Multi-objective evolutionary algorithms applied to non-intrusive load monitoring

Abstract The task of load disaggregation is inherently an optimization problem. Owing to the existence of noise level and electrical interference from neighboring systems, the real operating state of appliances is not the optimal solution for a single-objective function. However, most recent works weigh objective functions into a single one to construct an aggregate objective function to solve, and the weighted parameters for the different objective functions are sensitive to different datasets and are difficult to tune. Only using load data of appliances running individually to model, proposed method can identify several appliances with multiple operating modes operating simultaneously. A multi-objective load disaggregation model integrates more features including macroscopic features and microscopic features which help model to describe appliances from multiple perspectives. Five objective functions using active power, apparent power, reactive power, current waveform, and harmonics as load signatures are established to identify several electrical appliances. Proposed framework using multi-objective evolutionary algorithms for load disaggregation not only avoid adjusting weighted parameters, but also consider conflict among objectives. A problem-specific method during initialization is presented to deal with the problem that one type of appliance only works on one of these operating modes for a moment. To deal with the constraint on the number of appliances operating simultaneously, objective-rank assignment is applied. The load disaggregation is finally solved as a multi-objective problem by multi-objective evolutionary algorithms. Experimental results demonstrate the effectiveness of the proposed method for load disaggregation. The use of multi-feature methods significantly outperforms the methods using any single or two load signatures.