Prediction & Optimization Engines
Predictive telematics ML engine (LightGBM / XGBoost regressor) and MILP dock constraint optimizer (Google OR-Tools CP-SAT), streaming live.
ETA MAE
2.9min
Recompute cadence
60sec
Solver time
1046ms
Re-optimisations / hr
15
1. Predictive Telematics ML Engine
LightGBM / XGBoost regressor- Predicts precise arrival windows (ETÂ) by analyzing real-time GPS streaming data against historical traffic, weather, and driver behaviour models.
- Dynamically recalculates arrival probability matrices every 60 seconds to detect early/late arrival variances.
| Truck | ETÂ | Confidence | Variance | Signal |
|---|---|---|---|---|
| TRK-210 | 99 min | 88% | 0 min | Within tolerance |
| TRK-211 | 86 min | 97% | -12 min | Early — pull slot forward |
| TRK-212 | 74 min | 95% | -8 min | Early — pull slot forward |
| TRK-213 | 61 min | 84% | +7 min | Late — trigger reschedule |
| TRK-214 | 48 min | 78% | +13 min | Late — trigger reschedule |
| TRK-215 | 36 min | 84% | +2 min | Within tolerance |
2. MILP Dock Constraint Optimizer
Google OR-Tools CP-SAT / MILP solver- Solves complex multi-variable scheduling math to minimize truck wait time, ASRS buffer load variance, and AGV travel distances.
- Enforces hard operational constraints: cold chain dock doors, ASRS spur throughput caps, and AGV battery levels.
Cold chain dock affinity
Temperature-controlled loads restricted to D01, D02, D07
ASRS spur throughput cap
Max 72 moves/hr per aisle
AGV battery floor
No task dispatch below 30% state of charge
Labour shift coverage
Skill-matched operator per active door
Yard smoothing window
Max 6 trailers queued in yard