Rolling policy with recourse
This comparison combines the fixed first-year nominal plan with the rolling policy and annual operational deferral. It compares four cases on the same annual-factor paths:
static fixed plan;
fixed-future annual recourse;
rolling-only;
rolling recourse.
The first year is fixed. Later years are re-solved with the rolling persistence forecast and a two-year lookahead. In the recourse cases, the current year’s actions are deferred by lowest first-year risk reduction per hour when budget or capacity is exceeded.
Results
20 episodes, seed 7.
Policy |
Budget violation |
Completion P50 |
Mean completion |
Mean risk |
Cost P50 |
|---|---|---|---|---|---|
Static fixed |
95.0% |
100.0% |
100.0% |
0.0294 |
300,711 |
Fixed-future recourse |
0.0% |
93.1% |
93.3% |
0.0335 |
279,595 |
Rolling-only |
55.0% |
100.0% |
100.0% |
0.0293 |
300,236 |
Rolling recourse |
0.0% |
99.4% |
98.8% |
0.0349 |
297,451 |
Rolling recourse restores almost all first-year scope while remaining budget- and capacity-feasible. It costs more future treatment than fixed-future recourse and does not reduce mean risk in this run. The risk metric mixes the fixed first-year risk loss with re-optimized future years; it is not a like-for-like multi-year risk improvement.
590-segment scale test
A replicated 590-segment stress instance uses the same policy. The fixed plan is exact; annual rolling re-solves use the Fresh Solutions heuristic.
Policy |
Budget violation |
Completion P50 |
Mean risk |
Cost P50 |
|---|---|---|---|---|
Static fixed |
95.0% |
100.0% |
0.0291 |
3,008,496 |
Fixed-future recourse |
0.0% |
93.9% |
0.0329 |
2,811,913 |
Rolling-only |
55.0% |
100.0% |
0.0290 |
2,969,701 |
Rolling recourse |
0.0% |
99.6% |
0.0340 |
2,942,470 |
Runtime: fixed exact 47.1 s; rolling heuristic 87.2 s for 20 episodes. The instance is replicated, not a real 590-segment network, and the heuristic makes no global-optimality claim.
Limits
Fixed first year.
No weather, outage, recurrence, safety, or asset-owner priority model.
First-year risk ranking for deferral.
Rolling policy is a persistence-forecast heuristic, not a stochastic dynamic program.
20 episodes.