Use of matrices

After taking an active interest in the use of GP for an AI in the game
of Diplomacy, I have developed the Framework I think my AI will work
under.  An adaption of Sean Luke's SoccerSoftbots methodology.  The
main changes I am making are to the function set and the amount of GP
trees.

I found an explanation of distance calculations of varying polygons
at:
http://graphics.cs.ucdavis.edu/~okreylos/Private/AlgorithmCorner/PathFinding.html

I envisage the calculations as CM^N where N is the number of nodes to
iterate to.  I also see the use of this matrix methodology to be
equivalent to calculating Influence Maps with quasi-vectors (numbers
with a 'direction').
I took it to mean the matrix could be written as a function:
Matrix(f(a),x,y,z)
where a = object of interest
x = initial cost
y = move cost (+-/*SQR f(y))
z = N
Using the example from the link above could be written:
Matrix(f(Polygon),0,1,11)

Given this, I can construct 10 different matrix types as my Function
Set.

Each tree (program) that is executed results in a list of weight
vectors.
Multiple trees will be calculated at various times and compared.  The
lowest weight will determine the vector chosen.

It is this need (methodologically speaking) to compare weights that
requires multiple trees.  My AI at the moment requires 11 trees.  The
11 trees are broken into 3 groups.  My breeding regime will be adapted
such that only trees within groups can mate.

I am uncertain if this is too many trees, nor if this particular
method of  Influence Mapping is too inefficient.  Thus my question is
whether anyone has specific information on the construction of
Influence maps?