MATLAB: How to use structs as x input for Matlabs optimization methods

function handlesinput structsOptimization Toolboxstructstructures

My functions look like that:

function area_rectangle = area_rectangle(input_variables)
      if ~isfield(input_variables,'height') || ~isfield(input_variables,'width') 
          area_rectangle = NaN;
      else    
          area_rectangle = input_variables.width * input_variables.height;
      end

Suppose I want to use one of Matlabs optimization methods, e.g. fmincon, and pass a struct as input. How could I do that?

fmincon(@area_rectangle, input_variables, ...)
Error using fmincon (line 224)
FMINCON requires the following inputs to be of data type double: 'X0'.

Thanks in advance! Carolin

Best Answer

You would not do that. Remember that the optimization routines control the values of x themselves, adjusting its value until the optimum value is found. The optimization routines have no idea how to update a struct.

You would instead use one of the optimization routines that can work on vectors of values, and you would pass the values of the fields as elements of x. For example,

function area = calc_area(x)
  height = x(1,:);
  width = x(2,:);
  area = height .* width;
end

Related Solutions

MATLAB: How to implement an array of function handles that takes a struct as input and calculates the function value for each function handle

Hello Carolin, you can evaluate functions in a loop:

>> values = zeros(1,length(test));
>> for k = 1:length(test)
       values(k) = test{k}(input_variables,constants);
   end
>> values
values =
    6.0000    3.1416

or using cellfun():

>> values = cellfun(@(F)F(input_variables,constants),test)
values =
    6.0000    3.1416

cellfun looks more compact, but it is often harder to read and modify, especially when you are going back to the piece of code after a while.

MATLAB: Fmincon() power grid

Yes, my advice would be "Don't do that."

fmincon is a local minimizer, not a global minimizer, so it will get stuck in the first local minima it happens to see. However, power grid calculations would be expected to have a number of local minima. fmincon() is likely to get stuck.

You will likely need a global minimization technique. Some of the global minimizing routines can call upon fmincon() to get a more precise answer as the last step. I can also suggest a modification to the code of patternsearch() that allows it to call fmincon() at steps between in order to accelerate convergence; this would be a more advanced technique.

Best Answer

Related Solutions

MATLAB: How to implement an array of function handles that takes a struct as input and calculates the function value for each function handle

MATLAB: Fmincon() power grid

Related Question