Abstract:
Non-Intrusive Load Monitoring (NILM), also called, load disaggregation aims to infer load level electrical energy consumption from the aggregate power signal. Several challenges are limiting the deployment of NILM systems in residential and commercial buildings. In this work, we treat the case of energy disaggregation in the presence of non-targeted loads. Non-targeted loads in our context stand for electrical
loads for which we do not have labels during the training phase of the NILM algorithm. However, those loads may exist in a real-world scenario, and their power consumption adds to the aggregate power signal.
In this work, we present our load disaggregation method based on a multi-label classification approach and study the impact of non-targeted loads on the NILM disaggregation performance. We show that the nontargeted loads can negatively affect the disaggregation performance of NILM systems. We also found a significant correlation between the disaggregation performance impact and the overlapping coefficient between the targeted and non-targeted loads’ power distributions. Results are obtained using a publicly available dataset of power measurements
