Bayesian Energy Disaggregation At Substations With Uncertainty Modeling
Ming Yi, Meng Wang
Abstract
This article considers energy disaggregation at substations (EDS) where the objective is to estimate the consumption of each load from aggregate measurements, in which whether or not some loads are consuming power is unknown to the operator. The existing EDS method cannot provide any reliability measure of the disaggregation results, while the disaggregation accuracy can vary significantly for different data due to the volatility of loads such as the solar generation. This article proposes a Bayesian-dictionary-learning-based approach to disaggregate the loads and provides an uncertainty measure of the returned estimation. Our approach learns the probability distributions of the load patterns and the decomposition coefficients from recorded data with partial labels at the offline stage. In real-time disaggregation of the obtained aggregate data, our approach computes the mean and covariance of the probability distribution of each load consumption, estimates the load using the mean, and computes the uncertainty index based on the covariance. Numerical experiments indicate that our method achieves improved disaggregation accuracy over the existing EDS method, and the uncertainty index measures the reliability of the returned estimation accurately.