Neuro-fuzzy traffic signal control for queue length management

Queue length management is the need strategy at signalized intersection, in order to increase signal operation efficiency with preventing spillback and offer equity on both directions. The objective of this queue control is to minimize the difference of queue length and the summation of queue length at signalized intersection. To minimize the difference, we used the sum of queue volume and link demand volume when calculating green time ratio. We designed the queue control model with variable cycle length in order to minimize total queue volumes. It was necessary feedback control to calculate the optimal cycle length and green time ratio that the optimum could satisfy two goals; minimizing difference and summation both queue lengths. We devised the queue length management with neuro-fuzzy system in order to find optimal solution, due to self-learning ability of neural network and correction of uncertainty from fuzzy theory. It was necessary repeated learning features of neural network theory to know good signal timing with reducing total queue length and balancing both queue lengths. We evaluate the effectiveness to analyze this model on five traffic volume scenarios with compounding between saturated condition and oversaturated condition. We compared the results with Webster's green split method, and we extended the range of cycle length to calculate better solution and to know the ability of application. The analysis results revealed that this queue control model with neuro-fuzzy system had a good effect to reduce congestions and to keep balancing queue length on both directions.