Cognitive dynamic airspace management

This thesis focuses on optimisation methods in airspace management. A given airspace is typically divided into sectors where a pair of air traffic controllers are assigned to ensure safety in each sector. The controllers’ workload can be generally divided into interactions with (a) pilots in the given sector and (b) other controllers when the aircraft exits their sector and enters into another sector. These interactions are referred to the monitoring and coordination workloads respectively. This research work specifically studies the optimisation methods applied to the optimal design of sector shapes in four aspects. • Static state: The formulation of finding sector shapes have been mostly proposed in a single objective optimisation framework. However, given the conflicting nature of the considerations, the problem should be formulated as a multi-objective problem incorporating the preferences of the user. • Dynamic state: Re-sectorization strategies for reconfiguration or change in sector shapes have been suggested to handle the changing air traffic flow. However, the dynamic airspace sectorization problem (DAS) has not been explored as a system that has varying traffic flows over time. • Weather: Weather plays a significant role in the availability of airspace. However, most work that addresses changing weather conditions focuses on the re-routing of affected airways, which could, in turn, cause an imbalance of controllers workload. The combination of re-routing and resectorization should be considered simultaneously under the effect of limited available airspace. • Multi-objective solver: Most multi-objective solvers stem from evolutionary algorithms (EAs) due to its population-based nature. Multi-objective EAs are fundamentally stochastic in nature, hence solutions are likely to be irreproducible. For reproducible and reliable solutions, deterministic algorithms are preferable. In this thesis, a preference-based bi-objective optimisation model that optimises sector shapes for a given set of traffic flows and airspace is first presented. The two objectives are i) minimizing the standard deviation of the monitoring workload (within the sector) among pairs of controllers and ii) minimizing the total coordination workload between sectors. The proposed model aims to obtain traffic flow conforming sectors while equally distributing the monitoring workload among controllers as much as possible. Furthermore, preference-based methods were used to help the solver focus on the particular region of interest on the approximate Pareto front. The proposed preference-based strategy was found to obtain a wider range of feasible solutions when compared to a constraint-based strategy. Given dynamically changing traffic flows, a single set of sector shapes could not remain optimal over time. Hence, the airspace management problem should be deemed as a system over time, re-sectorizing sectors when needed. The cognitive decision making architecture for dynamic airspace sectorization (CDAS) is first proposed to answer the questions on when-to-do and how-to-do a resectorization. With a multi-objective framework, CDAS provides the decision maker with the predicted performance objectives (based on flight plans) of available optimal sector shapes (for selection) for the next time period. This could allow the decision maker to avoid a need for resectorization in the next time period. However, there are still some uncertainties present in the feasibility of the sector shapes in future time intervals. Focusing on the benefits of planning, the airspace management problem is fitted into a rolling horizon optimisation framework with a single objective optimisation model. This approach optimizes the sector shapes with the consideration of traffic flows in the next few time intervals. In comparison with the single time interval optimisation, the proposed method is able to provide better feasible solutions over a time horizon. On top of varying traffic flows, dynamic weather conditions can affect the availability of airspace. The Simultaneous Optimisation of the AirWay and AirSpace (SAWAS) is proposed to address this changing availability of airspace. The model seeks to balance the ATC monitoring workload, minimize the total coordination workload and maximize the similarity of sector shapes with the initial sectorization. An experimental study was performed to compare this methodology with a sequential design approach. It was found that the concurrent consideration of sector re-design and flow re-routing on the design objectives yields better and more optimal solutions. The reproducibility and reliability of solutions in real-world problems are not guaranteed with stochastic solvers. Therefore, a deterministic indicator-based multi-objective Multi-scale Search Optimisation (MSO) algorithm, Pareto-Aware Dividing Rectangles (PA-DIRECT) is proposed to tackle this issue. PA-DIRECT is benchmarked against non-dominance-based multi-objective MSO algorithm, MO-DIRECT and popular evolutionary algorithms on a bi-objective test set on the Comparing Continuous Optimisers (COCO) platform. The study results affirm the performance of PA-DIRECT in providing a high-quality approximate set, particularly for multi-modal problems. Further, PA-DIRECT is used to solve the aforementioned preference-based bi-objective optimisation model with general constraint handling techniques. In summary, the thesis first introduces a preference-based approach for the designing of optimal sector shapes. Next, two different frameworks that consider future traffic flows are proposed for DAS. Following that, the changing availability of airspace is tackled by the simultaneous optimisation of re-routing and resectorization. Last but not least, a deterministic multi-objective solver is developed to find optimal airspace sector shapes. Future works include: i). a reference-point-based MSO algorithm for optimizing sector shapes; ii). interactive multi-objective optimisation approaches for optimizing sector shapes; iii). extending the rolling horizon optimisation framework with a multi-objective optimisation model; iv). solving the rolling horizon optimisation framework with a reinforcement learning approach; v). implementing SAWAS as a module in the CDAS framework; and vi). simulations for a better measurement of controllers’ workload.Doctor of Philosoph