Greedy Deep Disaggregating Sparse Coding

This is the first study combining deep learning with sparse coding (dictionary learning) to solve the energy disaggregation problem. In the first layer, we learn a disaggregating dictionary. In subsequent layers, we learn sparsifying dictionaries. The first layer is learnt from the smartmeter data; subsequent layers are learnt from features obtained from the previous layer. For disaggregation, instead of using a single level of dictionary as the basis for each device, we use the multiple level dictionaries. Experimental results show that by going deeper, it is possible to improve disaggregation accuracy. Keywords— Energy Disaggregation, Dictionary Learning, Deep Learning,