Please use this identifier to cite or link to this item:
Title: Spatio temporal traffic pediction for a large-scale road network
Authors: Krishnamoorthy, Nardana
Keywords: Engineering::Electrical and electronic engineering
Issue Date: 2019
Abstract: In a country like Singapore, which is adapting the Smart Nation program, Intelligent Transport Systems (ITS) plays a key role. In order to improve the travel experience by proper planning and avoidance of traffic congestion, traffic prediction is the key. The potentials of deep neural networks for real-life applications of various domains can be viewed in the recent years. Traffic speed prediction models that learn and adapt to the spatio-temporal correlations of the traffic network are always effective. This thesis deals with the problem of traffic speed prediction by developing a hybrid model that combines the known benefits of both CNN and LSTM neural networks. The traffic data collected from Land Transport Authority Singapore was utilized to predict the traffic speed of road segments in an express highway. Initially road segments in the considered Singapore road network were grouped into four clusters by means of Kmean clustering based on their average speed. The road segments belonging to one of these clusters were then tested. The architecture of the hybrid model was tailored according to the requirements of the test road network. The performance of the optimal hybrid model was then compared with LSTM and CNN models, which showed that the hybrid model performed better in terms of MAPE (Mean Absolute Percentage Error) and MSE (Mean Squared Error). By adapting to both the temporal and the spatial domain, the hybrid model gives better results. This proves the superior ability of the hybrid model is applicable even on the complex road networks.
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Theses

Files in This Item:
File Description SizeFormat 
  Restricted Access
2.16 MBAdobe PDFView/Open

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.