A case study of developing personalized spatial cognitive road network and raster capable route finding algorithm for pedestrian evacuation behavior simulation

Abstract

Agent modeling simulation had been widely applied in pedestrian behavior study. However, different to vehicles that behave on linear vector road networks, the pedestrians should be considered as active dots that behave in two-dimensional raster road space during outdoor activities. Since taking into account a large number of cells in the raster space equally forming one link in the vector space, it is hardly efficient to apply traditional algorithms of path finding that aim to find a shortest path to simulations in raster spaces. In addition, the author argued that the assumption implicitly contained in those shortest route algorithms that the agents could search throughout the entire road network to obtain results could not be solid in real event. Therefore, two issues for pedestrian behavior simulation were addressed including an appropriately limited road network according to pedestrian finite spatial knowledge and the route finding algorithm under raster space conditions. In this paper, a field survey was conducted in a real event covering the involved area. By reviewing the survey, the author proposed to personalize the road network based on individual spatial cognition of participant respectively. All the roads in the study area were graded according to their perception of the participants involved in the event through the interview; the more often the road was identified by pedestrians participated, the higher level it was. Each pedestrian derived his own road network from the roads selected from different levels consisted with personal characteristics. The pedestrian agents acted in these separately independent road networks when performing a simulation. In this way, two adjacent agents probably could have different understandings of “shortest route” leading to the same destination, which better represented the diversity of the public in the real world. In the second part, the author attempted to develop a raster capable route finding algorithm utilizing the personalized spatial cognitive road network to simulate the pedestrian behavior. This algorithm profited by the essentials of the classic algorithms and served as a gateway from vector to raster. Two layers, cognition and implementation layers were constructed. On the cognition layer, the pedestrian agents recognized the roads as vector links. They searched their corresponding personalized spatial cognitive road network to find a route to the destination according to their spatial knowledge and behave pattern. The route probably consisted of several roads and several intersections. The intersections performed as check points on the agents' journey to the destinations. While detailing to the implementation layer, the intersections transformed from vector points to a cluster of raster cells. The pedestrian agents figured out the action route from current cell to a particular available intersection cell. Since the search was limited in a section of one road, the process could be practical acceptable in terms of the calculation time cost on a desktop PC. Through this paper, the author expected to make a step that the pedestrian behavior simulation will advance on integrating the human cognition from theoretical precision.