Automatic Cleaning of Time Series Data in Rural Internet of Things Ecosystems That Use Nomadic Gateways

Area of Science:

• Computer Science
• Electrical Engineering
• Agricultural Technology

Background:

• Rural Internet of Things (IoT) deployments are hindered by limited telecommunications infrastructure, impacting continuous data collection.
• Unmanned Aerial Vehicle (UAV) based nomadic computing offers a solution but faces challenges with intermittent data collection due to weather and limited transmission windows.
• Ground sensors in rural IoT systems have constraints including limited energy, low-power microcontrollers (MCUs), and short-range radio access technology (RAT) modules, complicating on-device data cleaning.

Purpose of the Study:

• To propose a comprehensive approach for implementing a computationally inexpensive anomaly detector for time-series data on constrained rural IoT end devices.
• To enable efficient on-device data cleaning, reducing data volume and ensuring transmission within limited windows.
• To address the challenges of data collection in rural IoT systems with intermittent connectivity and resource-limited sensors.

Main Methods:

• Developed a low-computational demand anomaly detection algorithm for time-series data.
• Utilized the physics of measured signals for anomaly detection, based on a simple anomaly model.
• Optimized the anomaly model parameters using popular artificial intelligence (AI) techniques.

Main Results:

• Successfully implemented an anomaly detector suitable for resource-constrained IoT devices.
• The proposed method effectively cleans sensor data, minimizing volume and improving transmission efficiency.
• Validated the solution over a 10-month vegetation period in a real-world Rural IoT system.

Conclusions:

• The developed anomaly detection approach significantly enhances the feasibility of rural IoT deployments by addressing data cleaning challenges on edge devices.
• The method's low computational requirements make it suitable for energy-constrained sensors with limited processing capabilities.
• This work provides a practical solution for reliable data acquisition in challenging rural environments, paving the way for more robust IoT applications.