Using Stream Data Processing for Real-Time Occupancy Detection in Smart Buildings

Area of Science:

• Building management systems
• Energy efficiency
• Occupant comfort
• Internet of Things (IoT)
• Machine learning

Background:

• Building management systems (BMS) aim to optimize energy efficiency and occupant comfort but often use static configurations mismatched to user preferences.
• Occupancy detection is crucial for energy management in residential and industrial buildings.
• Traditional data-driven occupancy detection methods use batch learning, unsuitable for dynamic, non-stationary environments.

Purpose of the Study:

• To investigate and evaluate non-stationary machine learning techniques for real-time occupancy detection.
• To predict the number of occupants in smart buildings using stream data processing.
• To assess the accuracy and resource performance of these algorithms.

Main Methods:

• Implementation and deployment of an integrated platform combining IoT technologies with stream machine learning.
• Testing and evaluation of three distinct machine learning algorithms designed for stream data processing.
• Performance metrics included prediction accuracy, CPU time, and RAM utilization.

Main Results:

• The developed approach achieved an occupancy prediction accuracy exceeding 83%.
• The system demonstrated efficient resource utilization, with minimal CPU and RAM usage per hour.
• The effectiveness of integrating IoT with stream machine learning for occupancy prediction was validated.

Conclusions:

• Non-stationary machine learning techniques are effective for real-time occupancy detection in smart buildings.
• The proposed IoT-integrated platform optimizes building energy management by accurately predicting occupant numbers.
• This approach enhances energy efficiency and occupant comfort while minimizing computational resource consumption.