Optimize Building Energy with AI-Powered Digital Twins

PID controllers in HVAC waste 30% more energy than necessary due to lack of adaptation to weather and occupancy. In a 10,000 m² office building, this results in 1.5–2 million rubles in annual overspending. Implementation cost for a similar building starts from 4 million rubles, yielding ROI in under 2 years. A digital twin solves the problem: it synchronizes with the physical object via sensors and optimizes engineering systems in real time. We have implemented over 50 projects—from office centers to manufacturing facilities. Certified engineers guarantee quality implementation and KPI achievement. Our solutions leverage AI for buildings, BIM intelligent building models, and ML for buildings to deliver accurate predictions.

Why a building digital twin is more efficient than traditional BMS

Traditional BMS operate on rigid schedules—night setback, fixed timetables. They don't adapt to actual occupancy, weather changes, or tariffs. A digital twin uses ML for forecasting: MPC (Model Predictive Control) reduces HVAC energy consumption by 15–25%, which is 3–5 times more efficient than PID controllers. Additionally, fault detection is up to 10 times faster compared to manual inspections. The system accounts for a 48-hour weather forecast, occupancy based on CO₂ and PIR, time-of-use tariffs—and automatically selects optimal setpoints.

Digital twin architecture

The model is built on four layers:

• Geometric: BIM model (Revit, IFC)—geometry, structures, engineering networks.
• Semantic: building ontology—rooms, zones, equipment, relationships.
• Real-time data: sensors, meters, SCADA.
• Predictive: ML models—future state, risks, recommendations.

Technology stack:

BIM (Revit/OpenBIM) → IFC conversion
    ↓
Knowledge Graph (Apache Jena, Stardog) — building ontology graph
    ↓
IoT Data Platform (ThingsBoard, AWS IoT Core) — sensor data
    ↓
Digital Twin Platform (Bentley iTwin, Azure Digital Twins, Siemens Xcelerator)
    ↓
ML/Analytics Engine (Python, PyTorch, scikit-learn)
    ↓
Dashboard (Grafana, custom React/3D WebGL)

The BIM model is converted into a Knowledge Graph, which links each sensor to a specific room and equipment. The IoT platform aggregates data via BACnet, Modbus, KNX. ML modules build forecasts, and dashboards visualize the state in real time. We provide energy consumption optimization through AI-driven control. Our IoT for buildings platform aggregates sensor data.

Thermal model and HVAC optimization

RC Thermal Network (physical model):

class ThermalZoneModel:
    def __init__(self, C_zone, R_wall, R_window, R_hvac):
        self.C = C_zone
        self.R_wall = R_wall
        self.R_window = R_window
        self.R_hvac = R_hvac

    def next_temperature(self, T_zone, T_outdoor, T_supply_air, Q_occupants, dt):
        Q_wall = (T_outdoor - T_zone) / self.R_wall
        Q_window = (T_outdoor - T_zone) / self.R_window
        Q_hvac = (T_supply_air - T_zone) / self.R_hvac
        Q_total = Q_wall + Q_window + Q_hvac + Q_occupants
        dT = Q_total / self.C * dt
        return T_zone + dT

MPC optimizes setpoints over a 24–48 hour horizon, minimizing energy cost subject to comfort constraints. The RC thermal building model is calibrated using historical data and real-time sensors. See full code in the original.

How we calibrate the building thermal model

RC parameters (C, R) are estimated from historical data via Bayesian optimization or scipy.optimize. For real-time, we use Kalman Filter. This allows the model to accurately predict thermal dynamics even with changing weather and occupancy.

Comparison of HVAC control methods

Digital twin vs. traditional BMS

How we integrate digital twin with your BMS

We do not replace existing equipment. The digital twin overlays the BMS/BAS via BACnet, Modbus, or KNX. Data from controllers flows into the IoT platform, where it is enriched with weather and occupancy data. ML models build forecasts, and optimized setpoints are sent back to the BMS. The entire cycle takes 5 to 15 seconds—fast enough for real-time operation. Get a preliminary efficiency assessment for your building in 1 day.

Project implementation stages

1. Building audit—collecting BIM, sensors, BMS data, analyzing current consumption.
2. Thermal model development—calibrating RC parameters, training ML models.
3. Platform deployment—Azure Digital Twins or Bentley iTwin, Knowledge Graph setup.
4. Integration with CMMS, ERP, tenant billing—transferring data to adjacent systems.
5. Staff training—working with dashboards, interpreting forecasts.
6. 12-month warranty—support and refinements as needed.

What's included

• Building audit report with efficiency assessment
• Calibrated thermal model (RC parameters)
• Digital twin platform (Azure Digital Twins or Bentley iTwin)
• Integration with existing BMS/BAS
• Custom dashboard (Grafana or 3D WebGL)
• Staff training (2 sessions)
• 12 months of technical support and updates

What KPIs do we guarantee?

Target metrics are fixed in the contract: energy savings (15–30%), reduction in emergency failures (up to 70%), temperature forecast accuracy (RMSE < 0.5°C). If KPIs are not met, rework at our expense. For a small facility (5,000 m²), average savings are about 0.8 million rubles/year.

Our experience

Our team consists of engineers with experience in AI/ML, certified in Microsoft Azure and Bentley iTwin. Over 50 facilities in Russia and CIS already use our solutions. Request a consultation on implementing a digital twin for your building. We will assess the project for free and prepare a commercial proposal.

Based on Wikipedia: Digital Twin

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.