Methodology
How the EV fleet demo is grounded
Synthetic Oakland depot fixtures, PG&E-style TOU bands, and a cached QPU trace for charger-queue micro-window replay.
Evidence
Sources
- PG&E B-19–style TOU bands (synthetic $/kWh and demand charge $/kW).
- Last-mile fleet sizing: 18 vans, 12 L2 bays, 24 stops, embedded distance matrix.
- Rank 3 use-case framing from demos/Demo_Use_Cases.md ($400–900/day peak avoidance).
Pipeline
Solver settings
- Routing: greedy nearest-neighbor VRP with energy feasibility per van.
- Classical: OR-Tools CP-SAT charger×slot assignment minimizing TOU + demand proxy.
- Hybrid: vehicle×charger×slot QUBO with peak-concurrency penalty; fixture QPU replay.
- QAOA defaults: reps=1, maxiter=8, shots=256; gated above QTANGL_EVFLEET_QAOA_MAX_VARIABLES (24).
Repository
Evidence files
- benchmarks/ev_fleet_results.csv
- demos/ev_fleet_charging/data/qpu_trace.json
- backend/app/ev_fleet/fixtures/calibration.md