A Two-phase BP neural network method to predict average delay of signalized intersection under multi-saturation traffic states

Urban intersections' average delay is a kind of basic data of the modern intelligent transportation system (ITS), used in real-time navigation, emergency traffic management and signal control. In this paper a Two-phase Back-Propagation (TBP) neural network model is introduced, which takes real-time volume, average speed and time occupancy as its inputs and outputs the intersection's average delay of the next time step. An important characteristic is its flexibility to multi-saturation traffic states. The method is tested and verified using data from VISSIM simulation platform, which achieved satisfactory results.