May 2026

Version 2026.5

---

Highlights

May marks a milestone for myLoadshift: over 150 VAV buildings are now actively load-shifting heating against ENTSO-E day-ahead electricity prices (delivered via the Vanta Energia signal) — load shifting is a working production solution at scale.

The Siemens integration is live, and buildings are being migrated to OpenMeteo as the new weather provider.

A significant AI safety improvement of the AI optimization. A significant change to the AI's guardrails ensures buildings are no longer affected by checks on components the AI doesn't actually control.

---

myPortal — Customers

• Cleaner graphs for AI-disabled signals — Signals where AI control is disabled no longer surface graph suggestions in myPortal, reducing visual noise.
• Timezone display preference — Users can set a preferred timezone for how times are displayed (default UTC).
• Sensor Map — type switcher added to the signal list, making it easier to filter by signal type.
• Insights climate chart — Outdoor temperature now displays a clean "no data" state when readings are unavailable, rather than rendering an empty series.

BackOffice & Integration - Partners

New integrations

• Siemens Building X integration — Now live.
• Fidelix — Discovery support added.
• OpenMeteo — Now the new weather data provider; existing buildings are being migrated, improving stability and predictions.

BackOffice

• Manual setpoint values — A new BackOffice page lets you create manual setpoint values for write processes.
• Batch delete empty components with clearer deletion status feedback.
• No writes to AI-disabled signals — The platform now refuses to send values to signals where AI control is disabled, preventing accidental overrides.

Smart Control

• AI guardrails — Previously, a group of sensors tied to a non-controlled system could still trigger guardrail crashes, meaning the AI could disrupt parts of the building it had no control over. The optimizer now only evaluates AI-managed components against guardrails, removing this class of crash.
• Heating with dew point — Dew point can now influence comfort bounds when relevant, improving setpoint quality in humid conditions. Important before summer and cooling season.
• Max-change heating anchored to latest write — Even when feedback lags slightly, reducing oscillation.
• Stability and accuracy pass across the optimizer: min/max validation now only applies to active AI-managed signals; duplicated signal names handled cleanly; weather-only rows no longer trigger spurious drop-row checks; verifies the data-storage connection before sending values.

myLoadshift

Load shifting is now a working, production-grade solution running across our VAV fleet.

• 150+ VAV buildings load-shifting against electricity price — All VAV buildings in optimization now use ENTSO-E spot prices (delivered via the Vanta Energia signal) to control heating, shifting load away from high-price hours.
• Load shift prices persisted to optimization tables — Prices are stored alongside optimization data, making downstream analysis and dashboards reliable.

Load-shifting moves HVAC energy use in time so the building consumes more during cheap hours and less during expensive ones. The optimizer does this by multiplying its energy penalties by a time-varying price curve, so cheap hours look "cheaper to use" and peak hours look "more expensive to avoid" — comfort bounds are unchanged.

The resulting policy and the price curve driving it are visualized here: the top panel shows group climate against the comfort bounds, and the Control Variable panel below overlays the optimized policy with the normalized load-shift price. The tooltip reports both the raw market price and the normalized multiplier applied at that timestep.

594KB

Myrspoven - Loadshift.pdf

PDF

Open

Data Platform

• Trino to Spark — Maintenance pipeline and AI vs Baseline now runs on Spark, with tuning for spot-instance survival on heavy tables.
• Overview aggregator — A new daily aggregator producing building-level analysis data, paving the way for richer Insights views.
• Optimization, metric, and meter tables partitioned by building — Significant query performance improvement for data platform and Insights.
• AI vs Baseline reimplemented on Spark, with optimization, metric, and meter tables now partitioned by building for significantly faster queries.

These changes aren't directly user-visible but enable richer capabilities and more reliable operation going forward.