Offset-free Data-Driven Predictive Control for Grid-Connected Power Converters in Weak Grid Faults

arXiv:2511.14337 · eess.SY, cs.SY · Submitted 2025-11-18 · Read on arXiv

Listen

Transcript

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

Rosa: Today's paper: "Offset-free Data-Driven Predictive Control for Grid-Connected Power Converters in Weak Grid Faults".

Dev: Grid-connected power converters encounter significant stability challenges during weak grid faults, when conventional PI-based controllers exhibit an oscillatory response and poor fault-ride-through performance.

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: So, we've just finished looking at the paper "Offset-free Data-Driven Predictive Control for Grid-Connected Power Converters in Weak Grid Faults," and it seems like they're tackling a pretty nasty problem: stability issues when power converters operate on weak grids during faults, where standard PI controllers just start oscillating.

Dev: Yeah, that's the core issue they're addressing; conventional PI-based controllers struggle with poor fault-ride-through performance in those weak grid conditions. What caught my eye right away was their solution involves swapping out those traditional outer PI loops for something data-driven and offset-free to regulate the DC link and PCC voltages.

Taro: From an autonomy standpoint, I'm curious how this translates when the grid itself is acting erratically; they're moving toward a system that doesn't rely on pre-defined physics models for control decisions. What does that mean for real-world operation outside of a perfect simulation?

Rosa: Well, the paper suggests their approach uses either data collected before a fault happens or data gathered during the fault itself to build input-output predictors, which allows them to get offset-free control without needing any physics-based modeling at all. That sounds like a huge simplification for deployment.

Dev: Exactly, and they show that by using this pre-fault offset-free DPC approach, they managed to double the critical equivalent grid impedance that the system can handle, and they also cut the root mean squared error during faults by a factor of forty compared to conventional PI control. That's a substantial improvement in accuracy under stress.

Taro: Doubling the handled impedance while cutting error by forty times is significant; it means this AI system could operate reliably in scenarios that were previously considered unstable or too demanding for standard hardware. Does this robustness extend to really harsh, unpredictable grid events?

Rosa: The paper does show that their regular iSPC controller manages to keep sustained oscillations even when only using pre-fault data and facing a critical grid reactance of zero point three five nine five p.u., which is almost double the value where other controllers like CC can still maintain some form of stable behavior.

Dev: That level of sustained oscillation management under severe fault conditions really tells us something about the stability margin they've opened up; it’s not just about avoiding immediate collapse, but maintaining a manageable state. Plus, they kept the computation times comparable to conventional PI control, which is crucial for real-time operation.

Taro: Maintaining that computational efficiency while achieving such a significant increase in fault handling capability is impressive because it means this predictive control can be implemented on standard GCPC microprocessors without needing massive processing power. What about the data requirements for this method?

Rosa: That’s an interesting point; they noted that the developed iSPC solution can handle large datasets, specifically measurements up to ten thousand DC-link and PCC voltage magnitude readings, which is much more scalable than other DPC methods like iDeePC.

Title and authors: Dev: Scalability is key for practical application because it means the system doesn't need an impossibly huge history of data just to maintain performance; they also showed that an analytical iSPC solution can be computed with a complexity similar to conventional PI control, which speaks directly to real-time loop rate concerns.

Taro: So, we have a predictive controller that handles large data sets and runs fast enough for standard hardware while demonstrating better stability in weak grid faults; what's the big picture impact this has on how we design resilient power infrastructure?

Rosa: The implication is that we can build converters that are significantly more fault-tolerant simply by using smart control based on past or current data, rather than relying solely on complex, model-based physics descriptions for every single fault scenario. This shifts the design focus toward data utilization.

Dev: I think the most immediate impact is in improving the reliability of power distribution systems connected to grids that are inherently unstable or prone to faults; this paper offers a practical way to enhance fault-ride-through capabilities without a massive increase in hardware complexity.

Taro: If we consider the broader context of other papers we've seen, like those on graph-based barrier functions or temporal logic verification, this predictive control method seems to be an implementation that actually works in a physical system under dynamic stress; it bridges the gap between theoretical safety guarantees and practical, high-performance operation.

Rosa: I agree; the fact that they identified the critical impedance limit—doubling it for a four percent grid voltage drop—gives us a concrete benchmark for how much better this control strategy is over existing methods in terms of handling system stress.

Dev: It's about moving from reactive control, which is what PI often does during faults, to proactive control that anticipates the required action based on observed data patterns. That anticipation is where the performance gain comes from when things go wrong.

Taro: So, for our listeners, this means future power systems could be designed with converters capable of surviving more severe grid disturbances without needing over-engineered protective measures just to maintain connection.

Rosa: That’s right; the paper "Offset-free Data-Driven Predictive Control for Grid-Connected Power Converters in Weak Grid Faults" gives us a concrete, data-driven tool that enhances fault tolerance significantly. We'll be taking a quick break and coming back after this to discuss how this predictive control compares to other advanced methods we've been looking at.

Dev: Stay tuned; we’re going to keep exploring the technical details of this paper and what it means for real-world control loops, right?

Taro: We'll be here shortly with more thoughts on the broader autonomy implications of such robust control systems.

The paper's summary: Rosa: So, to recap, this paper introduces an offset-free data-driven predictive control method for grid-connected power converters that tackles stability issues during weak grid faults by using pre-fault or fault-time data to predict the system's behavior without needing a physics model.

Dev: That’s right; it replaces those traditional PI controllers with this new predictive approach, and the big result they show is that it can handle double the critical grid impedance while cutting down error during faults by a factor of forty compared to standard PI control.

Taro: From an autonomy angle, that means when the grid starts acting weird, this system can anticipate those issues based on what it has already seen or what's happening right now, which is pretty cool for mission-critical applications where immediate reaction time matters.

Rosa: Exactly; the implication here is that we can build converters that are way more resilient to grid instability without having to rely on incredibly complex, slow model-based calculations during an emergency.

Dev: I'm focused on the engineering side—they managed to keep the computation time pretty close to what a conventional PI loop needs, which means we aren't introducing massive latency or processing overhead when we deploy this in hardware.

Taro: It’s about shifting the control paradigm from purely reactive to something that is predictive based on observed data patterns, which is much more useful when the environment itself is unpredictable.

Rosa: And they even show it can maintain sustained oscillations under pretty severe fault conditions, which speaks to a level of robustness we hadn't seen with this type of method before.

Dev: That sustained oscillation capability is interesting because it shows a deeper understanding of the system dynamics during failure, not just how to quickly stabilize it back to normal.

Taro: So, if we think about the broader world impact, this suggests that infrastructure connected to weak or unstable grids could become much more reliable simply by integrating this type of data-driven control strategy into their power conversion systems.

Rosa: It really does point toward a future where resilience in power distribution isn't just about bigger hardware, but smarter intelligence embedded directly into the control logic.

Dev: We need to keep looking at how fast these predictors update during rapid changes, because even with good data usage, latency is still a concern for high-speed fault recovery.

Taro: That leads us perfectly into the next part of our discussion on verifying such complex systems under extreme conditions and testing their real-world deployment limits.

The paper's improvements: Rosa: So, looking at the improvements in this paper, it seems like they've really nailed how to use data from before or during a fault to build predictors without needing any physical equations for control.

Dev: That’s right; the main improvement is that this data-driven predictive approach can handle twice the critical grid impedance compared to older methods, which directly translates to a much more capable system in weak grid scenarios.

Taro: From an autonomy standpoint, what's exciting is that this system doesn't need a perfect model of the entire grid dynamics; it just needs enough input/output data to make smart decisions when things get messy.

Rosa: It really means we can design power converters that are robust against unexpected grid behavior because they rely on learned patterns rather than brittle, pre-defined control laws.

Dev: The computational improvement is also significant because the analytical solution for this predictor has a complexity level similar to conventional PI control, which keeps the loop rate fast and low latency, which is essential for real-time stability.

Taro: That speed combined with the predictive capability suggests that autonomous systems connected to power grids could react to sudden disturbances much faster than current reactive control schemes allow.

Rosa: And they showed it can handle a substantial forty-fold reduction in error during faults, which means the power quality stays much better even when the grid is struggling.

Dev: That reduced error is fantastic for system health; it implies that the converter spends less time oscillating around an unstable operating point, which is a major failure mode we're trying to avoid.

Taro: It’s about making autonomous decisions based on empirical data rather than rigid rules, which could be very beneficial when deploying these converters in remote or unpredictable environments.

Rosa: Exactly; the paper’s conclusion emphasizes that this simple, analytical approach can be implemented on standard microprocessors without needing specialized, high-end hardware for the control loop itself.

Dev: That ease of implementation is a huge practical advantage; it lowers the barrier to deploying advanced fault-ride-through capabilities across a wider range of commercial power converter designs.

Taro: It makes robust control accessible, which is what we need when we're thinking about scaling up autonomous power delivery systems in areas where grid infrastructure might be less reliable.

Rosa: So, the big picture here is that this method provides a simple yet highly effective way to boost the reliability of power conversion devices operating in challenging grid conditions.

Dev: We'll need to watch how they handle constraint verification next, because while the data-driven part is strong, ensuring it respects physical limits under extreme stress is where we need more rigorous testing.

Taro: I agree; verifying those constraints will be the next big step in proving that this predictive control can operate safely when the environment truly misbehaves.

Conclusion: Rosa: So, to wrap things up, we’ve seen how this paper on "Offset-free Data-Driven Predictive Control for Grid-Connected Power Converters in Weak Grid Faults" shows that we can build much smarter fault handling for power converters by using data instead of complex models.

Dev: We established that this approach doubles the handleable grid impedance and cuts error by forty percent, all while keeping the control loop fast enough for real-time operation on standard hardware.

Taro: It’s really about giving autonomous systems a proactive way to manage connection stability when the external environment is unstable, which is a huge step for reliable deployment in unpredictable settings.

Rosa: That makes sense; we're moving toward control systems that are more adaptive to real-world grid fluctuations instead of just reacting to them with traditional methods.

Dev: I think the most important part for us as engineers is that this method simplifies the design process by removing the need for physics-based modeling, which reduces potential sources of error and complexity in hardware implementation.

Taro: And that simplification allows us to focus more on the actual operational constraints and safety verification, which we've been focusing on with papers like those on graph-based barrier functions.

Rosa: Indeed; the paper lays a solid foundation for how we can integrate data-driven methods into existing control architectures to create more resilient power infrastructure.

Dev: Moving forward, I'll be looking at how they handle those constraints mentioned in their future work to see if this analytical solution holds up under more rigorous testing scenarios.

Taro: I’m interested in seeing how they test the robustness of this system when it encounters the kind of complex, multi-faceted failures that come with real grid events.

Rosa: Well, that’s all for today's deep dive into this paper; we really have some exciting tools here for improving power converter resilience.

Dev: We'll be sure to follow up next week by looking at how these predictive algorithms stack up against formal verification methods in terms of safety guarantees.

Taro: That sounds like a great topic; understanding the trade-off between predictive performance and guaranteed safety is crucial for autonomous systems.

Department of Electrical Engineering, Eindhoven University of Technology

eess.SY, cs.SY

Submitted: 2025-11-18

Updated: 2026-09-28

Journal ref: Electric Power Systems Research, Volume 263, Article 113551, February 2027

DOI: 10.1016/j.epsr.2026.113551

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 81/100

The gist: Grid-connected power converters encounter significant stability challenges during weak grid faults, when conventional PI-based controllers exhibit an oscillatory response and poor fault-ride-through

Key concepts

Offset-free Data-Driven Predictive Control (DPC): This approach uses pre-fault or fault-time data to build input-output predictors for regulating DC link and PCC voltages without needing any physics models. It allows the system to regulate stability based on observed data patterns rather than relying on predefined physical equations.
Grid Impedance Handling: The paper shows that this DPC approach can handle twice the critical equivalent grid impedance compared to older methods. This means the converter can operate reliably in weaker grid conditions, such as those with a four percent grid voltage drop, which were previously considered unstable.
Computational Efficiency: A key advantage is that the analytical solution for this predictor has a complexity similar to conventional PI control. This keeps the computation time low and comparable to standard PI loops, which is essential for real-time operation on microprocessors without introducing significant latency.
Fault-Ride-Through Performance: This refers to how well the power converter maintains connection and stability when the grid experiences a fault. The new control method significantly improves this performance by reducing error during faults by a factor of forty compared to standard PI control.

Terminology

Robustness — 6× in this paper · explained in 5 episodes
Verification — 4× in this paper · explained in 2 episodes
Benchmark — 2× in this paper · explained in 2 episodes
Core — 2× in this paper · explained in 2 episodes
Accuracy — 1× in this paper · explained in 1 episode
Benchmarking — 1× in this paper · explained in 3 episodes
Formal Verification — 1× in this paper · explained in 2 episodes
Model Predictive Control (MPC) — 1× in this paper · explained in 5 episodes

Summary

Grid-connected power converters encounter significant stability challenges during weak grid faults, when conventional PI-based controllers exhibit an oscillatory response and poor fault-ride-through performance. This paper addresses this problem by replacing the conventional outer PI controllers that regulate DC-link and PCC voltages with an offset-free data-driven predictive controller. The developed algorithm leverages either pre-fault or fault-time data to construct inputoutput predictors, yielding offset-free control without the need for physics-based modelling. Simulation results show that pre-fault offset-free DPC doubles the critical equivalent grid impedance that can be handled and reduces the root mean squared error during faults by a factor of 40, while maintaining computation times comparable to conventional PI control. These findings demonstrate that the developed offset-free data predictive controller offers a simple, robust, and computationally efficient alternative to conventional control, significantly enhancing fault-ride-through capabilities of converters in weak grids.

The paper presents an implementation of integral subspace predictive control (iSPC) [17], an alternative offset-free DPC algorithm, for GCPCs in weak grid faults. In the developed architecture, iSPC replaces the conventional outer layer controlling the DC-link and PCC voltage magnitudes. To validate the improvement brought by iSPC, the critical grid impedance and voltage drop that can be handled by CC are first identified. Then it is shown that iSPC can handle double the impedance for a similar grid voltage drop. Compared to other DPC algorithms like iDeePC, the developed iSPC solution can handle large datasets, e.g. 10,000 DC-link and PCC voltage magnitude measurements. Regarding real-time implementation, an analytical iSPC solution can be computed with similar computational complexity as conventional PI control.

The main contributions of the paper are: "(i) reproduce oscillations that occur during weak grid faults for GSPC with conventional PI-controllers using a realistic grid model; (ii) implementation of iSPC for GCPC control; and (iii) validation of iSPC in critical faults and comparison with PI-controllers."

In the simulation results, regular iSPC consistently outperforms the other two controllers in both metrics (settling time and RMSE). Specifically, for PCC Voltage Magnitude during fault after fault, regular iSPC showed a settling time of 0.21 s and an RMSE of 7.45 × 10−3 p.u., whereas CC showed a settling time of 0.30 s and an RMSE of 1.26 × 10−2 p.u., and FT iSPC showed a settling time of 6.09 × 10−3 s and an RMSE of 6.09 × 10−3 p.u., respectively, in the same time window after fault clearance (Fig. 8). Furthermore, regular iSPC manages to maintain sustained oscillations despite the severe fault and only prefault data, demonstrating its robustness up to a critical grid reactance of Xg = 0.3595 p.u., which is almost double the critical value for which CC can maintain oscillatory but still (marginally) stable behaviour. The computations were performed in MATLAB/Simulink 2025a, and the Matlab function block that calculates the iSPC input took on average 0.242 ms, which is within the 1 ms sampling time.

The paper concludes that "a simple analytical iSPC controller, identified pre-fault under nominal conditions, can already substantially improve GCPC control and, given the time required for calculation, be implemented in standard GCPC microprocessor hardware. The limits of the iSPC controller with pre-fault data were determined to lead to a doubling of the critical equivalent grid reactance for a 4% grid voltage drop. Future work will include verifying the controller with constraints, testing an adaptive version of iSPC, and investigating the controller response with more advanced grid models." The paper also notes that the identification phase involves adding excitation (e.g., white noise) to u(k) and recording u(k) and y(k), followed by constructing Hankel matrices to obtain the predictor matrices Θ∗. In terms of data collection, two options are compared: fault-triggered iSPC (FT iSPC) and regular iSPC, with FT iSPC being closer to the fault dynamics but using less data and having delayed activation, while regular iSPC captures different dynamics but uses more data and is active from the start. The parameters P, Q, R, T, Tini, and N are tuned such that A higher P and Q make the controller more aggressive as the tracking error penalty is higher. Increasing R makes the controller less aggressive as faster-changing inputs are penalised. The general rules of thumb for tuning are stated as "N ≥ Tini ≥ nI:= nu + ny and T ≥ (Tini + N)nu + nI.

Improvements for AI systems

Based on the scientific paper, here are specific improvements for AI systems and what those improved systems can achieve:

Improve control robustness in Grid-Connected Power Converters (GCPCs) operating in weak grid faults by replacing conventional PI controllers with an offset-free Data-Driven Predictive Control (DPC) algorithm, specifically the Integral Subspace Predictive Control (iSPC) scheme.
Enable the improved AI system to handle critical grid impedance up to double the conventional control limit while reducing Root Mean Squared Error (RMSE) during faults by a factor of 40 compared to PI control.
Allow the controller to maintain stability and sustain oscillations under more severe fault conditions (up to a critical equivalent grid reactance nearly double that of conventional controllers) without immediate instability, demonstrating superior Fault-Ride-Through (FRT) capability.
Achieve real-time computational efficiency comparable to conventional PI control, enabling implementation on standard GCPC microprocessors despite the complexity of the predictive algorithm.
Improve data utilization by enabling the system to handle large datasets (e.g., 10,000 voltage magnitude measurements) with a more scalable approach compared to other DPC methods like iDeePC, which suffer from complexity scaling with dataset size.
Develop a control architecture that can utilize either pre-fault or fault-time data to construct input-output predictors directly from measurements, eliminating the need for physics-based models or state observers.
Implement a control scheme (regular iSPC) that outperforms both conventional PI control and fault-triggered iSPC (FT iSPC) in both settling time and RMSE during faults, leading to significantly faster recovery times after fault clearance.

The improved AI system can achieve:

A robust, highly reliable DC-link voltage and PCC voltage magnitude regulation for power converters in unstable or weak grid environments. It will actively maintain grid connection during severe faults (FRT) by rapidly adjusting control actions based purely on input/output data, leading to significantly reduced transient errors (RMSE) and faster system recovery compared to traditional methods.