Title: Energy Disaggregation with Semi-supervised Sparse Coding

Abstract: Residential smart meters have been widely installed in urban houses nationwide to provide efficient and responsive monitoring and billing for consumers. Studies have shown that providing customers with device-level usage information can lead consumers to economize significant amounts of energy, while modern smart meters can only provide informative whole-home data with low resolution. Thus, energy disaggregation research which aims to decompose the aggregated energy consumption data into its component appliances has attracted broad attention. In this paper, a discriminative disaggregation model based on sparse coding has been evaluated on large-scale household power usage dataset for energy conservation. We utilize a structured prediction model for providing discriminative sparse coding training, accordingly, maximizing the energy disaggregation performance. Designing such large scale disaggregation task is investigated analytically, and examined in the real-world smart meter dataset compared with benchmark models.