You are on page 1of 5

OPTIMIZATION OF AIRSPACE SECTORIZATION

Autors: Joan Albert, Ferran Morales


Grup: 4GM22

ABSTRACT
In this task, the main objective will be the implementation of the Genetic Algorithm
through a Matlab code (where a GA toolbox is provided) to find an optimal sectorization
in Barcelonas FIR. In the previous deliverable, we found a sectorization where the
controllers workload was not well balanced, and now, the created code will be used in
order to enhance the output results with the Genetic Algorithm and compute the
complexity of the studied airspace. It is important to recall that three traffic files are given,
and each one corresponds to the traffic information at a certain time of the day (morning,
midday and night).
INTRODUCTION
This project is going to be the second part of the first deliverable of GEAIO about airspace
sectorization, thus we will use the Matlab code we had already implemented in this task.
Now, the main aim is to use the Genetic Algorithm (GA) through Matlab in order to find
the best sectorization for Barcelona FIR. Achieving that goal, will let the controllers to
reduce the workload they have. In order to do that, we will have to consider the number
of aircraft in each sector and we will start to compute the complexity of the airspace to
better balance the controllers workload. Therefore, the workload function created in the
first deliverable will have to be modified because we will need to consider the static
characteristics of the studied airspace.
We can state that this algorithm is based on five steps: Initial Population, Evaluation,
Fitness Assigment, Selection and Reproduction. First, Genetic Algorithm creates an
initial population of random individuals. Second, these individuals are assesed and the
objective values are computed. Third, the objective values previously computed are used
in order to determine the Fitness Values. Fourth, individuals are selected and filtered
if they accomplish the Fitness Values. Finally, the algorithm creates new individuals
from the ones selected before (thorugh processes called crossover and mutation) and starts
again the cycle from Evaluation and so on.

Figure 1. Genetic Algorithm scheme.

OBJECTIVE FUNCTION
First, and before starting the implementation of Genetic Algorithm, an objective function
based on the number of aircraft in each sector because this is the factor we want to balance
(find an optimal solution). In order to do that, we have created a function called
ObjectiveFunction.m that will get the information we desire and it will have its useful
function inside the GA: the function will extract the lowest variance from the random
individuals created by the GA in order to select the best one. Once the best optimization
is found, the variance per sector will be computed as well as the amount of aircraft in
each sector to see if we have achieved our objective (that is, if we have better balanced
the amount of aircraft per sector from the sectorization we had performed in the previous
Deliverable).
FINDING AN OPTIMAL SECTORIZATION (TRAFFIC BASED)
Once we have created the function explained in the previous section, we are ready to
implement the Genetic Algorithm. In this section, we will only use the traffic data
provided (three files previously mentioned) to perform this task. First of all, some
conditions have to be imposed: each sector should not contain more than fifteen aircraft,
the number of sectors must be minimized and Barcelona FIR has to be divided at least in
two sectors in order to avoid too large areas. The cost of the best element versus the
number of iterations for each iteration of the GA will also be depicted.
FIRST SCENARIO: MORNING (07:55-08:10)
First of all, we will plot the cost of the GA performance (Best individual value versus
Generation) and then, we will plot the sectorization we have obtained. In this case, we
have obtained six sectors, but surpisingly we have not obtained less than fifteen aircraft
per sector as asked. The aircraft allocation we have obtained is the following: 24, 17, 19,
18 and 19. The variance value we have obtained is Var=9.06.

Figure 2. Cost plot of the GA for the first scenario.

Figure 3. Sectorization BCN FIR for first scenario.

SECOND SCENARIO: MIDDAY (11:55-12:10)


The cost plot will be again depicted followed by the sectorization of the Barcelona FIR.
In this case, we could fit the demanded aircraft per sector (less than fifteen) into three
sectors. The aircraft allocation obtained is the following: 8, 9 and 13. The variance value
we have obtained is Var=7, it indicates that aircraft are better allocated than the previous
scenario where the variance value was Var= 9.06.

Figure 4. Cost plot of the GA for the second scenario.

Figure 5. Sectorization BCN FIR for second scenario.

THIRD SCENARIO: NIGHT (21:55-22:10)


In this scenario we have obtained that there are no aircrafts overflying our airspace
domain, thus the cost of the GA will be zero (as depicted in Figure 6) and the sectorization
will be very simple: we will just have two sectors since we are asked to divide at least
into sectors the FIR (see Figure 7).Obviously, the variance will be also zero because there
are no aircraft inside the domain.

Figure 6. Cost plot of the GA for the third scenario.

Figure 7. Sectorization BCN FIR for third scenario.

FINDING AN OPTIMAL SECTORIZATION (STATIC PARAMETERS ONLY)


Now, the main aim is to modify the WorkloadFunction we created in the first deliverable
in order to not consider the aircraft per sector but taking into account the static parameters
of the airspace (that is for instance, FIR intersection points, Airways intersection points,
Transfer Points or Airways length). Therefore, now the workload will be considered as
a sum of several factors, such the ones previously mentioned, and each one will be
multiplied per a constant number not specificied. It is important to highlight that
depending on that values, the workload value will suffer considerable changes which will
modify our sectorization. As this constant factor gets higher, it will mean that we are
considering more important this factor: for example, if we choose a huge constant value
for FIR intersection it will mean that we are considering that this factor is quite relevant.

You might also like