Abstract
This doctoral research critically and comprehensively addresses the issue of traffic congestion at signalized road intersections by developing a predictive approach using traffic flow variables obtained from seven roadsites within the South Africa Road network. These roadsites are all interconnected to the N1 Allandale interchange, one of the busiest road networks in Africa. Traffic congestion is an important aspect of the road transportation system, and it plays a significant role in determining the traffic flow of vehicles at road intersections. Over the past century, there has been a dramatic increase in traffic congestion in developed and developing countries. Several models have been used to explain where, when, and how vehicles move through a network of road intersections. Recent developments have seen classical models adopted on major road interchanges and intersections to address traffic congestion. However, far little attention has been paid to developing soft computing techniques in modelling traffic flow at signalized road intersections. Another important research question this research investigates is a need to determine which traffic flow parameters in traffic datasets can be used to understand the traffic flow patterns needed to develop a model capable of predicting traffic flow at signalized road intersections. In this research, soft computing techniques such as ANN-PSO, ANFIS, and ANN models were developed for modelling vehicular traffic flow at a signalized road intersection using the vehicular speed of each category of vehicles on the road, traffic density, time, and traffic volume as input and output variables. Four hundred and thirty-four traffic datasets were obtained from the Allandale interchange (the N1 route in South Africa). The traffic data were collected from seven road intersections connecting to the N1 road network using inductive loop detectors, video cameras, and GPS-controlled equipment. The results obtained from this research has shown that the application of the ANN model on each of the seven (7) roadsites produced a testing performance of 0.99975, 0.99169, 0.99686, 0.99767, 0.99974, 0.99780, and 0.99629, the ANFIS model results show a testing performance of 0.9370, 0.9952, 0.9790, 0.9987, 0.9826, 0.9940 and 0.99140, and the ANN-PSO, which is a hybrid model as the ANFIS model has a testing performance of 0.98220, 0.98940,0.99460, 0.99770, 0.99710, 0.99300 and 0.99140. These results suggest that the approaches proposed in this study could be used to predict and analyse traffic flow with a relatively high level of accuracy. This research shows that the soft computing techniques developed in this research can model vehicular traffic flow at each road vi intersection. The evidence from this research suggests that the ANN, ANFIS, and ANN-PSO models are an appropriate predictive approach for modelling vehicular traffic flow at a signalized road intersection. Another significant finding from this research based on the results is that understanding traffic flow patterns that involved multiples nonlinear variables difficult to address using analytical formulations could be analysed using soft computing techniques. The modelling approaches proposed in this study will assist transportation engineers and urban planners in developing ways to use their respective country's datasets in understanding traffic flow patterns, what kind of traffic flow variables to use in developing their respective predictive models, and in designing a traffic control system for traffic lights at road intersections.
Ph.D. (Mechanical and Industrial Engineering Technology)