• No results found

Testing and Simulation

7.2 Simulation Overview

The primary objective of the simulation is to provide enough information to test the decision support system and evaluate the effects of decisions made, without providing unnecessary detail about aspects that will not affect it.

The simulation has to provide a series of problems for the decision support system to solve. In some ways this is similar to the rolling time window [117] or rolling horizon approaches [105, 106], since all of these approaches attempt to obtain an overall good schedule by solving a sequence of sub-problems. The main difference is that the sub-problems are presented by the simulation because there is only limited information available at any time, rather than in order to reduce the complexity of the problem to be solved.

The features and responsibilities of the simulation are as follows:

1. Create an appropriate sub-problem for the decision support system.

This means creating the input problem for the decision support system, thus specifying the current state of the airport departure system. The simulation uses real historic data, provided by NATS to generate the initial input data for the decision support system. The effects of decisions made by the decision support system are then used to modify the later input problems that are presented.

The input data is detailed in section 2.14 and must include:

• The positions of any aircraft within the holding area and the path planned for each one through the holding area.

• Information about any aircraft which recently took off.

• Predicted arrival times and arrival entrances for all aircraft currently taxiing around the airport towards the holding area.

• Details of all aircraft in the system, for example, the weight class, speed group and assigned departure route.

2. Accept the output from the decision support system under test.

At a minimum this means a planned take-off sequence. Predicted take-off times are also useful. The implemented simulation goes beyond this as it also receives details of how the output state was achieved, including the paths which were assigned to aircraft and even a snapshot of the expected positions of aircraft within the holding area at some point in the future.

3. Enact the decisions made in order to see the later effects of the decisions.

This means at least assigning paths through the holding area to aircraft and predicting the movement of aircraft.

In the implemented departure system simulation this task is delegated to the decision support system and the paths it assigned are used by the simulation.

4. Define a set of rules to specify how the state changes over time. For example: • Aircraft lined up for take-off will take-off.

• Aircraft within the holding area may move forwards through the holding area, along their assigned path, or may have to wait for other aircraft to overtake them first. • Aircraft on the taxiways will advance towards the holding area over time.

This means reducing the time the aircraft is expected to remain on the taxiways, and possibly reducing any simulated uncertainty or prediction error in the value passed to the decision support system.

5. Optionally, provide some simulation of the uncertainty in information with which a real decision support system must contend.

In the simulation described here, information about the exact times at which things will happen is maintained, but ‘estimated’ values can also be produced by adding prediction errors. It is, then, the inaccurate predicted values which are provided to the decision support system as input data at each stage. This simulates the fact that the decision support system would not necessarily have accurate information in advance.

7.3

Data Maintained By The Simulation

The problem presented by the simulation to the decision support system includes different levels of detail for aircraft, depending upon the current position of the aircraft in the departure system. The current position of an aircraft is determined by comparing the current simulation time with the historic times at which aircraft pushed back from the stand, the times the aircraft reached the holding area and the predicted take-off times provided by the decision support system. Aircraft will either be at their stands, on the taxiway, at the holding area, or in the air. Aircraft move forward through the states over time. The level of detail the simulation maintains for aircraft varies as the state changes.

7.3.1 Aircraft at the stands

No information is given to the decision support system about aircraft that are at their stands. Aircraft only enter the problem after they have pushed back from their stands. Until this point, the level of uncertainty currently involved in the time they will push back from their stand is such that there is little use in incorporating these aircraft into the system. As described in section 3.9, there is a possibility that this may change in future and such aircraft could then be taken into consideration.

7.3.2 Taxiing aircraft

Taxiing aircraft are dealt with at a relatively high level of abstraction and are only included in order to give some idea of what will be arriving at the holding area, when it will be arriving and where it will arrive. Experiments have shown that the decision support system performs better if it is given knowledge of the taxiing aircraft, thus enabling the search to have visibility of the effects upon these taxiing aircraft of decisions currently being made.

One common case that including taxiing aircraft can assist with is when there is a taxiing aircraft with a tight take-off time-slot (CTOT). This aircraft may require a straight run through the holding area in order to achieve take-off within the CTOT slot. Knowing about such aircraft in advance can ensure that other aircraft are not in the way when they arrive.

Another potential benefit is seen when there are multiple taxiing aircraft with similar departure routes. It is often useful to fit aircraft already in the holding area between these taxiing aircraft, avoiding the large separations. To do this, it may again be necessary for the first of these aircraft to overtake aircraft in the holding area.

Knowing about taxiing aircraft can reduce the amount of re-sequencing required of the aircraft already in the holding area. For example, knowing that a taxiing aircraft could be slipped into a gap in the take-off schedule means that it is not worthwhile eliminating the gap at the expense of other parts of the take-off sequence.

The following information is maintained for aircraft currently on the taxiways.

• The historic arrival time at the holding area. This is available in the historic input data. An expected arrival time is provided to the decision support system by optionally modifying the historic arrival time by a variable prediction error to simulate a level of uncertainty for the values. Prediction errors would be expected to decrease over time for any specific aircraft as it approaches the holding area. The simulation assumes that aircraft will leave their stand and arrive at the holding area at the times at which they did so historically, but the decision support system will not necessarily know these time accurately in advance due to the added prediction errors.

• For all aircraft on the taxiways, the holding area entrance at which the aircraft will arrive is predicted based upon the stand from which the aircraft left. For example, there are three entrances in the 27R holding area diagram given in figure 6.2 and they are labelled D, K and G. The simplest case assumes the Ground Movement Controller (GMC) directs aircraft to the closest entrance to the stand, as discussed further in section 8.2.7.

7.3.3 Aircraft within the holding area

More information is maintained for aircraft in the holding area as these aircraft are under the control of the runway controller or the decision support system in these experiments. The effects of decisions made for these aircraft have to be maintained.

The simulation uses the same directed graph representation of the holding area as is used by the decision support system. This allows the simulation to provide initial holding area positions of the aircraft to the decision support system and also allows the simulation to ask the decision support system to predict new positions for aircraft, as will be seen later.

The following information is available about these aircraft:

• The actual holding area arrival time and holding area entrance at which the aircraft arrived. This is important as the holding area delay is calculated based upon the arrival time, and the arrival entrance affects the traversal paths that can be taken and the possibilities for overtaking.

• Any holding area traversal path that was previously allocated. This is maintained as the decision support system is not permitted to change this for any aircraft that is already in the holding area, as described in section 7.4.6.

• The current position of the aircraft in the holding area. This is predicted by the simulation and depends upon the allocated path. It will change over time as the aircraft moves forward through the holding area. The simulation uses the results of the feasibility check to predict holding area positions, as described in section 7.7. An implemented decision support system could use ground radar data to determine the position of aircraft.

7.3.4 Aircraft which have taken off

Once aircraft have taken off, the only information that is needed is the take-off time and aircraft characteristics, so that the required separations can be calculated for future aircraft.

7.4

Simulation Loop

The simulation maintains a list of the aircraft that the decision support system should be aware of at the time, with the details described earlier for each aircraft.

For evaluation purposes, it is assumed that a controller will follow the advice of a decision support system. The simulation can then be used to evaluate the effects of various changes to the departure system, putting the decision support system in the place of the controller.

The simulation starts at the start of the dataset and performs the steps described in algorithm 17. In algorithm 6, TS refers to the current simulation time and the other variables

and constants are described in the explanation of the algorithm given below.

For each aircraft, a, the simulation needs to know the push-back time, Pa, of the aircraft,

the holding area arrival time, Ha, of the aircraft and, once the decision support system has

scheduled a take-off time for an aircraft, the predicted take-off time, Da, and the position, Oa,

Algorithm 17 The simulation loop

1: initialise simulation time, TS, to start time of the dataset under test

2: empty the set of aircraft SA to provide to the decision support system

3: while there are aircraft which have not taken off do

4: for each aircraft a in the dataset do

5: if (a /∈ SA) and a has not already taken off and been removed from SAthen

6: if the flag E is set or TS ≥ Pa then

7: if TS≥ (Ha− TP) then

8: add a to the set SA

9: end if

10: end if 11: end if

12: end for each

13: pass the set of aircraft SA to the decision support system and solve the take-off problem

14: advance TS by the current simulation step size, for example 60 seconds

15: for each aircraft a ∈ SA do

16: if TS≥ Ha then

17: aircraft a is within the holding area

18: retrieve the allocated traversal path for a from the decision support system and fix it for the aircraft, so it can no longer be changed in future

19: retrieve the predicted holding area position for a from the decision support system. This will be passed to the decision support system in future iterations

20: end if

21: retrieve the position Oa of the aircraft in the desired take-off sequence from the decision

support system

22: retrieve the predicted take-off time Dafor the aircraft from the decision support system

23: if TS≥ (Da+ TM) then

24: the aircraft has taken off long enough ago that it can be ignored so remove a from the set SA

25: else if TS ≥ (Da− TF) then

26: the aircraft is close to take-off. Fix the position, Oa, in the take-off sequence and the

take-off time, Da of aircraft a

27: end if

28: end for each

29: end while

30: end simulation and return the predicted take-off times for all aircraft

Configuration parameters are required in order to determine when the decision support system should be made aware of each aircraft. If aircraft are included before they push back, then the flag E should be set to true, otherwise it should be false to indicate that the system only considers aircraft after they push back. A planning horizon, TP, must be specified, determining

how early before holding area arrival the system will be given information about taxiing aircraft. The planning horizon is considered in more detail in chapter 8.

The take-off sequence is fixed for aircraft close to their time of take-off to allow for the fact that aircraft would need some time to line up and both controllers and pilots would need some firm knowledge of the future take-off sequence. The length of time during which the take-off time is assumed to be fixed is given by the variable TF. As the positions of these aircraft are

fixed, the searches have no flexibility to re-sequence them any more.

In order to maintain separations from aircraft which have taken off, an aircraft must remain in the system even after take-off until it can have no effect upon later take-offs. Let TM

be the memory time, the time after take-off for which the system must remember aircraft. A safe value of TM should equal or exceed the maximum duration of a separation rule. A value of

ten minutes was used in the experiments, allowing for the application of a large MDI, increasing the required separation on a departure route.

7.4.1 The problem passed to the decision support system

When the current problem is passed to the decision support system in step 13, all of the required details such as weight class, speed group and departure route are passed for each of the aircraft in the set SA. The decision support system is told the current state of the aircraft; whether it is

taxiing, at the holding area or has taken off. Additionally, where predicted holding area positions are known they are also passed, as are the details of any traversal paths that have been fixed and details of whether the position of the aircraft in the take-off sequence has been fixed.

The previously determined take-off sequence (if there is one) is also passed to the decision support system. As the search moves through the time period of the dataset, later searches use the partial schedule built by previous searches, with knowledge of any extra aircraft that arrived, to re-sequence the remaining aircraft. Eventually the aircraft will all have a fixed take-off time and a final schedule will have been created and can be evaluated. The outputs of the simulation are discussed in section 7.5, along with a description of the method used to evaluate these outputs.

7.4.2 Updating the internal model

The internal model of the departure system is updated in a number of ways. The positions of the aircraft in the holding area will move forwards over time, as updated predictions of positions are retrieved from the decision support system.

Aircraft will move through the departure system towards take-off as the simulation time is advanced. Although the decision support system may only receive an inaccurate, predicted holding area arrival time, the (real) arrival time for each aircraft that is known by the simulation is fixed, so eventually each aircraft will arrive at the holding area. At this time, the last traversal path that was allocated to an aircraft is fixed and cannot be changed.

The decision support system predicts take-off times for aircraft and provides these predictions to the simulation. The simulation assumes that the decisions of the decision support system are enacted, so aircraft will take-off when they are predicted to do so. Eventually, aircraft will reach the predicted take-off time and be assumed to take off. At some point before this time the position of the aircraft in the take-off sequence should be fixed.

7.4.3 Extra outputs from the decision support system

The decision support system has an important role to play in the simulation as it performs much of the prediction work. The decision support system already calculates a lot of information

in order to perform its task and rather than have the simulation re-calculate these things, the decision support system is responsible for storing the information it calculated for aircraft and providing it to the simulation when needed. This information includes a predicted position in the holding area, a predicted take-off time and the assigned holding area traversal path.

7.4.4 Predicting holding area positions

For each aircraft within the holding area, the current node the aircraft is occupying must be specified as one element of the input data for the decision support system. The positions have to be determined by the simulation and the implemented simulation does this by delegating the task to the decision support system.

The simulation tells the decision support system to re-play the feasibility test for the selected take-off sequence. At a relevant point in the re-play, the decision support system takes a snapshot of the positions of the aircraft and uses these positions as the predicted positions to supply to the simulation. This snapshot is taken at the point where the first aircraft in the holding area takes off or the first time the current time is advanced; whichever happens first. The implementation is actually more complex than this as it needs to ensure that the positions are realistic, so the system is considered in more detail in section 7.7.

7.4.5 Simulation step time

The simulation time is increased in pre-determined steps, as indicated by step 14 of algorithm 17. A sixty second increment was used for the experiments performed for this thesis. With a search time of a second, one second increments would be more realistic, simulating a constantly running decision support system, however, this would allow the decision support system multiple attempts to obtain a good schedule in response to changing situations and would therefore make evaluation of the performance of a single iteration much harder. The selected step size makes the problem much harder for a decision support system to solve.

7.4.6 Fixing the traversal path

The holding area traversal path is important. Once an aircraft has reached the holding area, changing the path it uses to traverse the holding area is more difficult than doing so before it

Related documents