Publication: Deep learning and system identification methods applied to power system state estimation
Loading...
Identifiers
Publication date
2023-06
Defense date
2023-09-21
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This dissertation proposes new topology identification methods for static and dynamic
state estimation applications and new approaches for the inertia estimation of traditional
and non-synchronous generators.
The proposed topology identification methods are based on deep neural network techniques and allow one to accurately estimate the topology configuration of an electric network.
The proposed approach exploits the input measurements of static state estimators
to provide the actual topology identification. Furthermore, with some adjustments, the
method can also be extended to dynamic state estimation applications and can be used in both a centralized and decentralized manner. The topology identification methods proposed in this dissertation have been tested in several benchmark networks, namely, the IEEE 14-, 39-, and 118-bus test systems. The proposed method proves to be reliable and shows remarkably low computational times that are compatible with real-time applications. Additionally, this thesis presents new inertia estimators based on system identification methods. Particularly, a recursive system identification method is proposed for estimating the inertia of both traditional and non-synchronous generators. The approach has an element-wide scope and can be used after large power imbalance events and when the system is under normal operating conditions. Apart from the inertia constant, the method also provides other useful parameters such as the primary frequency control gain and the damping coefficient of the generation unit. Subsequently, an iterative system identification method is proposed to carry out the inertia estimation with an area-wide scope. These inertia estimation methods have been implemented in modified versions of the IEEE 39-and 118-bus test systems, and both provide highly accurate estimations with very small computation burdens that make them suitable for online monitoring and control applications.
Description
Mención Internacional en el título de doctor
Keywords
Power systems, State estimation, Deep learning