| Seamless Machine Learning Models to Detect Faulty Solar Panels |
| Paper ID : 1057-ICEEM2023 (R1) |
| Authors: |
|
Muhammad Elgamal *1, Ahmed Abdelmaksooud2, Yehea Ismail3 1Center of Nanoelectronics and Devices, the American University in Cairo 2School of Science and Engineering The American University in Cairo 3Center of Nanoelectronics and Devices, The American University in Cairo |
| Abstract: |
| Photovoltaic energy is a renewable source that provides good opportunities for investments by manufacturing and deployment companies. Several management paradigms exist to monitor solar stations, most importantly industry 4.0 technology, which incorporates the internet of things (IoT) and machine learning (ML), among others, to provide commercial systems for fault detection and localization. We compare various ML models based on their capability of classifying faulty and functional solar panels from their measured current-voltage (IV) curves. Regularly, one needs to compare the measurements taken in the field to the values projected from the datasheets of the solar panel. This method is sensitive to the accuracy of measuring devices (i.e., characterizers) attached to each panel in our photovoltaic 4.0 monitoring technology. However, we depend solely on the historical measurements of each panel when it is faulty or functional (possibly by imitating operational faults), and we do not care about the quality of characterizers (an issue raised when we depend on datasheet performance), and then we can successfully detect operation faults. By depending on many features extracted from IV curves, we can use several ML algorithms. However, we can depend on lesser features and an even smaller training dataset. Among our tried models, we found random forest to be the most successful ML model for classification. Additionally, it is insensitive to the reduction of the training dataset and suitable for data augmentation techniques like SMOTE (Synthetic Minority Oversampling Technique). |
| Keywords: |
| Internet of Things; Industry 4.0; Random Forest; K-Nearest Neighbors; Decision Trees, Support Vector Machine, Single-Perceptron Classifier. |
| Status : Paper Accepted |
