Cluster Aggregated GAN (CAG): A Cluster-Based Hybrid Model for Appliance Pattern Generation

The Cluster Aggregated GAN (CAG) framework has been introduced to enhance the generation of synthetic appliance data for non-intrusive load monitoring. By employing a hybrid approach that utilizes specialized branches for intermittent and continuous appliances—where clustering is used for the former and an LSTM-based generator for the latter—CAG improves modeling fidelity and training stability. Experimental results on the UVIC smart plug dataset indicate that CAG outperforms existing GAN methods in realism, diversity, and stability, highlighting its potential for advancing privacy-preserving energy research.