The combination of digital twin technology and edge artificial intelligence has proven effective in reducing the operating costs of smart buildings. Edge AI plays a crucial role in processing data locally to optimize efficiency.
In commercial real estate and campus environments, “standby power consumption” (energy consumed by devices in standby or idle mode) can account for up to 32% of a building’s total energy consumption. Previous research has shown that up to one-third of office building electricity consumption originates from this standby power consumption.
For business leaders, the first step is often to audit these “always-on” devices to identify opportunities for immediate cost savings. While many businesses have adopted advanced metering, precise control at the outlet level remains challenging. This is due to the high coordination costs of managing distributed devices.
Engineers at the James Watt School of Engineering at the University of Glasgow have developed a prototype digital tool. The research team designed the tool to address this energy waste without compromising business continuity.
The Impact of Standby Power Consumption
Many organizations often consider standby power consumption negligible, but it actually has a significant impact on a company’s profit and loss statement.Research shows that various plugged-in devices, from monitors and workstations to servers, account for a large portion of building energy consumption. In student dormitories alone, standby power consumption can account for up to 33% of total electricity usage.
Dr. Ahmed Taha, a lecturer in Autonomous Systems and Connectivity at the James Watt School of Engineering who led the study, stated, “I firmly believe that small, collective actions on climate change can have a huge impact, and standby power consumption is clearly an ideal target for such actions.”
The challenge often lies in distinguishing whether a device is wasting energy while idle or in a necessary low-power state for a quick restart. Traditional binary control systems (timer-based on/off switching) often fail due to a lack of contextual information. This frustrates users and ultimately forces them to manually shut down the system.
The adoption of these control systems would increase if logic could consider user habits and return probabilities. Rather than simply relying on schedule-based on/off switching. Edge-Enabled Digital Twin (EEDT) systems for smart buildings address this problem. They create virtual representations of physical assets on local edge servers, upon which artificial intelligence can provide further insights and enable automation.
EEDT reduces the privacy risks associated with monitoring individual usage patterns by processing data locally rather than in the cloud. It ensures the low latency required for real-time control. Prioritizing local edge processing is crucial for addressing employee privacy concerns and unlocking the potential of artificial intelligence.
The core difference in this approach lies in abandoning rule-based automation in favor of “fuzzy logic” (a computational method based on truth values rather than traditional Boolean logic). The system acquires data from a network of smart energy sensors that transmit power information to a central server. The system uses the LoRaWAN protocol, which is widely adopted in IoT systems.
The prototype system employs a decision-making framework based on 27 optimization rules. Instead of simply cutting off power after a set time, the system calculates three specific metrics:
User Habit Score: Analyzes usage frequency and stability to understand user behavior.
Device Activity Score: Assesses whether a device is currently inactive by considering both standby time and the time since its last activity.
Confidence Score: Measures data reliability to ensure the system does not act based on incomplete information.
These inputs enable the digital twin to make flexible decisions about smart building assets: immediately shut down, delay the decision, notify users, or maintain the current state. When the system detects prolonged inactivity, it displays a prompt on the user’s screen. The prompt asks if they are working remotely or running background processes.
This approach aims to raise user awareness of their device’s idle time, encouraging more careful device use. While also preventing disruptions to normal workflows.
Results and Return on Investment
To validate the architecture, researchers deployed the system in a university research lab. They utilized smart plugs and environmental sensors to communicate via LoRaWAN.
The results provide a strong commercial justification for intelligent edge AI management based on digital twins. Deployment results showed a weekly power consumption reduction of approximately 40.14% per monitored workstation. For standby loads, the fuzzy decision framework achieved a power consumption reduction of up to 82%.
The economic benefits are evident when scaled up to broader smart building deployments. Under the UK electricity price cap taking effect in July 2025, deploying the system on 500 devices can save over £9,000 annually.
In addition to the direct energy savings, Dr. Taha highlighted a secondary economic benefit in terms of asset lifecycle management. “Secondly, by reducing device power consumption, organizations can reduce the need to replace older equipment with newer, more energy-efficient alternatives.”
This, in turn, can help businesses save on equipment costs in an increasingly challenging economic environment.” The technical implementation of such systems typically relies on a containerized edge architecture. The research team utilized Docker containers to host an MQTT message broker, Node-RED data parsing, and InfluxDB time-series storage. This technology stack enables “closed-loop” control, allowing the digital twin to not only monitor the physical world. It can also proactively intervene.
A necessary component for user acceptance is an “anti-oscillation filter.” In early automated systems, rapid switching (lag) between on/off states often led to hardware wear and user dissatisfaction. The EEDT system integrates cooling management and stability checks to ensure the stability and contextual rationality of equipment shutdown decisions.
The system also integrates a prediction module using Long Short-Term Memory (LSTM) deep learning. This model, trained with only two days of historical data, can predict energy consumption trends for the following day. Integrating these short-term prediction models allows facility teams to anticipate peak loads rather than react reactively.
Edge AI Digital Twins as Drivers: Making Buildings Truly Smart?
The next essential step in the development of smart buildings is transitioning from passive energy monitoring to edge AI-driven optimization. Digital twin technology enables this transition. While this research focuses on a university environment, organizations can directly apply its architecture to spaces that are not yet effectively managed.Examples include corporate offices, healthcare facilities, and industrial environments.
Dr. Taha added, “Achieving net-zero emissions requires a comprehensive approach to energy monitoring. Digital twin technology can become a crucial component of broader institutional strategies for reducing carbon footprints. The system achieves this by monitoring variables such as occupancy and temperature control. The team is currently exploring how this tool can help universities achieve their broader goal of net-zero emissions by 2030.
Scalable application requires addressing existing infrastructure issues. Relying on manually designed fuzzy rules (27 in this case) may limit its rapid scalability across different asset types. Future enterprise-level solutions may need to incorporate neurofuzzy learning techniques to automatically generate rules. These rules are based on the behavior of specific sectors.
The data needed to reduce energy costs resides in the network. The challenge today is no longer collecting this data and visualizing it using digital twins. Instead, it is about leveraging artificial intelligence to empower edge assets, enabling them to intelligently process this data. This truly achieves building intelligence.