New approaches to dynamic equivalent of active distribution network for transient analysis

• Neue Ansätze für das dynamische Äquivalent des aktiven Verteilungsnetzes für die Transientenanalyse

Wu, Xiang; Monti, Antonello (Thesis advisor); Moser, Albert (Thesis advisor)

1. Auflage. - Aachen : E.ON Energy Research Center, RWTH Aachen University (2016, 2017)
Book, Dissertation / PhD Thesis

In: E.On Energy Research Center : ACS, Automation of complex power systems 40
Page(s)/Article-Nr.: xviii, 138 Seiten : Illustrationen, Diagramme

Dissertation, RWTH Aachen University, 2016

Abstract

With the increasing amount of distributed generators at the distribution level, the dynamic behavior of active distribution networks (ADNs) will have a more significant influence on the overall electrical system. To perform transient analysis of such a large and complex system, it is neither practical nor necessary to apply a fully detailed system model. A technique for obtaining highly accurate yet simple equivalents for ADNs is becoming increasingly important. In this dissertation, three original equivalent models are proposed to cover the challenges posed by the size and complexity of power system transient analysis. A fixed-structure dynamic equivalent model (FDEM) is proposed by integrating four individual equivalent models (IEMs), which are derived from approximation of the physical models of electrical equipment. The FDEM appears as a sixth- order state space, which is much less complex than the original systems. It can be easily integrated into different tools as a modular component with to-be-edited parameters. The derivation of the IEMs and FDEM presents the first original contribution. An adaptive dynamic equivalent model (ADEM) is proposed by formulating an equivalence problem in terms of a Markov decision process problem, which is solved using a machine learning algorithm based on reinforcement learning. The structure of the ADEM is adaptive depending on measured data and it can be directly applied for on-line applications. It keeps not only a simple equivalent model form but also brings flexibility for an equivalent model structure. The transformation of the equivalence problem to Markov decision process problem and the learning skills of the ADEM are the second original contribution. A random forest-based dynamic equivalent model (RF-DEM) is proposed by introducing randomized learning framework with feedbacks from outputs, which trains the relationship between inputs and outputs using RF as the supervised learning algorithm. The RF-DEM takes advantage of easy implementation and does not require electrical modeling and approximation knowledge for deriving the equivalent models. The design of the RF-DEM forms the third original contribution.

Institutions

• Institute for Automation of Complex Power Systems [616310]
• E.ON Energy Research Center [080052]