Aprender com o Passado: Apoio à Negociação Automática nos Centros de Controlo Operacionais

The process of planning and scheduling the flights of an airline consists of several steps, some of which are prepared several months in advance. Even though, having a great plan is as important as keeping it, this task can be quite demanding due to unexpected events (disruptions) that can occur close to the operation date. Such problems can lead to delays and/or cancellation of flights, if nothing is done to prevent it. In the Laboratorio de Inteligencia Artificial e Ciencia da Computacao (LIACC) is being developed a project called MASDIMA (Multi-Agent System for Disruption Management), in collaboration with TAP Portugal. This project addresses the problem of managing disruptions by implementing a multi-agent system that uses automatic negotiation to solve problems on cooperative distributed environments, applied to the scenario of Airlines Operation Control (AOC). The aim of this dissertation is to incorporate in MASDIMA, an additional software layer, on the set of agents responsible by the generation, analysis and decision regarding new solutions, so they can learn from the past. Therefore, it was investigated a way of having the system solve current problems based on its knowledge of similar situations occurred in the past. In order for this to become a reality we created a new methodology called Case-based Reasoning Distributed for Dynamic Environments (CBR-DDE), based on Case-based Reasoning (CBR). Using this methodology we are able to resolve problems, learning from the past, on the AOC. Using metrics such as the average response time of the system to a new case and the quality of the proposed solutions, the results obtained with CBR-DDE methodology are compared with those obtained by human operators on TAP Portugal Operational Control Center and the pre-existing MASDIMA approaches. We conclude that the objectives of this work were achieved.