Adaptive blind control using deep learning methods

Journal of Physics Conference Series, vol. 3140, pp. 112009

Abstract

Abstract Automated shading systems can be disruptive and unsatisfactory for occupants when they fail to meet individual comfort needs and user preferences. Conventional control strategies—such as cut-off angle and sensor-based systems—often fail to capture diverse user preferences. This paper presents a data-driven approach to predict shade position using deep learning, aiming to improve the adaptability and occupant satisfaction of automated shading. A preliminary analysis was conducted on manual blind use patterns collected from 63 offices in an institutional building. K-means clustering was applied to identify the behavioural-based clusters within the offices, resulting in 3 distinct clusters with a Silhouette Score of 0.59. A Multi-Layer Perceptron (MLP) model was trained and tested to predict blind positions using 16 input features—including behavioral, indoor climate, weather, time-related and contextual features. Observations revealed that occupants often adjusted the blind to either fully open or closed positions, most often in the morning. Six groups of difference scenarios were tested to find the optimal MLP model. Model performance improved when behavioral cluster information was included as an additional input feature, achieving the highest F1 score of 0.90 and a balanced accuracy across all classes. The findings highlight the potential of data-driven machine learning control strategies to overcome conventional control limitations. Further studies are needed to implement and evaluate the approach in real-world settings to ensure user comfort and acceptance.

Authors 3

1. Ghadeer Derbas

   Forschungszentrum Jülich · Palestine Technical University - Kadoorie

   Affiliation as printed

   Faculty of Engineering and Technology, Palestine Technical University, Tulkarm, Palestine

   Forschungszentrum Jülich GmbH, Institute of Climate and Energy Research, Energy Systems Engineering (ICE-1), Juelich, Germany

2. André Xhonneux

   Forschungszentrum Jülich

   Affiliation as printed

   Forschungszentrum Jülich GmbH, Institute of Climate and Energy Research, Energy Systems Engineering (ICE-1), Juelich, Germany

3. Dirk Müller Aachen E.ON Energy Research Center Institute for Energy Efficient Buildings and Indoor Climate

   RWTH Aachen University · Forschungszentrum Jülich

   Affiliation as printed

   E.ON Energy Research Center, Institute for Energy Efficient Buildings and Indoor Climate, RWTH Aachen University, Aachen, Germany

   Forschungszentrum Jülich GmbH, Institute of Climate and Energy Research, Energy Systems Engineering (ICE-1), Juelich, Germany