On Electrical Load Disaggregation using Recurrent Neural Networks

In a residential setting, Load disaggregation (LD) is about obtaining appliance-specific operational details in terms of time and power consumption by processing aggregate power consumption data. The disaggregated load information helps utilities to categorize customers based on their usage patterns, facilitating optimal demand response design. Further, LD helps customers to know about their energy-consuming behavior, which is beneficial in reducing the consumption. To be able to provide appliance-specific consumption patterns for aforementioned goals, apart from accurate load identification, estimates of energy consumption of appliances of interest are necessary. In short, it is essential to cull out operational waveform of each of the requisite appliance. Towards this end, very few results have been reported in the literature related to estimating the operational wave-forms, even for large power consuming appliances. In this work, we address this problem using a deep-learning architecture with Recurrent Neural Networks (RNN) variants like Long-Short Term Memory networks (LSTM) and Generalized Recurrent Unit networks (GRU). In addition, a simple but effective technique in pre-processing of the aggregated data is proposed and implemented to identify and reconstruct the consumption pattern of low-power consuming appliances like Refrigerator.