Aerospace and Electronic Systems - May 2019 - 17

Karahasanovic et al.
Table 1.

T-Test Comparison of ATCOS versus Algorithm
Run 1

paired t-test
(p-value)

Controllers (s)

D(%)

algorithm (s)

Avg. taxi time

0.000001

283.02

183.516

À35.2

Avg. punctuality

0.010514

243.32

80.59

À66.9

Avg. taxi time

0.000000

299.15

199.32

À33.4

Avg. punctuality

0.000290

190.17

80.44

À57.7

Avg. taxi time

0.00001

298.93

199.79

À33.2

Avg. punctuality

0.00506

144.30

59.74

À58.6

Run 2

Run 3

Because there were no other disturbances or external
influences, other than those from the ATCOs or algorithm
themselves, the difference in optimization can be explained
by how coordinated the different decisions were. The
results show that the algorithm improved the overall result
in terms of both punctuality and taxi time by integrating
both departure and surface routing management, giving
rise to a more holistic evaluation of local decisions.
An important byproduct of this optimization is that it
leads to a reduction in the number of airplanes simultaneously moving on the runway and taxiways, which, in
turn, decreases the risk of collisions. For example, in Scenario 2, the ATCOs needed to guide up to 11 airplanes at
the same time. In contrast, the algorithm almost halved
the maximum workload, only guiding up to six airplanes
at once (see Figure 3). Similar to the airspace sector
workload indicators, the number of moving aircraft being
actively monitored by the ATCOs can be seen as an
adequate workload indicator. A reduction in the number
of airplanes moving simultaneously means that less
communication with pilots is necessary. This allows more

Figure 3.
Number of moving airplanes over time in Scenario 2.

MAY 2019

time for monitoring the safety of the airport. Similar
observations can be made for the other two scenarios.
Given these results, it is expected that this difference will
expand even more when the number of flights handled by
the airport increases.
An increased integration of arrival scheduling with
departure and surface routing is expected to yield even
better overall results. Further investigation showed that
the algorithm currently performs better for departures that
it does for arrivals. From a practical standpoint, arrivals
are less controlled by the receiving airport than they are
by the departing airport and actual flight operations. This
is especially true in the conducted case study, in which the
arrivals landed exactly at their scheduled times. Because
of this, the algorithm is configured to prioritize departing
flights over arriving flights.
If ATCOs could have access to solution proposals generated by the holistic optimization algorithm and the associated tool, it could be an important enabler for airports to
reach a higher safety level, due to the following reasons:
First, the algorithm works on a mathematical model that
inherently adheres to all safety rules (e.g., separation). In
contrast, ATCOs, as a group of decision makers, lack the
ability to jointly consider time and space with mathematical precision. This is a drawback because ATCOs need to
resolve safety issues during execution, when the issues are
about to occur. Second, the results have shown that the
number of airplanes moving simultaneously can be
reduced substantially through the use of optimization technology. Such a reduction might also lead to a decreased
risk of collisions.
The average duration of a complete optimization run,
without a "warm" start and for which all flights from each
scenario are considered, is 15 s on a laptop with an Intel i7
Central Processing Unit (CPU), four cores, and 4 GB
of random-access memory (RAM). This was the first

IEEE A&E SYSTEMS MAGAZINE

Aerospace and Electronic Systems - May 2019

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - May 2019