Prediction architecture of deep learning assisted short long term neural network for advanced traffic critical prediction system using remote sensing data

• Main Authors: Wei Wang, R. Dinesh Jackson Samuel, Ching-Hsien Hsu
• Format: Article
• Language: English
• Published: Taylor & Francis Group 2020-07-01
• Series: European Journal of Remote Sensing
• Subjects: data extraction; neural convolutional short long term and dynamic nonlinear traffic
• Online Access: http://dx.doi.org/10.1080/22797254.2020.1755998

This paper presents a Neural Convolutional Network (NCN) based approach for learning traffic as images and predicting high accuracy network-wide broad traffic speed. In the recent past, images describe time and space of traffic flow, where a 2-dimensional time-space matrix is used to convert space dynamics. In recent years, neural networks have been widely used for the prediction of short term traffic, where the description is covered by an NCN in two consecutive steps: abstract data extraction and network-wide traffic forecast. This paper proposes Prediction Architecture of Neural Convolutional Short Long Term Network (PANCSLTN) for the purpose of effectively capturing dynamic nonlinear traffic systems with deep learning assistance. The PANCSLTN can resolve the problem of backdated decay error via memory blocks and shows superior prediction capacity for time series with long-time dependency. Moreover, PANCSLTN can determine the optimum time laggards automatically and the travel data from Beijing microwave traffic detectors which are used for model the training and testing to validate the effective PANCSLTN using remote sensing technique. PANCSLTN can deliver the most accurate and stable prediction performance compared to different topologies in dynamical neural resealing networks or other dominant parametric and nonparametric algorithms during experimental analysis.