01590332

Component

In recent decades, short term traffic flow forecasting has been regarded as a critical aspect of developing the proactive traffic management and control systems. In this regard, many methods have been developed, and among these methods, artificial neural network (ANN) approach has been considered as a potentially plausible forecasting model for many transportation applications and has been investigated extensively over the decades. In the artificial neural networks, the learning process is an indispensable component for fine-tuning the model and hence to improve the prediction. However, the effects of training algorithms on the performance of ANN based short term traffic flow forecasting have not been adequately revealed. Therefore, in this end, the purpose of this paper is to compare the typical training algorithms that have been applied in the multi-layer perceptron (MLP) ANN model. In this paper, four typical training algorithms are selected for the MLP model, including gradient descent (GD), Levenberg-Marquardt (LM), Bayesian regularization (BR), and Fletcher-Reeves conjugate gradient (CGF). The forecasting performances were investigated using real world traffic flow data in terms of aggregated measures, disaggregated measures, and computation efficiency. The results show that in terms of forecasting accuracy, the models based on LM, BR, and CGF show similar performances and outperform the model based on GD. In terms of computation efficiency, the model based on BR method shows the best performance. Combined, ANN model based on BR can generate desirable short term traffic flow forecasting performances in terms of both accuracy and computation efficiency. Based on the study, future work is recommended on investigating or fine-tuning other well-known forecasting algorithms.

This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.

01584066

16-3100

English

17p

2016

Transportation Research Board 95th Annual Meeting

Digital/other

Figures; References; Tables

Accuracy; Algorithms; Neural networks; Traffic flow; Traffic forecasting

Computational efficiency

Data and Information Technology; Highways; I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2016 Paper #16-3100

TRIS, TRB, ATRI

Jan 12 2016 5:23PM