[2512.22287] Cluster Aggregated GAN (CAG): A Cluster-Primarily based Hybrid Mannequin for Equipment Sample Era

View a PDF of the paper titled Cluster Aggregated GAN (CAG): A Cluster-Primarily based Hybrid Mannequin for Equipment Sample Era, by Zikun Guo and a pair of different authors

View PDF
HTML (experimental)

Summary:Artificial equipment information are important for growing non-intrusive load monitoring algorithms and enabling privateness preserving vitality analysis, but the shortage of labeled datasets stays a major barrier. Current GAN-based strategies have demonstrated the feasibility of synthesizing load patterns, however most current approaches deal with all units uniformly inside a single mannequin, neglecting the behavioral variations between intermittent and steady home equipment and leading to unstable coaching and restricted output constancy. To deal with these limitations, we suggest the Cluster Aggregated GAN framework, a hybrid generative method that routes every equipment to a specialised department primarily based on its behavioral traits. For intermittent home equipment, a clustering module teams related activation patterns and allocates devoted mills for every cluster, guaranteeing that each widespread and uncommon operational modes obtain sufficient modeling capability. Steady home equipment observe a separate department that employs an LSTM-based generator to seize gradual temporal evolution whereas sustaining coaching stability by means of sequence compression. In depth experiments on the UVIC good plug dataset show that the proposed framework persistently outperforms baseline strategies throughout metrics measuring realism, variety, and coaching stability, and that integrating clustering as an lively generative part considerably improves each interpretability and scalability. These findings set up the proposed framework as an efficient method for artificial load technology in non-intrusive load monitoring analysis.