Distribution system state estimation for system identification and network model validation: An experience on a real low voltage network

Distribution network data in utility databases are known to present multiple issues that may lead to problematic results when used in physics-based engines, e.g., leading to constraint violations in (optimal) power flow. This paper discusses the application of state and parameter estimation methods to a real low voltage network, where power and voltage time series from digital meters are used to improve the utility’s network data. Good input data are crucial for the advanced decision support tools that are needed to manage networks with increased shares of low carbon technology.

Conventional state and parameter estimation methods leverage measurements from a single (or few) time stamp(s) to detect sparse, local data errors or sudden changes in the system (e.g., a line being de-energized). The methods in this paper differ in that their goal is to estimate “historical” states and reconstruct system parameters from scratch for all users and branches. This is possible through the augmentation of conventional state vectors (i.e., voltage phasors) to include asset properties (e.g., phase connectivity), and binding the asset states as time-independent throughout the time series.

Discussions of real-life experiences are uncommon, but valuable to highlight the differences between working with synthetic or field data. For example, the main contribution of this work rests in exploring the use of state estimation for the statistical validation of data-driven models for real networks, for which the ground-truth is not available (contrary to the case of synthetic data).