Smart sensors are increasingly used in agriculture to monitor environmental conditions and support data-driven decision-making. However, traditional sensor implementations face critical challenges related to power consumption, especially in remote farms—such as pitaya plantations—where access to electricity and ongoing maintenance is limited. This paper presents a smart energy management system for agricultural sensor nodes integrating a machine learning model for adaptive sampling and a batching strategy to optimize energy usage. A lightweight Stochastic Gradient Descent (SGD) regressor trained on temperature dynamics runs on-device to predict the sampling interval ($T_s$). In parallel, the node adjusts the number of buffered samples as the battery state of charge ($SOC$) decreases, reducing Long Range (LoRa) transmissions. Field experiments show that the proposed approach reduces energy consumption by 77.8% compared with fixed-interval sampling, while maintaining good temperature fidelity with Mean Absolute Error (MAE) of 0.537 °C for temperature reconstruction.

This document presents the design and implementation of a platform that detects the ON/OFF state of any electronic device and show this information through an Internet web site despite the absence of any Internet network. The device state is defined through a current sensor, the information is saved with date in a backup memory, and then, a Short Message Service (SMS) message containing the ON/OFF state is sent to a massive message platform on the Internet. After that, the information is collected in a custom web platform where the user can access through the Internet and see the electrical status of the device. Additionally, the platform works even when there is not AC power supply.