Digital Twin Model of a Solar Panel for Intelligent Monitoring Using Artificial Neural Network

Abstract

The growing demand for renewable energy has sparked interest in optimizing the performance and reliability of solar panels and solar-powered installations. Reliability, a key quality of any power system, can be improved through early fault detection and predictive maintenance in solar power plants. Detecting faults early in solar panels significantly reduces maintenance costs and minimizes downtime, making efficient and cost-effective monitoring methods essential for enhancing the reliability of solar power systems. In this paper, a digital twin prototype of a 30W solar panel was developed for intelligent monitoring, fault detection, and fault classification using MATLAB/Simulink and Bayesian regularization Artificial Neural Networks. The digital twin consists of several layers: a hardware layer consisting of voltage, current, temperature, humidity, and light sensors interfaced with an Arduino microcontroller; a data synchronization layer that transfers data from the microcontroller to Google Cloud Logging using serial communication, a GET request in a Python script, and a corresponding JavaScript code on Google Cloud Platform to capture and store the data; an analytics layer implemented in the MATLAB Machine Learning Toolbox; and a virtual representation layer in Simulink for real-time visualization and monitoring. This real-world data was collected and pre-processed before training nineteen artificial intelligence models. The dataset was split into 80 % for training and 20 % for the testing sets. The models’ performance was measured using metrics such as Root Mean Square Error, Mean Square Error, and coefficient of determination. The Bayesian regularization model achieved the lowest root mean square error and an overall prediction accuracy of 81 %, outperforming linear regression, support vector machines, and other gaussian process regression models. After the fault detection by the chosen algorithm, these faults were categorized into three based on the deviations between predicted and measured voltage and current profiles. The digital twin demonstrated how Bayesian regularization improves generalization performance on limited PV datasets, while providing a virtual replica capable of real-time monitoring. This research provides a scalable low-cost solution for intelligent monitoring in solar installations, particularly in large-scale solar PV solutions.