Page - 380 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

If the agreement is not reached before the deadline, the AOC supervisor ends the negotiation with the compatible offer that has the most favourable votes. The AOC supervisor can employ a time-based concession strategy in order to decide which offer to announce. I.e., he first offers the best solution according to his own evaluation function and concedes over time. That is, in the next round he offers his second best solution, then his third best and so forth Specialist Agents’ Acceptance Strategy: When the AOC supervisor makes an offer, the utility of this offer, U(o) is calculated by each specialist agent. If the utility of the given offer is greater than the threshold value, the specialist agent accepts the offer; otherwise, it votes to reject it. Note that agents determine their threshold value before the negotiation starts. Before negotiation starts, each specialist agent determines the importance of their evaluation criteria by setting weights directly, or using the Analytic Hierarchy Process [7] to estimate the weights. By using pairwise comparison such as “Key Performance Area (KPA) 1 is two times as important as KPA2”, agents can estimate the weight values of each criterion (e.g., w_KPA1=0.7). Note that the sum of those weight values is equal to one. The preferences of each agent can be modelled by an additive utility function where the utility of an offer is estimated by the weighted sum of evaluation values for each criterion. These evaluation values can be estimated by using domain knowledge. For example, if the agent estimates EV_KPA1(o)= 0.8 and EV_KPA2(o)=0.6 and the criterion weights are w_KPA1=0.7 and w_KPA2=0.3, then the utility of the given offer is calculated as U(o)= 0.7*0.8+0.3*0.6. 3. Agent-Based Modelling Developing the AOC agent-based model is performed in four major steps. In step 1, the agents are identified. Since the purpose of the simulation model is to compare different coordination strategies, the main agents are those human operators involved in managing the disruption and the decision support systems they use. Thus, all actors were modelled as proactive agents. The complete list of agent types is provided in Appendix A. Once the key agents have been identified, their behaviour in the context of the considered scenario is accurately specified in the step 2 based on data and interviews with experts. Subsequently, interactions between the agents are implemented in the modelling environment in step 3 and the model is verified and executed in step 4. To represent the agents and their behaviour we employ a generic temporal-causal modelling approach [8]. The ontology used can be found in Appendix A. 3.1. Case Study In order to assess the impact of the four policies (P1-P4) we will consider a challenging AOC scenario that is well described and evaluated in [7] and includes an overview of the flights being monitored by the airline controller at the time of disruption [9]. The scenario concerns a mechanical problem with an aircraft at Charles de Gaulle (CDG) airport, aiming for a long-haul flight to a fictitious airport in the Pacific, which is indicated by the code PCF. In [7], this scenario was considered by a panel of AOC experts. They developed several alternatives, and subsequently identified the best solution, which was to re-route the flight from CDG to PCF and to include a stop-over in Mumbai (BOM). In S.Bouarfaetal. /AMulti-AgentNegotiationApproach forAirlineOperationControl380