Enabling Artificial Intelligent Virtual Sensors in an IoT Environment.

Area of Science:

• Computer Science
• Artificial Intelligence
• Internet of Things (IoT)

Background:

• Increasing demand for sensors in IoT and smart cities escalates infrastructure costs and maintenance.
• High sensor density leads to performance degradation in device communication, monitoring, and orchestration.
• Physical sensors require significant installation and upkeep, posing scalability challenges.

Purpose of the Study:

• To propose virtual sensors, powered by machine learning (ML) models, as a cost-effective and efficient alternative to physical sensors.
• To evaluate the accuracy and applicability of various ML models for sensor replacement in IoT environments.
• To develop and test a lightweight IoT platform capable of running both physical and virtual sensors.

Main Methods:

• Comparison of fourteen distinct machine learning models to identify optimal candidates for virtual sensor implementation.
• Design and deployment of virtual sensors (ML models) and physical sensors within a smart home environment.
• Development of a custom, lightweight IoT platform on a Raspberry Pi to host and manage sensor operations.

Main Results:

• Extensive simulations demonstrated the feasibility and accuracy of selected ML models as virtual sensors.
• The custom IoT platform successfully integrated and executed both physical and virtual sensor functionalities.
• Evaluation in a smart home scenario confirmed the promising performance of the virtual sensor methodology.

Conclusions:

• Virtual sensors offer a viable solution to mitigate the challenges associated with the proliferation of physical sensors in IoT.
• Machine learning models can effectively replace physical sensors, reducing costs and enhancing system performance.
• The proposed lightweight IoT platform provides a practical foundation for deploying virtual sensor technology in smart environments.