Non-intrusive load monitoring (NILM) is a technique used for effective and cost-efficient electricity consumption management. This thesis presents two different NILM methods. One of them employs a hybrid model of convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) for low-frequency data disaggregation, and the oher one utilizes a graph neural network (GNN) along with a long short-term memory (LSTM) network for load prediction, both combined with attention mechanism. The first study is adept at extracting temporal and spatial features from low-frequency power data, enhanced by an attention mechanism for event detection and load disaggregation. We conduct simulations using the publicly available low-frequency REDD dataset to assess our model’s performance. The proposed approach exhibits superior accuracy and computational efficiency compared to existing methods. The second study explores NILM load prediction, integrating a GNN to represent complex time correlations between appliances, forming a graph-based foundation for feature extraction. The outcome is coupled with LSTM for temporal pattern capturing and attention processes for focusing on key information. The results confirm the effectiveness of this approach in predicting load and uncovering hidden power consumption patterns. Both studies contribute significantly to the field of NILM, offering advanced methodologies for energy management in smart homes.