Intelligent long-term traffic forecasting system using deep neural network

다국어 초록 (Multilingual Abstract)

An intelligent transportation system is a next-generation traffic system that embraces the latest information and communication technology. As the traffic system becomes large and complex, artificial intelligence technology is essential for providing safe and fast traffic services. From various artificial intelligence technology, deep learning shows the best performance for learning complex models. Therefore, research on deep learning-based intelligent transportation system applications is carried out by researchers.
This study reports a deep learning based novel intelligent traffic forecasting system. The forecasting system can predict the traffic conditions of highways several months ahead, reflecting weather, time, and road structure information for a more precise forecast. The internal architecture of the system adopts the modules of a neural network and enables data separation during computation. Thus, the system presents the dependable traffic forecast, even under conditions, e.g., rush hours, weather changes.
For the practical traffic forecast service, faster training and prediction are necessary. I embrace parallel computing to speeding up the traffic forecasting system. For parallelizing the training and prediction of the predictors, training and forecasting tasks are distributed to the computing nodes.
The forecasting accuracy of the framework is evaluated for different cases, such as ordinary days, vacations, and bad weather days. Comparing with other traffic prediction models including other machine learning based model and the time series model, the proposed scheme achieves improved prediction. In experiments with multiple computing nodes, execution times of both training and forecasting processes are reduced.

목차 (Table of Contents)

• ABSTRACT 4
• Contents 6
• List of figures 8
• List of tables 12
• 1. Introduction 13
• 1.1. Literature review 15
• 1.2. Outline of the dissertation 17
• 1.3. Lists of symbols and notations 18
• 2. Design of traffic forecasting framework 20
• 2.1. Data used in the traffic forecasting system 20
• 2.2. Procedures of the traffic forecasting system 24
• 3. Implementation of traffic forecasting framework 27
• 3.1. Preliminaries of deep neural network 27
• 3.2. Structure of traffic predictor 34
• 3.3. Speeding up of traffic forecasting 45
• 4. Experimental results and discussions 54
• 4.1. Experiments on prediction performance 55
• 4.2. Experiments on boosting up traffic forecasting system 86
• 5. Conclusion and future works 90
• Reference 93
• 국문 초록 98