Defining by nature "control" variables as "state" variable on the scenario tree

[bitanalui]

I am solving a Multistage Stochastic Optimization problem, for dynamic cascade reservoir optimization. To incorporate non-anticipativity, release (control variable) needs to be decided prior to the realization of inflows at t, that is nodes belong to period t-1 . It is suggested to treat release as state variable.
Is this implementation correct?

@variable(subproblem, release[r=1:6] >= Cascade_system[r].release_min, SDDP.State, initial_value = Cascade_system[r].release_initial)

and then it appears in the following constraint:

@Constraints(subproblem,begin
storage_equa[r = 1:6],
storage[r].out == storage[r].in + (inflow[r] + sum(release[j].in*A[r,j] for j=1:r))*1.9835
end)

release[j].in is the release at node that belongs to period t-1.

[odow]

Without the full details, hard to offer advice. But seems correct.

See https://odow.github.io/SDDP.jl/stable/guides/access_previous_variables/#Access-a-decision-from-N-stages-ago

One thing you need to be careful of is that stage t is feasible for any release that might be chosen in stage t-1.

[bitanalui]

Hello Oscar,
Thank you, by full details do you mean the complete model?
Here I am attaching a complete model with a small input file.
With the balance constraint as formulated with release[j].in we face infeasibility issue.
ControlvsStateVariable.zip

[odow]

Closing because it seems like @bitanalui and @bolbolim are the same people, and because the infeasibility is likely the same as #646.

Let's keep the discussion to a single post please.

[bitanalui]

Hi Oscar, we are not the same people, but we are trying to address and solve two issues of one model.

Thank you!