Cai, C.; (2010) Adaptive traffic signal control using approximate dynamic programming. Doctoral thesis , UCL (University College London).

Preview

PDF 20164.pdf Download (2MB)

Abstract

This thesis presents a study on an adaptive traffic signal controller for real-time operation. An approximate dynamic programming (ADP) algorithm is developed for controlling traffic signals at isolated intersection and in distributed traffic networks. This approach is derived from the premise that classic dynamic programming is computationally difficult to solve, and approximation is the second-best option for establishing sequential decision-making for complex process. The proposed ADP algorithm substantially reduces computational burden by using a linear approximation function to replace the exact value function of dynamic programming solution. Machine-learning techniques are used to improve the approximation progressively. Not knowing the ideal response for the approximation to learn from, we use the paradigm of unsupervised learning, and reinforcement learning in particular. Temporal-difference learning and perturbation learning are investigated as appropriate candidates in the family of unsupervised learning. We find in computer simulation that the proposed method achieves substantial reduction in vehicle delays in comparison with optimised fixed-time plans, and is competitive against other adaptive methods in computational efficiency and effectiveness in managing varying traffic. Our results show that substantial benefits can be gained by increasing the frequency at which the signal plans are revised. The proposed ADP algorithm is in compliance with a range of discrete systems of resolution from 0.5 to 5 seconds per temporal step. This study demonstrates the readiness of the proposed approach for real-time operations at isolated intersections and the potentials for distributed network control.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as JSONCSVXML

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item