Rikard Strid, ASHRAE Member

Skolfastigheter i Stockholm AB (SISAB), an organization responsible for the management of 600 public schools and preschools in Stockholm, Sweden, recognized the necessity of open systems to mitigate vendor lock-in. This early strategic decision proved instrumental in establishing the structured data environment required for the successful deployment of autonomous artificial intelligence (AI) agents within physical building assets.

SISAB manages a portfolio of 600 educational properties in Stockholm, encompassing a total area of approximately 1.8 million square meters (m²). These facilities accommodate a daily population of approximately 200,000 individuals.

Prior to this initiative, the organization faced the industry-wide challenge of managing a disparate system park comprising various proprietary control systems. This heterogeneity of technology hindered comprehensive portfolio-level optimization. A strategic decision was made to mandate that all new and existing system communication must adhere to the open standard of BACnet (ASHRAE Standard 135), moving away from siloed, proprietary solutions.Strategic Implementation: Establishing a Common Protocol

The core objective was to secure long-term control over the building systems. The implementation was guided by three principles:

1. Data Ownership: Data generated by the building systems is to be owned by the facility management organization, not the vendor.
2. System Independence: The ability to replace individual controllers or sensors without necessitating a complete system replacement.
3. Standardization: The establishment of a unified naming convention and network topology.

Standardization on BACnet/IP resulted in a uniform digital infrastructure where all data points (e.g., temperature, fan speed, damper position) are accessed consistently, irrespective of the hardware manufacturer. This structured, consistent data layer is considered critical for the system’s scalability and future development.Transition from Reactive Monitoring to Proactive Management: The Building Management System (BMS)

The initial operational phase involved the deployment of a centralized Building Management System (BMS) platform. This system is fully BACnet-compliant and serves as the central control and visualization interface for the technical building systems. By aggregating millions of data points in real-time, the BMS enabled facility engineers to transition from reactive fault-finding to remote, proactive system diagnostics and oversight.Foundational Layer for Autonomous AI Agents

The subsequent development phase focused on enabling machine-based action on the standardized data set. The architecture utilized the structured data from the BMS to build an AI platform. This platform, developed in partnership with external entities, consumes the consistent data from the BMS for advanced analysis, optimization, and intelligent control.

Scalable implementation of AI in building systems necessitates structured data. An AI model requires unambiguous context to optimize a system; for example, it must distinguish whether a data point represents a classroom temperature or a pump pressure. The consistent data layer provided by the BACnet-compliant infrastructure is the prerequisite for the AI platform’s operational effectiveness.

Applications and Future Development of the AI Platform

Current applications of the AI platform include:

• Real-time Optimization: The system utilizes internal temperature data (from the BMS) and external weather forecasts to proactively adjust heating curves. This functionality minimizes peak power consumption while maintaining occupant comfort.
• Predictive Maintenance: Patterns in actuator movements and other system parameters are analyzed to provide early warnings of potential component failure.

Future development areas for autonomous AI agents include:

• Self-Correction: Agents will be developed to cross-correlate anomalous sensor readings (e.g., carbon dioxide) with other data points (e.g., occupancy sensors) to identify and compensate for sensor malfunctions by simulating a “virtual” value until physical replacement is possible.
• Grid Interaction: Buildings are envisioned to operate as autonomous agents in the energy market, buying and selling flexibility based on real-time data from the control systems.

Conclusion

The SISAB case study demonstrates that the successful path toward advanced AI implementation in buildings is contingent not upon algorithm complexity, but upon the establishment of robust, open protocols. By standardizing on BACnet and instituting a clear data structure via the BMS and a subsequent AI layer, the properties have transitioned from passive enclosures to active, data-driven platforms. This approach confirms the fundamental requirement for structured data to enable future autonomous building operations.—–System Metrics

• Total Facilities: 600
• Connected Air Handling Units: 3,500
• Connected Heating Systems: 1,500
• Connected Occupancy Temperature and CO2 Sensors: 25,000