Generative neural networks for providing pseudo-measurements in electric power distribution systems

Abstract

The success of automation and control functions envisioned for smart distribution networks depends on reliable real-time network supervision. This task is performed by the distribution state estimator, responsible for processing a set of measurements received by the supervisory control and data acquisition (SCADA) system. In smart grids, the advanced measurement infrastructure (AMI) allows to collect regular readings of consumer voltage and power measurements—this can complement the few measurements (coming from the SCADA system) usually available for monitoring the distribution network and benefit the state estimation process. However, due to communication bottlenecks, such measurements are available only on an hourly basis. In order to circumvent the lack of real-time measurements this paper investigates the application of different neural network models—AutoEncoder, Contractive AutoEncoder, and Variational AutoEncoder—and proposes a methodology to generate AMI pseudo-measurements to complement SCADA measurements when only the latter are available for processing. Simulations performed with a 34-bus distribution system illustrate the proposed methodology, and the results obtained confirm its potential for pseudo-measurement provision.