Load Forecasting using GNN-LSTM Attention Mechanism with Low-Frequency Data

Non-intrusive load monitoring (NILM) in smart home applications provides insights into household energy usage patterns. This paper presents a NILM methodology using a graph neural network (GNN), a long short-term memory (LSTM) network, and an attention mechanism. It also investigates the temporal correlation between household appliances. The objective of the proposed learning-based approach is to represent the complex dependencies between appliances and their power usage. We capture the interaction between appliances in the form of a graph by integrating a GNN, which serves as the foundation for more effective feature extraction. An LSTM network is then implemented to capture temporal patterns. The attention process, on the other hand, focuses on the most crucial information to further boost the prediction performance. We utilize heat maps to develop a better understanding of how household appliances perform and correlate over time. These visualizations provide insight into usage patterns and power consumption sequences by giving information about the temporal interconnections between appliances. Experimental results demonstrate the effectiveness of the proposed approach in accurate load prediction by uncovering hidden patterns within household appliance data.