LSTM-Based Path Prediction for Effective Sensor Filtering in Sensor Registry System.

Area of Science:

• Computer Science
• Ubiquitous Computing
• Wireless Sensor Networks

Background:

• The Internet of Things (IoT) relies on ambient sensors for intelligent services.
• Mobile devices use Sensor Registry Systems (SRS) for sensor metadata and direct connections.
• Location-based sensor filtering in SRS is prone to GPS inaccuracies, causing connection failures.

Purpose of the Study:

• To improve the success rate of connections between mobile devices and ambient sensors in IoT environments.
• To address the challenge of GPS inaccuracies affecting location-based sensor filtering.
• To introduce a novel dual collaboration strategy for enhanced sensor filtering.

Main Methods:

• Proposed a dual collaboration strategy combining GPS readings with historical trajectory predictions.
• Implemented Long Short-Term Memory (LSTM)-based path prediction for trajectory analysis.
• Updated the SRS evaluation approach using a Monte Carlo-based simulation flow to measure service provision rate.

Main Results:

• The LSTM-based path prediction effectively compensates for GPS location abnormalities.
• The proposed dual collaboration strategy significantly improves the probability of successful mobile device-to-sensor requests.
• Empirical studies validated the effectiveness of the LSTM-based model as a sensor filtering solution.

Conclusions:

• The dual collaboration strategy enhances the reliability of sensor filtering in IoT systems.
• LSTM-based path prediction is a robust method for overcoming GPS errors in mobile sensing.
• This approach leads to a higher service provision rate in location-aware IoT applications.