How to use four Gaussian process regression trained machine learning models as objective functions in multiobjective optimization?

Question

Saurabh Sharma 2023-12-18

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2061797-how-to-use-four-gaussian-process-regression-trained-machine-learning-models-as-objective-functions-i

回答： Ayush Aniket 2023-12-22

I have trained four Gaussian process regression machine learning models with each model having six input variables and one output variable. In all four models, the input variables are same but output variables are different. Out of four output variables, I want to minimise three variables and maximise one variable, all four at same time. So, I have four objectives. I am using multiobjective optimization and I want to use the 4 trained models as objective functions. How can I use these 4 models as objective functions. If there is a way to get the trained model in a single mathematical equation form, then it will also be helpful. For constraints, I have 6 inequalities, each inequality having lower and upper bounds, each inequality for each input variable. Code: fun = @(X) [modell1,modell3,modell4,modell5];

A = [1,0,0,0,0,0;0,1,0,0,0,0;0,0,1,0,0,0;0,0,0,1,0,0;0,0,0,0,1,0;0,0,0,0,0,1]; b = [5;5;5;1;1;1]; Aeq = []; beq = []; lb = [1,1,1,0.4,0.4,0.4]; ub = [5,5,5,1,1,1];

x0 = [0,0,0,0,0,0];

[x fval]=fmincon(fun,x0,A,b,Aeq,beq,lb,ub)

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Ayush Aniket 2023-12-22

在 MATLAB Online 中打开

Hi Saurabh,

As per my understanding you want to use four trained Gaussian process regression models with six input variables as objective functions in a multiobjective optimization problem in MATLAB. For this purpose, you need to create a function that takes the decision variables as input and returns the predictions of the GPR models as the objectives as shown below:

function objectives = myObjectiveFunction(X, modell1, modell2, modell3, modell4)
    % Predictions from the trained GPR models passed as parameters
    obj1 = predict(modell1, X);
    obj2 = -predict(modell2, X); % Negate to maximize
    obj3 = predict(modell3, X);
    obj4 = predict(modell4, X);
    
    objectives = [obj1, obj2, obj3, obj4];
end

Since you want to minimize three of the outputs and maximize one, you should negate the output of the model you wish to maximize. This function can then be passed to an optimizer that supports multiobjective optimization such as 'gamultiobj' in MATLAB.

You can modify your code after defining the objective function as shown below:

% Create an anonymous function that captures the trained models and passes them to the objective function
objectiveWithModels = @(X) myObjectiveFunction(X, modell1, modell2, modell3, modell4);
% Run the multiobjective optimization using the anonymous function
options = optimoptions('gamultiobj', 'PlotFcn', @gaplotpareto, 'PopulationSize', 50, 'MaxGenerations', 30);
[x, fval] = gamultiobj(objectiveWithModels, 6, A, b, Aeq, beq, lb, ub, options);