*Important notice: This news reports on an unedited version of an accepted paper and is awaiting final editing. Therefore, the paper should not be regarded as conclusive or treated as established information.
An adaptive lighting-control system has been developed and experimentally validated for intelligent buildings, employing artificial neural networks. The system was deployed on low-cost embedded hardware to optimize energy consumption while maintaining required illuminance levels. Researchers published their findings in Scientific Reports.
Study: Adaptive lighting control in intelligent buildings using artificial neural networks: design, implementation, and experimental validation. Image Credit: zhu difeng/Shutterstock.com
Intelligent Building Lighting Challenges
Lighting systems account for a significant portion of energy consumption in commercial and office buildings, and intelligent control is essential for optimizing energy use while maintaining comfort.
Traditional lighting controls often rely on fixed schedules or simple threshold-based systems, which are insufficient in responding dynamically to variable daylight and occupancy changes. Recent advances in artificial intelligence (AI), particularly artificial neural networks (ANNs), have demonstrated potential for adaptive lighting control through data-driven predictive models.
Download a free PDF of this news story here!
However, many ANN-based solutions remain limited to simulations or laboratory studies and require high-performance computing resources, restricting their deployment in real operational environments. This study addresses these gaps by developing a computationally efficient, ANN-based lighting controller deployable on low-cost embedded platforms, validated through long-term field experiments.
ANN Controller Design and Deployment
The research methodology integrates several critical phases: data acquisition, neural-network training, system integration, and experimental validation in a real intelligent building at the University of Žilina’s Research Center.
Data collection plays a pivotal role in capturing diverse input parameters such as daylight irradiance levels, occupancy patterns, and user illumination preferences, which form the training dataset for the ANN.
The control system architecture employs a feedforward neural network trained offline on historical sensor data to predict optimal artificial lighting levels to meet a minimum work-plane illuminance of 500 lux, as stipulated by EN 12464-1. Data preprocessing filtered noise and accounted for variations in illumination, ensuring the model could generalize effectively under variable conditions.
An embedded control unit (Loxone Miniserver) serves as the deployment platform for the ANN inference engine, enabling real-time decision-making without dependence on external computing resources. The control strategy continuously adjusts dimming levels based on ANN predictions, integrating daylight harvesting and occupancy detection data to optimize energy usage while maintaining visual comfort.
The collection of operational sensor data, including illuminance and occupancy detection, supports adaptive model updates and helps the system respond to nonlinear environmental variations and human factors. The experimental setup involved both an ANN-controlled test room and a reference room operating legacy manual lighting to facilitate comparative analysis.
Energy Savings and Performance
Over a four-month continuous operational period, extensive mined data confirmed that the ANN-based controller effectively optimized lighting energy consumption while maintaining the required illuminance levels.
Energy savings in the range of 7.4% to 12.5% were consistently achieved during fully automated 24/7 operation, with savings of approximately 8.5% under occupancy-driven conditions. This is a significant improvement over conventional three-step DALI-based lighting control systems that rely on predefined switching thresholds.
The ANN model showcased high predictive accuracy (R2 = 0.91) and robustness in managing the dynamic interplay of daylight variability, occupancy fluctuations, and user preferences. The mined-data-driven approach enabled smooth dimming transitions, reducing abrupt illumination changes common in traditional systems and thus supporting system stability.
Furthermore, the mining of sensor feedback enabled the system to minimize the fraction of occupied time when illuminance fell below the required threshold from 27.9% to 12.7%, confirming improved lighting quality and energy efficiency. The system’s decentralized and modular control framework supports scalability with minimal communication overhead, suggesting it is suitable for deployment across multiple zones or buildings.
Challenges addressed include the computational constraints of embedded platforms, accuracy and stability in sensor data collection, and the ability to generalize the model across variable lighting and occupancy conditions without explicit physical daylight modeling. By separating offline training from online inference, the architecture maintains low runtime complexity while leveraging rich operational datasets mined during building use.
Compared to other AI-based lighting control research, often limited to simulations or systems that require powerful external processors, this study stands out through its real-world embedded deployment and field validation.
Practical Implications and Future Directions
This study demonstrates that mining real operational data to train and implement an artificial neural-network-based adaptive lighting controller is a viable and effective approach for intelligent buildings. By deploying the ANN inference on a low-cost embedded platform within a live office environment, the research bridges the gap between theoretical AI lighting controls and practical application in commercial settings.
The system achieved meaningful energy savings of up to 12.5% under continuous operation and maintained minimum illuminance standards, improving visual comfort and reducing unnecessary energy consumption.
In conclusion, data-driven approaches combined with ANN predictive capabilities offer a scalable and economically favorable solution for lighting energy management in modern intelligent buildings.
Future extensions should explore multi-objective optimization, incorporating additional comfort metrics and interactions with HVAC and shading systems to build more holistic, user-centered smart building controls.
Journal Reference
Belany P., Sedivy S., et al. (2026). Adaptive lighting control in intelligent buildings using artificial neural networks: design, implementation, and experimental validation. Scientific Reports. https://www.nature.com/articles/s41598-026-58858-4.