| dc.description.abstract | Tunnel traffic congestion can increase the risk of traffic accidents, tunnel fires, and environmental effect. Despite numerous studies on traffic forecasting using deep learning, research on tunnel traffic remains limited.Utilizing traffic flow data from the Norwegian Public Road Administration, this thesis analyzes the applicability of recurrent neural networks for tunnel traffic prediction. The data is retrieved from different sources and traffic sensors near or inside the tunnels are selected through a geo-spatial analysis. The recurrent neural network is designed to be trained on either a single tunnel or several tunnels. Furthermore, based on their geographical location and population density, the tunnels are classified as urban or sub-urban. Based on the results of the experiments and the sample of tunnels used,the recurrent neural network outperformed the baseline for urban tunnels in terms of root-mean-squared-error. However, the performance advantage
was not significant for sub-urban tunnels. The addition of features such as temporal features and category features provided no significant results.These findings are discussed in the final sections of the thesis. | |
