How can I use some of the class properties as optimization parameters while keeping others fixed ?

Hi,

I have a class that contains the model parameters required to model a fuel cell. While modeling the fuel cell, I pass an object instance of this class to the functions where these parameters are used for calculations. Now, I want to use same model/functions and perform optimization/calibration on fuel cell. This would require me to vary some of the model parameters while keeping others as fixed vlaues. The split between the number of fixed and varying parameters can also change.

As per my limited understanding, If I pass the object with all the properties to the optimizer function then all the parameters would be varied. So, my question is: how can I divide the properties of the class object into two sets: fixed and varying and then pass them to the optimizer.

I have looked into the answer at link below that highlights the method of using wrapper functions. But I am still not sure how do apply this idea when the parameter set is contained inside a class object.

https://in.mathworks.com/matlabcentral/answers/304275-how-to-control-the-number-of-fitted-parameters-when-using-lsqcurvefit

Thanks.

[Ameer Hamza]

Can you show your code? How are you creating the object which is passed to the optimization algorithm?

[Abhijay Awasthi]

This is the class object that contains the model parameters. I want to use a set of these properties as optimization handles while keeping others fixed.

classdef ElectrolysisParameters
%ELECTROLYSISPARAMETERS  
%   Class containing the set of model parameters used in electrolysis model.

properties (Constant = true)
    Treference = 298, % reference temperature , in K
end

%% Electrolyzer geometrical parameters
properties
    % Electrode active surface area
    electrodeSurfaceArea; % in m2
    
    % Electrode porosity, dimensionless
    anodePorosity;
    cathodePorosity;
    
    % Electrode tortuosity, dimensionless
    anodeTortuosity;
    cathodeTortuosity;
    
    % Electrode thickness
    anodeThickness; % in m
    cathodeThickness; % in m
    
    % Electrode Channel to catalyst layer distance
    anodeLCC; %m
    cathodeLCC; %m  
    
    % Membrane/Separator parameters
    membraneThickness; % m
    membraneTortuosity; % dimensionless
    membranePorosity; % dimensionless
    membraneWettability; % dimensionless
      
    % Electrode roughness factor
    anodeRoughnessFactor;
    cathodeRoughnessFactor;
end

%% Electrolysis process parameters
properties       
    % Reference exchange current density
    i0refAnode, % A/m2
    i0refCathode; %A/m2
    
    % Charge transfer coefficients
    alphaAnode; % dimensionless
    alphaCathode; % dimensionless
    
    % Free activation energy
    deltaGCAnode; % J/mol
    deltaGCCathode; % J/mol
    
    % Limiting current density
    iLim; % Limiting current density at 100% bubble coverage
        
    % Bubble zone width
    betaAnode; % in m
    betaCathode; % in m
    
    %Mean pore radius of electrode surface, (=2*porosity/(specific surface area*bulk density))
    rPore; % in m
    
    % Electrolyte concentration
    molElectrolyte; % Molality of KOH solution, 7.65 for 30% KOH    
end

%% Methods
methods
    % Constructor function
    function obj = ElectrolysisParameters(electrolyzerName)
        if nargin == 0 
            % load default parameter set
        else
            switch electrolyzerName
                case "HRI"
                    %% load HRI parameter data
                    % Electrode active surface area
                    obj.electrodeSurfaceArea = 300e-4; % in m2
                    
                    % Electrode porosity, dimensionless
                    obj.anodePorosity = 0.3;
                    obj.cathodePorosity = 0.3;
                    
                    % Electrode tortuosity, dimensionless
                    obj.anodeTortuosity = 3.65;
                    obj.cathodeTortuosity = 3.65;
                    
                    % Electrode thickness
                    obj.anodeThickness = 0.2e-2; % in m
                    obj.cathodeThickness = 0.2e-2; % in m
  
                    % Electrode Channel to catalyst layer distance
                    obj.anodeLCC = 0.125e-2; %m
                    obj.cathodeLCC = 0.125e-2; %m
                    
                    % Membrane/Separator parameters
                    obj.membraneThickness = 0.05e-2; % m
                    obj.membraneTortuosity = 2.18; % dimensionless
                    obj.membranePorosity = 0.42; % dimensionless
                    obj.membraneWettability = 0.85; % dimensionless
                    
                    % Electrode roughness factor
                    obj.anodeRoughnessFactor = 1.25;
                    obj.cathodeRoughnessFactor = 1.05;
                    
                    % Reference exchange current density
                    obj.i0refAnode = 1e-11 * 1e6; % A/m2
                    obj.i0refCathode = 1e-3 * 1e6; %A/m2
                    
                    % Charge transfer coefficients
                    obj.alphaAnode = 1.65; % dimensionless
                    obj.alphaCathode = 0.73; % dimensionless
                    
                    % Free activation energy
                    obj.deltaGCAnode = 80 * 1e3; % J/mol
                    obj.deltaGCCathode = 51 * 1e3; % J/mol
                    
                    % Limiting current density
                    obj.iLim = 300*1e3; % in A/m2 Limiting current density at 100% bubble coverage
                    
                    % Bubble zone width
                    obj.betaAnode = 0.045e-2; % in m
                    obj.betaCathode = 0.045e-2; % in m
                    
                    %Mean pore radius of electrode surface, (=2*porosity/(specific surface area*bulk density))
                    obj.rPore = 0.1e-6; % in m
                    
                    % Electrolyte concentration
                    obj.molElectrolyte = 7.65; % Molality of KOH solution, 7.65 for 30% KOH
  
                otherwise
                    % load custom electrolyzer parameter set
            end
        end
    end    
end

end

[Walter Roberson]

No, when you pass a class object to the Mathworks optimization functions as the initial value, then the optimization function will error. The optimization functions only allow numeric values (or, in obscure cases, symbolic expressions that are convertible to numeric)

[Ameer Hamza]

How are you calling the optimization function using this class?

[Abhijay Awasthi]

ahh, Ok. Thank you very much for this clarification. Would you have any tips for me on how to use parameters in this class object as optimization handles? Any workaround by creating numeric arrays ?

[Abhijay Awasthi]

Ameer, Thanks for replying. I am still developing the code for using this class inside an optimizer. But I think, based on Walter's answer, this approach would not work.

Answer 1

[Walter Roberson]

The split between the number of fixed and varying parameters can also change.

For each different combination, write a accessory function that accepts and object and a vector of values, and splits the values in pieces and sets the object properties to the extracted parts and returns the resulting object.

Then write an objective function that is parameterized to accept a vector of values and the "prototype" object and a function handle (to the accessory function.) The objective function should pass the vector of values and the prototype to the accessory function, getting back an object, and then the objective function should calculate the "cost" of that particular object.

As far as the optimization functions are concerned, they are calling a function that accepts a vector of values. The anonymous function captures the prototype and the accessory function handle, calls the accessory function to build the new object, and costs that out.

Depending on your needs, it might be possible to construct just one accessory function that takes a data structure that describes how to do the splitting, such as a struct with fields named for properties to set, and whose values are indices of values to take from the x vector and pass into the set() routine

[Abhijay Awasthi]

Thanks a lot. I understand the approach. Will try this out.