AI Heating Supply Optimization System Development

Development of an AI System for Heating Supply Optimization

Heating supply is the largest expense item for housing utilities in Russia. An AI system optimizes heating network operation: forecasts heat load, automatically adjusts temperature schedule, and reduces gas/heat consumption by 8-15%.

Building Heat Balance

Physical model:

Q_loss = U_building × A × (T_indoor - T_outdoor) + Q_ventilation
Q_needed = Q_loss - Q_solar_gain - Q_internal_gain

Building U-value (thermal conductivity)—key parameter. Determined from heat meter data + historical temperatures (regression).

Thermal inertia: Building doesn't respond instantly to outdoor temperature changes. Temporal lag: 1-6 hours depending on building mass. ML model explicitly accounts for this lag.

Heating Load Forecasting

Input data:

heating_features = {
    # Weather (main driver)
    'temp_outside': outdoor_temperature,
    'temp_forecast_6h': temperature_6h_ahead,
    'wind_speed': wind_speed,  # convective losses
    'solar_radiation': ghi,     # passive solar heating

    # Building/network
    'temp_indoor_setpoint': 22.0,
    'building_heat_loss_coeff': U_building,
    'thermal_mass': building_thermal_mass,

    # Historical
    'heat_demand_lag_1h': heat_demand_1h_ago,
    'heat_demand_lag_24h': heat_demand_24h_ago,

    # Context
    'hour': hour_of_day,
    'is_occupied': occupancy_schedule,  # working hours vs. night
    'day_type': encode(workday_weekend_holiday)
}

Models:

• RC-model (Resistance-Capacitance): physical heat balance model. Parameters identified from ASKUET data.
• ML (LightGBM): better captures anomalies (wind through cracks, unexpected insulation failures)
• Hybrid: RC-model + ML residual correction

Accuracy: MAPE 3-6% for hourly forecast 24 hours ahead.

Temperature Schedule Optimization

Traditional CT temperature schedule: dependency of supply temperature on outdoor air temperature—fixed curve in ITP regulator.

AI optimization:

def optimal_supply_temperature(T_outdoor, T_indoor_target, Q_predicted,
                                hydraulic_state, network_losses):
    """
    Minimize: gas_consumption(T_supply)
    Subject to: T_indoor >= T_target for all consumers
    """
    # Hydraulic network model → temperature at each consumer
    # as function of T_supply and flows
    T_consumer = hydraulic_model(T_supply, flow_rates)
    constraint = T_consumer.min() >= T_indoor_target

    # Optimize
    result = minimize_gas(T_supply, constraints=[constraint])
    return result.x

Weather-based regulation with forecast:

• Classic: adjustment by current T_outdoor
• AI: adjustment by T_outdoor 2-3 hours ahead (accounting for building thermal inertia)

This prevents overheating with warming and overcooling with sudden cold.

Automatic ITP Control

ITP (Individual Heat Point)—control point for building:

Controlled parameters:

• Heat carrier supply temperature
• Flow (via control valve)
• Hot water supply mode

SCADA/Control System:

• ITP controllers: Siemens PLC / Oven PLC
• Protocols: Modbus TCP, MQTT for IoT sensors
• SCADA: ZENON, IntegraTooll

ML decision model for ITP: RL agent controls valve, receiving observation: T_indoor, T_supply, T_outdoor_forecast. Reward: -energy_consumed with T_indoor >= setpoint.

Loss Detection and Emergency Response

Thermal loss analysis: Comparison: heat supplied by source vs. heat received by consumers. Difference = network losses. Abnormal loss increase → possible pipeline emergency.

def detect_network_leak(supply_heat, return_heat, consumer_receipts):
    theoretical_losses = supply_heat - consumer_receipts
    actual_losses = supply_heat - return_heat  # by meters
    unexplained_loss = actual_losses - theoretical_losses

    if unexplained_loss / supply_heat > 0.05:  # >5% sudden losses
        alert("Possible network emergency, localize by section")

Network segmentation: Hydraulic network model + anomaly detection → loss section localization to 200-500 m.

GIS integration: QGIS / ArcGIS + pipeline database → anomaly visualization on map → dispatcher sees specific section.

System metrics:

• Gas savings: 8-15% with AI control vs. fixed schedule
• Complaints about overheating/overcooling: 50-70% reduction
• Heating load forecast MAPE: < 5%
• Emergency localization time: from 4-8 hours to 30-60 minutes

Timeline: basic forecasting system + automatic temperature schedule—6-8 weeks. Full system with RL ITP control and emergency detector—4-5 months.