An object-oriented neural network approach to short-term traffic forecasting

https://doi.org/10.1016/S0377-2217(00)00125-9Get rights and content

Abstract

This paper discusses an object-oriented neural network model that was developed for predicting short-term traffic conditions on a section of the Pacific Highway between Brisbane and the Gold Coast in Queensland, Australia. The feasibility of this approach is demonstrated through a time-lag recurrent network (TLRN) which was developed for predicting speed data up to 15 minutes into the future. The results obtained indicate that the TLRN is capable of predicting speed up to 5 minutes into the future with a high degree of accuracy (90–94%). Similar models, which were developed for predicting freeway travel times on the same facility, were successful in predicting travel times up to 15 minutes into the future with a similar degree of accuracy (93–95%). These results represent substantial improvements on conventional model performance and clearly demonstrate the feasibility of using the object-oriented approach for short-term traffic prediction.

Introduction

The prediction of short-term traffic conditions is a vital component of advanced traffic management and information systems which aim to influence travel behaviour, reduce traffic congestion, improve mobility and enhance air quality. Traffic prediction models can be used to provide metropolitan traffic control centres with an automated tool for anticipating the congestion that may arise on road facilities and its expected duration. This information can then be provided to drivers in real-time to give them realistic estimates of travel times, expected delays and alternative routes to their destinations. Providing drivers with this information is believed to have the potential to alleviate traffic congestion and enhance the performance of the road network.

Traffic information provided to drivers may conceptually fall into one of three categories: historical, current and predictive. Historical information describes the state of the transportation system during previous time periods. Current information is the most up-to-date information about traffic conditions. A number of currently available intelligent transport systems (ITS) technologies allow for the provision of this information in real-time at intervals less than 20 seconds on a 24-hour basis. Predictive information, on the other hand, falls into two distinct categories: strategic and short-term. Strategic information is mainly needed for major decisions on road planning and includes prediction of traffic flows and conditions months or years into the future. In contrast, short-term predictive information often has a horizon of only a few minutes and is therefore more suited to implementation in traffic management and information systems.

Most of today’s traffic control systems rely mainly on historical and current traffic data as a basis for traffic management actions. The performance of these systems is constrained because they lack the predictive capabilities. Ideally, traffic conditions should be anticipated and actions should be planned accordingly. Since drivers’ decisions are affected by expected network conditions, it is also clear that the most useful type of information for a driver faced with travel choices is reliable predictive information. Drivers making travel decisions in the absence of predictive information are implicitly projecting future conditions from the historical and current information available to them. Therefore, predictions of what traffic conditions are likely to be in a few minutes’ time (e.g., 5–10 minutes into the future) are needed for effective traffic management and information systems.

This paper describes the development and evaluation of an object-oriented neural network model that was developed to predict speed at a detector station up to 15 minutes into the future and shows how similar models were developed for freeway travel time estimation. The results reported in this paper clearly demonstrate the feasibility of implementing this approach for freeway short-term traffic forecasting and the potential for its implementation in other real-time applications such as arterial travel time estimation and automatic incident detection.

Section snippets

Previous research work

A number of traffic prediction algorithms have been developed or proposed over the last two decades. Their structure varies in the degree of sophistication, complexity and data requirements. Inductive loop detectors, embedded in the freeway pavement, are typically used to obtain the traffic data needed for these algorithms. Some of the most widely used traffic prediction models are those based on spectral analysis, ARIMA time-series models, Box–Jenkins analysis and Kalman filtering (Clark et

Data for model development and evaluation

The neural network models presented in this paper were developed using field data collected from four inductive loop detector stations installed on a 1.5-km section of the Pacific Highway between Brisbane and the Gold Coast (Fig. 1). Inductive loop detectors were installed at approximately 500 m intervals to collect speed and flow data from each detector station (Lam et al., 1996). The raw data used in this study were collected over a 5-hour period on 2 days in April 1995. The 5-hour period

Neural network training

In order to develop a neural network model to perform traffic prediction, the network needs to be trained with historical examples of input–output data. This paper will present the results for speed prediction only. Speed measurements from the current time interval (t0) at a given detector station form the input to the neural network model. The output of the model comprises the speed measurements at the same station at some future time interval tn.

Model performance evaluation

Based on the results reported in Table 2, the TLRN model was selected for further investigation. Table 3 below lists the performance results for nine TLRN models that were developed to provide speed forecasts for prediction horizons up to 15 minutes. The average percent errors (computed as the percentage difference between the actual and predicted speed) are based on the testing data set and are indicative of the generalisation performance of the model when applied in the field. These results

Freeway travel time estimation

The effectiveness of a wide range of ITS control strategies designed to alleviate traffic congestion depends heavily on the accuracy, credibility and reliability of travel time estimates. Dynamic estimates of travel time can be used to inform drivers of expected delays and travel times to their destinations and are important inputs into a number of real-time applications such as automatic incident detection and dynamic route guidance.

The next section describes briefly how the modelling

Conclusions

The results reported in this paper clearly demonstrate the superior predictive performance of the dynamic neural network architectures (e.g., TLRN, recurrent and hybrid networks) compared to the static classifiers (e.g., MLPs). These results also demonstrate the feasibility of using the object-oriented neural network approach for short-term traffic forecasting. The models described in this paper were capable of predicting speed data up to 5 minutes into the future with a high degree of accuracy

Acknowledgements

The author wishes to thank Mr. Richard Lam from the Department of Main Roads in Queensland and Mr. Ron Roper from ARRB Transport Research Ltd. for providing the data for this study. The author also wishes to thank the anonymous referees for their valuable comments and suggestions.

References (8)

There are more references available in the full text version of this article.

Cited by (353)

  • Explainable Stacking-Based Learning Model for Traffic Forecasting

    2024, Journal of Transportation Engineering Part A: Systems
View all citing articles on Scopus
View full text