Non-Invasive Load Recognition Method Based on Single-Channel Multi-Feature Fusion Color Coding

To resolve the issue of low recognition accuracy for load devices with similar V-I trajectories in non-intrusive load monitoring (NILM) systems, this paper suggests a single-channel multi-feature fusion-based method for generating color images of voltage-current (V-I) trajectories that incorporates the V-I trajectory features, power features, and harmonic features of the load. This effectively compensates for the absence of current harmonic features. On this basis, the AlexNet model and transfer learning technology are employed to recognize loads, and the algorithm's convergence rate is accelerated by utilizing its pre-trained parameters. Experiments on the PLAID and WHITED public dataset show that the proposed method is 97.9% and 99.1% accurate at recognizing objects on that dataset. This proves that the method works.