A Novel Learning Heuristic Applied for Computationally Hard Managerial Decision Making and Transportation Operations Control

摘要

Planning and scheduling transportation operations is considered computationally hard and time demanding. We consider a complex network of interconnected links across a set of locations and a set of vehicle movements on these links to fulfil some traffic demands. In such a set-up, basically two types of computational solutions are sought: one is a decision making, if a link (and/or a vehicle) is to be included in a route and second, is computation of a specific time duration during which a link and a vehicle are allocated to fulfil a traffic demand. Such planning efforts are undertaken at a strategic level. At the operational level, the planned schedules are executed. For any unforeseen reasons, if a planned schedule gets disrupted, while en-route, the entire planning efforts turn futile. The situation turns an emergency, as the available time to restore the original schedule is very small and it also knocks down other transport schedules that follow. Using an available approach to reschedule disrupted transportation services, we developed a learning architecture that is capable of learning and creating a knowledge base from the solved instances of disruption resolution. When a new disruption happens, the learner deduces the applicability of any solved instances in the knowledge base to quickly resolve the disruption. If not, the disruption is resolved completely and the resolution is added to the existing knowledge base. The combined abilities of the agents based architecture include monitoring, identification, deduction, decision making, and resolution execution, by learning, storing solution seeds and facilitating quick resolutions. The architecture and the application were developed using JADE Giovanni et al. (2007) toolkit on Ubuntu OS; we verified the deployability of the system on a Windows machine. The agent architecture and empirical results are presented in this paper. Our proposed architecture is faster than traditional methods and novel as an evidence of agent capabilities in solving real-life complex problems and safety-critical domains.