The creation, development and modification of complex systems is becoming increasingly commonplace, but it remains a costly and time consuming activity to predict the behaviour of such systems due to their data intensive nature. Such systems are often dynamic in nature and may not always lend themselves to quantitative behavioural analysis. Air Traffic Management (ATM) systems are examples of these complex, dynamic systems and the European and US systems are both currently undergoing redevelopment initiatives; Single European Sky and NextGen respectively. The current ATM climate, which sees these large scale redevelopments, provides a system that would benefit from an understanding of the relationships and interdependencies which it comprises.
The focus of this research is therefore to develop an analysis that identifies these relationships and interdependencies whilst taking into account the problem characteristics associated with a complex ATM system. The analysis does this by invoking a change in an element within the architecture and measuring the deviation experienced by the other elements, with the deviation termed as impact; where an element is defined as being any part of the system whose state or operational ability is of interest to the user, and which is capable of impacting, or being impacted by, other elements.
Impact, in this context, is defined as being the measure of how a system element deviates from a pre-defined “normal” operational mode due to a change in the operation of another element existing within the same architecture, or due to an interaction with an external environmental condition.
Approach to Modeling
After some investigation, it was decided that the agent-based platform, JACK, would be a good fit. The impact analysis is being applied through system operational simulations and therefore needs to model all elements of the system. Within the ATM system which will be tested, the system elements may have the ability to make decisions based on their current information and this must be demonstrated in the software tool. Agent based modeling incorporates this concept with the agent construct being able to choose which plan to follow to achieve its goals. Furthermore, this modeling approach has the capability of inter-agent communications meaning agents can share information to achieve a goal that mirrors real world systems where two or more system elements must work together.
The JACK platform has the necessary features required to create a model capable of performing impact analysis on a complex system.
For this study, it was assumed that there is a single controller within each airport responsible for take-off and landing and a controller in the enroute sector responsible for directing aircraft between the two airport sectors. A further assumption was that the airports each had a single runway. Elements required in this system include 2 airports, 3 sectors, 3 controllers and 5 aircraft.
- The airport is modeled as an object with two states; available and unavailable, a reference to the availability of the runway.
- Controllers are modeled as objects with a variable stating the number of aircraft each currently handles.
- The sectors are modeled by applying a location to each aircraft; therefore an aircraft can be either in Belfast, Enroute or Liverpool.
- The aircraft are modeled in the system as agents with a “flying” capability.
This capability construct posts the “Request_flight_start” event which is in turn handed by the “Flight” plan construct. This is shown below for the Aircraft agent.
In modeling this impact analysis it is necessary to include a construct responsible for monitoring the system and calculating the impact during the scenario timeframe. This is referred to as the “Impactor” and is modeled using the agent construct. This agent has the “Impacting” capability which posts the “Start_monitoring_Impact” event that is handled by the agent using the “Impact_calculation” plan. The structure of the Impactor agent is shown below.
The model is compiled and run using the JACK compiler tool, which prompts the user to enter the appropriate information including flight details. The system is then initiated with the aircraft and impactor agents being created. Each agent has its own unique flight details as variable within the construct and therefore will wait until the designated time of departure before requesting take-off. If the runway is available the aircraft will enter the take-off phase with duration of five minutes. During this time the runway is unavailable to other aircraft for any purpose. Should the runway be unavailable the aircraft will wait for 1 minute before repeating the request. Upon completion of this phase the aircraft moves into the enroute stage of flight which is 30 minutes for Liverpool to Belfast and 40 for Belfast to Liverpool. The aircraft will then request to land at the destination airport which again will depend on the availability of the runway with the same procedure for repeated request being applied as in take-off. The landing phase takes a further 5 minutes and then the flight is completed.
The impact aspect of the system is also started when the user-defined inputs have been entered. The Impactor will calculate the impact of each of the elements in minute intervals throughout the system operations until the status of the system becomes “Completed”.
Results and Discussion
The characteristics of a complex system make them difficult to understand or predict behaviour and whilst there are current approaches that have been used to aid this they tend to be costly and time consuming. This paper has presented an analysis approach and method for analyzing complex systems which was to be simple and qualitative, reducing the computational burden. The analysis approach was deemed to have the function of identifying elemental relationships and interdependencies, and analyzing the impact of one element on any or all of the elements within a system.
The model that has been created has shown that it was possible to understand more about the simplified ATM system, the way it operates and predict its behaviour in reference to the scenarios that were applied. This demonstrates the potential of applying such an analysis to a larger, more complex system.
AOS would like to thank Gary Davies (QUB) for providing the content of this article.