constrained pso : how to set initial population ?

hi, i tried

options = optimoptions('particleswarm', 'InitialPopulation', tous_xi);

where tous_xi contains initial positions, and i get the error :

'InitialPopulation' is not an option for PARTICLESWARM.

Can you help me to set that initial population? Thank you

[genevois pierre]

hi again, when i try :

options = optimoptions('particleswarm', 'InitialSwarmMatrix', tous_xi);

i get an error ... Can you help me again ? Thank you

[Alan Weiss]

Please show us the error that MATLAB throws so we have a chance of helping you.

Alan Weiss

MATLAB mathematical toolbox documentation

[genevois pierre]

when i use

options = optimoptions('particleswarm', 'InitialSwarmMatrix', tous_xi);

i get this error :

Error using optimoptions (line 105) 'InitialSwarmMatrix' is not an option for PARTICLESWARM. A list of options can be found on the PARTICLESWARM documentation page.

in this doc, i see that the option for initial population reads

InitialSwarm Initial population or partial population of particles.

When i try, accordingly,

options = optimoptions('particleswarm', 'InitialSwarm', tous_xi);

I get this error :

Error using psocheckinitialpopulation (line 103) Problem is infeasible due to nonlinear constraints

Error in pso (line 338) [state,options] = psocheckinitialpopulation(state,...

It seems to switch to the unconstrained standard Matlab's pso ... Thank you for your help

[Walter Roberson]

Which MATLAB version are you using?

[genevois pierre]

hi, i use R2015b

[Walter Roberson]

In R2015b, the option was 'InitialSwarm'

"Problem is infeasible due to nonlinear constraints"

That can happen. Are you certain that your constraints can be satisfied?

[genevois pierre]

hi, i switched from my laptop (Matlab R2015b) to my office's PC (Matlab R2013a) : i will use Matlab R2013a from now on. Does this change in Matlab version changes something to your last message ?

The constraints can be satisfied, because i get no problem when i use ga instead of pso.

With Matlab R2013a, when i run

options_pso = psooptimset(@pso);

options_pso = psooptimset('InitialPopulation', tous_xi);

[x, fval, exitflag] = pso(@mapse_vu_sd, 7, [], [], [], [], ...
    lx, ux, @mapse_CNL_sd_c, options_pso)

the error is

Error using psocheckinitialpopulation (line 103)
Problem is infeasible due to nonlinear constraints

Error in pso (line 338)
    [state,options] = psocheckinitialpopulation(state,...

Error in essai_pso_matlab (line 40)
[x, fval, exitflag] = pso(@mapse_vu_sd, 7, [], [], [], [], ...

but since there is no error due to the 2 lines that use psooptimset, it seems to mean that 'InitialPopulation' is not a problem ... Thank you again for your help.

[Walter Roberson]

pso() and psooptimset() are not Mathworks routines. They appear to be https://www.mathworks.com/matlabcentral/fileexchange/38224-co-blade-software-for-analysis-and-design-of-composite-blades or https://www.mathworks.com/matlabcentral/fileexchange/25986-constrained-particle-swarm-optimization

The Mathworks routine is particleswarm() introduced in R2014b, which does not support constraints.

Someone would have to study the constrained pso() code to determine why it does not feel that the constraints can be satisfied. I do not feel that is something I would have time to do.

Answer 1

[Walter Roberson]

InitialSwarmMatrix is the option

Answer 2

[jonas_h]

Did you find a solution yet? I have the same problem. I am also using this PSO https://de.mathworks.com/matlabcentral/fileexchange/25986-constrained-particle-swarm-optimization because I need the nonlinear constraints.

I get the same error and thought I made a mistake but with ga everything is working fine (much slower but at least it is working). I tried the pso several times and sometimes it is working (maybe in 2/10 cases) In the other cases there is this error:

Error using psocheckinitialpopulation (line 103)
Problem is infeasible due to nonlinear constraints
Error in pso (line 338)
    [state,options] = psocheckinitialpopulation(state,...

There are possibilieties where there constraints can be satiefies. Otherwise it would not be working in some cases.

After this I tried to set the initialpopulation:

psooptimset(...InitialPopulation',startwerte);

"startwerte" should be a vector of my starting values (the function contains 12 variables which should be optimized). With the starting values the constraints are 100% satisfied. This did not help very much. maybe it is working more often now but not everytime.

Is this a bug in the PSO? I think then I have to use the ga. I would like to use the PSO because it is much faster in my case.

Thanks!

[Walter Roberson]

It would help to have specific functions (and inputs) to test with.

Answer 3

[genevois pierre]

I get the same problem ... probably due to the call to fmincon in psocheckinitialpopulation, line 99.