Chance-constrained capacity
This extends the sampled budget chance constraint to capacity: at most two of 20 training paths may exceed budget and at most two may exceed capacity in each year. Both constraints are sampled; this is not a distributional guarantee.
Results
Plan |
5-year cost P10-P90 |
Mean risk |
Budget violation |
Capacity violation |
|---|---|---|---|---|
Chance capacity |
199,709-248,168 |
0.0516 |
22.0% |
0.0% |
Chance budget only |
198,743-247,028 |
0.0516 |
21.0% |
0.0% |
Nominal balanced |
269,679-333,777 |
0.0294 |
93.0% |
0.0% |
Chance capacity nominal cost: 222,076 CAD; 170 actions.
Chance capacity does not improve the out-of-sample result. Budget violation is 22% versus 21% for budget-only chance, risk is unchanged, and out-of-sample capacity violation is already 0% in the test set. The budget-only plan is better in this test.
Interpretation
The sampled capacity relaxation does not help because the test paths do not show capacity violations and the budget constraint remains tighter. The experiment is a negative result for this parameter set.
A better next step would be an operational recourse model or a distributional chance constraint with more training paths.
Limits
Sampled constraints, not distributional.
20 training paths; 100 test paths.
Shared static decisions; no recourse.
No risk, weather, outage, safety, asset-owner priority, or field calibration.