link a custom creation function

Hello,

i am using optimization toolbox and i want to link m.file that contains my initial population matrix .

% Input Data
M= 1;
I=2;
J = 9;
P_Prop_left = [3200,2690,4500,2900,2300,2000,1000,500,0];
P_EL =        [2190,3100,4100,900,2000,2300,125.5,2125.57,82.19];
P_EL_Max = max(P_EL);
P_Prop_max = max(P_Prop_left);
nvars = 30


% Initial values functions
P_mcr_m_Target_initial = (max(P_Prop_left)*1.2) ./ (M.* ones(1,M))
P_mcr_i_Target_initial = (P_EL_Max * 1.1) ./ (I .* ones(1,I))
Ld_left_initial = ( P_Prop_left + (0.5 .* P_EL)) ./ P_mcr_m_Target_initial';
Ld_left_initial (Ld_left_initial > 0.9) = 0.85
Ld_left_initial (Ld_left_initial < 0.25) = 0  % make it Like a constraint later on
Ld_left_initial = reshape (Ld_left_initial', 1, M*J)
P_BTot_initial= P_mcr_m_Target_initial' * Ld_left_initial
PTI_initial = P_Prop_left - ( P_mcr_m_Target_initial *  Ld_left_initial) ;
PTO_initial = -1 .* PTI_initial;
PTO_initial(PTO_initial<0) = 0
PTI_initial(PTI_initial<0) = 0
LF_initial = (P_EL - 2 .* PTO_initial + 2 .* PTI_initial ) ./ (I .* P_mcr_i_Target_initial')
LF_initial = reshape( LF_initial', 1, I*J)
% IP Initial Population matrix of size 1x NoOfVariables
IP = [P_mcr_m_Target_initial, P_mcr_i_Target_initial, Ld_left_initial, LF_initial]

my m file returns a matrix of size 1x nvars.

Q1: can i consider this matrix as initial population matrix?

in my understanding, this creates only one individual of the whole population, which may be not enoght. if yes,

Q2: how i create a whole population close to the values returned by the IP .

Q3: and what the differnece between using Initial Population matrix and create a custom population creation function

i suppose to use uniform creation function , or constraint dependant creation function, and use the IP matrix on (initial population) cell on the optimization toolbox

Answer 1

[Abdolkarim Mohammadi]

1) "If you have a partial initial population, meaning fewer than PopulationSize rows, then the genetic algorithm calls CreationFcn to generate the remaining individuals."

https://www.mathworks.com/help/gads/genetic-algorithm-options.html#f14223 (Population options)

2) You can do it by repeating your individual to make it a matrix, and then add a noise to the elements.

InitialPopulationMatrix = repmat(IP, [PopulationSize,1]);
InitialPopulationMatrix(2:end,:) = InitialPopulationMatrix(2:end,:) + a*randn(PopulationSize-1,nvars);

Where a is the standrad deviation of the random Normal numbers to be added as noise. However, an initial population with low diversity (high similarity between the individuals, low a) increases the chance of ga() getting stuck in local minimum. Initial population should cover the whole landscape, and not concentrating in a small portion of it.

3) ga() calls the creation function to create the intial population matrix. If you provide no initial population, the creation function creates all of the initial population. if you provide an initial population matrix with fewer individuals, ga() will create the rest to have exactly an initial population with PopulationSize individuals. If you provide all of the individuals, creation function does nothing.

[Nourhan Elsayed]

thank you very much.it works very well.

i have done some modification, as part of the decision variables are on the range of 5000:7000, and the othe part of decision variables are on the range of 0:1, i hade to use two values for 'a',

i refer to 'a' by StD1 and StD2

StD1 = 500;
StD2 = 0.2;
InitialPopulationMatrix(2:end,1:M+I) = InitialPopulationMatrix(2:end,1:M+I) + StD1*randn(PopulationSize-1,M+I);
InitialPopulationMatrix(2:end,M+I+1:end) = InitialPopulationMatrix(2:end,M+I+1:end) + StD2*randn(PopulationSize-1,nvars- M - I);

that gives me a good range of values, except the negative ones that are created due to the 'randn' operator. however, i can fix this by anyway.

thank you Abdolkarim

[Abdolkarim Mohammadi]

I'm glad I could help. I want to emphasize that the diversity of the initial population is important. For your case, I suggest to fill just about 10-20 percent of the initial population with the valued you obtained by adding noise to your preferred individual. Let ga() choose the rest using uniform distribution between the lower bound and upper bound. It assures an acceptable diversity in the initial population. Global optimization problems have multiple minima, which may be located far away from your preferred individual.

[Nourhan Elsayed]

yeah, thats what exactly happening, i always stuck on the same fitness value however i change the options.

in that case, i will provide only part of the Initial Population Matrix, which is done by adding this line to the code

 
options = optimoptions('ga','InitialPopulationMatrix',InitialPopulationMatrix)

and use ga default options to generate the other part of poulation and set the creation function options to 'uniform distribution'

it is confused to use both at the same time . do i understand well?

[Nourhan Elsayed]

yes i found my answer here

InitialPopulationMatrix specifies an initial population for the genetic algorithm. The default value is [], in which case ga uses the default CreationFcn to create an initial population. If you enter a nonempty array in the InitialPopulationMatrix, the array must have no more than PopulationSize rows, and exactly nvars columns, where nvars is the number of variables, the second input to ga or gamultiobj. If you have a partial initial population, meaning fewer than PopulationSize rows, then the genetic algorithm calls CreationFcn to generate the remaining individuals

https://www.mathworks.com/help/gads/genetic-algorithm-options.html?s_tid=srchtitle#f14223

Thank you