Aggregately Regularized Multi-task Matrix Factorization for Household Energy Breakdown

Abstract

Household energy breakdown aims to disaggregate the monthly energy consumption into appliance level usage. It is an important but challenging issue due to the cost of hardware deployments. Existing approaches shed light on decomposing the energy in a non-intrusive way and utilizing matrix factorization. However, traditional matrix factorization methods overlook the relations among appliances and aggregations. In this paper, we propose an novel aggregately regularized Multi-task model, Non-negative Matrix Factorization (MultiNMF), to address this issue. By combining the per-appliance tasks with regularizations, MultiNMF can simultaneously infer the appliance level energy usage for users. The model is evaluated on both synthetic and real world datasets with different settings, and the experimental results demonstrate the effectiveness of our approach.