⚡ electrical engineering

Synthesizing Voltage Ride-Through Controllers for Data Centers

This paper introduces SolVRT, a formal methods-based system that automatically synthesizes correct-by-construction voltage ride-through controllers for data centers to ensure grid-code compliance, while also providing diagnostic capabilities to identify and resolve infeasibility through targeted hardware or workload modifications.

Original authors: Wayne Wang, Archit Bhatnagar, Tongyuan Miao, Saniya Kalamkar, Wenqi Cui, Inigo Incer, Ang Chen

Published 2026-08-10

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Wayne Wang, Archit Bhatnagar, Tongyuan Miao, Saniya Kalamkar, Wenqi Cui, Inigo Incer, Ang Chen

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 the power grid as a giant, invisible ocean that flows through our cities, carrying electricity to everything from your phone charger to the massive servers that run the internet. For a long time, scientists and engineers treated big power users like data centers as simple, passive things—like a bucket that just sits there and gets filled. But recently, these "buckets" have grown so huge that if they suddenly tip over and spill, they can cause the whole ocean to crash. This is the world of Voltage Ride-Through (VRT). Think of it like a surfer's rule: when a wave (a voltage drop) hits, you don't just bail off your board and swim to shore; you have to stay on, paddle hard to keep your balance, and get back to cruising speed once the wave passes. If too many surfers jump off at once, the ocean gets chaotic, and the whole beach might get wrecked. The challenge is that data centers are made of sensitive electronics that naturally want to bail out the second things get shaky. The question is: how do we teach these digital giants to stay calm, keep their balance, and ride the wave without crashing the whole system?

This is where a new system called SolVRT comes in. The authors, a team of researchers, decided to stop guessing how data centers should behave and start using "formal methods"—a fancy way of saying they used strict, math-based logic to design a perfect controller. Instead of letting engineers tweak knobs by hand and hoping for the best, they wrote the rules of the road (the grid codes) into a precise language that a computer can understand and check. They built a digital twin of a massive 200 MW data center and asked the computer to solve a puzzle: "Can you find a set of instructions that keeps this data center connected and stable during a voltage crash?"

The results were fascinating. When the data center had enough hardware power, SolVRT successfully synthesized a controller that acted like a master surfer, keeping the data center connected and stable through voltage dips that would have caused a normal system to shut down instantly. But here is the real magic: when the computer said "No, this is impossible with the current equipment," it didn't just give up. It acted like a detective, tracing exactly why it failed. It found a "conflict frontier"—a specific point where the data center's hardware was too weak to do two things at once: keep its internal voltage high enough to stay alive and keep drawing enough power to satisfy the grid rules.

The paper then used this detective work to offer two very different solutions. First, it showed that if you upgrade the hardware (specifically, making the power converters 50% bigger, from 25 MVA to 38 MVA), the data center could ride through any wave the grid threw at it. Second, it showed a "software-only" fix: if you simply tell the data center to run at only 32% of its maximum power, it could also survive the waves, but only if the waves weren't too deep. The authors found that while turning down the workload helps, it's a temporary patch; upgrading the hardware is the only way to guarantee the data center can handle the worst storms. In short, SolVRT proves that we can mathematically prove whether a data center can survive a power crisis, and if it can't, it can tell you exactly what to buy or change to make it safe.

Technical Summary: Synthesizing Voltage Ride-Through Controllers for Data Centers

Problem Statement
Data centers (DCs) have evolved into massive, fluctuating electrical loads that are highly sensitive to power quality. Traditionally, DCs protect their internal equipment by disconnecting from the grid during voltage disturbances (sags or swells). However, simultaneous disconnection of multiple large DCs can destabilize the power grid, exacerbating the initial fault. To mitigate this, grid operators (e.g., ERCOT, Energinet) have introduced Voltage Ride-Through (VRT) codes. These codes mandate that DCs remain connected during specific voltage disturbances, maintain a minimum active power draw (a "power floor"), and recover to pre-fault power levels within a defined deadline.

The core challenge is that designing controllers to satisfy these coupled temporal and operational requirements is difficult. Current DC responses rely on hand-tuned heuristics and individual device protections, which offer no formal guarantee of compliance. Furthermore, grid codes are written in natural language, making them ambiguous and difficult to analyze formally. There is a lack of principled methods to synthesize controllers that are guaranteed to meet these codes or to diagnose why a specific facility cannot comply.

Methodology: SolVRT
The authors propose SolVRT, a system that uses formal methods to synthesize grid-code-compliant VRT controllers. The approach treats VRT requirements as an "assume-guarantee" specification, a framework common in safety-critical domains like aerospace.

Formal Specification (STL):
The system introduces a Domain-Specific Language (DSL) to translate natural language grid codes into Signal Temporal Logic (STL). This formalization captures the three core obligations of VRT:
- Connection: The internal bus voltage ( $v_i$ ) must remain above a trip threshold during the disturbance.
- Power Floor: The aggregate active power ( $p$ ) must stay above a calculated floor (often proportional to the sag depth and pre-fault power) during the event.
- Recovery: The power draw must return to a target percentage of the pre-fault level within a specific deadline after the voltage recovers.
  The specification explicitly defines assumptions (the grid voltage disturbance envelope) and guarantees (the DC's response).
Controller Synthesis:
SolVRT encodes the STL specification, a model of the DC's internal power topology (using the LinDistFlow linearized power-flow approximation), and hardware constraints (converter apparent power limits) into a controller synthesis problem.
- The system searches for a control input (active and reactive power injections from UPS/BESS converters) that satisfies the STL guarantees under the worst-case grid voltage trace.
- If a solution exists, the system synthesizes a Model Predictive Controller (MPC) that is "correct-by-construction," meaning it is mathematically guaranteed to satisfy the grid code for the given model and disturbance.
Diagnosis and Provisioning:
If the synthesis query is infeasible (no controller exists), SolVRT performs a diagnostic analysis:
- Conflict Frontier: It computes a boundary representing the maximum active power floor the facility can sustain while staying connected at various sag depths.
- Infeasibility Attribution: It identifies the "Irreducible Infeasible Subsystem" (IIS) to determine which specific clauses (e.g., connection vs. power floor) are in conflict.
- Provisioning Recommendations: It calculates the minimal hardware or workload changes required to restore feasibility. This includes increasing converter apparent power ratings, adjusting interconnection reactance, or capping pre-fault workload power.

Key Contributions

Formalization of VRT: The paper is the first to formalize data center VRT requirements as a controller synthesis problem using STL, moving beyond informal text to machine-checkable logic.
Synthesis Algorithm: It provides an algorithm that either produces a compliant MPC controller or proves that no such controller exists for a given facility and grid code.
Diagnostic Framework: It introduces the "conflict frontier" abstraction to diagnose infeasibility, isolating conflicting requirements and quantifying the exact hardware or workload modifications needed to achieve compliance.
DSL for Grid Codes: The development of a specification language that makes voltage event assumptions, response timers, and power floors explicit and parameterized.

Results
The system was evaluated using a closed-loop simulation of a 200 MW data center connected to a 140-bus transmission system (NPCC model).

Effectiveness: The synthesized controller successfully rode through voltage sags of varying depths (0.83, 0.70, and 0.54 pu) that caused uncontrolled facilities to trip. It maintained internal voltage just above the trip threshold and met active power floor and recovery requirements.
Comparison: Unlike penalty-based controllers (which rely on hand-tuned weights and offer no compliance guarantees), the synthesized controller guaranteed compliance by construction.
Diagnosis: When applied to a facility with undersized hardware (25 MVA converters), the system correctly identified infeasibility. It traced the failure to a conflict between the connection requirement and the power floor requirement.
Provisioning: The system determined that increasing the converter rating from 25 MVA to 38 MVA (or scaling interconnection reactance) would restore compliance. Alternatively, capping the pre-fault workload to 32% of rated capacity would remove the conflict, though it would not solve the fundamental hardware limitation for deeper sags.

Significance
The paper claims that SolVRT provides a rigorous, machine-checkable interface between grid operators and data center operators. By formalizing grid codes, the system eliminates ambiguity in compliance requirements. It allows operators to verify compliance before deployment and provides actionable, quantitative recommendations for hardware upgrades or workload management when compliance is unattainable with current infrastructure. The authors position this work as a step toward increasing the resilience of cross-infrastructure interactions between the power grid and critical digital loads.

Technical Summary: Synthesizing Voltage Ride-Through Controllers for Data Centers

More like this