From ComplexWorldWiki
Jump to: navigation, search

DCI-4HD2D is formally the extension of the CASSIOPEIA project.

Problem statement

The goal of the CASSIOPEIA project was to improve the way future ATM operational concepts are modelled. The solution proposed entailed using a single software platform that contains all the elements necessary to model operational scenarios at different scales, allowing the researchers to provide further details to its elements which are saved for future scenarios if needed.

The main hurdle towards modelling the operational scenarios was designing the complexity of the system. Complexity should be understood here as the interaction of multiple elements which have a non-linear behaviour, driving the ultimate state of the scenario. The complexity of the system cannot be achieved using existing tools in the ATM domain through model design parameters, due to the associated non-linearities and the impossibility of deriving analytical models. The use Agent Based Modelling (ABM) allowed us to model the elements interacting, which will create the complex system.

The Flight Path 2050 presents Europe’s Vision for Aviation for the future. In what refers to air traffic management, this vision includes concrete goals for the punctuality of flights and capacity of the air traffic management system. Additionally, the document adds a concrete goal in what refers to passenger mobility, stating that 90% of the passengers should be able to travel door-to-door in Europe within 4 hours. After the success of the CASSIOPEIA project, an extension was approved to apply the platform to study the impact of Dynamic Cost Indexing on the ATM system on the 4 hour Door-to-Door objective.

Passenger mobility is the ultimate goal of the air transport system, which mission is to transport passengers and freight, not airplanes. However, punctuality is currently mostly measured as aircraft operations performance. Moreover, most air traffic management technology improvements are targeting aircraft punctuality and not passenger punctuality. Passenger punctuality depends critically on passenger connectivity, as a missed connection impacts very negatively in passenger mobility performance.

It was deemed necessary to study the impact of the following factors on passenger mobility:

  • Dynamic Cost Indexing,
  • SESAR operational enablers such as direct routing, extended approach manager, and collaborative decision making processes,
  • Operational ground improvements and airline operators buffer reductions,
  • Passenger gate-to-gate times, specially transit passengers in the context of the 4 hours door-to-door challenge,
  • Future passenger compensation regulations uptake, and
  • Airline schedule recovery strategies considering passenger connectivity as well as hard and soft passenger costs.

It was of great interest also the possibility to study the impact of these factors on other areas like operators' economics, and environmental impact in terms of emissions.


The CASSIOPEIA DCI-4HD2D project extension studied how changing the trajectory of each aircraft to either minimise fuel consumption or to minimise time to destination can be used as a adaptability mechanism, to work together with other ATM improvements, to address passenger connectivity. Understanding how this mechanism, known as Dynamic Cost Indexing (DCI) , increases the adaptability of the system, required the analysis, design and implementation of a complex software system as a collection of interacting, autonomous agents.

Agent based modelling (ABM) allowed the researchers to describe the behaviour of the different agents involved on the operations at the hub airport studied in a separated and detailed manner. When running the simulations with the different agents and their individual behaviour, a global emergent behaviour of the system was obtained.

The project modelled the operations of one day at a major European hub airport, for which the price of the tickets as well as the connections for each passenger were known. The schedule was modified to simulate the application of ground operations improvements, and SESAR improvements. The operations were modelled in such a way that each aircraft calculated its DCI based on a different strategy several times during its flight, considering also the connecting flights' costs, and the airline changed the departure time of outbound connecting flights when necessary to wait for passengers. This way each aircraft took into account every type of cost when applying a DCI change. Different scenarios were designed to permute the different variables of the system:

  • Operational improvements applied
  • DCI strategy
  • Fuel cost
  • System delay

Key results
Cost-related conclusions:

• A direct relationship between the application of ground improvements by reducing airline buffers on ground by 20% and an increase in airline cost has not been proven. Future work should study further reduction of the buffers to identify the optimum buffer for each operation. • Higher fuel costs leads to less flights deciding to increase the speed. However, the total delay experienced is similar and the passenger costs do not increase. • Application of cost optimisation strategies would reduce airline cost between 0.5% and 0.7%. This reduction is observed to be obtained by increasing the number of outbound connecting flight performing wait-for-pax and the duration of the waiting time to avoid passengers missing connections. When the amount of passengers claiming compensation increases, the optimised strategy improvement is reduced to 0.2%. • The initial delay in the system plays an important role on the total cost that airlines experience. The savings in fuel that are observed at the AMAN decrease as the delay increases and the extra costs in fuel and passenger costs increases when the system has higher delay.

Delay related conclusions:

  • The average gate-to-gate time for passengers improves marginally when ground improvements are implemented (an average reduction of 0.3% with respect to 2010 traffic trajectories); when adding SESAR improvements, the average gate-to-gate time is reduced in average 4.5% as routes are shortened.
  • The application of airline cost optimisation strategy increases gate-to-gate time in average 1.1%. However, it is important to look at the difference between connecting and non-connecting pax. The optimisation strategy increases non-connecting gate-to-gate time by 0.4%; this is due to the increase of wait-for-pax time and translates into a reduction of 0.8% for connecting passengers due to a reduction of passengers missing connections. Results show that aircraft waiting for passengers increase significantly when applying optimisation (aircraft applying wait for pax increase from 1.7% to 6.5%). Also, average waiting time increases from 7 min to 13-14 min (for nominal-high delay in optimisation strategy). Outbound flights waiting for passengers benefit from the possibility of applying DCI on their turn.
  • Increasing the number of passengers claiming compensation will reduce the amount of passengers missing connections when the aircraft operator costs are optimised (8.5% in nominal delay and 15% in high delay)

Efficiency related conclusions

  • In nominal conditions (nominal delay) with airlines optimising their strategies, the increment in fuel cost will make airlines fly slower, reducing fuel consumption and emissions by approximately 25%.
  • An optimisation that allows speed variations lead to lower emissions (i.e., lower fuel consumptions) than current operations. The main reason for this is that the optimisation is considering the total cost, including fuel consumptions, while in the current operations, even if a few flights recover delay, there is no assessment on the fuel that that recovery will represent. Moreover, in the optimised strategy wait-for-passengers seems to be playing a role as important as speed variations to minimise airline operations costs. The extra cost of fuel due to speed increments lead to small speed increments on the optimised scenarios.
  • Higher initial delays lead to lower holding delay but the biggest difference is between current and optimised strategies. In the optimised strategy the delay increases significantly. Part of the reason might be due to speed selected to save fuel at the AMAN phase in the optimised strategy. This might be the reason behind why not extra delay is saved on the optimised strategy. In the modelling of current operations, the first slot available is assigned to the arriving flights regardless of the potential fuel usage.

Future steps

Further analysis

The focus of the project was the analysis of delay recovery strategies and how their optimisation could improve airline costs and reduce passenger delay. This has derived into a series of metrics and indicators to provide the information the researchers thought useful, and explored certain relationships among these metrics. However, further work could be executed by exploring other metric dependencies among the existing indicators, generating further indicators to explore more dependencies, or even changing the focus of the model as explained later in the model and platform enhancement. Further analysis could be conducted to explore the sensitivity and stability of the solutions; for example, the capacity of the airport has been set so that it did not generate significant delay as that was not the focus of the project, but different arrival capacities could be assessed.
Model and platform enhancement

The conclusions of the model invite to explore in further detail the efficiency of the AMAN. The runway throughput is controlled by the AMAN and there is room for improvement, since its time horizon could be extended further than 60 minutes. This would entail designing new algorithms to better negotiate the arrival time of the flights and maintain the flexibility required for last minute changes, especially for those flights which are closed to the destination airport, which need an arrival slot subject to change. The AMAN slot assignment algorithm could also be improved using learning algorithms through which it could predict delays and modify slots accordingly.

Currently the optimisation strategy considers modifying the arrival times of flights in a downstream manner, however, while it would be challenging, it could be possible to implement algorithms for modification of upstream flights, meaning that different inbound flights optimise their cost index based on the modification of other inbound and common outbound flights. This, however, would create a ripple effect, which entails a computation challenge for the model.

The cost distribution for the different airlines is usually quite different; therefore, it would seem appropriate to include differences on the strategies based on the operator classification.

From the computer science perspective, it would be a great improvement the development of automated data analysis, which could be used for the vast amount of results obtained.

Concluding remarks

The CASSIOPEIA platform has proven very useful to assess scenarios where different individual agents play a role, such as the one presented in DCI-4HD2D.

Each flight in the model does not just compute its cost index dynamically, but also computes it in collaboration with the rest of the flights which in turn update their own strategies. In this sense flights under the same operator act as a network, or rather as a system of systems. On top of that, since decisions and proposals are continuously shared and updated between flights sharing passengers, feedback loops appears increasing the complexity of the system. Agent Based Modelling has proven a suitable tool to model these interactions and ultimately reveal some emergent behaviours not expected from the initial strategies. These behaviours cannot be model with other techniques and therefore, to the best of our knowledge, this combination of functionalities far exceeds what any other ATM platform of this kind can realistically model.

The project has focused on two main performance areas, cost and delay, gaining a deep understanding on how those two performance areas relate to each other and are impacted by different operational improvements or traffic conditions. Speed variations and AMAN holding delay have also been found as important indicators of the efficiency of the model. Additionally, other efficiency factors such as emissions can be derived of the results.

Finally, it is worth mentioning that the variety and complexity of the functionalities implemented increased significantly the effort and duration initially allocated. Nevertheless, the scope of the project has not been compromised and the results obtained, as well as the potential for its enhancements and future applications to other scenarios, exceed the initial expectations of the project team.

Personal tools