Development of an AI fleet management system Fleet Management AI transport
Transport fleets—bus depots, taxi companies, and car-sharing networks—are faced with the challenge of optimizing vehicle utilization and maintenance. The AI system combines telematics, predictive maintenance, and operational dispatching.
Prediction of technical failures
CAN-bus telematics:
Modern buses generate thousands of parameters via the OBD-II/J1939 protocol. Key parameters for forecasting include: - Coolant, oil, and transmission temperatures - Turbocharger and injection pressure - Engine speed and its variance - Brake pad wear (based on braking intensity)
from sklearn.ensemble import RandomForestClassifier
from sklearn.preprocessing import RobustScaler
import pandas as pd
import numpy as np
class FleetFailurePredictor:
"""Прогноз отказа компонентов транспортного средства за 7–14 дней"""
COMPONENTS = ['engine', 'transmission', 'brakes', 'suspension', 'electrical']
def extract_rolling_features(self, telemetry_df, window_hours=24):
"""Агрегация телематики за скользящее окно"""
features = {}
freq = f'{window_hours}H'
for col in ['engine_temp', 'oil_temp', 'rpm', 'fuel_rate']:
rolled = telemetry_df[col].rolling(freq)
features[f'{col}_mean'] = rolled.mean()
features[f'{col}_std'] = rolled.std()
features[f'{col}_max'] = rolled.max()
features[f'{col}_trend'] = rolled.apply(lambda x: np.polyfit(range(len(x)), x, 1)[0])
# Аномалии: сколько раз параметр выходил за 3σ за последние сутки
for col in ['engine_temp', 'oil_temp']:
mean, std = telemetry_df[col].mean(), telemetry_df[col].std()
features[f'{col}_spike_count'] = (
(telemetry_df[col] > mean + 3*std).rolling(freq).sum()
)
return pd.DataFrame(features)
Predictive maintenance metrics for transport: - Recall failures within 7 days: target >85% - False positive rate: <20% (every fifth alert is a real problem or not) - Lead time: average 5–8 days before failure
Optimizing the release schedule
Problem: With 200 buses and 50 routes, which buses should be deployed, on which routes, with what driver schedule?
Vehicle-Shift Scheduling Problem: - Driver works 8-10 hours with a 45-minute break (Labor Code of the Russian Federation, ECHR for intercity) - Bus requires 1 hour of washing + maintenance between shifts - Minimize: number of buses + driver overtime
OR-Tools CP-SAT: for a fleet of 100–300 units, solution in 30–120 seconds.
Operational dispatching
Interval Alignment (Bus Bunching Prevention):
Classic problem: two buses traveling together after a delay. Algorithm: - Real-time GPS headway monitoring - When merging: first bus — holding (delay at the stop) - ML model predicts optimal holding time: too long = passenger irritation, too short = bunching again
Redistribution on failure:
Bus breakdown on route → the system offers: 1. The nearest backup bus from the depot (ETA calculation) 2. Reroute the bus from a parallel route 3. Notify passengers via boards and the app
Economics and Analytics
Fleet Management KPIs:
| Метрика | Типичные значения | Цель с AI |
|---|---|---|
| Техническая готовность парка | 82–88% | 92–96% |
| Пробег до внепланового ТО | 8 000–12 000 км | 18 000–25 000 км |
| Расход топлива л/100км | базис | -8–12% |
| Время оборота на маршруте | базис | -5–8% (регуляризация интервалов) |
Fuel control:
Consumption rate = f(route, load, weather). Deviation >15% → suspected fuel drain: - Fuel level correlation with GPS route and load - ML classifier: natural fluctuations vs. fuel drain - Alert with geolocation of suspected incident
Development period: 4–6 months for a predictive maintenance + dispatching + analytics system with integration into the existing MIS/TMS of a motor transport company.