Despite extensive research on non-intrusive load monitoring(NILM), existing models face challenges in accurately predicting multiple operational states of appliances, leading to significant decreases in prediction accuracy. To address this issue, this paper proposes non-intrusive load monitoring model based on multi-algorithm fusion. Firstly, the REDD low-frequency dataset is preprocessed using time-based interpolation and oversampling. Secondly, the model employs graph convolutional networks(GCN) and convolutional neural networks(CNN) to extract power features, which are then fed into a self-attention mechanism and long short-term memory(LSTM) networks. This effectively captures the key features of the input signals, thereby improving the prediction accuracy for appliances with multiple operational states. Finally, simulation verification is conducted using the preprocessed REDD low-frequency dataset. The experimental results indicate that the proposed model outperforms comparative models in terms of MAE, SAE, and R^{\mathrm{2}}, demonstrating its effectiveness in load disaggregation.