AI-based optimization of resource consumption at the enterprise
Operating expenses for energy, water, and materials account for 20-40% of the cost of production in industry. ML-based resource optimization offers a specific ROI: reducing energy consumption by 10-20% and material loss by 5-15% without reducing output.
Multi-resource optimization
Resource Types:
- Electricity: peak consumption, tariffs by time zones, reactive power
- Thermal energy: steam, hot water, process furnaces
- Water: industrial water supply, cooling, process needs
- Compressed air: leaks, network pressure
- Raw materials and supplies: yield optimization, losses during transitions
Predictive consumption monitoring
SCADA/MES integration:
# Источники данных: OPC-UA, Modbus, MQTT с ПЛК
consumption_features = {
'power_kw_5min': sensor_readings['main_meter'],
'production_units_h': mes_data['throughput'],
'specific_consumption': power_kw / production_units, # кВт/единица
'ambient_temp': weather_api['temperature'],
'shift_code': calendar['shift'], # A/B/C shift + maintenance
'product_type': mes_data['current_sku'] # тип продукции влияет на потребление
}
Baseline and anomalies:
- Specific consumption (kWh/ton of product) is a key KPI
- The model predicts expected consumption at current output
- Deviation > 10% from expected → signal of inefficiency or leakage
Electrical Load Management
Demand Response и Peak Shaving:
def peak_shaving_schedule(production_plan, electricity_tariffs, battery_soc):
"""
Перенос гибкой нагрузки (сжатый воздух, насосы, морозильники)
из пиковых часов в ночные
"""
peak_hours = [hour for hour in range(24) if electricity_tariffs[hour] > peak_threshold]
off_peak = [hour for hour in range(24) if electricity_tariffs[hour] < off_peak_threshold]
# Гибкая нагрузка: сжатый воздух можно накапливать в ресивере
# Морозилки: тепловая инерция позволяет отключить на 30-60 минут
# Дробление: дробилки, мельницы — можно сместить на ночь
return shifted_schedule
Power Factor Correction: Reactive power = penalty tariffs. ML identifies equipment with low cos(φ) and recommends compensation (capacitor banks).
Daily Consumption Estimate: SARIMA + external regressors (production plan, temperature, day of the week) → consumption forecast for the next day to purchase electricity on the wholesale market at favorable prices.
Optimization of technological processes
Compressor Systems: Compressed air is one of the most energy-intensive utilities (10-30% of electricity consumption). Optimization:
- Network pressure: every extra 0.1 bar = +0.5% consumption
- Leak detection for nighttime compressor consumption (production is stopped - leaks are visible)
- Optimal load distribution between compressors of different sizes
def compressor_dispatch(demand_m3_min, compressors):
"""
Оптимальный выбор комбинации компрессоров для покрытия спроса
Минимизация удельного потребления кВт/(м³/мин)
"""
best_combination = None
min_power = float('inf')
for combo in all_combinations(compressors):
total_capacity = sum(c.capacity for c in combo)
if total_capacity >= demand_m3_min:
total_power = sum(c.power_at_load(demand_m3_min / total_capacity) for c in combo)
if total_power < min_power:
min_power = total_power
best_combination = combo
return best_combination
Furnaces and thermal processes:
- Optimization of the fuel/air ratio (excess air → flue gas losses)
- Prediction of temperature profile → minimum heating time with the required quality
- Heat recovery: when and how to use heat exchangers
Water resources management
Water Cascade Optimization: Water of varying quality is used at different stages of the process. The goal is to minimize fresh water consumption by reusing:
# Water pinch analysis (аналог тепловых pinch-методов)
# Стоки одной стадии = потенциальный ввод для другой (если примеси совместимы)
water_network = WaterPinch(
process_streams=streams,
freshwater_cost=cost_per_m3,
treatment_costs=treatment_cost_matrix
)
optimal_reuse = water_network.optimize()
Cooling Tower optimization:
- Cycles of concentration (cooling water concentration): balance between water savings and scale/corrosion risks
- Fan speed optimization based on wet bulb temperature and production load
- Prediction of Legionella risk based on temperature and biochemical parameters
Monitoring and reporting
Energy Management Dashboard:
- Online monitoring of specific consumption by workshops
- Comparison with a benchmark (similar companies in the industry)
- Trends for the week/month/year
- Top 5 sources of inefficiency
ISO 50001 support: The system generates energy baselines, EnPIs (Energy Performance Indicators) and documentation for ISO 50001 certification.
Timeframe: SCADA connection, basic monitoring of specific consumption, and anomalies – 4-5 weeks. Demand Response, compressor dispatch, water cascade optimization, and ISO 50001 reporting – 3-4 months.