An Energy-Efficient Hybrid LoRa–Wi-Fi Architecture for Real- Time Water Quality Monitoring and Machine Learning-Based Trend Forecasting
Water quality management in large-scale institutional infrastructures faces significant challenges due to the high latency of manual sampling and the energy–connectivity trade-offs in traditional IoT deployments. This paper proposes HydroSense AI, a robust three-tier IoT framework designed for real-time multi-parameter water quality monitoring and predictive analytics. The system integrates a heterogeneous sensing layer (pH, TDS, turbidity, and temperature) with a hybrid communication architecture, utilising Long Range (LoRa) technology for low-power transmission over long ranges (manufacturer-rated for line-of-sight distances of up to 16 km, and validated up to 2 km within a dense campus environment in this study), bridged via an ESP32-based gateway to the cloud. To address the critical issue of energy autonomy in remote sensing nodes, we implement a hardware-synchronised duty-cycling mechanism using a DS3231 Real-Time Clock (RTC), enabling precise deep-sleep scheduling and significantly extending battery operational life. Beyond data acquisition, the framework incorporates AI-driven trend-forecasting and anomaly-detection models to provide early warnings of water degradation through a Telegram-integrated alert system. Experimental validation over an extended deployment period demonstrates high measurement stability, with the forecasting model achieving a one-step (10-min) normalised RMSE of 0.0063 (equivalent to 0.033 pH units) for pH and 0.0298 (17.0 ppm) for TDS on a held-out test partition; a benchmark against persistence and ARIMA baselines is also provided. A complete measured energy decomposition of the deployed node is reported: hardware-synchronised duty cycling reduces the quiescent current to 18.2 μA, and with a 12 s acquisition window at 112 mA on a 10-min cycle, the mean current is 2.26 mA, corresponding to an estimated 46 days of unattended operation on a 2500 mAh cell. Critically, the acquisition window accounts for 99.2% of the per-cycle energy budget and the sleep interval for only 0.8%, so quiescent current—the figure of merit most often reported as evidence of low-power design—is shown not to be the binding constraint for sensor-dominated nodes of this class. The results indicate that the proposed hybrid architecture offers a 99.8% packet delivery ratio for sustainable water management.