An Improved YOLOv8-XGBoost load rapid identification method based on multi-feature fusion

• A novel load identification method leveraging time-frequency domain features is proposed, enhancing accuracy for similar loads. • An optimized YOLOv8 network reduces parameters, accelerates convergence, and improves time-domain feature recognition efficiency. • The YOLOv8-XGBoost model integrates frequency-domain features, achieving 100% recognition accuracy through time-frequency fusion. • The proposed model demonstrates superior generalization, high accuracy across datasets, and low computational resource demands. Existing non-intrusive load monitoring (NILM) approaches face challenges including limited identification accuracy, computationally intensive architectures, and constrained generalization performance. To address these issues, this paper proposes an Improved YOLOv8-XGBoost rapid load identification method based on multi-feature fusion. First, the temporal features of load current data are extracted using the Markov Transition Field (MTF) algorithm through image encoding methods. Additionally, the Fast Fourier Transform (FFT) is employed to extract the maximum frequency component and average frequency amplitude as corresponding frequency-domain features. Subsequently, the Improved YOLOv8 network is utilized for preliminary recognition of temporal domain images. Finally, to resolve the low discriminability issue of similar loads, the recognition results of the Improved YOLOv8 network are fused with selected frequency-domain features using the eXtreme Gradient Boosting (XGBoost) algorithm for training to achieve the final identification results. Validation on both high-frequency and low-frequency datasets demonstrates that the proposed method exhibits strong generalization capability, achieving an identification accuracy of over 99%.