A key concern in transportation planning and traffic management is the ability to forecast traffic flows on a street network. Traffic flows forecasts can be transformed to obtain travel time estimates and then use these as input to travel demand models, dynamic route guidance and congestion management procedures. A variety of mathematical techniques have been proposed for modeling traffic flow on a street network. Briefly, the most widely used theories are: -Kinetic models based on partial differential equations that describe waves of different traffic densities, -deterministic models that use nonlinear equations for the estimation of different car routes, -large scale simulation models such as cellular automata and, -stochastic modeling of traffic density at distinct points in space. One problem with these approaches is that the traffic flow process is characterized by nonstationarities that cannot be taken into account by the vast majority of modeling strategies. However, recent advances in statistical modeling in fields such as econometrics or environmetrics enable us to overcome this problem. The aim of this work is to present how two statistical techniques, namely, vector autoregressive modeling and dynamic space-time modeling can be used to develop efficient and reliable forecasts of traffic flow. The former approach is encountered in the econometrics literature, whereas the later is mostly used in environmetrics. Recent advances in statistical methodology provide powerful tools for traffic flow description and forecasting. For a purely statistical approach to travel time prediction one may consult Rice and van Zwet (2002). In this work, the authors employ a time varying coefficients regression technique that can be easily implemented computationally, but is sensitive to nonstationarities and does not take into account traffic flow information from neighboring points in the network that can significantly improve forecasts. According to our approach, traffic flow measurements, that is count of vehicles and road occupancy obtained at constants time intervals through loop detectors located at various distinct points of a road network, form a multiple time series set. This set can be described by a vector autoregressive process that models each series as a linear combination of past observations of some (optimally selected) components of the vector; in our case the vector is comprised by the different measurement points of traffic flow. For a thorough technical discussion on vector autoregressive processes we refer to Lutkerpohl (1987), whereas a number of applications can be found in Ooms (1994). Nowadays, these models are easily implemented in commercial software like SAS or MATLAB; see for example LeSage (1999). The spatial distribution of the measurement locations and their neighboring relations cannot be incorporated in a vector autoregressive model. However, accounting for this information may optimize model fitting and provide insight into spatial correlation structures that evolve through time. This can be accomplished by applying space-time modeling techniques. The main difference of space-time models encountered in literature with the vector autoregressive ones lies in the inclusion of a weight matrix that defines the neighboring relations and places the appropriate restrictions. For some early references on space-time models, one could consult Pfeifer and Deutsch (1980 a,b); for a Bayesian approach, insensitive to nonstationarities we refer to Wikle, Berliner and Cressie (1998). In this work, we discuss how the space-time methodology can be implemented to traffic flow modeling. The aforementioned modeling strategies are applied in a subset of traffic flow measurements collected every 15 minutes through loop detectors at 74 locations in the city of Athens. A comparative study in terms of model fitting and forecasting accuracy is performed. Univariate time series models are also fitted in each measurement location in order to investigate the relation between a model's dimension and performance. References: LeSage J. P. (1999). Applied Econometrics using MATLAB. Manuscript, Dept. of Economics, University of Toronto Lutkerpohl H. (1987). Forecasting Aggregated Vector ARMA Processes. Lecture Notes in Economics and Mathematical Systems. Springer Verlag Berlin Heidelberg Ooms M. (1994). Empirical Vector Autoregressive Modeling. Springer Verlag Berlin Heidelberg Pfeifer P. E., and Deutsch S. J. (1980a). A three-stage iterative procedure for Space-Time Modeling. Technometrics, 22, 35-47 Pfeifer P. E., and Deutsch S. J. (1980b). Identification and Interpretation of First-Order Space-Time ARMA models. Technometrics, 22, 397-408 Rice J., and van Zwet E. (2002). A simple and effective method for predicting travel times on freeways. Manuscript, Dept. of Statistics, University of California at Berkeley Wikle C. K., Berliner L. M. and Cressie N. (1998). Hierarchical Bayesian space-time models. Environmental and Ecological Statistics, 5, 117-154
[1]
Joe Whittaker,et al.
TRACKING AND PREDICTING A NETWORK TRAFFIC PROCESS
,
1997
.
[2]
Gwilym M. Jenkins,et al.
Time series analysis, forecasting and control
,
1971
.
[3]
Daniel B. Fambro,et al.
Application of Subset Autoregressive Integrated Moving Average Model for Short-Term Freeway Traffic Volume Forecasting
,
1999
.
[4]
D L Streiner,et al.
An Introduction to Multivariate Statistics
,
1993,
Canadian journal of psychiatry. Revue canadienne de psychiatrie.
[5]
J. Ramos,et al.
Space-Time Modeling of Vector Hydrologic Sequences
,
1986
.
[6]
Helmut Lütkepohl,et al.
Introduction to multiple time series analysis
,
1991
.
[7]
Moshe Ben-Akiva,et al.
Dynamic Traffic Prediction for Motorway Networks
,
1996
.
[8]
Billy M. Williams.
Multivariate Vehicular Traffic Flow Prediction: Evaluation of ARIMAX Modeling
,
2001
.
[9]
Helmut Lütkepohl,et al.
Forecasting aggregated vector ARMA processes
,
1987
.
[10]
J. Paul Elhorst,et al.
Dynamic models in space and time
,
2010
.
[11]
C. Granger.
Investigating Causal Relations by Econometric Models and Cross-Spectral Methods
,
1969
.
[12]
Ir J W C Van Lint,et al.
ROBUST FREEWAY TRAVELTIME PREDICTION WITH STATE-SPACE NEURAL NETWORKS
,
2003
.
[13]
Matthew G. Karlaftis,et al.
A multivariate state space approach for urban traffic flow modeling and prediction
,
2003
.
[14]
P. Pfeifer,et al.
Independence and sphericity tests for the residuals of space-time arma models
,
1980
.
[15]
Erik van Zwet,et al.
A simple and effective method for predicting travel times on freeways
,
2004,
IEEE Transactions on Intelligent Transportation Systems.
[16]
Stuart Jay Deutsch,et al.
Variance of the Sample Space-time Autocorrelation Function
,
1981
.
[17]
Paolo Toth,et al.
Advanced Methods in Transportation Analysis
,
1997
.
[18]
G. C. Tiao,et al.
An introduction to multiple time series analysis.
,
1993,
Medical care.
[19]
P. Young,et al.
Time series analysis, forecasting and control
,
1972,
IEEE Transactions on Automatic Control.
[20]
Stuart Jay Deutsch,et al.
Space-Time ARMA Modeling With Contemporaneously Correlated Innovations
,
1981
.
[21]
C. Granger,et al.
Aggregation of Space-Time Processes
,
2001
.
[22]
S. P. Hoogendoorn,et al.
Freeway Travel Time Prediction with State-Space Neural Networks: Modeling State-Space Dynamics with Recurrent Neural Networks
,
2002
.
[23]
Marius Ooms.
Empirical Vector Autoregressive Modeling
,
1994
.
[24]
P. Pfeifer,et al.
A Three-Stage Iterative Procedure for Space-Time Modeling
,
1980
.
[25]
Gary A. Davis,et al.
Nonparametric Regression and Short‐Term Freeway Traffic Forecasting
,
1991
.
[26]
Stuart Jay Deutsch,et al.
Variance of the Sample Space-Time Correlation Function of Contemporaneously Correlated Variables
,
1981
.
[27]
P. Pfeifer,et al.
Identification and Interpretation of First Order Space-Time ARMA Models
,
1980
.
[28]
S. Ghon Rhee,et al.
Pacific-basin capital markets research
,
1990
.