Data-driven current control of modular multilevel converter-based HVDC transmission system

arXiv:2611.07734 · eess.SY, cs.SY · Submitted 2026-11-05 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We're looking at the paper: "Data-driven current control of modular multilevel converter-based HVDC transmission system".

Dev: Here's the gist of the paper. Model-based design (MBD) is a mature framework for designing the current controllers of modular multilevel converters (MMCs) in high-voltage direct current (HVDC) transmission systems.

Rosa: This paper studies a data-driven (DD) current control method based on integral Subspace Predictive Control (iSPC), synthesized from input–output data measured in a high fidelity real-time digital simulator to propose an initial guideline for its systematic application to MMC current control.

Dev: High-voltage direct current (HVDC) systems increasingly rely on modular multilevel converters (MMCs) due to their high efficiency, scalability, robustness, and low harmonic distortion.

Rosa: The performance and safe operation of MMCs strongly depend on the design of their control systems, particularly the current controllers.

Paper summary: Rosa: So, we're wrapping up this look at the paper "Data-driven current control of modular multilevel converter-based HVDC transmission system." It’s about taking those complex modular multilevel converters in high voltage direct current systems and designing their current controllers using a data driven approach.

Dev: Yeah, yeah, it’s about taking those complex modular multilevel converters in high voltage direct current systems and designing their current controllers using a data driven approach. Basically, they synthesized an integral subspace predictive control method directly from input output measurements taken during real time simulation.

Taro: That means they didn't just rely on standard math equations for the model; they used actual operational data to build a better one.

Rosa: It’s about moving away from purely first principle modeling, where you try to guess every single interaction between the layers of the control system.

Dev: The paper found that this data driven approach was able to capture some cross coupling effects between output currents and those circulating currents that traditional models often miss.

Taro: That's significant because those interactions are usually where things go wrong in real hardware, especially under noisy conditions or when the digital delays kick in.

Rosa: The implication is that instead of just trying to build a perfect theoretical model from scratch, you can start with what you actually measure during testing.

Dev: They also proposed some concrete guidelines for tuning this method, telling people how to choose the prediction horizon and what kind of excitation signal they need.

Taro: It shows that this isn't just some abstract idea; it’s a practical tool that can be applied systematically to design these current controllers.

Rosa: So, the big takeaway here is that using real data can give you a much more robust starting point for your predictive controller design than just relying on textbook state space models.

Dev: It opens the door for designing systems where you account for those real world imperfections right from the beginning of the control synthesis process.

Taro: And since they are planning to test this method directly on a real time digital simulator, we gotta watch that next step closely.

Conclusion: Rosa: So we’re wrapping up this look at "Data-driven current control of modular multilevel converter-based HVDC transmission system."

Dev: Yeah, yeah, it’s about taking those complex modular multilevel converters in high voltage direct current systems and designing their current controllers using a data driven approach.

Taro: Basically, they synthesized an integral subspace predictive control method directly from input output measurements taken during real time simulation.

Rosa: That means they didn't just rely on standard math equations for the model; they used actual operational data to build a better one.

Dev: It’s about moving away from purely first principle modeling, where you try to guess every single interaction between the layers of the control system.

Taro: The paper found that this data driven approach was able to capture some cross coupling effects between output currents and those circulating currents that traditional models often miss.

Rosa: That's significant because those interactions are usually where things go wrong in real hardware, especially under noisy conditions or when the digital delays kick in.

Dev: The implication is that instead of just trying to build a perfect theoretical model from scratch, you can start with what you actually measure during testing.

Taro: They also proposed some concrete guidelines for tuning this method, telling people how to choose the prediction horizon and what kind of excitation signal they need.

Rosa: It shows that this isn't just some abstract idea; it’s a practical tool that can be applied systematically to design these current controllers.

Dev: So, the big takeaway here is that using real data can give you a much more robust starting point for your predictive controller design than just relying on textbook state space models.

Taro: It opens the door for designing systems where you account for those real world imperfections right from the beginning of the control synthesis process.

Rosa: And since they are planning to test this method directly on a real time digital simulator, we gotta watch that next step closely.

Delft University of Technology

eess.SY, cs.SY

Submitted: 2026-11-05

Updated: 2026-11-05

Comments: 8 pages, submitted to IET AC and DC Power Transmission (ACDC)

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 79/100

The gist: Model-based design (MBD) is a mature framework for designing the current controllers of modular multilevel converters (MMCs) in high-voltage direct current (HVDC) transmission systems.

Key concepts

Model-based design (MBD)
A mature framework used to design controllers by exploiting mathematical models and control theory. It systematically synthesizes controllers with desired properties like stability, using linear state-space models as the preferred approach for MMCs.
Data-driven predictive control (DPC)
A predictive control method that relies on identified model or measured input-output data instead of explicit identification. It uses Hankel matrices constructed from collected I/O data to estimate unknown system matrices, allowing for prediction and control without a separate identification step.
Integral Subspace Predictive Control (iSPC)
A specific data-driven predictive control algorithm developed for MMC current control. It utilizes input-output data to build a dynamic model and uses this model within a predictive framework to generate the optimal control signal, aiming for better performance than standard methods.
Prediction Horizon (N)
The number of future time steps the controller looks ahead when making decisions. The study found that selecting the optimal prediction horizon is crucial; increasing it improves control performance but increases computational load. Tuning this parameter is a key guideline proposed by the research.

Terminology

Summary

Model-based design (MBD) is a mature framework for designing the current controllers of modular multilevel converters (MMCs) in high-voltage direct current (HVDC) transmission systems. p. 0

Where the paper says it

reyesdreke@tudelft. nl Keywords: DATA-DRIVEN CONTROL, MODULAR MULTILEVEL CONVERTER, HIGH VOLTAGE DC Abstract Model-based design (MBD) is a mature framework for designing the current controllers of modular multilevel converters (MMCs) in high-voltage direct current (HVDC) transmission systems.

This paper studies a data-driven (DD) current control method based on integral Subspace Predictive Control (iSPC), synthesized from input–output data measured in a high fidelity real-time digital simulator to propose an initial guideline for its systematic application to MMC current control. p. 0
Where the paper says it

Data-driven (DD) methods can deliver less biased models that capture parameter uncertainties, control-loop interactions, and grid dynamics, yet they remain under-exploited for controlling MMC currents. This paper studies DD current control method based on integral Subspace Predictive Control (iSPC), synthesized directly from input–output data measured in a high fidelity real-time digital simulator (RTDS®), and proposes an initial guideline outlining the main steps for its systematic application to MMC current control.

Background and Motivation

High-voltage direct current (HVDC) systems increasingly rely on modular multilevel converters (MMCs) due to their high efficiency, scalability, robustness, and low harmonic distortion. The performance and safe operation of MMCs strongly depend on the design of their control systems, particularly the current controllers. p. 0

Where the paper says it

In high-voltage direct current (HVDC) systems, modular multilevel converters (MMCs) have become the preferred solution for power conversion owing to their high efficiency, scalability, robustness, and low harmonic distortion [2]. However, the performance and safe operation of MMCs strongly depend on the design of their control systems, particularly the current controllers [3].

Model-based design (MBD) is a mature framework for designing the current controllers of MMC where mathematical models and control theory are exploited to systematically synthesize controllers that offer desired closed-loop properties like stability and robustness. p. 0
Where the paper says it

However, the performance and safe operation of MMCs strongly depend on the design of their control systems, particularly the current controllers [3]. A mature framework for designing the current controllers of MMC is model-based design (MBD), where mathematical models and control theory are exploited to systematically synthesize controllers that offer desired closed-loop properties, such as stability, and robustness [4].

Linear state-space (SS) models are preferred in MBD due to their good trade-off between accurate dynamics prediction and computational complexity. p. 0
Where the paper says it

Among the examples of MBD, we can highlight optimal control [5], robust control [6], and model predictive control (MPC) [7]. Here, linear state-space (SS) models are the preferred model structure due to their good trade-off between accurate dynamics prediction and computational complexity.

However, first-principle SS (fSS) modeling often neglects practical effects such as digital delays, filtering, and interactions between control layers which can degrade performance or induce instability. p. 0
Where the paper says it

Despite these advantages, the effectiveness of MBD strongly depends on the accuracy of the underlying model. When using first-principle SS (fSS) modeling, models rely on small-signal theory and often neglect practical effects such as digital delays, filtering, and interactions between control layers (including phase-locked loop), which may degrade performance or even induce instability [10].

This has motivated a shift toward data-driven (DD) modeling where frequency scanning is used to generate input–output (I/O) data that allows building less biased models. p. 0
Where the paper says it

When using first-principle SS (fSS) modeling, models rely on small-signal theory and often neglect practical effects such as digital delays, filtering, and interactions between control layers (including phase-locked loop), which may degrade performance or even induce instability [10]. These shortcomings have motivated a shift toward data-driven (DD) modeling, where frequency scanning is often used to generate input–output (I/O) data that allows to build less biased model, see e.

System Dynamics and Data Acquisition

MMCs are comprised of 3 legs with two arms, in which N submodules (SMs) are connected in series to an equivalent resistor (Rm) and inductor (Lm). The dynamics of the MMC currents are governed by Kirchhoff’s voltage law as shown in equations (1a) and (1b). p. 1

Where the paper says it

The arms are named as upper and lower arm, denoted by n ∈ {u, l}, and the phases are denoted by m ∈ {a, b, c}. Then, Kirchhoff’s voltage law shows that MMC currents are governed by the transformer (TF) (V r m), + 3 3 3 - + - SM SM SM (a) Arm (b) Submodule Fig.

The augmented model is derived to enforce zero-error tracking when following step references, resulting in the state-space form x+ = Ax + Bu. p. 2
Where the paper says it

Operator diag(·) defines a diagonal matrix where the diagonal elements are its argument. The augmented model is derived to enforce zeroerror tracking when following step references.

Data-Driven Predictive Control Methodology

Data-driven predictive control (DPC) follows the MBD-MPC principle but relies on an identified model or directly on measured I/O data rather than an explicit identification step. DPC builds on two key concepts: Hankel matrix and persistence of excitation. p. 2

Where the paper says it

The I/O data are collected through an SI experiment, in which an excitation signal applied to the arm voltages forms the input sequence u˜ and the measured currents (i o dq, ic dq) form the output sequence y˜ (see Fig. DPC builds on two key concepts: Hankel matrix and persistence of excitation.

The I/O data collected during SI are used to construct Hankel matrices Up, Yp, Uf, and Yf. p. 2
Where the paper says it

The I/O data collected during SI, i. , (˜y, u˜), are used to construct the following Hankel matrices: Up = H0 [Tini ,T ] (˜u), Yp = H1 [Tini ,T ] (˜y), Uf = H Tini [N,T ] (˜u), Yf = H Tini +1 [N,T ] (˜y), (7) where Tini is the "past horizon".

The estimation of the unknown matrix Θ can be carried out by solving the least-squares problem defined in equation (10) to find Θ∗. p. 3
Where the paper says it

According to (9) future outputs Yk are described as a linear combination of previous I/O data and δUk. By relying on this relation, the estimation of the unknown matrix Θ can be carried out by solving the least-squares problem: min Θ Yf − Θ   Up Yp Uf   2 Fro , (10) with solution Θ∗ , i.

Performance and Design Guidelines

The iSPC performance is studied in a point-to-point, two-terminal MMC-based HVDC link using the CIGRE DCS1 RTDS benchmark. The prediction horizon N proved to be the most influential design parameter, as increasing N yielded the best trade-off between control performance and computational burden. p. 5

Where the paper says it

Using these I/O data, we designed the iSPC and proposed a guideline for its systematic application, covering the design of the excitation signal, the selection of Tini and T, and the tuning of the prediction horizon. Apart from data quality, the prediction horizon N proved to be the most influential design parameter: increasing N yielded the best trade-off between control performance and computational burden, whereas increasing T beyond 20 Tini brought no substantial improvement in low to moderate noise conditions.

Increasing T beyond 20 Tini yielded no substantial improvement in low to moderate noise conditions. p. 5
Where the paper says it

Using these I/O data, we designed the iSPC and proposed a guideline for its systematic application, covering the design of the excitation signal, the selection of Tini and T, and the tuning of the prediction horizon. Apart from data quality, the prediction horizon N proved to be the most influential design parameter: increasing N yielded the best trade-off between control performance and computational burden, whereas increasing T beyond 20 Tini brought no substantial improvement in low to moderate noise conditions.

The paper proposes guidelines for tuning the iSPC covering the selection of the prediction horizon, data length, and excitation signal type. p. 1
Where the paper says it

DeePC, showing more accurate tracking and lower sensitivity to noise at a lower computation time. Based on extensive MATLAB/Simulink simulations, we further propose guidelines for tuning the iSPC, covering the selection of the prediction horizon, the data length, and excitation signal type.

Comparative Analysis

Compared with DeePC under identical tuning, iSPC responds faster but with a higher overshoot while achieving a roughly 13 times lower tracking MSE at the cost of a moderately higher control effort. The average computation time for iSPC in MATLAB was 1. p. 5

Where the paper says it

In MATLAB environment, the average computation time∗ per control step is 1. 15 ms for the iSPC and 268 ms for DeePC.

15 ms per control step, whereas for DeePC it was 268 ms per control step. p. 5
Where the paper says it

In MATLAB environment, the average computation time∗ per control step is 1. 15 ms for the iSPC and 268 ms for DeePC.

Future work will implement the iSPC directly on RTDS. p. 5
Where the paper says it

However, the iSPC structure allows the OCP to be solved with Hildreth’s algorithm [25] coded in C++, which has proven effective for MMC control [7]. Future work will therefore implement the iSPC directly on RTDS.

p. 5
Where the paper says it

However, the iSPC structure allows the OCP to be solved with Hildreth’s algorithm [25] coded in C++, which has proven effective for MMC control [7]. Future work will therefore implement the iSPC directly on RTDS.

The integral of the absolute control input (IAU) for iSPC was 7. p. 5
Where the paper says it

The integral of the absolute control input (IAU) is 7. 7846 pu for the iSPC and 6.

7846 pu, compared to 6. p. 5
Where the paper says it

The integral of the absolute control input (IAU) is 7. 7846 pu for the iSPC and 6.

3353 pu for DeePC. p. 5
Where the paper says it

7846 pu for the iSPC and 6. 3353 pu for DeePC.

Conclusion and Future Work

The SI experiment revealed notable cross-coupling between the output and circulating currents dynamics which were often neglected in fSS models. The dSS model identified from I/O data was proposed to predict the current dynamics more accurately. p. 5

Where the paper says it

The SI experiment revealed notable cross-coupling between the output and circulating currents dynamics, often neglected in fSS models. Consequently, dSS model identified from I/O data were proposed to predict the current dynamics more accurately.

The iSPC design and systematic application guideline covered the design of the excitation signal, selection of Tini and T, and tuning of the prediction horizon. p. 5
Where the paper says it

Consequently, dSS model identified from I/O data were proposed to predict the current dynamics more accurately. Using these I/O data, we designed the iSPC and proposed a guideline for its systematic application, covering the design of the excitation signal, the selection of Tini and T, and the tuning of the prediction horizon.

Future work will implement the iSPC directly on RTDS. The paper is supported by Horizon Europe project PROSECCO and NWO Veni project SAFE-GRID. p. 6
Where the paper says it

Future work will therefore implement the iSPC directly on RTDS. 6 Acknowledgements The work is supported by Horizon Europe project PROSECCO, under grant agreement 101160687, and by NWO Veni project SAFE-GRID, with project number 20248.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Data-driven model generation can capture dynamics neglected by first-principle state-space models, such as cross-coupling between the output and circulating current loops, switching and modulation ripple, leading to a more accurate prediction of current dynamics.

  2. The integral Subspace Predictive Control (iSPC) can be implemented to replace only the Current Control Loop PI controllers in Hierarchical Control Architectures (HCAs), achieving a roughly 13 times lower tracking MSE at the cost of a moderately higher control effort, while reducing the computation time by more than two orders of magnitude.

  3. The iSPC design guideline provides systematic steps for application, specifically covering the selection of the prediction horizon, the data length, and excitation signal type, allowing for optimized trade-offs between robustness and computational burden based on Signal-to-Noise Ratio (SNR).

  4. The system identification process can be improved by using random binary sequences superimposed on white noise to generate input data that captures the steady-state behavior while simultaneously improving the data PE compared to multisine signals.

  5. The system can perform direct optimal control without an explicit identification step, as seen in Data-Enabled Predictive Control (DeePC), which is described as a direct method that solves the OCP using the data directly, without an explicit identification step.

Abstract

Model-based design (MBD) is a mature framework for designing the current controllers of modular multilevel converters (MMCs) in high-voltage direct current (HVDC) transmission systems. In MBD, linear state-space models are preferred for their good trade-off between accurate dynamics prediction and simplicity. However, first-principle derivations often neglect relevant MMC dynamics, which can hinder controller performance. Data-driven (DD) methods can deliver less biased models that capture parameter uncertainties, control-loop interactions, and grid dynamics, yet they remain under-exploited for controlling MMC currents. This paper studies DD current control method based on integral Subspace Predictive Control (iSPC), synthesized directly from input--output data measured in a high fidelity real-time digital simulator (RTDS), and proposes an initial guideline outlining the main steps for its systematic application to MMC current control. The proposed iSPC is compared against state-of-the-art DD solution, and the findings are validated through simulation studies in MATLAB/Simulink.

Related papers