Historical Analysis

Years of BDX trend data is a dataset, not just a record. We apply ML techniques to long-horizon historical data to surface operational patterns, seasonality, and anomalies that point-in-time dashboards miss.

Output

Illustrative chart — synthetic data shown for demonstration.

What we use it for

• Load-profile clustering — automatically grouping days into weekday / weekend / holiday / event-day / anomaly clusters without hand-labeling.
• Seasonal decomposition of utility and equipment performance to separate weather effects from operational change.
• Anomaly detection across multi-year windows — identifying days, weeks, or zones that deviate from learned norms.
• Change-point detection to pinpoint when a building's behavior shifted (commissioning, retrofit, schedule change, fault).
• Portfolio benchmarking — comparing similar buildings on normalized metrics to prioritize engineering attention.

Approach (high level)

• Unsupervised clustering (k-means, DBSCAN, time-series-specific distance metrics) on daily load shapes and operating envelopes.
• Isolation forests and residual-based scoring for multivariate anomaly detection.
• Bayesian change-point detection on key KPIs.
• Results feed directly into the automated reporting and dashboard examples elsewhere in this gallery.

Proprietary implementation

The full analytical pipeline — including the specific feature engineering and thresholds we use on client sites — is part of BuildingLogiX's internal tooling.