REPOSITORY > RESULTS

Abstract

This dissertation investigates the problem of observability in power distribution networks.
A recent increase in the share of renewable energy sources has increased uncertainty and
variability on the generation part of the power equation. Due to lower nominal powers,
these sources are normally connected to distribution networks, which poses technical challenges
to reliable network operation. In the dissertation, we first the examine the accurate
modeling of a power network. We proposed a unified three-phase branch model which
can represent all the relevant components of a power distribution network (overhead lines,
underground cables, transformers in any connection, tap changers in any connection, and
voltage regulators). The presented model enables a straightforward construction of the
network-admittance matrix, i.e. the network model. The network model is crucial in the
later processing with network observability algorithms. In relation to the model, we have
reviewed the methods for obtaining model parameters, which comprise methods for self
and mutual series impedances, methods for self and mutual shunt admittances, and the
calculation of connection matrices.
Further, we have reviewed various state estimators. Since we are dealing with the
estimation of deterministic parameters, the reviewed estimators belong to the so-called
class of classic estimators. Maximum Likelihood Estimation (MLE) and Weighted Least
Squares (WLS) estimation are discussed in detail. Besides the MLE and WLS estimators,
we also discuss estimators that exhibit a robustness or insensitivity to bad data present in
the input set of measurements. From this class of robust estimators, the Schweppe Huber
Generalized-M (SHGM) estimator and Least Absolute Value (LAV) are discussed in detail.
The reviewed estimators are compared based on calculation times and statistical efficiency
for dierent measurement configurations on the reference IEEE distribution feeder.
Further, the dissertation discusses sensitivities to small model and measurement errors.
It is frequent in practice that actual conductor lengths dier from the modeled ones. The
conducted sensitivity analysis determines the uncertainty intervals in the estimated state
variables from the uncertainty intevals in the input model variables. For analysis, the
approach with perturbation of the Karush-Kuhn-Tucker conditions was leveraged, which
yields all the specified sensitivities at once.
In the nal part of the dissertation, the application of the state estimation solution
to a real-world power distribution network is discussed. Besides the implementation of
the algorithm itself, the real-word implementation requires many other support functions
architected into a workable solution. The software architecture of the solution and the detailed
process of implementation are presented. The dissertation also presents the concept
of a low-cost synchrophasor measurement device, suitable for implementation in power
distribution networks.