app.state.arrivals

Arrival detection from GPS: when did a vehicle actually reach each planned stop.

A port of ml/src/features/tabular.gps_history with the same constants: the prediction model was trained on arrivals found by exactly this rule, so the backend’s current deviation must use it too.

For a moment T, walk the stops planned in [T − 1 h, T + 5 min] (skipping manually-filled ones) in order. The arrival at a stop is the GPS fix closest to it, within ARR_RADIUS_M, inside the window [plan − 7 min, plan + 12 min] ∩ (previous arrival, T] — and only once the vehicle has since moved LEAVE_M further away (it has left the stop). Dwell = time spent within ARR_RADIUS_M.

Functions

detect_arrivals(ts, lon, lat, sched, T)

Arrivals at stops planned in [T − 1 h, T + 5 min], from cleaned telemetry (ts sorted, int seconds, only fixes with ts <= T matter; invalid positions are NaN).

Classes

Arrival(pos, plan_s, arrival_s, delay_s, dwell_s)

class app.state.arrivals.Arrival(pos: 'int', plan_s: 'int', arrival_s: 'int', delay_s: 'float', dwell_s: 'float')

Базовые классы: object

__init__(pos: int, plan_s: int, arrival_s: int, delay_s: float, dwell_s: float) → None

arrival_s: int

delay_s: float

dwell_s: float

plan_s: int

pos: int

app.state.arrivals.detect_arrivals(ts: numpy.ndarray, lon: numpy.ndarray, lat: numpy.ndarray, sched: VehicleSchedule, T: int) → list[Arrival]

Arrivals at stops planned in [T − 1 h, T + 5 min], from cleaned telemetry (ts sorted, int seconds, only fixes with ts <= T matter; invalid positions are NaN).