Pre-classification-based multi-model fusion non-intrusive load identification

To tackle the challenges posed by the limited volume of current electric load datasets and the inability of non-intrusive load monitoring algorithms to effectively balance accuracy and efficiency, this paper proposes a deep learning-based pre-classification multi-model fusion approach for non-intrusive load identification. Initially, a dynamic harmonic admittance model is employed to generate diverse, complex, and more realistic electric load data, enhancing the representativeness of the dataset. This generated data is subsequently fed into a correction network for calibration, ensuring that the data aligns closely with real-world scenarios. Building on this foundation, a pre-classification load identification fusion model is constructed, utilizing decision trees to effectively categorize electric loads. The proposed pre-classification multi-model fusion non-intrusive load identification algorithm integrates convolutional neural networks (CNN) and Transformer architectures, achieving a balanced trade-off between identification accuracy and operational efficiency. The overall performance of the algorithm was rigorously evaluated using a real dataset that was supplemented with model-generated data. Experimental results demonstrate that the proposed method significantly enhances identification accuracy and outperforms existing algorithms in the field. This research not only contributes a novel approach to load identification but also paves the way for more effective non-intrusive monitoring solutions in various applications.