AI Fuel Consumption Optimization System Development

Development of an AI system for fuel consumption optimization

Fuel accounts for 30–40% of a transport company's operating costs. The AI system reduces consumption through three methods: route optimization, driver coaching on driving style, and predictive engine maintenance.

Fuel consumption modeling

Physical model of consumption:

Fuel consumption depends on the resistance force: - Aerodynamic drag: increases proportionally to v³ - Rolling resistance: proportional to mass and speed - Inertial losses: braking = release of stored kinetic energy - Terrain: climbs require additional work against gravity

import numpy as np

def fuel_model_physics(
    route_segments,   # [(distance_m, grade_pct, speed_limit_kmh)]
    vehicle_params,   # {'mass_kg', 'Cd', 'A_frontal', 'Crr', 'engine_eff'}
    actual_speeds=None
):
    """
    Физическая модель расхода топлива по маршруту.
    Возвращает л/100км для заданного профиля скоростей.
    """
    rho_air = 1.2  # кг/м³
    g = 9.81
    m = vehicle_params['mass_kg']
    Cd = vehicle_params['Cd']  # аэродинамич. коэффициент (~0.35 для TIR)
    A = vehicle_params['A_frontal']  # м² (~8 для TIR)
    Crr = vehicle_params['Crr']  # коэффициент качения (~0.006)
    eta = vehicle_params['engine_eff']  # КПД привода (~0.35)

    total_fuel_j = 0
    total_dist_m = 0

    for dist_m, grade_pct, speed_kmh in route_segments:
        v = (actual_speeds or speed_kmh) / 3.6  # м/с
        grade = grade_pct / 100

        F_aero = 0.5 * rho_air * Cd * A * v**2
        F_roll = Crr * m * g * np.cos(np.arctan(grade))
        F_grade = m * g * np.sin(np.arctan(grade))

        F_total = F_aero + F_roll + F_grade  # только движение вперёд
        if F_total < 0:  # спуск — можно рекуперировать (для EV) или мотор-тормоз
            F_total = 0

        # Работа = сила × расстояние
        work_j = max(0, F_total) * dist_m
        fuel_energy_j = work_j / eta

        total_fuel_j += fuel_energy_j
        total_dist_m += dist_m

    diesel_energy_density = 35.8e6  # Дж/литр
    fuel_liters = total_fuel_j / diesel_energy_density
    return fuel_liters / (total_dist_m / 1000) * 100  # л/100км

ML-correction of the physical model:

The physical model does not take into account real-world conditions: engine temperature, injector wear, and asphalt type. ML (XGBoost) builds a residual model: δ = actual - physical_model. The final model: ŷ = physical(x) + ML(x).

Eco-driving system

Driving style scoring:

Each driving event is classified and contributes to the eco-score:

Drivers get a personal dashboard and real-time push recommendations: - "800m downhill ahead - release the accelerator" - "Speed 98 km/h - 88 km/h is more advantageous"

Gamification: monthly ranking + bonus for the top 20% of eco-drivers.

Route optimization with fuel criteria

The shortest route doesn't always equal the least fuel-consuming. ML fuel cost estimates for each route: - SRTM terrain: total elevation gain (climbs = consumption) - Road type: highway (optimal cruising speed) vs. city traffic (lots of starts) - Historical traffic: time spent in traffic with the engine running

Typical result: a route 5% longer, but 8-12% more economical.

Monitoring technical losses

Abnormally high flow = technical signal:

• Leaking injector: increased consumption under normal driving conditions - Ignition system malfunction: misfiring → incomplete combustion - Tire pressure: underinflated tires +2–4% consumption

LSTM-Autoencoder with normalized fuel consumption (l/100km with adjustments for terrain and load) → anomalies → detailed diagnostics in the service center.

Development time: 2–4 months for eco-driving scoring, consumption anomalies and route optimization with integration into telematics (Wialon, OMNICOMM, AutoGRAPH).