What this comparison shows
The Genetic Algorithm and Pathy solve the same aircraft scheduling problem through different optimization strategies. This benchmark compares the schedules they produced as the planning horizon grew from one day to three days.
All cost values are reported in RecOpt dollars, the shared modeled cost unit used to evaluate schedule quality. Lower total cost represents a lower combined objective value for the cost components and penalties included in this test.
Each value comes from one run, not an average across repeated runs. The results describe these three scenarios and do not establish that either engine is universally faster or lower-cost.
Scenario-by-scenario summary
Complete coverage
Both engines assigned every flight. Pathy completed in 1 minute with a total cost of 23,332, compared with 3 minutes and 26,240 for the Genetic Algorithm.
Closely matched results
Both engines left 2 flights and 125 block minutes unassigned. Pathy completed in 12 minutes with a total cost of 273,896, compared with 13 minutes and 277,255.
Different tradeoffs
Pathy left 3 flights and 175 block minutes unassigned, compared with 4 flights and 300 minutes. Pathy completed in 33 minutes with a total cost of 460,694; the Genetic Algorithm completed in 30 minutes with a total cost of 549,063.
Total modeled cost
Total cost includes the applicable unassigned-flight penalty as well as the cost categories shown in the detailed table. The unassigned-flight penalty is included in the total but is not broken out as a separate row in this benchmark.
Total cost in RecOpt dollars
Bar lengths use the same scale across all three scheduling horizons.
Optimization duration
Runtime remained close as the horizon grew. Pathy completed sooner in the one- and two-day scenarios, while the Genetic Algorithm completed sooner in the three-day scenario.
Elapsed time in minutes
Bar lengths use the same scale across all three scheduling horizons.
Complete benchmark results
| Metric | Genetic Algorithm | Pathy | ||||
|---|---|---|---|---|---|---|
| 1 day | 2 days | 3 days | 1 day | 2 days | 3 days | |
| Optimization duration | 3 min | 13 min | 30 min | 1 min | 12 min | 33 min |
| Unassigned flights | 0 | 2 | 4 | 0 | 2 | 3 |
| Total cost (RecOpt dollars) | 26,240 | 277,255 | 549,063 | 23,332 | 273,896 | 460,694 |
| Unassigned block minutes | 0 | 125 | 300 | 0 | 125 | 175 |
| Utilization cost | 16,992 | 25,330 | 52,399 | 13,775 | 31,730 | 52,368 |
| Fuel cost | 8,248 | 17,175 | 26,664 | 8,057 | 17,166 | 26,576 |
| Other penalties | 1,000 | 12,250 | 40,000 | 1,500 | 12,000 | 61,250 |
| Sparse cost | 0 | 0 | 0 | 0 | 500 | 3,000 |
| Total flights | 461 | 945 | 1,401 | 461 | 945 | 1,401 |
Methodology and limitations
- Every scenario contains the same 82 aircraft.
- The paired runs used identical flight data, hardware, operational constraints, and cost weights.
- Each result represents one run. The benchmark does not show averages or run-to-run variation.
- Optimization duration is reported in elapsed minutes.
- Total cost is reported in RecOpt dollars and includes the applicable unassigned-flight penalty.
- The unassigned-flight penalty is included in total cost but is not listed as a separate cost component in the table.
- Passenger cost, refleeting cost, minimum turn-time cost, maximum turn-time cost, and schedule-difference cost for reoptimization were excluded.
These results reflect the algorithm versions used for this comparison. Both engines are continually improved in speed, solution quality, and supported features, so future measurements may differ.