Residential electric load disaggregation for low-frequency utility applications

Recent load disaggregation approaches take advantage of artificial intelligent techniques and require low sampling frequency. From utility perspective, intrusive data for training are not available due to privacy and the sampling frequency may be too low to recognize meaningful signatures. This paper proposes a 1-hour frequency disaggregation algorithm for real and reactive energy without knowing what appliances are in a specific house/apartment. The proposed algorithm is particularly developed concerning utility's constraints. A database of typical types of appliances in BC, Canada is built. Based on BC Hydro's smart meter data over a year, the most probable appliances in a specific house/apartment are firstly inferred through likelihood maximization and energy consumption matching. The disaggregation is then implemented by an integer multi-objective Genetic Algorithm tuned by appliance dependence rules. The results show that despite of high uncertainty, more than 50% of energy consumption could be disaggregated for random houses/apartments.