A New Methodology for Traffic Flow Forecasting: Dynamic Wavelet Neural Network Xiaomo Jiang Graduate Research Associate, Dept. of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, 470 Hitchcock Hall, 2070 Neil Ave., Columbus, OH, 43210, USA. Jiang.98@osu.edu Hojjat Adeli Lichtenstein Professor, Dept. of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, 470 Hitchcock Hall, 2070 Neil Ave., Columbus, OH, 43210, USA. Adeli.1@osu.edu Abstract In this study, a novel nonparametric dynamic wavelet neural network methodology is presented for forecasting traffic flow. The model incorporates the self-similar, singular, and fractal properties discovered in the traffic flow using discrete wavelets packet transform (DWPT). A DWPT-based approach combined with a wavelet coefficients penalization scheme and soft-thresholding is also developed for denoising the traffic flow. The concept of wavelet frame is introduced and exploited in the forecasting model to provide flexibility in the design of wavelets and to add extra features such as adaptable translation parameters desirable in traffic flow forecasting. The methodology has been validated using actual freeway traffic flow data. 1. Introduction There has been a steady increase in both rural and urban freeway traffic in recent years resulting in congestion in many freeway systems. The freeway traffic congestion can no longer be dealt with simply by extending more highways for economical and environmental reasons 0. Intelligent transportation systems (ITS) provide solutions for alleviating the increasing congestion problems through optimum use of existing traffic network to manage the traffic congestion. Accurate and timely forecasting of traffic flow is of paramount importance for effective management of traffic congestion in ITS, which requires a detailed understanding of the properties of traffic flow. Some researchers have investigated the characteristics of traffic flow in freeways [2]~[7]. Over the past decade, a number of papers have been published on the application of neural network models for forecasting traffic flow taking advantage of their ability to capture the indeterministic and complex nonlinearity of time series [4][10]~[17]. However, the existing neural network-based approaches have their inherent shortcomings such as lack of an efficient constructive model (for example, requiring arbitrary selection of the number of hidden nodes), slow convergence rate resulting in excessive computation time, and entrapment in a local minimum as pointed out by Adeli and Hung [18] and others. Besides, these methods do not incorporate the characteristics of traffic flow, resulting in poor reliability of traffic flow forecasting model. To achieve high accuracy in the traffic flow forecasting model, in the opinion of the authors, one has to adopt more than a purely statistical approach. Rather, the model has to incorporate the dynamics of the traffic flow. To the authors’ best knowledge there is no effective method for the long-term traffic flow forecasting which may be any period from a few hours to one day, one week, one month or more. In order to develop an efficient forecasting model suitable for both short-term and long-term traffic flow, the following rationale and features are considered in developing the model: (a) Traffic flow is highly complex and not amenable to accurate mathematical modeling. Therefore, nonparametric methods and adaptive algorithms are required to learn and recognize patterns in an effective manner. (b) Prior knowledge of flow behavior should be used wherever possible to simplify the algorithm and improve performance. (c) The forecasting algorithm must be capable of real-time operation. Therefore, computationally intensive operations must be avoided. It has been demonstrated recently that adroit integration of wavelets with neural networks can result in a powerful approach for pattern recognition with enhanced feature detection capability [19]~[27]. Our review of the literature produced no research article on the use of wavelets for traffic flow forecasting. In this study, a novel dynamic time-delay recurrent wavelet neural network (WNN) model is presented for forecasting traffic flow. The model