Intensive Data-Driven Model for Real-Time Observability in Low-Voltage Radial DSO Grids

Increasing levels of distributed generation (DG), as well as changes in electricity consumption behavior, are reshaping power distribution systems. These changes might place particular stress on the secondary low-voltage (LV) distribution systems not originally designed for bi-directional power flows. Voltage violations, reverse power flow, and congestion are the main arising concerns for distribution system operators (DSOs), while observability in these grids is typically nonexistent or very low. The present paper addresses this issue by developing a method for nodal voltage estimation in unbalanced radial LV grids (at 0.4 kV). The workflow of the proposed method combines a data-driven grey-box modeling approach with generalized additive models (GAMs). Furthermore, the proposed method relies on experimental data from a real-world LV grid in Denmark and uses data input from only one measuring device per feeder. Predictions are evaluated by using a test data set of 31 days, which is more than twice the size of the training data set of 13 days. The prediction results show high accuracy at root mean squared errors (RMSEs) of 0.002-0.0004 p.u. The method also requires a short computation time (14 s for the first stage and 2 s for the second stage) that meets requirements for the practical, real-time monitoring of DSO grids.

Automated summary

This paper develops a method for nodal voltage estimation in unbalanced radial low-voltage grids using a data-driven grey-box modeling approach. The method relies on experimental data from a real-world LV grid in Denmark and uses data input from only one measuring device per feeder. The prediction results show high accuracy at root mean squared errors (RMSEs) of 0.002-0.0004 p.u. and a short computation time that meets requirements for practical, real-time monitoring of distribution system operator (DSO) grids.