Active and Reactive Power Sequences for Energy Disaggregation with Deep Learning Models

Non-Intrusive Load Monitoring (NILM) aims to decompose the aggregated power signal of a household smart meter into the power consumption of individual appliances. This information about consumption can potentially lead to energy savings by changing consumer behavior or facilitating the development of flexibility at the residential level. Deep learning methods have been shown to be the most successful in solving this problem. Nevertheless, these methods focus, most of the time, only on the measurement of the active power neglecting other available information such as reactive power. This paper evaluates the impact of adding reactive power sequences in such methods. In particular, we evaluate the impact of reactive power sequences in two state-of-the-art methods: one updated sequence-to-point Convolutional Neural Network (CNN) that we designed and one Variational Autoencoder (VAE) from the literature. Our experiments show that the addition of reactive power improves the performance of both algorithms on the UK-DALE dataset.