You are on page 1of 7

AIR TRAFFIC RUNWAY ALLOCATION PROBLEM USING

ARTMAP (ART1)

Krishan Kumar1
1
Invertis Institute of Management Studies, Bareilly, India (kumar_krishana@yahoo.com)

Ravendra Singh2
2
Department of CSIT,IET, MJP Rohilkhand University, Bareilly, India (rsingh_iet@rediffmail.com)

Zubair Khan3
3
Department of Computer Science, IIET, Bareilly, India (zubair_762001@yahoo.com)

Ajai Indian3
3
Invertis Institute of Management Studies,Bareilly (indian_ajai123@rediffmail.com)

ABSTRACT
The biggest of the ATC challenges is the continuing rise in traffic volumes and which
is increasing the movements of airbuses around the airport due to busy runways. This
air traffic congestion is increasing the running cost of flights due to unwanted
movement of airbuses around the airport and it will also increase the chances of the
airbus collision. Such type of problem can be solved either by increasing runway’s
capacity which is very costly or by means of providing some artificial intelligence
solutions to optimize the utilization of existing capacity. So for optimum utilization of
existing resources, some researchers have given the genetic algorithm based solutions
but these solutions are not able to minimize the average arrival delay of the air traffic
less than 11.25 minutes. In this paper we have proposed Adaptive Resonance Theory
ART (ART1) algorithm to produce optimize solution. Our result shows that the
average arrival delay of 9 flights using ARTMAP has been minimized to 10.55
minutes.

KEYWORDS- ATC (Air Traffic Control), ATM (Air Traffic Management), Adaptive
resonance theory

1 INTRODUCTION detection and alert systems to reduce cost & delay, as


Today as we know that Air Traffic is going to well as improved ground guidance systems will
increase day by day and creates traffic congestion, increase situational awareness both in the cockpit and
delay of flights at airport and hence loss to air on the ground. Trial and error or cut and try methods
industry due to increase in running cost. However, for finding the answers to a problem are well accepted
continuing rise in traffic volume can be solved by in every field of science. The success rate of the cut
increasing the existing capacity and will require and try approaches depends greatly on the experience
investment in new automated systems and the user has with the problem under consideration.
infrastructure. Improving the current systems will Artificial Intelligence areas like Artificial Neural
provide a short-term solution. Artificial Intelligence Networks, Fuzzy Logic, & Genetic Algorithm
techniques constitute an optimized methodology constitute an optimization methodology effective for
effective for solving discontinuous, non-convex, non- solving discontinuous, non-convex, non linear or non-
linear, or non-analytic problems. analytic problems. This research explores the
Airport ground operations and en-route ATM application of such techniques to a non-analytic
Safety is being enhanced through the implementation event-related air traffic control problems that of run-
of improved tools for enhancement in conflict
way assignment, sequencing and scheduling of arrival
flights at an airport with multiple runways.
Today survey of Federal Aviation Administration
(FAA) shows that average arrival and departing
traffic both are experiencing delays of 15 minutes or
more, and a lot of research [3],[4],[5],[6] has been
done till now. In order to minimize the fuel
consumption and total cost for a flight, we have
planned a new tool in java language using ART map
which solves the problem of congestion and landing
on the airport in a most economical and efficient
manner. This tool will proved to be a great assistance
for ground controllers and provide them optimized
solutions. The major problem of landing at the busiest
runways can be solved by calculating the total waiting
time and fuel cost an aircraft spend while waiting to
land on runway. This method can be applied to
modern ATC’s where ground controlling is the major
cause of concern and hence we can minimize the
average delay. In this paper there is a busy airport Figure 1: Nine aircrafts are waiting to land on three
with three runways. runways

2 PROBLEM DESCRIPTION A simple scenario has been arbitrarily generated


for evaluation of the two schemes with nine flights
Consider a busy airport with three runways and 9 and three runways. For identification purposes, the
flights are waiting to land. The ATC problem call signs of the flight numbers from 1 to 9 are given
discussed in this paper has concerns with the runway by s1, s2, s3, s4, s5, s6, s7, s8, s9.
assignment and scheduling of x arrival flights to y Matrix for eta values (in minutes) is given by
runways.

2.1 Assumptions 16 11 8
The problem is formulated with the following 4 16 8
conditions: 5 6 7
1. For each of flight arrival, an expected time of 15 6 17
arrival (eta) to each runway is available as the eta [9][3] = 15 11 13
minimum time to get to the runway. Any restriction 16 8 9
in aircraft performance and flight pattern is 4 16 13
assumed to have been included in the determination 15 6 7
of these expected times of arrival (eta) values. The 5 6 7
schedule time of arrival (sta) of a flight to a runway
can not be earlier than the corresponding eta. Where the rows represent the flight numbers and
2. All the flights scheduled to land on a runway have columns represent runway numbers.
to observe specific separation rules between the sta [ 9]= {16, 27, 28, 19, 23, 21 ,25 ,20, 24}
leading aircraft and the trailing aircraft based on the
aircraft types (We have taken 9 values in the above array, which
means we are taking only one value out of three
3. When a flight is scheduled to a runway, the delay values on three runways)
as a result of the schedule is defined as the
difference between the schedule time of arrival 3 METHODOLOGY USED
(sta) and the expected time of arrival (eta) among
all the runways for that flight. Earlier in 1999 Chang and P.K. Menon used four
[3] genetic search schemes, out of which section third
to explore the different characteristics associated with
the different base to formulate the genetic search. match.) If the ratio of ║x║ and ║s║ is greater than or
Genetic search methods represent a logical approach equal to the vigilance parameter, the weights (top
to cut and try search methods. Continuing down and bottom up) for the winning cluster unit are
improvements in the computational power of desktop adjusted.
machines make genetic search method a practical However, the ratio is less than the vigilance
methodology for finding the answers to difficult parameter; the candidate unit is rejected, and another
problems. But these solutions are not able to candidate unit must be chosen. The current winning
minimize the delay less than 11.25 minutes. So we cluster becomes inhibited, so that it cannot be chosen
are proposing an Adaptive Resonance Theory based again as a candidate on this learning trial, and the
solution to minimize the delay by using ART1. activations of the F1 units are reset to zero. The same
input vector again sends its signals to the interface
3.1 Algorithm units, which again send this as the bottom-up signal to
the F2 layer, and the competition is repeated (but
A discussion of the choice of parameter values
without the participation of any inhibited units).
and initial weights follows the training algorithm. The
The process continues until either a satisfactory match
notation we use is as follows:
is found (a candidate is accepted) or all units are
inhibited. The action to be taken if all units are
n- number of components in the input vector.
inhibited must be specified by the user.
m- maximum number of clusters that can be
formed.
3.3 Training Algorithm
bij-bottom-up weights (from F1(b) unit Xi to F2
unit Yj). The training algorithm an ART1 net is presented
tji- top-down weights (from F2 unit Yj to F1 unit next. A discussion of the role of the parameters and
Xi). an appropriate choice of initial weights follows.
ρ- vigilance parameter. Step0. Initialize parameters:
s- binary input vector (an n-tuple).
x-activation vector for F1 (b) layer (binary). L > 1,
║x║ norm of vector x, defined as the sum of the 0 < ρ ≤ 1.
components xi. Initialize weights:
0 < bij(0) < L / L- 1 + n ,
3.2 Description tji(0) = 1.
Step1. While stopping condition is false, do
A binary input vector s is presented to the F1 (a)
Steps 2-13.
layer, and the signals are sent to the corresponding X
Step2. For each training input, do steps 3-12.
units. These F1 (b) units then broadcast to the F2
Step3. Set activation of all F2 units to zero.
layer over connection pathways with bottom-up
Set activations of F1(a) units to input
weights. Each F2 unit computes its net input, and the
vector s.
units compete for the right to be active. The unit with
Step4. Compute the norm of s:
the largest net input sets its activation to 1; all others
║s║= ∑ si
have an activation of 0.We shall denote the inbox of
Step5. Send input signal from F1(a) to the F1(b)
the unit as j. This winning unit becomes the candidate
layer: xi = si.
to learn the input pattern. A signal is then sent down
from F2 to F1 (b) (multiplied by the top down
Step6. For each F2 node that is not inhibited:
weights). The X units (in the interface portion of the
If yj ≠ -1, then
F1 layer) remain “on” only if they receive nonzero
yj = ∑ bij*xi.
signals from both the F1 (a) and F2 units [figure (2)].
The norm of the vector x (the activation vector Step7. While reset is true, do step 8-11.
for the interface portion of F1) gives the number of Step8. Find J such that yJ ≥ yj for all nodes j.
components in which the top-down weight vector for If yj = -1, then all nodes are inhibited and
the wining F2 unit tj and the input vector s are both 1.
(This quantity is sometimes is referred to as the
Figure2: Basic Architecture of ART1

bij: bottom-up weights (from F1(b) unit Xi to F2 unit


this pattern can not be clustered.
Yj).Used to store different clusters values. Permissible
Step9. Recompute activation x of F1 (b):
range is given by
xi = si*tji. 0 < bij(0) < L / (L – 1 + n) sample value 1 /( 1 + n).

Step10. Compute the norm of vector x: tji: top-down weights (from F2 unit Yj to F1 unit
║x║ = ∑ xi. Xi).Used to store runways assignment for different
Step11. Test for reset: flights.
If ║x║ / ║s║ < ρ, then yj = -1 (inhibited ρ - vigilance parameter.(For deciding the learning
node J) (and continue executing Step 7 again) node).
If ║x║ / ║s║ ≥ ρ,
s - binary input vector (an n-tuple). Input array to
Then proceed to Step 12. store different input values.
Step12. Update the weights for node j (fast learning):
x - activation vector for F1 (b) layer (binary).
bij(new) = L*xi / L-1 + ║x║ , tji (new) = xi.
Output array to decide the learning node.
Step13. Test for stopping condition.
║x║ - norm of vector x, defined as the sum of the
3.4 Parameters Used components xi.
Full algorithm is coded by using core java.
n: number of components in the input vector. Used
as a sequence of flights.
Now the algorithm works as follows in our case:
m: maximum number of clusters that can be formed.
Initialize parameters:
Used as a runway assignment.
L =50
ρ=0.8
Initialize weights:
bij(0) =0.2
tji(0) = 1 4 CONCLUSION AND RESULTS
Our main objective of allocating runways to
We have taken 9 flights in all and 3 runways at
flights (table2) and hence minimizing congestion and
an airport. Each flight is presented as an input pattern
minimizing the arrival delays in minutes (table3,
one by one in a sequence and the top down weight
figure3) is successfully achieved. The use o Adaptive
matrix is updated in which each row is taken as the
Resonance Theory (ART1) instead of Genetic
runway number. The flight sequence is defined by the
Algorithm has drastically reduced the arrival delay
following table1:
and hence cost. The computation time required is
reduced 6000 generations to 50 generations as
compared with genetic techniques (table4, figure4)
Binary Input pattern Flight Number
and result so obtained are more accurate and
0001 flight1 (s1) optimum. Implementations for all the iterations is
0010 Flight2 (s2) done by using java development tool kit (jdk1.6)and
on Pentium IV processor. Implementation time is 30
0011 Flight3 (s3)
minutes for 50 iterations.
0100 Flight4 (s4) For the particular sequence 918653472 at 50
0101 Flight5 (s5) generations following runway assignment is found
(shown by table2)
0110 Flight6 (s6)
0111 Flight7 (s7)
1000 Flight8 (s8) S. No Sum4 (as per Flight Allocated
input sequence) Number Runway
1001 Flight9 (s9) 1 4 s1 R1
2 15 s2 R3
3 15 s3 R3
Table1: Input Sequence Pattern as per Flight 4 15 s4 R3
Sequence 5 15 s5 R3
6 15 s6 R3
7 15 s7 R3
After the completion of one sequence pattern
8 15 s8 R3
different delay values are calculated for that pattern. 9 15 s9 R3
Flights are assigned to the runways on the basis of top
down weight matrix. Flights are arranged in sequence Table2: Runway Allocation
starting from first row for first flight, second row for
second flight and so on. Now whatever the value of Air Traffic Control System Command Center
the runway number in that row, the flight is assigned says that general arrival/departure delays are 15
to that runway. That sequence pattern to be selected is minutes or more. Departures are experiencing air taxi
on the basis of delay value that is calculated and is delays of 16 to 45 minutes and also arrivals are
then compare with the other delay values and the experiencing airborne holding delays of 16 to 45
minimum one is selected. minutes. Traffic destined to this airport is being
We have taken the following assumption in our case delayed at its departure point. Departures are
for runway numbers: experiencing taxi delays greater than 45 minutes
and/or arrivals are experiencing airborne holding
Runway 1->0001, 0100, 0111 delays greater than 45 minutes. So we can say that our
Runway 2->0010, 0101, 1000 results are far better.
Runway 3->0011, 0110, 1001
5 FUTURE WORK
and rest of the sequence left up to 1111.
This means that if the value in the first row of top This technique can be further implemented to air
down matrix is 0101 then flight number 1 will land at and local control of ATC to provide the optimum air
runway 2 and so on. traffic control at the busiest airport. Also the first
The delay values are calculated by subtracting module, which we implemented with this problem
the expected time of arrival (eta) for a flight at a can be combined with second module by using
runway and scheduled time of arrival (sta) for that adaptive resonance theory (ART2) to get the best air
flight on the scheduled traffic simulation tool. Further work will concentrate
in refining the modeling and the global criteria to
optimize, taking into account for example take-off
sequencing needs of approach sectors or priority
levels for slotted departures. Average Landing Time Chart Using ART map

16
14
Sequence of Flights and

Delay(in minutes)
12
Generations Corresponding Average Delay in 10
(Iterations) Minutes 8
5 485291673:15 6
10 148569327:15 4
2
15 163297548:14.5555555
0
20 362579481:14.5555555 0 20 40 60 80 100 120
25 167834592:14.5555555 Generations
30 854173692:14.2222222
35 275194368:14
40 879156342:14 Figure3: Results according to our proposed ART1
45 498175362:14
50 918653472:13.7777777
55 315864972:13.7777777 35
60 437295168:12.1111111

Average Arrival Delay(minutes)


30 30.25
65 781569423:12.1111111
70 187569423:11.2222222 25
75 871564932:11.2222222 20 19
80 381659247:10.5555555
15
85 718596432:10.5555555 13.5 13.5 13
11.25
90 841569723:10.5555555 10
95 381549628:10.5555555 5
100 921567413:10.5555555
0
0 2000 4000 6000 8000
Table3: Delay according to our proposed ART1
Generations

Figure4: Results according to V.H.L. Chang


& et. al. [3]

Generations Fitness value (Delay)


1000 30.25 REFERENCES
2000 19
[1] Gail A.Carpenter, Stephen Grossberg, Adaptive
3000 13.5
Resonance Theory MAP.
4000 13.5 [2] Yang Yuying, Shi Xizhi, Li Guoyi, Rules from
5000 13 fuzzy adaptive resonance theory map, Progress in
6000 11.25 Natural Science May 1999, Vol.9 No.5, p.382.
[3] V.H.L. Chang, Crawford, P.K.Menon, Air Traffic
Control Using Genetic Search Techniques,
Table4: Delay according to V.H.L.Chang & et. al.[3] Optimal synthesis palo alto California, May 1999,
Vol.9 No.5, p.382.
[4] Nathan L. Kleiman, Stacy D. Hill, Simulation
optimization of air traffic delay cost, IEEE Winter
Simulation Conference, Proceedings of the 30th
Conference on Winter simulation Washington,
D.C., United States, Pages: 1177–1182,Year of
Publication: 1998,ISBN:0-7803- 5134-7.
[5] Delahaye and Elliot, Genetic Algorithm for Air
Traffic Assignment, European ECAI 94, 11th
Conference on Artificial Intelligence.
[6] B. A. Peters, J. S. Smith, D. J. Medeiros, and M.
W. Rohrer, eds. David W. Hutchison, Stacy D.
Hill, Simulation Optimization of Airline Delay
with Constraints, Proceedings of the 2001
Winter Simulation Conference.
[7] Heinz Erzberger, The Automated Airspace
Concept, 4th USA/Europe Air Traffic
Management R&D Seminar, December 2001.
[8] Erich Gamma, Richard Helm, Ralph Johnson, and
John Vlissides. Design Patterns: Elements of
Reusable Object-Oriented Software. Addison-
Wesley, Reading, Massachusetts, 1995.
[9] Tom G. Reynolds and R. John Hansman.
Conformance monitoring approaches in current
and future air traffic control environments. 21st
Digital Avionics Systems Conference, October
2002.
[10] David C. Zhang. Collaborative arrival planner:
Its design and analysis using object modelling.
Master’s thesis, Massachusetts Institute of
Technology, May 2000.
[11] H. Rezberger, T.J. Davis and S. Green, De
sign of center-TRACON Automated System,
Proceedings of the 56th AGARD
Symposium on Machine Intelligence in Air
Traffic Management, Berlin Germany, 1993
pp 11-1-11-12
[12] Gail A. carpenter, Stephen Grossberg John H.
Reynolds, ARTMAP: Supervised Real-Time
Learning and Classification of Nonstationary
Data by a Self-Organizing Neural Network

You might also like