Optimization

Optimizing a snapshot

Nosy always minimizes objectives. To maximize an expression, minimize its negative instead.

The objective passed to optimize! can be any scalar affine expression: the usual objective is the total cost of a snapshot, but metrics and arithmetic can be combined freely.

Examples:

optimize!(snapshot, cost(snapshot))
optimize!(snapshot, cost(snapshot, :fuel) + cost(snapshot, :startup))
optimize!(snapshot, -capacity(snapshot, "gasplant"))

Because optimization is always minimization, minimizing -capacity(snapshot, "gasplant") maximizes that capacity.

Optimizing multiple snapshots

optimize! can also optimize a vector of snapshots that share the same simulation. This is useful for more advanced calculations, such as stochastic programming or optimized pathways.

Example:

optimize!([snapshot1, snapshot2], cost(snapshot1) + 2 * cost(snapshot2))

Infeasible problems

Some problems are infeasible, generally due to conflicting constraints. In that case, you can identify the conflicting constraints using the conflicts function. Nosy generates variables with explicit names to facilitate debugging. Please note: the IIS of an infeasible problem is not necessarily unique, and this step can be iterative.

Extract results

After optimizing a snapshot, the usual workflow is to extract the results, meaning generating a snapshot populated with the values of the optimal solution. The generated result is a replica of the problem snapshot, but with values instead of variables and expressions.

optimize!(snapshot, cost(snapshot))
result = extract(snapshot)

Exporting a problem to file

Before running the optimizer, optimize! calls finalize! if the snapshot has not already been finalized. Finalization adds the node constraints, checks connections, and cleans up the problem, including variable bounds. If you want to export the finalized optimization problem before solving it, set the objective and call finalize! explicitly before writing the JuMP model:

finalize!(snapshot)
JuMP.set_objective(model(sim(snapshot)), JuMP.MIN_SENSE, cost(snapshot))
JuMP.write_to_file(model(sim(snapshot)), "example.mps")

If your problem is already optimized, there is no need to finalize! the snapshot, you can call JuMP.write_to_file directly.