A representational learning assisted matrix factorization approach for electrical load disaggregation

Electrical Load Disaggregation or LD plays a pivotal role in energy management as it enables consumer-segmentation and therefore aids targeted demand response. In this work, the authors propose a novel generic two-stage Restricted Boltzmann Machine (RBM) assisted Matrix Factorization framework for the purpose of Load Disaggregation of low sampled smart meter data. The proposed framework uses representational learning due to RBMs in the first stage and a Least Square Solver in the second stage to obtain the disaggregated estimates. We observe that the use of appropriate mapping function is critical to the accurate appliance load reconstruction in such two-stage learning frameworks. Experimental evaluation of the proposed framework has been carried out on open data sets to demonstrate reconstruction efficiency.