Development of an Agent-based Collective Behaviour Model for Simulating Pedestrian Movement

Pedestrian movement inside a building is an important issue in today’s mega buildings.The improper spatial design of a building can cause overcrowding or even catastrophicstampedes. It can also cause fatalities when evacuating the building during a fire.Pedestrian movement in a building can be analysed using computer simulations duringbuilding design. Currently, discrete models (the cellular automata (CA) model) andcontinuous models (the social force model and its variants) are the two main approachestaken in modelling microscopic pedestrian movement. Both can be considered as agent-basedmodels. Each pedestrian is an agent who changes his or her movement accordingto the physical changes in the neighbouring environment. The behaviour of each agent isgoverned by a set of rules developed from different theories in mathematics and physics.The parameters of these simulation models must be calibrated using real pedestrianmovement records. Pedestrian movement is a kind of human behaviour. A pedestrianreceives information through his or her vision and makes decisions accordingly.However, the decisions may not only depend on the physical information about thesurrounding environment, they may also be governed by many psychological factors inresponse to what is seen.The sufficiency of the current mathematical models used to simulate pedestrianmovement is debatable. This project proposes to model pedestrian movement inversely.Instead of developing mathematical models to describe the movement behaviour, we willtrain a statistical learning model with actual pedestrian movement and informationabout the pedestrians’ surrounding environment, such as the movement of neighbouringpedestrians and the geometrical relationships between pedestrians and neighbouringobstacles and exits. The model will learn the highly nonlinear correlations between themovement of a pedestrian and his or her neighbouring environment from the collectedreal data. Once the model has been trained, it will be used as the agent of the pedestrianin a CA framework to establish a complete agent-based collective behaviour model forsimulating pedestrian movement. This model will be verified using different realpedestrian movement records. This approach inversely develops a pedestrian movementsimulation model and circumvents the difficulties in traditional mathematical modellingof pedestrian movement.