⚡ electrical engineering

Robust Tuning of Model Predictive Control for MMC-Based High-Voltage Power Systems

This paper proposes a robust Model Predictive Control tuning method for Modular Multilevel Converter-based HVDC systems that solves a convex linear optimization problem to compute optimal weighting matrices, thereby ensuring stability and improved performance under model uncertainties without increasing online computational burden.

Original authors: Victor Daniel Reyes Dreke, Rahul Rane, Aleksandra Lekić

Published 2026-06-11

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Victor Daniel Reyes Dreke, Rahul Rane, Aleksandra Lekić

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a massive, high-voltage power grid as a giant, complex orchestra. The musicians are the Modular Multilevel Converters (MMCs), which are the instruments responsible for moving electricity from one place to another. The conductor is the control system, whose job is to keep the music playing smoothly, ensuring the right amount of power flows without any instruments breaking or the tempo getting out of hand.

Here is a simple breakdown of what this paper does, using everyday analogies:

1. The Problem: The Orchestra is Unpredictable

In the real world, these power instruments aren't perfect. The "strings" (resistors and inductors) might stretch or shrink slightly due to heat or manufacturing quirks. The "sheet music" (the grid frequency) might change speed.

The Challenge: If the conductor (the controller) tries to play based on a perfect, ideal score, but the instruments are slightly out of tune or the room temperature changes, the music can get chaotic. The current might spike (like a violin string snapping) or the system might become unstable.
Current Solutions:
- Old School (PI Controllers): Like a conductor who just yells "Louder!" or "Quieter!" based on what they hear right now. It works okay, but it struggles with complex, fast changes.
- Advanced (Model Predictive Control - MPC): This is like a conductor who looks ahead at the next few measures of music and plans the perfect moves. However, tuning this "future-looking" conductor is very hard. If you tune it wrong, it might plan a move that causes a crash, or it might be too sensitive to small changes in the room.

2. The Solution: A "Robust" Conductor with a Safety Net

The authors propose a new way to tune this "future-looking" conductor (the MPC) so it is robust.

The Analogy: Imagine you are teaching a robot to walk on a tightrope.
- A standard robot might fall if the wind blows even a little bit.
- This paper's method teaches the robot to walk on the tightrope while expecting strong gusts of wind. It calculates the perfect walking pattern that keeps the robot safe even if the wind (uncertainty) hits it.
How they did it:
1. First, they built a "Safety Coach" (Robust Matching Controller): They designed a simple, super-safe controller that knows exactly how to handle the worst-case wind gusts. This controller guarantees the system won't crash, but it might be a bit slow or conservative.
2. Then, they taught the "Future-Looking Conductor" (MPC) to copy the Safety Coach: Instead of guessing how to tune the complex MPC, they used math to make the MPC behave exactly like the safe "Safety Coach" when things are going well, but with the added benefit of looking ahead.
3. The Magic Trick: They turned this tuning process into a convex optimization problem. Think of this as a puzzle where there is only one perfect solution, and a computer can find it quickly and reliably. It's not a game of "try and guess"; it's a guaranteed mathematical solution.

3. The Result: Better Music, No Extra Work

Performance: When they tested this on a digital simulator (a virtual power grid), the new method performed better than the old standard methods. It handled the "wind gusts" (uncertainties) much better, keeping the power flow smooth and safe.
Efficiency: Usually, making a system "robust" (safe against errors) requires heavy computing power, like asking a supercomputer to do extra math every second. This paper's method is clever because it does all the heavy math before the system starts running (offline). Once the system is running, it doesn't need to do any extra work. It's like preparing a perfect meal in advance so the chef can just serve it quickly when the customer arrives.

Summary

The paper presents a new recipe for tuning the brain of a high-voltage power system.

The Issue: Power systems are messy and unpredictable; standard controllers often fail when things change.
The Fix: They created a method to automatically tune a smart controller (MPC) by first designing a "bulletproof" backup plan and then teaching the smart controller to mimic that safety.
The Benefit: The system becomes much more reliable and resistant to errors, without slowing down the computer that runs it.

They proved this works by simulating a real-world power link between two cities, showing that their method keeps the power flowing smoothly even when the equipment isn't perfect.

Technical Summary: Robust Tuning of Model Predictive Control for MMC-Based High-Voltage Power Systems

Problem Statement
Modular Multilevel Converters (MMCs) are a dominant topology in modern High-Voltage Direct Current (HVDC) transmission systems, particularly for integrating renewable energy. However, MMC dynamics are characterized by strong multivariable coupling, strict operational constraints, and significant uncertainties (e.g., component tolerances, grid impedance variations, and unmodeled dynamics). While Model Predictive Control (MPC) offers a promising framework for handling these constraints explicitly, its practical application is hindered by the difficulty of tuning weighting matrices ( $Q, R, P$ ) to ensure stability and robustness. Conventional tuning methods, such as those based on Linear Quadratic Regulator (LQR) theory, often rely on heuristic processes that do not inherently guarantee robust stability in the presence of model uncertainties. Furthermore, existing robust MPC formulations (e.g., tube-based or min-max MPC) often impose prohibitive computational burdens, limiting their use to simpler converter topologies rather than real-time HVDC systems.

Methodology
The paper proposes a systematic, automatic tuning method for MPC that ensures robust performance under bounded model uncertainties without increasing the online computational burden. The methodology follows a three-step design process:

System Modeling: The MMC is modeled using an uncertain discrete-time state-space realization (SSR) in the $dqz$ reference frame. System parameters (inductances, resistances, frequency) are decomposed into nominal and uncertain components, with uncertainties assumed to lie within a bounded polytopic convex set. An augmented state-space model is utilized to enable offset-free tracking of piecewise-constant setpoints.
Robust Matching Controller (RMC) Design: A static state-feedback controller is first designed to be robust against the defined uncertainty set. This design utilizes Linear Matrix Inequalities (LMIs) to enforce discrete-time Lyapunov stability and to embed a robust positively invariant ellipsoid within the admissible state and input constraint sets. An optimization problem is solved to maximize the domain of attraction (DOA) of this RMC.
MPC Tuning via Controller Matching: The core contribution is an automatic tuning algorithm that computes the optimal MPC weighting matrices ( $Q, R, P$ ) by matching the unconstrained response of the MPC to the pre-computed RMC. This is formulated as a convex linear optimization problem. The objective is to minimize the deviation between the MPC's unconstrained optimal control law and the robust RMC gain across the vertices of the uncertainty polytope. This approach ensures that the resulting MPC inherits the robustness properties of the RMC while maintaining the standard MPC structure.

Key Contributions

Robust Tuning Framework: The paper introduces a method to compute MPC weighting matrices that guarantee closed-loop stability and robustness against bounded parametric uncertainties, addressing the gap between robust control theory and practical MPC implementation.
Convex Optimization Formulation: The tuning process is cast as a convex linear optimization problem, ensuring global optimality and reproducibility. Unlike heuristic tuning, this method systematically derives matrices that satisfy phase and gain margin requirements specified by Transmission System Operators (TSOs).
Computational Efficiency: The proposed method enhances robustness without increasing the online computational burden. The robustness is achieved during the offline design phase (tuning), allowing the online controller to remain a standard, unconstrained MPC formulation.
Validation: The approach is validated on a real-time digital simulator (RTDS) using a point-to-point HVDC system model based on the CIGRE benchmark.

Results
The proposed method was tested on a real-time digital simulator (RTDS) model of a point-to-point HVDC system. The results demonstrated that the proposed robustly tuned MPC outperformed conventional LQR-based tuning. Specifically, the proposed method guaranteed closed-loop stability and robust performance under parametric uncertainty, effectively enforcing phase and gain margin requirements. The offline computation times for the RMC design and the MPC tuning were reported as 1.93 seconds and 0.139 seconds, respectively, on a standard processor, indicating the feasibility of the design process.

Significance and Claims
The authors claim that this work addresses a critical gap in the literature: the lack of reported uses of robust MPC in real-time HVDC systems due to design complexity and computational constraints. By leveraging a matching-control approach combined with an LMI-based robust controller, the paper provides a systematic design tool that ensures robustness without sacrificing the computational efficiency required for real-time implementation. The method enables the direct enforcement of stability margins under uncertainty, offering a more reliable alternative to heuristic tuning or complex robust MPC formulations for MMC-based HVDC applications.

1. The Problem: The Orchestra is Unpredictable

2. The Solution: A "Robust" Conductor with a Safety Net

3. The Result: Better Music, No Extra Work

Summary

More like this