capacity model and safety estimation model, a reasonable scheme of pedestrian organization is put forward. Finally, a case study is adopted to apply the proposed models.

Accurate and Efficient Interdependent Timing Model for Flip-Flop in Wide Voltage Region

Abstract: The interdependency of the setup and hold constraints and clock-to-q (c2q) delay for flip-flops (FF) has been studied in static timing analysis (STA) to facilitate timing closure and improve circuit performance. However circuits operating in near-threshold voltage (NTV) region pose severe challenge to the interdependent timing modelling due to its significant nonlinear relation and much wider constraint range compared with super-threshold voltage (STV) region. In this paper an accurate and efficient modelling approach is proposed by employing artificial neuron network (ANN) to characterize the interdependency among the setup time hold time and c2q delay of FF in wide voltage region. Experimental results show the prediction errors of c2q delay with varied setup and hold constraints are lower than 2.1% and 1.8% in NTV and STV regions with 3.51x and 1.09x accuracy improvement compared with the prior work. Besides the modelling efficiency is validated by 1.41 x/1.10x simulation cost reduction and 21.08x/4.14x library storage saving for NTV/STV domain with accuracy advantage.

Distance-Based Congestion Pricing with Day-to-Day Dynamic Traffic Flow Evolution Process

Abstract: This paper studies the distance-based congestion pricing in a network considering the day-to-day dynamic traffic flow evolution process. It is well known that, after an implementation or adjustment of a new congestion toll scheme, the network environment will change and traffic flows will be nonequilibrium in the following days; thus it is not suitable to take the equilibrium-based indexes as the objective of the congestion toll. In the context of nonequilibrium state, prior research proposed a mini–max regret model to solve the distance-based congestion pricing problem in a network considering day-to-day dynamics. However, it is computationally demanding due to the calculation of minimal total travel cost for each day among the whole planning horizon. Therefore, in order to overcome the expensive computational burden problem and make the robust toll scheme more practical, we propose a new robust optimization model in this paper. The essence of this model, which is an extension of our prior work, is to optimize the worst condition among the whole planning period and ameliorate severe traffic congestions in some bad days. Firstly, a piecewise linear function is adopted to formulate the nonlinear distance toll, which can be encapsulated to a day-to-day dynamics context. A very clear and concise model named logit-type Markov adaptive learning model is then proposed to depict commuters’ day-to-day route choice behaviors. Finally, a robust optimization model which minimizes the maximum total travel cost among the whole planning horizon is formulated and a modified artificial bee colony algorithm is developed for the robust optimization model.