Machine learning deployment for energy monitoring of Internet of Things nodes in smart agriculture

Abstract

SummaryLow‐Power Wide‐Area Network technologies, such as LoRa, are gaining popularity in the agricultural sector for field deployment. The crucial factors in these devices are their range and power efficiency. The energy consumption of a LoRa wireless sensor network is predominantly affected by transmission parameters like carrier frequency, bandwidth, transmit power, spreading factor, and coding rate. Incorrect chosen transmission parameters can lead to a reduction in the battery life of end nodes, requiring frequent battery replacements—a situation undesirable for field deployment. This study introduces a machine learning deployment in the form of a web application designed to monitor the energy consumption of end nodes in LoRa wireless sensor networks. The research initially employs 12 regression models, including Linear, Random Forest, K‐Nearest Neighbours, Decision Tree, Support Vector, Lasso, Ridge, AdaBoost, Gradient Boost, XGBoost, CatBoost, and LightGBM models. The findings of the study reveal that the LightGBM model surpasses other models in accurately predicting the energy consumption of Internet of Things (IoT) nodes, leading to its selection for the web application. This machine learning web application can be implemented in a programmable Long Range Wide Area Network (LoRaWAN) gateway to effectively monitor the energy consumption of IoT end nodes in the agricultural sector.