| |
| |
| |
|
|
| function optimize!(model::JuMP.Model, method::XavQiuWanThi2019.Method)::Nothing |
| if !occursin("Gurobi", JuMP.solver_name(model)) |
| method.two_phase_gap = false |
| end |
| function set_gap(gap) |
| JuMP.set_optimizer_attribute(model, "MIPGap", gap) |
| @info @sprintf("MIP gap tolerance set to %f", gap) |
| end |
| initial_time = time() |
| large_gap = false |
| has_transmission = false |
| for sc in model[:instance].scenarios |
| if length(sc.isf) > 0 |
| has_transmission = true |
| end |
| if has_transmission && method.two_phase_gap |
| set_gap(1e-2) |
| large_gap = true |
| end |
| end |
| while true |
| time_elapsed = time() - initial_time |
| time_remaining = method.time_limit - time_elapsed |
| if time_remaining < 0 |
| @info "Time limit exceeded" |
| break |
| end |
| @info @sprintf( |
| "Setting MILP time limit to %.2f seconds", |
| time_remaining |
| ) |
| JuMP.set_time_limit_sec(model, time_remaining) |
| @info "Solving MILP..." |
| JuMP.optimize!(model) |
|
|
| has_transmission || break |
|
|
| @info "Verifying transmission limits..." |
| time_screening = @elapsed begin |
| violations = [] |
| for sc in model[:instance].scenarios |
| push!( |
| violations, |
| _find_violations( |
| model, |
| sc, |
| max_per_line = method.max_violations_per_line, |
| max_per_period = method.max_violations_per_period, |
| ), |
| ) |
| end |
| end |
| @info @sprintf( |
| "Verified transmission limits in %.2f seconds", |
| time_screening |
| ) |
|
|
| violations_found = false |
| for v in violations |
| if !isempty(v) |
| violations_found = true |
| end |
| end |
|
|
| if violations_found |
| for (i, v) in enumerate(violations) |
| _enforce_transmission(model, v, model[:instance].scenarios[i]) |
| end |
| else |
| @info "No violations found" |
| if large_gap |
| large_gap = false |
| set_gap(method.gap_limit) |
| else |
| break |
| end |
| end |
| end |
| return |
| end |
|
|
| explanation = """ |
| What is the Two-Phase Gap? |
| Phase 1: |
| The solver is allowed to terminate early, accepting a looser optimality gap (e.g., 1%). This means the solver can stop as soon as it finds a solution within 1% of the best possible bound. This phase is fast and is used to quickly identify major issues, such as transmission constraint violations. |
| |
| Phase 2: |
| If no violations are found (or after enforcing them), the solver is run again, but now with a tighter optimality gap (e.g., 0.1% or whatever is specified in method.gap_limit). This ensures the final solution is of high quality. |
| |
| Benefit: |
| This approach saves time: you avoid spending a long time finding a very tight solution to a model that may need to be changed anyway (due to constraint violations found in screening). |
| |
| Why is Gurobi Treated Differently? |
| Gurobi supports dynamic adjustment of the MIP gap and time limit during the solve, and is robust to repeated changes in these parameters. |
| Other solvers (like HiGHS, CBC, etc.) may not support dynamic gap adjustment as reliably, or may not respond as well to repeated changes in gap/time limit during iterative solves. |
| |
| In the code: |
| If the solver is not Gurobi, the two-phase gap feature is disabled (method.two_phase_gap = false). |
| If Gurobi is used, the code will: |
| Set a loose gap (e.g., 1%) for the first phase. |
| After constraint screening, if no violations are found, it tightens the gap and resolves for a high-quality solution. |
| """ |
