A neural network based modeling and simulation of bicycle conflict avoidance behaviors at non-signalized intersections

The modeling and simulation of bicycle conflict avoidance behaviors with other vehicles (motor-carother bicyclespedestrians) and the obstacle (safety island, fence etc.) at non-signalized intersections is of great importance in junction analysis. However, it is very difficult to simulate the conflict avoidance behaviors of individual bicycle because of the great variations in the cycling behaviors, conflict characteristics and traffic environment. A computer-based four-layered back propagation neural network (NN) model was developed for bicycle conflict avoidance behaviors modeling. The NN model was trained, validated with field data and then compared with Social Force model. Results showed that the NN model could produce reasonable estimates for individual bicycle conflict avoidance behaviors at non-signalized intersections.