Football Game Algorithm
FGA
Overview
Football Game Algorithm (FGA) is one of the
most recent sports based search and
optimization methods.
It simulates football players’ behavior during
a game for finding the best positions to score
a goal under the supervision of the team’s
coach
Terms Used in FGA
Idealized Rules
1. Pseudo time of the algorithm will be considered as the
overall time of the match. The members of the population
are considered as football players and search or design
space is considered as the football pitch.
2. All the football players belongs to one team that is
considered like a team that is losing the game or always
wants to score more goals and they are attacking all long of
the game.
3. It is considered that all the players of opposing team are
positioned around the peaks of a minimization problem that
makes these places unwilling for the attacking team (the
coach and the players).
Initialization
The initial population is equivalent to the
initial formation of the players on the pitch
The initial fittness values attributed to each
player indicates the quality of the player’s
position
The general movements of the players
Normally and without coaching effects on the
players’ movement, their movements is
composed of two terms.
First term is a simple random walk, and
second term is a movement toward the ball.
Every player moves randomly from his last
position to find a better position.
Also he moves toward the player that has the
ball to receive it and make a better chance
for scoring a goal
The general movements of the players
the new position of the player at time step t is
given by:
Where ε ∈[-1,1] and β ∈[0,1] are random numbers drawn
from a uniform distribution and α > 0 is the step size
which should be related to the scales of the problem of
interest.
is the position of the player who has the ball at time
step t .
The ball will be passed randomly between the players and
the players in the better positions have more chance to
receive the ball.
the random walk term can easily be extended to other
distribution such as Levy fights.
Also a further improvement on the convergence of the
algorithm is to vary the randomization parameter α so
that it decreases gradually as the optima are
approaching.
For example, we can use
Where θ ∈(0,1] is the randomness reduction constant.
α 0 is the initial randomization parameter which should be related to
the scales of the problem of interest
Coaching
In addition to general movement of the
players, the team coach uses his findings
during the match to guide the players and put
them in the better positions.
The coach memory (CM) is actually the
memory of the algorithm to save best
positions and their corresponding fitness
values.
The coach memory size (CMS) will be chosen
as a fraction of the population size.
Coaching: Strategy 1: Attacking
In this strategy the coach pushes the back
players like
defenders and midfielders forward and
toward the best positions.
Technically speaking, we consider an
imaginary Hyper Sphere (HS) with Hyper
Radius (HR) that the position of the best
member of the population will be set as its
center
a Hyper Radius Limitation Value (HRLV) will
be defined that will be decreased gradually as
the iterations proceed.
every member of the population has a hyper
distance (HD) from the best position.
Members with higher distance value in
comparison with HRLV will be pushed toward the
nearest best positions.
According to the optimization terminology we can
call this strategy Hyper Radius Penalty method.
Where γ ∈(0,1) is the reduction constant of HRLV
Coaching strategy 2 : Substitution
In this strategy the coach uses substitution
option to change weaker players with better
ones.
Actually in this strategy weaker players that
have high corresponding fitness values will be
replaced with other players around the
nearest best position, according to the coach
memory
we define a Fitness Limitation Value (FLV)
that will be decreased accordingly as the
iterations proceed.
As can be seen in the figure, every member of the
population which has grater fitness value than FLV
will be replaced with another one around the
nearest best solution.
we name this strategy Fitness Penalty method
