Road Traffic Flow Research Articles

Hierarchical User-Driven Trajectory Planning and Charging Scheduling of Autonomous Electric Vehicles

Autonomous electric vehicles (A-EVs), regarded as one of the innovations to accelerate transportation electrification, have sparked a flurry of interest in trajectory planning and charging scheduling. In this regard, this work employs mobile edge computing (MEC) to design a decentralized hierarchical algorithm for finding an optimal path to the nearby A-EV parking lots (PLs), selecting the best PL, and executing an optimal charging scheduling. The proposed model makes use of unmanned aerial vehicles (UAVs) to assist edge servers in trajectory planning by surveying road traffic flow in real time. Furthermore, the target PLs are selected using a user-driven multiobjective problem to minimize the cost and waiting time of A-EVs. To tackle the complexity of the optimization problem, a greedy-based algorithm has been developed. Finally, charging/discharging power is scheduled using a local optimizer based on the PLs’ real-time loads, which minimizes the deviation of the charging/discharging power from the average load. The obtained results show that the proposed model can handle charging/discharging requests of on-move A-EVs and bring fiscal and nonfiscal benefits for A-EVs and the power grid, respectively. Moreover, it observed that user satisfaction in terms of traveling time and traveling distance is increased by using the edge-UAV model.

How does trip frequency distort time value to impact congestion charging schemes?

Abstract A notable feature of a city or region with close economic and social connections with its neighbours is its highly mixed local and external traffic, and in some cases the external traffic volume is almost as high as that of local traffic. Whilst local traffic volume may be largely made up of the same regular local commuters making frequent trips, external traffic from outside of the city (region) may not be the same people making regular trips to/from the city. However, from a large pool of people making infrequent trips to/from the city, the existence of external traffic is proven by data from the licence plate recognition system of road vehicles in Changde, China. The function of value of time correlated with the income/wage rate and trip frequency is exploited and verified statistically. The time value distorted by trip frequency is defined as perceived time value (PTV), which also influences the way travellers perceive any travel impedance such as congestion delay and toll charges. This paper analyses the price of anarchy (POA) when explicitly considering the travel frequency of the trip-makers and their PTV, and compares with previous analysis without considering travel frequency. We show that when travel frequency is considered, the optimal toll of congested road pricing schemes which converts road traffic flow from user equilibrium into system optimization is much lower than that without considering travel frequency. The cost of licence plate auction cannot be treated as a congestion toll, which is only a threshold of vehicle ownership. That travellers choose routes by PTV rather than TV (time value) is proven by an example of Heishipu Bridge in Changsha, Hunan Province, China.

Urban Road Traffic Flow Prediction with Attention-Based Convolutional Bidirectional Long Short-Term Memory Networks

Accurate traffic prediction is critical for industry practitioners and researchers in intervening and dredging future traffic in advance to avoid traffic congestion. Considering that most prediction models fail to effectively capture the complex nonlinearity of traffic data and thus cannot obtain satisfactory prediction results, we propose a novel deep-learning architecture for traffic flow prediction, called AC-BLSTM (attention-based convolutional bidirectional long short-term memory). The proposed model captures traffic information through multilayer network architectures composed of convolutional bidirectional long short-term memory (conv-BiLSTM) network and attention mechanism. The spatiotemporal features of traffic flow are extracted by convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) network. Then attention mechanism combines the outputs of CNN and BiLSTM to assign corresponding weights to the features extracted at different times. In addition, we employ a parallel sub-module structure to model three temporal properties of traffic flow, that is, weekly, daily, and recent dependencies. Finally, the results of these three parts are fused to predict the traffic flow values through the fully connected (FC) layers. Experiment results using a real-world urban road traffic dataset demonstrate that compared with other competing models, the proposed model has better prediction performance.

Implementing the CNOSSOS-EU correction near traffic light junctions and roundabouts

Under the Environmental Noise Directive (END), the Round 4 strategic noise maps are to be calculated using the methodology set out in Directive 2015/996 (as amended) (CNOSSOS-EU). As part of the road traffic source emission model, a correction for the effect on noise levels due to the acceleration and deceleration of vehicles shall be applied near crossings with traffic lights and roundabouts. As part of the revision of the EPA guidelines on strategic noise mapping in Ireland, a technical investigation was carried out to identify the effect of the correction factor on road traffic noise emissions, and determine whether there was a practical approach to implement the correction factor within the upcoming noise mapping. This paper will provide an overview of the results, including the change in noise level emissions for low speed low flow roads, and high speed high flow road. It will also present a practical approach to preparing the road traffic flow models in GIS based on available datasets.

Sensitivity of Input Parameter on CNOSSOS-EU Road Emission Levels

The United Kingdom Department for Food and Rural Affairs (Defra) commissioned a series of studies investigating the sensitivity of the CNOSSOS-EU noise assessment method. CNOSSOS-EU presents challenges in terms of input data accuracy and availability. For this reason, the studies were commissioned to support data decision making and quantify potential uncertainty in Defra's national noise model. The quality framework set out in Directive 996/2014 requires uncertainty in road emission levels at source to correspond to an uncertainty of ±2dB(A). By calculating emissions across a range of values for input parameter within various scenarios in which all other parameters were kept constant, changes in road traffic noise emissions were observed. This was performed separately for each of the five CNOSSOS-EU vehicle categories as well as for a selection of hypothetical composite road traffic flows. A similar analysis was then performed for a series of road surface types. The results were used to indicate which input parameters road traffic noise emissions were most sensitive to. The results identified that uncertainty in traffic flow speed has a tolerance of ±11 kmh-1 for cars and two-wheelers, and ±24 kmh-1 for HGVs in order to meet the quality framework. Significant variation in emissions were found across surface types, indicating the importance of obtaining representative road surface data. The study found that traffic volumes should be within 60% of their true value. The effect of road gradient was found to be dependent on flow composition and speed. Road traffic noise emissions were found to be insensitive to changes in air temperature. Composite flow analysis using English traffic statistics with various proportions of HGVs showed that these assumptions become more important on lower speed roads, while the proportion of two-wheelers is relatively insignificant with respect to overall noise emissions.

Different types of criteria for dealing with anomalous noise events in urban environments under stable road traffic flow conditions

The management of anomalous noise events in the assessment of environmental noise of specific sources is an important issue for determining the actual exposure of the population to noise pollution. Of all the sources currently considered in the European Noise Directive, road traffic is the most important in cities, both in terms of its temporal and spatial presence, as it is generally maintained 24 h a day and affects almost all streets in urban environments. This paper proposes two types of criteria for dealing with these events in urban environments under stable vehicle flow conditions. The results show the convenience of using each type of the suggested criteria for the elimination of anomalous noise events depending on the objective of the research. Those exclusion criteria based on applying a threshold to the average sound level, may be considered the best among the proposed options to study the variation of the annual average LAeq,1h. However, to study the relation between road traffic noise and temperature, a novel proposal for anomalous event elimination criteria based on dividing the full temperature range into small intervals may be a relevant approach. Overall increases of 8 %, respectively, 14 %, in the explanation of variability were found for the criteria C6 (successive application of the thresholds ±10 dB, ±6 dB and ±3 dB at each temperature interval of 3 °C) with respect to the cases where anomalous sound events were not excluded, respectively, using criteria C1 from the scientific literature. The application of this type of criteria for the management of anomalous noise events could also be useful for the study of the relation between road traffic noise and other objective and subjective variables in cities, especially in those areas with a greater daily thermal oscillation.

Dynamic emission characteristics and control strategies of air pollutants from motor vehicles in downtown Beijing, China

This study combined the real-time monitoring and investigation of traffic flows to comprehensively analyze the road traffic flow and vehicle structure in downtown Beijing. A dynamic emission inventory of motor vehicle air pollutants in downtown Beijing in 2021 was established, and the impact of these emissions on air quality was simulated and quantified, and different emission reduction control scenarios were proposed to evaluate their environmental improvement effects and explore measures to mitigate the impact of pollution emissions. The results show that the high traffic flow and the structure of the motor vehicle emissions in downtown Beijing are the main causes of severe motor vehicle pollution. Monitoring data shows that traffic flow in central Beijing is dominated by small passenger vehicles, while the vehicle mix is better than in other regions, with 72.0% of vehicles meeting "National V" or higher emission standards. However, to achieve higher air quality goals, further reducing vehicle emissions is necessary. Based on dynamic traffic flow, the average daily emissions of nitrogen oxides (NOX), particulate matters(PM2.5) and volatile organic compounds(VOC) from motor vehicles in central Beijing are 17.7 tons, 0.6 tons and 14.0 tons, respectively, accounting for 23.0% of the city's average daily motor vehicle emissions. If a zero-emission zone for motor vehicles were implemented in central Beijing, the annual average emission level of pollutants will be reduced by 10.4% to 21.0%. The designation of ultra-low emission zones for motor vehicles could be effective in improving the air quality in the center of Beijing.

ФОРМИРОВАНИЕ ОНЛАЙН-КАРТ ЗАГРЯЗНЕНИЯ АТМОСФЕРНОГО ВОЗДУХА Г. ИЖЕВСКА, ФОРМИРУЕМОГО ВЫБРОСАМИ ПЕРЕДВИЖНЫХ ИСТОЧНИКОВ

The study deals with the assessment of spatio-temporal dynamics of atmospheric pollution caused by emissions from road traffic flows. Nitrogen dioxide concentrations were calculated in the atmospheric air near Udmurtskaya street of the city of Izhevsk. The work was based on a mathematical model created for the calculation of emissions and their dispersion taking into account the dynamics of traffic flows, wind conditions. As the initial data, we used video recordings from continuous automatic recording of the traffic flow intensity. When calculating the emission of pollutants, we applied the model presented in the Methodology for Determining Emissions of Pollutants into the Atmospheric Air from Mobile Sources for Performing Summary Calculations of Atmospheric Air Pollution (approved by Order of the Ministry of Natural Resources of the Russian Federation No. 804 of November 27, 2019). To build a scattering calculation model, we used Methods for Calculating Dispersion of Emissions of Harmful (Polluting) Substances in the Atmospheric Air (approved by Order of the Ministry of Natural Resources of the Russian Federation No. 273 of June 6, 2017). To automate the process of calculating the scattering of impurities in the atmospheric air, a program code was created using the Go programming language. The data obtained by means of the implementation of the program code were visualized in the form of a dynamic map of the concentration distribution posted on the website https://komfortsreda.udsu.ru/ in the section Izhevsk/Atmospheric air/Dynamics of NO2 concentrations. This study deals with a new promising direction in the assessment of natural environments – assessment performed through online modeling. Emission dispersion modeling and assessment of the current state of atmospheric air are important as contributing to the general characterization of the ecological state of an urban area.

基于扩展计划行为理论的驾驶员跟驰意向分析

Road traffic flow will show the phenomenon of mixed driving of autonomous vehicles and traditional vehicles in the future. In order to explore psychological factors that affect traditional vehicle drivers' intention to follow autonomous vehicles in the human-machine hybrid driving environment, three variables of driver's understanding, risk perception, and acceptance of autonomous vehicles are introduced to construct an extended theoretical framework of planning behavior. Three hundred thirty-one subjective evaluation data are obtained through questionnaire survey. The internal consistency and reliability of the data are tested by SPSS and AMOS softwares. Finally, the path analysis and mediation effect analysis are carried out by using structural equation model to test relationship among the influencing factors. The results show that the structural equation model of driver's car-following intention based on the theory of extended planning behavior has good explanatory power for the driver's car-following intention in the mixed driving environment. Driver's behavioral attitude, subjective norm and perceived behavioral control of autonomous vehicles have a significant positive and direct effect on driver's car-following intention. Driver's risk perception and acceptance of autonomous vehicles have a significant indirect effect on driver's car-following intention through intermediary variables, in which the effect of risk perception is negative, and the effect of acceptance is positive. The degree of understanding has both a significant positive direct effect and a significant positive indirect effect on driver's car-following intention. The research results can be used as the basis for vehicle interaction behavior analysis in the human-machine hybrid driving environment.

Road-Type Classification with Deep AutoEncoder.

Machine learning algorithms are among the driving forces towards the success of intelligent road network systems design. Such algorithms allow for the design of systems that provide safe road usage, efficient infrastructure, and traffic flow management. One such application of machine learning in intelligent road networks is classifying different road network types that provide useful traffic information to road users. We propose a deep autoencoder model for representation learning to classify road network types. Each road segment node is represented as a feature vector. Unlike existing graph embedding methods that perform road segment embedding using the neighbouring road segments, the proposed method performs embedding directly on the road segment vectors. The proposed method performs embedding directly on the road segment vectors. Comparison with state-of-the-art graph embedding methods show that the proposed method outperforms graph convolution networks, GraphSAGE-MEAN, graph attention networks, and graph isomorphism network methods, and it achieves similar performance to GraphSAGE-MAXPOOL.

Definisanje dnevnih kapaciteta za potrebe planiranja puteva u Republici Srbiji

In every edition of the Highway Capacity Manual (HCM) until 2016, three levels of capacity analysis were defined: operational, project and planning analysis. The level of detail of each analysis is a function of purpose, breadth of access and available information. Planning analysis determines capacity in the earliest stages of planning, when small amount of data and information is available. The planning procedure is based on the AADT forecast and macro traffic simulation. Often, the usage of simulation models can lead to errors, and for these reasons, when analyzing traffic demands, it is necessary to carry out a check when more informations are available. Planning analyzes from the HCM manual are generally focused on broad issues such as initial problem identification, road design concept, and when alternatives are compared. These analyzes are suitable for evaluating proposed system measures, such as heavy vehicle lane usage control, highway ramp metering, and other applications of intelligent transportation systems. The methodology recommended in HCM implies the use of default input values, i.e. table of generalized values. Given that the recommended capacity values are based on typical road profiles and traffic flow characteristics in the USA, the paper provides a detailed capacity analysis at the planning level for local conditions - four standard geometric two-lane and highway profiles. Capacities are calculated for different values of design hourly volume, design speed, peak hour factor, commercial vehicle impact and terrain type, in order to compare with the generalized capacity tables from HCM2010. The comparative analysis indicates differences in the calculated capacity values, and for that reason the obtained values are recommended for use in model simulations for local conditions in the Republic of Serbia.

Intelligent Based Real Time Traffic Monitoring in Smart Cities

PC-based Intelligent Traffic Monitoring System is unable to deal with the on-going issues surrounding congestion. The current traffic light models are not suited to tackle problems such as traffic jams, ease of access for emergency vehicles and prevention of accidents. In order to counteract these issues, we propose ‘PC-based Intelligent Traffic Monitoring System with Real-time Analysis for Smart Cities’. The system consists of a network of cameras installed at strategic locations across the city, which capture images of vehicles and pedestrians. These images are then processed by intelligent algorithms, which analyse the data and provide real-time traffic information. The system is also equipped with a real-time monitoring interface, which displays the traffic data in an easy-to-understand format. The system is designed to help cities optimize their traffic flow and reduce congestion. By analysing real-time traffic data, city officials can make informed decisions about traffic management and can take immediate action to alleviate congestion. This project will focus on two aspects of implementation. First and foremost, to make traffic light controlling more efficient, image processing alongside with embedded system, will be used. This system will intelligently decide when to alternate signals based on the emergency vehicle on each lane which will increase road capacity and traffic flow. Secondly the system favours school students by giving green signal to give them free passage. It is a matter of concern, as emergency vehicles are being categorized as priority vehicles, so their waiting time should be minimum.

Deep Q‐Learning Network Model for Optimizing Transit Bus Priority at Multiphase Traffic Signal Controlled Intersection

When multiple bus vehicles send priority requests at a single intersection, the existing fixed‐phase sequence control methods cannot provide priority traffic request services for multiphase bus vehicles. In view of the conflict of multiphase bus priority requests at intersections, the priority vehicle traffic sequence is determined, which is the focus of this study. In this paper, a connected vehicle‐enabled transit signal priority system (CV‐TSPS) has been proposed, which uses vehicle‐infrastructure communication function (V2I) technology to obtain real‐time vehicle movement, road traffic states, and traffic signal light phase information. By developing a deep Q‐learning neural network (DQNN), especially for optimizing traffic signal control strategy, the public transit vehicles will be prioritized to improve their travel efficiency, while the overall delay of road traffic flow will be balanced to ensure the safe and orderly passage of intersections. In order to verify the validity of the model, the SUMO traffic analysis software has been applied to simulate real‐time traffic control, and the experimental results show that compared with the traditional timing signal control, the loss time of vehicles is reduced by nearly 40%, and the cumulative loss time per capita is reduced by nearly 43.5%, and a good control effect is achieved. In the case of medium and low densities, it is better than the solid scheduled traffic control scheme.

Impact of autonomous vehicles on air quality and energy efficiency of road traffic flows – a case study in an urban roundabout

Due to the likely market penetration of autonomous vehicles (AVs) in the short-medium term, several studies have analysed their environmental impact due to changes in demand patterns or market models. However, there is little research on the operational impact of AVs on air quality. This work aims to explore several scenarios of Avs penetration with different driving patterns and anticipate the potential impacts of traffic performance, emissions, and air quality. An urban intersection was used as a case study, namely a roundabout located in a critical context from the air pollution exposure point of view. Results show that the introduction of Avs may lead to higher travel times at peak hours and increased emissions, mainly due to AVs with defensive behaviour. Regarding air quality, the introduction of AVs led to a degradation of air quality, especially during peak hours.

A Time Series Data Prediction Model Based on Adaptive Weighted LSTM

Financial time series prediction has always been a hot topic in the field of statistics learning. Aiming at the step selection problem of LSTM time series prediction model, this paper proposes an adaptive weighted LSTM model based on model average method. The model average is mainly reflected in two aspects: On the one hand, the proposed method takes intraday price information into account. Firstly, functional and nonlinear information of intraday price series are extracted through functional principal component analysis and kernel principal component analysis, and then Bagging is used to fit the residual sequence generated by the original LSTM model. On the other hand, the proposed method integrates the information of the model under different time Windows by using the weight based on distance correlation coefficient, and adaptively solves the step size selection problem, so as to improve the effectiveness of the overall model. The actual data analysis results show that the proposed method can effectively improve the prediction accuracy of the original LSTM model and has a certain robustness. Due to the flexibility of the proposed method, it can be used in time series prediction such as energy consumption prediction, environment detection and road traffic flow monitoring.

Effect of Different Plant Communities on Fine Particle Removal in an Urban Road Greenbelt and Its Key Factors in Nanjing, China

Determining the relationships between the structure and species of plant communities and their impact on ambient particulate matter (PM) is an important topic in city road greenbelt planning and design. The correlation between the distribution of plant communities and ambient PM concentrations in a city road greenbelt has specific spatial patterns. In this study, we selected 14 plant-community-monitoring sites on seven roads in Nanjing as research targets and monitored these roads in January 2022 for various parameters such as PM with aerodynamic diameters ≤10 µm (PM10) and PM with aerodynamic diameters ≤2.5 µm (PM2.5). We used a spatial model to analyze the relationship between the concentrations of ambient PM10 and PM2.5 and the spatial heterogeneity of plant communities. The consequences revealed that the composition and species of plant communities directly affected the concentrations of ambient PM. However, upon comparing the PM concentration patterns in the green community on the urban road, we found that the ability of the plant community structures to reduce ambient PM is in the order: trees + shrubs + grasses > trees + shrubs > trees + grasses > pure trees. Regarding the reduction in ambient PM by tree species in the plant community (conifer trees > deciduous trees > evergreen broad-leaved trees) and the result of the mixed forest abatement rate, coniferous + broad-leaved trees in mixed forests have the best reduction ability. The rates of reduction in PM10 and PM2.5 were 14.29% and 22.39%, respectively. We also found that the environmental climate indices of the road community, temperature, and traffic flow were positively correlated with ambient PM, but relative humidity was negatively correlated with ambient PM. Among them, PM2.5 and PM10 were significantly related to temperature and humidity, and the more open the green space on the road, the higher the correlation degree. PM10 is also related to light and atmospheric radiation. These characteristics of plant communities and the meteorological factors on urban roads are the foundation of urban greenery ecological services, and our research showed that the adjustment of plant communities could improve greenbelt ecological services by reducing the concentration of ambient PM.

Deep Reinforcement Learning Car-Following Model Considering Longitudinal and Lateral Control

The lateral control of the vehicle is significant for reducing the rollover risk of high-speed cars and improving the stability of the following vehicle. However, the existing car-following (CF) models rarely consider lateral control. Therefore, a CF model with combined longitudinal and lateral control is constructed based on the three degrees of freedom vehicle dynamics model and reinforcement learning method. First, 100 CF segments were selected from the OpenACC database, including 50 straight and 50 curved road trajectories. Afterward, the deep deterministic policy gradient (DDPG) car-following model and multi-agent deep deterministic policy gradient (MADDPG) car-following model were constructed based on the deterministic policy gradient theory. Finally, the models are trained with the extracted trajectory data and verified by comparison with the observed data. The results indicate that the vehicle under the control of the MADDPG model and the vehicle under the control of the DDPG model are both safer and more comfortable than the human-driven vehicle (HDV) on straight roads and curved roads. Under the premise of safety, the vehicle under the control of the MADDPG model has the highest road traffic flow efficiency. The maximum lateral offset of the vehicle under the control of the MADDPG model and the vehicle under the control of the DDPG model in straight road conditions is respectively reduced by 80.86% and 71.92%, compared with the HDV, and the maximum lateral offset in the curved road conditions is lessened by 83.67% and 78.95%. The proposed car following model can provide a reference for developing an adaptive cruise control system considering lateral stability.

Identifying Causes of Traffic Crashes Associated with Driver Behavior Using Supervised Machine Learning Methods: Case of Highway 15 in Saudi Arabia

Identifying the causes of road traffic crashes (RTCs) and contributing factors is of utmost importance for developing sustainable road network plans and urban transport management. Driver-related factors are the leading causes of RTCs, and speed is claimed to be a major contributor to crash occurrences. The results reported in the literature are mixed regarding speed-crash occurrence causality on rural and urban roads. Even though recent studies shed some light on factors and the direction of effects, knowledge is still insufficient to allow for specific quantifications. Thus, this paper aimed to contribute to the analysis of speed-crash occurrence causality by identifying the road features and traffic flow parameters leading to RTCs associated with driver errors along an access-controlled major highway (761.6 km of Highway 15 between Taif and Medina) in Saudi Arabia. Binomial logistic regression (BNLOGREG) was employed to predict the probability of RTCs associated with driver errors (p < 0.001), and its results were compared with other supervised machine learning (ML) models, such as random forest (RF) and k-nearest neighbor (kNN) to search for more accurate predictions. The highest classification accuracy (CA) yielded by RF and BNLOGREG was 0.787, compared to kNN’s 0.750. Moreover, RF resulted in the largest area under the ROC (a receiver operating characteristic) curve (AUC for RF = 0.712, BLOGREG = 0.608, and kNN = 0.643). As a result, increases in the number of lanes (NL) and daily average speed of traffic flow (ASF) decreased the probability of driver error-related crashes. Conversely, an increase in annual average daily traffic (AADT) and the availability of straight and horizontal curve sections increased the probability of driver-related RTCs. The findings support previous studies in similar study contexts that looked at speed dispersion in crash occurrence and severity but disagreed with others that looked at absolute speed at individual vehicle or road segment levels. Thus, the paper contributes to insufficient knowledge of the factors in crash occurrences associated with driver errors on major roads within the context of this case study. Finally, crash prevention and mitigation strategies were recommended regarding the factors involved in RTCs and should be implemented when and where they are needed.

Meta-Extreme Learning Machine for Short-Term Traffic Flow Forecasting

The traffic flow forecasting proposed for a series of problems, such as urban road congestion and unreasonable road planning, aims to build a new smart city, improve urban infrastructure, and alleviate road congestion. The problems encountered in traffic flow forecasting are also relatively difficult; the reason is that traffic flow forecasting is uncertain, dynamic, and nonlinear. It is challenging to build a reliable and safe model. Aiming at this complex and nonlinear traffic flow forecasting problem, this paper proposes a solution of an ABC-ELM model optimized by an artificial bee colony algorithm to solve the above problem. It uses the characteristics of the artificial bee colony algorithm to optimize the model so that the model can better and faster find the optimal solution in space. Moreover, it also uses the characteristics of the limit learning machine to quickly deal with this nonlinear specific problem. Experimental results on the Amsterdam road traffic flow dataset show that the traffic flow prediction model proposed in this paper has higher prediction accuracy and is more sensitive to data changes.

Research on Improved Traffic Flow Prediction Network Based on CapsNet

Traffic flow prediction is the basis and key to the realization of an intelligent transportation system. The road traffic flow prediction of city-level complex road network can be realized using traffic big data. In the traffic prediction task, the limitation of the convolutional neural network (CNN) for modeling the spatial relationship of the road network and the insufficient feature extraction of the shallow network make it impossible to accurately predict the traffic flow. In order to improve the prediction performance of the model, this paper proposes an improved capsule network (MCapsNet) based on capsule network (CapsNet). First, in the preliminary feature extraction stage, a depthwise separable convolutional block is added to expand the feature channel to enrich channel information. Subsequently, in order to strengthen the reuse of important features and suppress useless information, channel attention is used to selectively reinforce learning of extended channel information so that the network can extract a large number of high-dimensional important features and improve the ability of network feature learning and expression. At the same time, in order to alleviate the feature degradation during training and the channel collapse problem easily caused by deep convolution, a shortcut connection, and a modified linear bottleneck layer structure are added to the convolution layer. The bottleneck layer adds the depth convolution and channel attention connection to the residual block of the network. Finally, the deep local feature information extracted from the improved convolutional layer is vectorized into the form of a capsule, which can more accurately model the details of road network attributes and features and improve the model expression power and prediction performance. The network is tested on the Wenyi Road dataset and the public dataset SZ-taxi. Compared with other models, the evaluation indicators of MCapsNet are better than other models in the tests of different time periods and predictors. Compared with CapsNet, the RMSE index of MCapsNet is reduced by 10.50% in the full period of Wenyi Road, 4.66% in the peak period, 9.78% in the off-peak period, and 6.07% in the SZ-tax dataset. The experimental results verify the effectiveness of the model improvement.