Optimizing a function with a vector as input

Hi everyone,

I have a function with an input of 6 values (6 periods of power) and only one output, the price. I would like to optimize the output value (minimize it) changing the input, but I have found no information about how to optimize with a vector as an imput instead of an scalar. I already tried with fminsearch or optimproblem, but none worked.

Thank you in advance!

Answer 1

[Matt J]

There is no reason why an OptimizationVariable can't be vector-valued, e.g.,

P=optimvar('P',[6,1]);

size(P)

ans = 1×2

6 1

[Víctor García]

Then the problem should be in how I call the function: (being P a vector 1x6)

prob.Objective = function(P);

but I don't see why there should be a problem there

the error is the following:

The following error occurred converting from optim.problemdef.OptimizationExpression to single:

Conversion to single from optim.problemdef.OptimizationExpression is not possible.

[Matt J]

I don't know what your revised code looks like, but you definitely can't do things like this:

P=optimvar('P',[6,1]);
single(P)

Error using single
Conversion to single from optim.problemdef.OptimizationVariable is not possible.

[Matt J]

You also can't insert OptimizationVariables into numeric matrices, as in:

P=optimvar('P');

C=zeros(1,3,'single');
C(1)=P;

Unable to perform assignment because value of type 'optim.problemdef.OptimizationVariable' is not convertible to 'single'.

Caused by:
Error using single
Conversion to single from optim.problemdef.OptimizationVariable is not possible.

To be honest, I'm not sure why you are using optimvar at all, as opposed to the solver-based framework. Your constraints are not very complicated, and your objective function is. That's exactly the opposite of when the problem-based framework is useful.

[Víctor García]

How can I write then the prob.Objective function? Let's say you want to minimize a function f output (which is a scalar), and the input is that: P=optimvar('P',[6,1]); Wouldn't it be:

prob = optimproblem('Objective',@function(P);

[x, fval] = solve(prob)

Thank you Matt!

[Matt J]

You have much more freedom in the way you write the objective function if you use the solver-based framework

https://www.mathworks.com/help/optim/ug/types-of-objective-functions.html