Research on Training-less Metaheuristic-based Energy Disaggregation Considering Mined Load Operational Constraints

A smart grid is one of the promising paradigms of artificial intelligence of things (AIoT), which enables bidirectional communication between the utility side and the consumer side to achieve demand-side management (DSM) using demand response programs.To achieve effective DSM, energy disaggregation (i.e., nonintrusive load monitoring), a cost-effective virtual sensing technique in appliance-level load monitoring, can disaggregate circuit-level/ building-level power consumption into appliance-level power consumption in a practical field of interest.In this paper, we present a training-less metaheuristic-based residential energy disaggregation approach for discerning appliance-level power consumption from power consumption via a combinatorial optimization search based on a genetic algorithm, whereby the algorithm is developed in consideration of time-series load profiling as a penalty term declared and used to incorporate appliance constraints by penalizing constraint violations.The algorithm executed with the penalty term during the evolutionary process for the energy disaggregation purpose can speed up and target accurately its evolutionary convergence.The presented genetic algorithm-based energy disaggregation approach is practically evaluated using a publicly available reference dataset-the electrical-end-use dataset.