Enhanced adaptive neuro-fuzzy inference system for the control of vehicle traffic at road intersections

Abstract

The conventional method of traffic light signal control at intersections present undesirable performance issues which include unintelligent signal pre-set. These issues do not give control based on traffic demand at the intersection. Improvements on the fixed-time system led to the use of adaptive control systems that utilise artificial intelligence techniques, reinforcement learning and hybrid systems among others. Mamdani-based fuzzy logic controllers are fast but lack accuracy whilst Sugeno-based fuzzy logic controllers give accurate results with complexities. Moreover, the fuzzy logic controllers are poor learners and artificial neural network controllers are good learners but exhibit poor adaptation to changing traffic parameters. Adaptive neuro-fuzzy controllers designed so far, based on Sugeno fuzzy inference system, give more accurate results but are computationally expensive and time consuming in taking decisions. In this research, an enhanced adaptive neuro-fuzzy controller is developed based on Mamdani fuzzy logic control and Sugeno adaptive neuro fuzzy inference system algorithms in order to address the trade-off between accuracy and computational complexity. Data from Legon-Adenta-Madina Old Road Agbogba intersection in Accra, were utilised to design the enhanced controller together with three other controllers namely, Sugeno adaptive neuro-fuzzy inference system controller, Mamdani-based and Sugeno-based fuzzy logic controllers. These were applied to a first-in first-out queuing system for the purpose of comparison. Computer simulations were performed with MATLAB/Simulink software using SimEvent blocks. The two adaptive neuro-fuzzy controllers gave under dense congestion, a cumulated throughput of 627 vehicles in 2 seconds with about 5 seconds delay in red phase; while under less dense congestion, gave out a cumulative throughput of 96 vehicles in 2 seconds with a delay of 1.8 seconds in red phase. The superiority of the enhanced controller over the Sugeno type resides in solving the computational complexity problem (i.e. the enhanced controller has both accuracy and simplicity). This implies that, the enhanced controller operated on 8 rules compared to 125 rules for Sugeno type. However, the robustness of the enhanced controller needs to be field confirmed so that it is deployed in Ghana to reduce vehicle traffic congestion and incessant delays at road intersections.