MOGA designed neural networks for non-intrusive load monitoring

The deployment of smart meters in residential buildings allowing for fine-grained monitoring of electricity consumption has led to new opportunities for managing and controlling energy consumption. One of the fundamental steps for increasing efficiency and saving energy in residential buildings is to have a comprehensive overview of how energy is actually consumed in the buildings. Non-intrusive load monitoring (NILM) enables a breakdown of the total energy consumption measured at a single point into specific appliance contributions. This paper proposes a NILM approach based on radial basis function neural networks designed by a multi-objective genetic algorithm (RBFNN-MOGA). The design data are selected using an approximate convex hull algorithm based on a low frequency sampling rate, thus enabling the use of low-cost smart meters. The results of the test conducted showed that the proposed framework achieves very good performance in terms of F1 score, ranging from 68% to 100%, and estimation accuracies between 0.73 and 0.99.