GAF-TCN NILM: A Novel Approach to Non-Intrusive Load Monitoring Using Image Analysis with Gramian Angular Field and Temporal Convolutional Networks

Accurately disaggregating household energy consumption into individual appliance usage is a vital aspect of Non-intrusive Load Monitoring (NILM), crucial for optimizing energy efficiency and reducing overall consumption. This paper introduces a novel approach that integrates image analysis techniques using Gramian Angular Field (GAF) and Temporal Convolutional Networks (TCN) for energy disaggregation. By transforming time-series energy data into images, this method captures complex patterns and temporal dependencies associated with appliance-level energy use, often missed by traditional approaches. Our approach is tested on three common household appliances of two houses across two widely recognized datasets, REDD and UK-DALE. The performance of the GAF and TCN model is compared against two traditional neural network models that use raw time-series data. Key metrics, including Mean Absolute Error and R-squared score, are used to assess the accuracy and reliability of each model. Results show that the image analysis-based approach significantly enhances energy disaggregation precision, consistently outperforming traditional models across both datasets. The transformation of time-series data into images enables TCN to better capture underlying patterns, leading to more accurate predictions and a deeper understanding of appliance usage. This research underscores the potential of combining image analysis with advanced neural networks to improve non-intrusive load monitoring. The superior performance of the proposed method highlights its value as a powerful tool for more effective energy monitoring, offering new insights for developing smarter, more sustainable energy systems.