Browsing by Author "Mbabali, Frank"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Machine learning-based monitoring and performance analysis of a laboratory solar photovoltaic system using an internet of things framework.
    (Busitema University, 2026) Mbabali, Frank
    Solar photovoltaic (PV) systems are increasingly important for electricity generation in Uganda, but their performance varies with environmental and operational conditions, while existing monitoring provides limited integration of real-time sensing and advanced data analysis. This experimental study aimed to design and implement an Internet of Things (IoT)enabled monitoring framework integrated with machine learning for real-time monitoring and performance analysis of a laboratory PV system. Specifically, the study sought to: (i) design and implement an IoT system for real-time acquisition, transmission and monitoring of environmental and electrical parameters; (ii) develop and evaluate exploratory XGBoost and LSTM tasks for PV power prediction and time-series forecasting; and (iii) evaluate relationships between environmental conditions and directly measured electrical performance. The experimental platform comprised a 40 W PV module, ESP32 microcontroller, seven environmental and electrical sensors, Wi-Fi/MQTT communication and HiveMQ cloud storage. Data were acquired at approximately 5-second intervals across five monitoring sessions between May and August 2026, producing 1,729 valid observations, of which 1,001 were used for analysis. XGBoost was evaluated using a chronological cross-session holdout and linear-regression baseline, while LSTM was retained as an exploratory time-series prototype; graphical and statistical analyses were conducted. The IoT framework successfully acquired, transmitted, stored and visualized monitoring data. However, XGBoost did not generalize to an unseen session (R² = −0.571), while the linear baseline also performed poorly (R² = −0.109), indicating insufficient data diversity for deployable prediction. The irradiance channel was not calibrated, and the battery-buffered topology prevented validation of conventional instantaneous efficiency and performance-ratio measures. The study concludes that the IoT architecture provides a functional foundation for laboratory PV monitoring, but the machine-learning and performance indicators remain exploratory. Future work should recalibrate irradiance sensing, extend monitoring to multi-week or multi-season periods, log battery state of charge, and validate fault detection using labelled fault events before largerscale deployment.