Use of sequential learning for short-term traffic flow forecasting

Accurate short-term traffic flow forecasting has become a crucial step in the overall goal of better road network management. Previous research [H. Kirby, M. Dougherty, S. Watson, Should we use neural networks or statistical models for short term motorway traffic forecasting, International Journal of Forecasting 13 (1997) 43–50.] has demonstrated that a straightforward application of neural networks can be used to forecast traffic flows along a motorway link. The objective of this paper is to report on the application and performance of an alternative neural computing algorithm which involves ‘sequential or dynamic learning’ of the traffic flow process. Our initial work [H. Chen, S. Clark, M.S. Dougherty, S.M. Grant-Muller, Investigation of network performance prediction, Report on Dynamic Neural Network and Performance Indicator development, Institute for Transport Studies, University of Leeds Technical Note 418, 1998 (unpublished)] was based on simulated data (generated using a Hermite polynomial with random noise) that had a profile similar to that of traffic flows in real data. This indicated the potential suitability of dynamic neural networks with traffic flow data. Using the Kalman filter type network an initial application with M25 motorway flow data suggested that a percentage absolute error (PAE) of approximately 9.5% could be achieved for a network with five hidden units (compared with 11% for the static neural network model). Three different neural networks were trained with all the data (containing an unknown number of incidents) and secondly using data wholly obtained around incidents. Results showed that from the three different models, the ‘simple dynamic model’ with the first five units fixed (and subsequent hidden units distributed amongst these) had the best forecasting performance. Comparisons were also made of the networks’ performance on data obtained around incidents. More detailed analysis of how the performance of the three networks changed through a single day (including an incident) showed that the simple dynamic model again outperformed the other two networks in all time periods. The use of ‘piecewise’ models (i.e. where a different model is selected according to traffic flow conditions) for data obtained around incidents highlighted good performance again by the simple dynamic network. This outperformed the standard Kalman filter neural network for a medium-sized network and is our overall recommendation for any future application.     

A car-following model considering asymmetric driving behavior based on long short-term memory neural networks

Asymmetric driving behavior is a critical characteristic of human driving behaviors and has a significant impact on traffic flow. In consideration of the asymmetric driving behavior, this paper proposes a long short-term memory (LSTM) neural networks (NN) based car-following (CF) model to capture realistic traffic flow characteristics by incorporating the driving memory. The NGSIM data are used to calibrate and validate the proposed CF model. Meanwhile, three characteristics closely related to the asymmetric driving behavior are investigated: hysteresis, discrete driving, and intensity difference. The simulation results show the good performance of the proposed CF model on reproducing realistic traffic flow features. Moreover, to further demonstrate the superiority of the proposed CF model, two other CF models including recurrent neural network based CF model and asymmetric full velocity difference model, are compared with LSTM-NN model. The results reveal that LSTM-NN model can capture the asymmetric driving behavior well and outperforms other models.     

基于灰色神经网络的输油管道运行费用预测

输油管道运行费用的预测在原油运输中有着重要意义,文中将灰色预测模型与神经网络预测模型结合起来,建立灰色神经网络预测模型,对输油管道运行费用进行预测。灰色神经网络预测模型充分发挥了灰色预测模型和神经网络预测模型样本少、计算速度快的优点。计算结果表明：灰色神经网络与EBP神经网络相比,预测模型精度高,计算量小,收敛速度快。     

Long short-term memory neural network for traffic speed prediction using remote microwave sensor data

Neural networks have been extensively applied to short-term traffic prediction in the past years. This study proposes a novel architecture of neural networks, Long Short-Term Neural Network (LSTM NN), to capture nonlinear traffic dynamic in an effective manner. The LSTM NN can overcome the issue of back-propagated error decay through memory blocks, and thus exhibits the superior capability for time series prediction with long temporal dependency. In addition, the LSTM NN can automatically determine the optimal time lags. To validate the effectiveness of LSTM NN, travel speed data from traffic microwave detectors in Beijing are used for model training and testing. A comparison with different topologies of dynamic neural networks as well as other prevailing parametric and nonparametric algorithms suggests that LSTM NN can achieve the best prediction performance in terms of both accuracy and stability.     

Hybrid artificial neural network and locally weighted regression models for lane-based short-term urban traffic flow forecasting

ABSTRACT

In recent years, there has been considerable research interest in short-term traffic flow forecasting. However, forecasting models offering a high accuracy at a fine temporal resolution (e.g. 1 or 5?min) and lane level are still rare. In this study, a combination of genetic algorithm, neural network and locally weighted regression is used to achieve optimal prediction under various input and traffic settings. The genetically optimized artificial neural network (GA-ANN) and locally weighted regression (GA-LWR) models are developed and tested, with the former forecasting traffic flow every 5-min within a 30-min period and the latter for forecasting traffic flow of a particular 5-min period of each for four lanes of an urban arterial road in Beijing, China. In particular, for morning peak and off-peak traffic flow prediction, the GA-ANN 5-min traffic flow model results in average errors of 3–5% and most 95th percentile errors of 7–14% for each of the four lanes; for the peak and off-peak time traffic flow predictions, the GA-LWR 5-min traffic flow model results in average errors of 2–4% and most 95th percentile errors are lower than 10% for each of the four lanes. When compared to previous models that usually offer average errors greater than 6–15%, such empirical findings should be of interest to and instrumental for transportation authorities to incorporate in their city- or state-wide Advanced Traveller Information Systems (ATIS).     

Classification modeling approach for vehicle dynamics model‐integrated traffic simulation assessing surrogate safety

To assess safety impacts of untried traffic control strategies, an earlier study developed a vehicle dynamics model‐integrated (i.e., VISSIM‐CarSim‐SSAM) simulation approach and evaluated its performance using surrogate safety measures. Although the study found that the integrated simulation approach was a superior alternative to existing approaches in assessing surrogate safety, the computation time required for the implementation of the integrated simulation approach prevents it from using it in practice. Thus, this study developed and evaluated two types of models that could replace the integrated simulation approach with much faster computation time, feasible for real‐time implementation. The two models are as follows: (i) a statistical model (i.e., logit model) and (ii) a nonparametric approach (i.e., artificial neural network). The logit model and the neural network model were developed and trained on the basis of three simulation data sets obtained from the VISSIM‐CarSim‐SSAM integrated simulation approach, and their performances were compared in terms of the prediction accuracy. These two models were evaluated using six new simulation data sets. The results indicated that the neural network approach showing 97.7% prediction accuracy was superior to the logit model with 85.9% prediction accuracy. In addition, the correlation analysis results between the traffic conflicts obtained from the neural network approach and the actual traffic crash data collected in the field indicated a statistically significant relationship (i.e., 0.68 correlation coefficient) between them. This correlation strength is higher than that of the VISSIM only (i.e., the state of practice) simulation approach. The study results indicated that the neural network approach is not only a time‐efficient way to implementing the VISSIM‐CarSim‐SSAM integrated simulation but also a superior alternative in assessing surrogate safety. Copyright © 2014 John Wiley & Sons, Ltd.     

Modeling traffic noise in a mountainous city using artificial neural networks and gradient correction

Different regions have established traffic noise prediction models to adapt to their particular environmental characteristics. This paper aimed to develop a traffic noise prediction model for mountainous cities. In China, the traffic noise prediction model HJ 2.4-2009, which itself is based on the sound pressure level corrected for roadway gradients (RGs), has been receiving widespread acceptance. On the basis of the model in HJ 2.4-2009, the RG correction coefficient was proposed to modify the original model and a per-vehicle noise prediction model was built using a multilayer feedforward artificial neural network (ANN) model. The data collected from a municipal road of a hilly city, Chongqing, was used to train and validate the ANN model. The predictor variables comprised the per-vehicle noise value, vehicle type, vehicle velocity, and roadway gradient. The results showed that the modified HJ 2.4-2009 model incorporating the gradient correction coefficient achieved a significantly higher R² for mountainous cities than the original model. Besides, the ANN-based noise prediction model achieved considerable accuracy improvement over the empirical predictive equations.     

A novel passenger flow prediction model using deep learning methods

Currently, deep learning has been successfully applied in many fields and achieved amazing results. Meanwhile, big data has revolutionized the transportation industry over the past several years. These two hot topics have inspired us to reconsider the traditional issue of passenger flow prediction. As a special structure of deep neural network (DNN), an autoencoder can deeply and abstractly extract the nonlinear features embedded in the input without any labels. By exploiting its remarkable capabilities, a novel hourly passenger flow prediction model using deep learning methods is proposed in this paper. Temporal features including the day of a week, the hour of a day, and holidays, the scenario features including inbound and outbound, and tickets and cards, and the passenger flow features including the previous average passenger flow and real-time passenger flow, are defined as the input features. These features are combined and trained as different stacked autoencoders (SAE) in the first stage. Then, the pre-trained SAE are further used to initialize the supervised DNN with the real-time passenger flow as the label data in the second stage. The hybrid model (SAE-DNN) is applied and evaluated with a case study of passenger flow prediction for four bus rapid transit (BRT) stations of Xiamen in the third stage. The experimental results show that the proposed method has the capability to provide a more accurate and universal passenger flow prediction model for different BRT stations with different passenger flow profiles.     

Short‐term traffic flow prediction with linear conditional Gaussian Bayesian network

Traffic flow prediction is an essential part of intelligent transportation systems (ITS). Most of the previous traffic flow prediction work treated traffic flow as a time series process only, ignoring the spatial relationship from the upstream flows or the correlation with other traffic attributes like speed and density. In this paper, we utilize a linear conditional Gaussian (LCG) Bayesian network (BN) model to consider both spatial and temporal dimensions of traffic as well as speed information for short‐term traffic flow prediction. The LCG BN allows both continuous and discrete variables, which enables the consideration of categorical variables in traffic flow prediction. A microscopic traffic simulation dataset is used to test the performance of the proposed model compared to other popular approaches under different predicting time intervals. In addition, the authors investigate the importance of spatial data and speed data in flow prediction by comparing models with different levels of information. The results indicate that the prediction accuracy will increase significantly when both spatial data and speed data are included. Copyright © 2016 John Wiley & Sons, Ltd.     

Short-Term Prediction of Travel Time using Neural Networks on an Interurban Highway

The main purpose of this study was to investigate the predictability of travel time with a model based on travel time data measured in the field on an interurban highway. Another purpose was to determine whether the forecasts would be accurate enough to implement the model in an actual online travel time information service. The study was carried out on a 28-kilometre-long rural two-lane road section where traffic congestion was a problem during weekend peak hours. The section was equipped with an automatic travel time monitoring and information system. The prediction models were made as feedforward multilayer perceptron neural networks. The main results showed that the majority of the forecasts were close to the actual measured values. Consequently, use of the prediction model would improve the quality of travel time information based directly on the sum of the latest measured travel times.     

Short‐term highway traffic flow prediction based on a hybrid strategy considering temporal–spatial information

Short‐term traffic flow prediction is fundamental for the intelligent transportation system and is proved to be a challenge. This paper proposed a hybrid strategy that is general and can make use of a large number of underlying machine learning or time‐series prediction models to capture the complex patterns beneath the traffic flow. With the strategy, four different combinations were implemented. To consider the spatial features of traffic phenomenon, several different state vectors including different observations were built. The performance of the proposed strategy was investigated using the traffic flow measurements from the Traffic Operation and Safety Laboratory in Wisconsin, USA. The results show the overall performance of hybrid strategy is better than a single model. Also, incorporating observations from adjacent junctions can improve prediction accuracy. Copyright © 2017 John Wiley & Sons, Ltd.     

Cellpath: Fusion of cellular and traffic sensor data for route flow estimation via convex optimization

A new convex optimization framework is developed for the route flow estimation problem from the fusion of vehicle count and cellular network data. The issue of highly underdetermined link flow based methods in transportation networks is investigated, then solved using the proposed concept of cellpaths for cellular network data. With this data-driven approach, our proposed approach is versatile: it is compatible with other data sources, and it is model agnostic and thus compatible with user equilibrium, system-optimum, Stackelberg concepts, and other models. Using a dimensionality reduction scheme, we design a projected gradient algorithm suitable for the proposed route flow estimation problem. The algorithm solves a block isotonic regression problem in the projection step in linear time. The accuracy, computational efficiency, and versatility of the proposed approach are validated on the I-210 corridor near Los Angeles, where we achieve 90% route flow accuracy with 1033 traffic sensors and 1000 cellular towers covering a large network of highways and arterials with more than 20,000 links. In contrast to long-term land use planning applications, we demonstrate the first system to our knowledge that can produce route-level flow estimates suitable for short time horizon prediction and control applications in traffic management. Our system is open source and available for validation and extension.     

Estimating historical hourly traffic volumes via machine learning and vehicle probe data: A Maryland case study

This paper focuses on the problem of estimating historical traffic volumes between sparsely-located traffic sensors, which transportation agencies need to accurately compute statewide performance measures. To this end, the paper examines applications of vehicle probe data, automatic traffic recorder counts, and neural network models to estimate hourly volumes in the Maryland highway network, and proposes a novel approach that combines neural networks with an existing profiling method. On average, the proposed approach yields 24% more accurate estimates than volume profiles, which are currently used by transportation agencies across the US to compute statewide performance measures. The paper also quantifies the value of using vehicle probe data in estimating hourly traffic volumes, which provides important managerial insights to transportation agencies interested in acquiring this type of data. For example, results show that volumes can be estimated with a mean absolute percent error of about 21% at locations where average number of observed probes is between 30 and 47 vehicles/h, which provides a useful guideline for assessing the value of probe vehicle data from different vendors.     

International experience with speed limits during and prior to the energy crisis of 1973–4

This paper presents an artificial neural network (ANN) based method for estimating route travel times between individual locations in an urban traffic network. Fast and accurate estimation of route travel times is required by the vehicle routing and scheduling process involved in many fleet vehicle operation systems such as dial‐a‐ride paratransit, school bus, and private delivery services. The methodology developed in this paper assumes that route travel times are time‐dependent and stochastic and their means and standard deviations need to be estimated. Three feed‐forward neural networks are developed to model the travel time behaviour during different time periods of the day‐the AM peak, the PM peak, and the off‐peak. These models are subsequently trained and tested using data simulated on the road network for the City of Edmonton, Alberta. A comparison of the ANN model with a traditional distance‐based model and a shortest path algorithm is then presented. The practical implication of the ANN method is subsequently demonstrated within a dial‐a‐ride paratransit vehicle routing and scheduling problem. The computational results show that the ANN‐based route travel time estimation model is appropriate, with respect to accuracy and speed, for use in real applications.     

An efficient realization of deep learning for traffic data imputation

Traffic data provide the basis for both research and applications in transportation control, management, and evaluation, but real-world traffic data collected from loop detectors or other sensors often contain corrupted or missing data points which need to be imputed for traffic analysis. For this end, here we propose a deep learning model named denoising stacked autoencoders for traffic data imputation. We tested and evaluated the model performance with consideration of both temporal and spatial factors. Through these experiments and evaluation results, we developed an algorithm for efficient realization of deep learning for traffic data imputation by training the model hierarchically using the full set of data from all vehicle detector stations. Using data provided by Caltrans PeMS, we have shown that the mean absolute error of the proposed realization is under 10 veh/5-min, a better performance compared with other popular models: the history model, ARIMA model and BP neural network model. We further investigated why the deep leaning model works well for traffic data imputation by visualizing the features extracted by the first hidden layer. Clearly, this work has demonstrated the effectiveness as well as efficiency of deep learning in the field of traffic data imputation and analysis.     

基于神经网络的无信号路口人车交通行为预测分析

以无信号灯路口人车交通行为为研究对象,对行人和机动车辆在无信号灯路口的整体交通行为进行分类预测。在对路口现场交通情况进行拍摄后,用电脑的分帧技术对所需要的数据进行提取和分类,而后建立BP神经网络模型,确定神经网络的输入变量与输出变量。将样本数据导入神经网络并进行训练和测试后,得出行人和车辆过街类型的分类准确率,并且通过准确率所达到的标准来证明了BP神经网络模型的可行性。     

Modelling dual carriageway lane changing using neural networks

Neural networks offer a potential alternative method of modelling driver behaviour within road traffic systems. This paper explores the application of neural networks to modelling the lane-changing decisions of drivers on dual carriageways. Two approaches are considered. The first, preliminary approach uses a prediction type of neural network with a single hidden layer and the back propagation learning algorithm to model the behaviour of an individual driver. A series of consecutive time-scan traffic patterns, which describe the driver's environment and changes over time as the selected vehicle travels along a link, are input to the neural network, which then predicts the new lane and position of the vehicle. Training data are collected from a human subject using an interactive driving simulation. The trained neural network successfully exhibited the rudiments of driving behaviour in terms of lane and speed changes. A major disadvantage of this approach was the difficulty in recording real-life data, which are required to train the neural network, for individual drivers. The second approach concentrates specifically on lane changing and makes use of a learning vector quantization classification type of neural network. Input to the neural network still consists primarily of time-scan traffic patterns, but the format is changed to facilitate the possibility of data acquisition using image processing. The neural network output classifies the input data by determining the new lane for the vehicle concerned. Performance in both testing and training was very good for data generated by the rule-based driver-decision model of a microscopic simulation. Performance in testing was less satisfactory for data taken directly from a road and highlighted the need for extensive data sets for successful training.     

Short-term speed predictions exploiting big data on large urban road networks

Big data from floating cars supply a frequent, ubiquitous sampling of traffic conditions on the road network and provide great opportunities for enhanced short-term traffic predictions based on real-time information on the whole network. Two network-based machine learning models, a Bayesian network and a neural network, are formulated with a double star framework that reflects time and space correlation among traffic variables and because of its modular structure is suitable for an automatic implementation on large road networks. Among different mono-dimensional time-series models, a seasonal autoregressive moving average model (SARMA) is selected for comparison. The time-series model is also used in a hybrid modeling framework to provide the Bayesian network with an a priori estimation of the predicted speed, which is then corrected exploiting the information collected on other links. A large floating car data set on a sub-area of the road network of Rome is used for validation. To account for the variable accuracy of the speed estimated from floating car data, a new error indicator is introduced that relates accuracy of prediction to accuracy of measure. Validation results highlighted that the spatial architecture of the Bayesian network is advantageous in standard conditions, where a priori knowledge is more significant, while mono-dimensional time series revealed to be more valuable in the few cases of non-recurrent congestion conditions observed in the data set. The results obtained suggested introducing a supervisor framework that selects the most suitable prediction depending on the detected traffic regimes.     

A dynamic Bayesian network model for real-time crash prediction using traffic speed conditions data

Traffic crashes occurring on freeways/expressways are considered to relate closely to previous traffic conditions, which are time-varying. Meanwhile, most studies use volume/occupancy/speed parameters to predict the likelihood of crashes, which are invalid for roads where the traffic conditions are estimated using speed data extracted from sampled floating cars or smart phones. Therefore, a dynamic Bayesian network (DBN) model of time sequence traffic data has been proposed to investigate the relationship between crash occurrence and dynamic speed condition data. Moreover, the traffic conditions near the crash site were identified as several state combinations according to the level of congestion and included in the DBN model. Based on 551 crashes and corresponding speed information collected on expressways in Shanghai, China, DBN models were built with time series speed condition data and different state combinations. A comparative analysis of the DBN model using flow detector data and a static Bayesian network model was also conducted. The results show that, with only speed condition data and nine traffic state combinations, the DBN model can achieve a crash prediction accuracy of 76.4% with a false alarm rate of 23.7%. In addition, the results of transferability testing imply that the DBN models are applicable to other similar expressways with 67.0% crash prediction accuracy.     

Forecasting current and next trip purpose with social media data and Google Places

Trip purpose is crucial to travel behavior modeling and travel demand estimation for transportation planning and investment decisions. However, the spatial-temporal complexity of human activities makes the prediction of trip purpose a challenging problem. This research, an extension of work by Ermagun et al. (2017) and Meng et al. (2017), addresses the problem of predicting both current and next trip purposes with both Google Places and social media data. First, this paper implements a new approach to match points of interest (POIs) from the Google Places API with historical Twitter data. Therefore, the popularity of each POI can be obtained. Additionally, a Bayesian neural network (BNN) is employed to model the trip dependence on each individual’s daily trip chain and infer the trip purpose. Compared with traditional models, it is found that Google Places and Twitter information can greatly improve the overall accuracy of prediction for certain activities, including “EatOut”, “Personal”, “Recreation” and “Shopping”, but not for “Education” and “Transportation”. In addition, trip duration is found to be an important factor in inferring activity/trip purposes. Further, to address the computational challenge in the BNN, an elastic net is implemented for feature selection before the classification task. Our research can lead to three types of possible applications: activity-based travel demand modeling, survey labeling assistance, and online recommendations.