La géosimulation orientée agent : un support pour la planification dans le monde réel
|Abstract:||Planning becomes complex when addressing uncertain situations. Accurate predictions remain a hard task for human planners. The Simulation-Based Planning approach consists in associating planning and simulation. Each generated plan is simulated in order to be tested and evaluated. The most appropriate plan is kept. The problem is even more complex when considering spatial constraints. For example, when fighting a wildfire, dozers build a firebreak to stop fire propagation. They have to take into account not only the fire spread but also the terrain characteristics in order to move easily. We propose an agent-based geosimulation approach to assist such planners with planning under strong spatial constraints in a real large-scale space. Forest fire fighting is a typical problem involving planning within an uncertain real world under strong spatial constraints. We use this case to illustrate our approach (ENCASM). The approach consists in drawing a parallel between the Real Environment RE (i.e. a forest in fire) and the Simulated Environment SE (i.e. a virtual reproduction of the forest). Spatial data within the SE should absolutely come from a GIS (Geographic Information System) for more realism. Real planners such as firefighters or dozers are simulated using software agents which reason about the space of the SE. To achieve a sufficient spatial awareness (taking into account all terrain’s features), agents have advanced capabilities such as perception. Using a multiagent geosimulation approach, we can generate a realistic simulation of the plan so that human decision makers can visualize the probable consequences of its execution. They can thus evaluate the plan and adjust it before it can effectively be executed. When the plan is in progress and in order to maintain coherence between RE and SE, we keep track in the SE of the real planners’ positions in the RE (using georeferencing technologies). We periodically replan the rest of the plan starting from the current position of the real planner. This is done in order to anticipate any problem which could occur due to the dynamism of the RE. We thus enhance the process of the classical Distributed Continual Planning DCP. Finally, the agents must replan as soon as an unexpected event is reported by planners within the RE. Since plans in the studied case (forest fires) are mainly paths, we propose a new approach based on agent geosimulation to solve particular Pathfinding problems. Besides, our approach highlights the benefits of the agent-based geo-simulation to the collaboration of both humans and agents. It thus shows: • How spatial cognitions of both spatially aware agents and human planners can be complementary. • How agent-based geo-simulation can complement human planning skills when addressing complex problems. Finally, when applying our approach on firefighting, we use MAGS as a simulation platform and Prometheus as a fire simulator. The main contributions of this thesis are: 1. An original architecture (ENCASMA) for the design and the implementation of applications (typically, natural disasters applications) in real, dynamic and large-scale geographic spaces. 2. An agent-based approach for particular Pathfinding problems (within real and spatially constrained environments and under qualitative constraints). 3. An enhancement of the DCP (particularly, the continual process) approach in order to overcome some limits of the classical DCP. 4. A practical solution for a real and complex problem: wildfires fighting. This new solution aims to assist experts when planning firefighting actions and monitoring the execution of these plans.|
|Document Type:||Thèse de doctorat|
|Open Access Date:||11 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.