Identification of Electrical Devices Using Machine Learning: A Study in Collaboration with Origin AS

This thesis aimed to develop reliable methods for identifying electrical devices in a household based on aggregate electrical signals from a main electric monitor. The study involved creating or obtaining a suitable dataset, implementing the NILM procedure, utilizing machine learning methods to determine their effectiveness in learning device usage patterns, and creating a practical application using these models. Random Forest, RNN/LSTM, and CNN models were employed, with the Random Forest model achieving the highest overall accuracy. The RNN/LSTM model showed strong potential in capturing temporal dependencies. Conversely, the CNN model did not perform well, indicating that further research is needed to optimize its application to NILM tasks. The study highlighted the importance of using comprehensive datasets that include multiple devices simultaneously, allowing the models to learn interactions and overlapping usage patterns. While the development of a practical application was initiated, it was not fully realized. Despite some limitations, such as the lack of adaptability testing across different households, the findings provide a foundation for future studies and practical applications in household energy management.