A Study of Non-Intrusive Load Recognition Based on Improved Unthresholded Recurrence Graph

The Non-Intrusive Load Monitoring (NILM) technology is a pivotal tool for estimating the energy consumption of individual appliances by analyzing aggregate meter data, thereby facilitating energy conservation and enhancing power management efficiency. However, as the diversity and operational modes of electrical loads expand, the precision of load identification methods faces significant challenges. To tackle these challenges, our research introduces an improved unthreshold recurrence graph (UTRG) technique. This approach leverages current-voltage correlation coefficients combined with a sigmoid activation function to refine the UTRG, thereby enhancing the differentiation of primary and secondary features. Additionally, it incorporates a nonlinear element to augment the distinctiveness of load signatures. Our experimental evaluation on the PLAID and LILACD public datasets demonstrates that the proposed method significantly improves load identification accuracy, outperforming the most recent algorithms in the NILM domain.