Towards Zero-Inertia Power Systems: Stability Analysis, Control & Physics-Informed Neural Networks

Large-scale integration of Renewable Energy Sources (RES) as well as the installation of new High Voltage Direct Current (HVDC) lines pose new technical challenges concerning the stability of the electrical power system, the accurate system modeling and the real-time dynamic security assessment. This is due to the lack of system inertia, limited overload capability of power electronic
devices and fast power electronic interfaces. To ensure the secure and reliable operation of power systems with high penetration of RES and HVDC interconnectors, advanced control methods and new operational tools are required. The research aims of this PhD thesis are twofold: first, it investigates the effects of controller design of converters on the power system stability and proposes control designs and metrics to enable the secure integration of converter-based resources. Second, this thesis investigates analysis tools for time-domain simulations by evaluating the appropriateness of existing reduced complexity models for converters and by proposing the use of physics-informed neural networks which can provide fast and accurate solutions. Overall, this work approaches the fundamental challenge of low- and zero-inertia power systems from two distinct angles and proposes pragmatic solutions. The first fundamental challenge concerns the replacement of conventional synchronous-based generation by renewable energy sources. This reduces the level of rotational inertia and introduces uncertainty in the dynamic behavior of the power grid due to their continuously varying power infeed. Consequently, the level of system inertia obtains a time-varying profile, which influences the frequency stability of the system. Two strategies are considered to tackle the problem of frequency stability. The first strategy leverages tools from control system theory and proposes a structured robust frequency control design that accounts for the impact of low inertia on frequency dynamics and aims to reduce the Rate Of Change Of Frequency (ROCOF) and frequency
nadir. The second strategy focuses on utilizing HVDC systems interconnecting asynchronous areas and studies the exchange of frequency reserves in order to limit the maximum ROCOF and frequency nadir. Particularly, we extract expressions for the ROCOF, frequency nadir and maximum steady-state frequency deviation as a function of the system inertia and components’ control parameters. These expressions are incorporated into the unit commitment problem as constraints. This enables the dispatch of enough generator-based units and the procurement of enough frequency reserves, to maintain the frequency stability of the system and allow for secure integration of RES into the power grid. Moreover, considering that converter-based resources are called to participate in grid-supporting services, we study how their operation mode, namely grid-forming and grid-following converter, affects the system stability and dynamic performance. Having as control objectives to avoid (i) induced instabilities caused by the converters and (ii) the saturation of converters in the event of a contingency, we propose methods for appropriate tuning of the converters’ control parameters. This allows the safe and reliable integration of converter-based resources to power grid. The second fundamental challenge relates to the system analyses based on time-domain simulations, which is a critical tool for power system operators. To ensure the secure power system operation, transmission system operators evaluate a large number of scenarios that correspond to hundreds of thousands of operating points and different types of contingencies. Due to the increase of
the number of generation units, mainly due to the large penetration of RES, the number of these scenarios dramatically increases, which in turn increases the computation time required for performing these dynamic security assessment studies. To this end, two methodologies are proposed. The first concerns the appropriateness of Root Mean Squared (RMS) models to assess the security of a power system with high penetration of RES. RES are usually connected to the grid over power electronic converters which introduce faster dynamics to the system response, that RMS models cannot always capture. Considering the key role grid-forming converters will play in the near future, we propose a loop-shaping control design, and derive sufficient conditions that determine the appropriateness of the RMS models of different classes of grid-forming converters, when simulating events where the power balance is disturbed, e.g. loss of load or generation. The second direction draws from recent developments in machine learning and neural networks for predicting solutions to systems of partial differential equations, ordinary differential equations and differential algebraic equations. This thesis proposes, for the first time, physics informed neural networks for power system applications and demonstrates how they can provide solutions for a system of differential algebraic equations at a fraction of the time required by traditional numerical solvers, while maintaing high accuracy.