The effect of HVDC lines in power-grids via Kuramoto modelling

Technical Summary: The Effect of HVDC Lines in Power-Grids via Kuramoto Modelling

Problem Statement
Modern power grids, traditionally modeled as large synchronous machines based on rotating generators, are becoming increasingly vulnerable due to the integration of renewable energy sources and the reduction of rotational inertia. This reduction diminishes the system's ability to dampen perturbations in power flow and frequency, leading to potential instabilities, synchronization loss, and cascade failures (blackouts). While High Voltage Direct Current (HVDC) lines are known to efficiently transmit power over long distances and can segment grids to limit cascade propagation, their impact on synchronization dynamics and the potential for adverse effects (such as the Braess paradox) requires deeper investigation beyond linear approximations. The paper addresses the need to understand how replacing specific AC links with HVDC connections affects synchronization stability and cascade failure behavior in a non-linear, large-scale regime.

Methodology
The authors employ a numerical study using the adaptive second-order Kuramoto model (KM) to simulate the dynamics of the European high-voltage (HV) power grid (EU2016 dataset, $N=13,478$ nodes). The model incorporates inertia ($I$), damping ($\alpha$), and non-linear phase interactions, moving beyond the linearized swing equations often used in electrical engineering.

Key methodological components include:

• Model Dynamics: The system is governed by second-order differential equations where nodes represent oscillators with self-frequencies drawn from a Gaussian ensemble. The model includes a threshold-based line failure mechanism: if the power flow $|\sin(\theta_j - \theta_i)|$ exceeds a threshold $T=0.99$, the line is permanently removed, triggering a cascade.
• HVDC Implementation: The study simulates the replacement of 36 seabed cables connecting the Scandinavian peninsula to the rest of Europe (approx. 0.2% of total links) with HVDC lines. Two primary modeling approaches are tested:
  1. Static HVDC: Links are removed from the adjacency graph, and fixed power injections/withdrawals are applied at the endpoints, effectively decoupling the phases of the two sub-networks.
  2. Adaptive HVDC: A new term is added to the equations of motion, $f(\dot{\theta}_j - \dot{\theta}_i) \cdot D_{ij}$, where the power flow depends on the local frequency difference. Various activation functions ($f$) are tested, including linear, $\tanh(x)$, $\text{softsign}(x)$, and $\text{sign}(x)$.
• Simulation Protocol: Simulations are run on GPU and CPU clusters using adaptive step-size integrators (Bulirsch-Stoer and Runge-Kutta). The system is initialized in a phase-synchronized state, thermalized to a steady state, and then perturbed by a random line removal to observe cascade evolution. Metrics include the Kuramoto order parameter ($R$), universal order parameter ($R_{uni}$), frequency variance ($\Omega$), and average cascade size ($\langle N_f \rangle$).

Key Contributions

1. Non-Perturbative Analysis: The study provides a dynamical simulation of the second-order Kuramoto model on a realistic, large-scale network, capturing non-linear effects and large phase differences that linear models miss.
2. Spectral Dimension Context: The results are interpreted through the lens of spectral dimension ($d_s \approx 2.34$ for the EU grid). The authors relate synchronization improvements to the finite-size scaling behavior of second-order Kuramoto models on graphs with $d_s < 4$, where the critical coupling strength $K_c$ diverges with system size.
3. Comparative HVDC Modeling: The paper systematically compares static HVDC segmentation against various adaptive control strategies (based on frequency differences), offering a physicist's perspective on grid stability complements to existing electrical engineering load-flow studies.

Results

• Synchronization and Stability:
  • Static HVDC: Replacing AC links with static DC connections improves phase synchronization (higher $R$) and reduces cascade sizes for intermediate coupling strengths. However, this comes at the cost of increased frequency spread ($\Omega$) because DC lines do not transmit phase or frequency information, effectively creating two independent sub-networks with higher individual frequency variances.
  • Adaptive HVDC: Adaptive methods generally improve steady-state synchronization. The $\text{sign}(x)$ activation function (step-function) provides the most robust stabilization across all coupling values, eliminating the frequency Braess paradox observed in static models at high coupling. However, this method introduces discontinuities that drastically increase relaxation times (by orders of magnitude) and cause slow convergence.
  • Linear and Smooth Activations: Linear, $\tanh$, and $\text{softsign}$ functions improve synchronization for lower coupling values ($K < 6000$) but fail to mitigate cascade sizes at higher couplings, showing a crossover behavior. Narrowing the linear activation interval pushes the saturation point lower, reducing cascade sizes.
• Cascade Behavior: While HVDC generally reduces cascade sizes compared to the full AC model, the effectiveness depends on the coupling strength and the specific HVDC control strategy. The adaptive methods show promise but suffer from long relaxation times.
• Braess Effects: The study notes that varying total transmitted power at large and small global couplings can lead to drawbacks in frequency spread and Braess effects, particularly when fluctuations are small, causing a "freezing" in dynamics.

Significance and Claims
The authors claim that their work provides a deeper, physics-based complement to existing electrical engineering analyses of power grids. By solving the adaptive second-order Kuramoto model without linear approximations, they demonstrate that:

• The spectral dimension of the grid ($d_s < 4$) fundamentally influences synchronization transitions, suggesting that segmenting the grid (via HVDC) into smaller sub-graphs can improve synchronization stability for a given coupling strength.
• Adaptive HVDC lines, particularly those using frequency-difference-based control, are more efficient in the steady state than static replacements but incur a penalty of very long relaxation times.
• The trade-off between phase synchronization and frequency spread is a critical factor; while HVDC improves phase alignment, it can degrade frequency coherence.

The paper concludes that while HVDC segmentation is a viable strategy for mitigating cascade failures and improving synchronization in specific regimes, the choice of control strategy (static vs. adaptive) involves significant trade-offs between stability, relaxation time, and frequency coherence. The authors modestly suggest that future work should investigate the effects of additive stochastic noises and compare frequency distributions with real-world data.