Elsevier

Neurocomputing

Volume 167, 1 November 2015, Pages 3-7
Neurocomputing

Urban traffic flow forecasting through statistical and neural network bagging ensemble hybrid modeling

https://doi.org/10.1016/j.neucom.2014.08.100Get rights and content

Abstract

In this paper we show a hybrid modeling approach which combines Artificial Neural Networks and a simple statistical approach in order to provide a one hour forecast of urban traffic flow rates. Experimentation has been carried out on three different classes of real streets and results show that the proposed approach outperforms the best of the methods it puts together.

Introduction

Transportation is a wide human-oriented field with diverse and challenging problems waiting to be solved. Characteristics and performances of transport systems, services, costs, infrastructures, vehicles and control systems are usually defined on the basis of quantitative evaluation of their main effects. Most of the transport decisions take place under imprecision, uncertainty and partial truth. Some objectives and constraints are often difficult to be measured by crisp values. Traditional analytical techniques were found to be ineffective when dealing with problems in which the dependencies between variables were too complex or ill-defined. Moreover, hard computing models cannot deal effectively with the transport decision-makers׳ ambiguities and uncertainties. In order to come up with solutions to some of these problems, over the last decade there has been much interest in soft computing applications of traffic and transport systems, leading to some successful implementations [1]. The use of soft computing methodologies for modeling and analyzing traffic and transport systems is of particular interest to researchers and practitioners due to their ability to handle quantitative and qualitative measures, and to efficiently solve complex problems which involve imprecision, uncertainty and partial truth. Soft computing can be used to bridge modeling gaps of normative and descriptive decision models in traffic and transport research. Transport problems can be classified into four main areas: traffic control and management, transport planning and management, logistics, design and construction of transport facilities. The first category includes traffic flow forecasting which is the topic tackled in this work. This issue has been faced by the soft computing community since the nineties [4], [5], [6], [7], [8], [9], [10] up today [12], [13], [14], [11] with Artificial Neural Networks (ANNs) [2], [3]. As example, among the most recent work [14] focuses on traffic flow forecasting approach based on Particle Swarm Optimization (PSO) with Wavelet Network Model (WNM). Pamula et al. [11] review neural networks applications in urban traffic management systems and presents a method of traffic flow prediction based on neural networks. Bucur et al. [12] propose the use of a self-adaptive fuzzy neural network for traffic prediction suggesting an architecture which tracks probability distribution drifts due to weather conditions, season, or other factors. All the mentioned applications have one feature in common: they use one single global model in order to perform the prediction. Therefore, the main novelty of the proposed work is to combine different heterogeneous models in order to get a meta-model capable of providing predictions more accurate than the best of the constituent models. In our work we firstly composed of a neural networks ensemble with a simple statistical model and compare the results over the one hour forecast, then we improved ensembling model with BAGGING. Results shown highlight a remarkable decrease of error through the BAGGING learning phase.

Section snippets

Basic model

In order to perform a meaningful comparison for the forecasting, a basic model should be introduced in order to quantify the improvement given by more intelligent and complex forecasting techniques. For seasonal data a basic model might be defined asxt=xtswith S being the appropriate seasonality period. This model gives a prediction at time t presenting the value observed exactly a period of S steps before. For this work we put the value of S=1 which corresponds to the previous hour. It means

Experimentation

In this paragraph we test and compare the methods presented in the previous section. The test case has concerned the short term traffic flow rate of three different streets, shown in Table 1, located in the town of Terni (about 90 km north of Rome). The data set is made of 3 months (13 weeks) of measurement corresponding to 2184 hourly samples.

The data set has been partitioned into training/testing and validation made respectively of 10 and 3 weeks each. We firstly present the result obtained

Conclusions

In this paper we showed a novel hybrid modeling approach which combines Artificial Neural Networks and a simple statistical approach in order to provide a one hour forecast of urban traffic flow rates. Experimentation has been carried out on three different classes of real streets and results showed that the proposed approach clearly outperforms the best of the methods it puts together achieving a prediction error lower than 3%. The reason for that is that the neural ensembling model is capable

Fabio Moretti received in 2009 the B.Sc. degree in Computer Science and Automation from the University of Roma Tre in Rome, Italy where he is currently Ph.D. student. He is a research fellow in ENEA (Energy New technologies and sustainable Economic development Agency). His research interests includes data fusion, data mining, evolutionary computation, optimization and computer vision applied to energy saving issues. He is currently involved in several projects concerning Smart Cities

References (21)

  • Y. Liu et al.

    Ensemble learning via negative correlation

    Neural Netw.

    (1999)
  • Erel Avineri, Soft computing applications in traffic and transport systems: a review, in: Soft Computing: Methodologies...
  • M.A. Arbib

    The Nandbook of Brain Theory and Neural Nnetworks

    (1995)
  • S. Haykin, Neural Networks, A Comprehensive Foundation, second ed., Prentice Hall, New Jersey,...
  • D. Canca, J. Larrañeta, S. Lozano, L. Onieva, Traffic intensity forecast in urban networks using a multilayer...
  • M.S. Dougherty, M. Cobbett, Short term inter-urban traffic forecasts using neural networks, in: Proceedings of the...
  • S. Ishak, P. Kotha, C. Alecsandru, Optimization of dynamic neural network performance for short-term traffic...
  • D. Park et al.

    Spectral basis neural networks for real-time travel time forecasting

    J. Transp. Eng.

    (1999)
  • C. Taylor, D. Meldrum, Freeway traffic data prediction using neural networks, in: Proceedings of the 6th VNIS...
  • J.W.C. Van Lint, S.P. Hoogendoorn, H.J. Van Zuylen, Freeway travel time prediction with state–space neural networks,...
There are more references available in the full text version of this article.

Cited by (0)

Fabio Moretti received in 2009 the B.Sc. degree in Computer Science and Automation from the University of Roma Tre in Rome, Italy where he is currently Ph.D. student. He is a research fellow in ENEA (Energy New technologies and sustainable Economic development Agency). His research interests includes data fusion, data mining, evolutionary computation, optimization and computer vision applied to energy saving issues. He is currently involved in several projects concerning Smart Cities activities, in particular buildings diagnostics, control and optimization and public lighting control.

Stefano Pizzuti, Dr. Master degree in Information Sciences (final voting : 107/110) obtained in 1996 at University of Rome ‘La Sapienza’. Since 1997 researcher at ENEA in the fields of advanced monitoring and control systems applied to energy production plants. In the last five years the main research activities have focused on Smart Cities application, namely smart building network management, smart lighting, smart communities and integrated infrastructures. Involved as work-package and task leader in several national smart cities projects and in the Joint Program EERA Smart Cities. Author of more than 60 national and international publications, program committee member of several international conferences and reviewer of many Impact Factor international journals.

Stefano Panzieri received the Laurea degree in Electronic Engineering in 1989 and the Ph.D. in Systems Engineering in 1994, both from the University of Roma "La Sapienza". Since February 1996 he is within the Engineering Department of University of "Roma Tre", as Associate Professor. Research interests are in the field of industrial control systems, robotics and sensor fusion. He is author of more the one hundred papers involving mobile and industrial robots. In particular, in the area of mobile robots, some attention has been given to the problem of navigation in structured and unstructured environments with a special attention to the problem of sensor based navigation and sensor fusion. His research interests includes Interdependency Modeling; Modeling and simulation of complex systems; SCADA vulnerabilities; Data fusion; Distributed algorithms in Sensor Networks; Smart Energy Management; Building Automation Systems.

Mauro Annunziato is currently Director of the Smart Energy Division of the Energy Dept. of ENEA (95 researchers). The activities of the Division include the research and development of new technologies for smart cities, sustainable mobility, critical infrastructures, energy efficiency, smart buildings, smart homes, intelligent systems, robotics, public lighting, smart appliances, ICT city platform. More than 120 scientific publications on Journals, book chapters and international conferences. More than 60 seminars, academic teachings (masters), many participations in Scientific Committees of Int. Journals, Conferences and Associations, organization of workshops and editor of journal issues, many national and international TV/newspaper interviews. More that 5000 web links on international web pages referring to works of M. Annunziato

View full text