🔢 mathematics

Resilience Enhancement of Electricity-Computing Interdependency System against Contingency

This paper proposes an Electricity-Computing Interdependency System (ECIS) model and a recovery optimization strategy that dynamically adjusts data center power quotas based on system states and infrastructure interdependencies, demonstrating a resilience improvement of over 20% during power shortages.

Original authors: Minqiao Zheng, Zejun Yang

Published 2026-07-28

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Minqiao Zheng, Zejun Yang

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 city as a giant, living organism. For decades, we've known that this organism needs electricity to keep its heart beating and its lights on. But in the last few years, something new has happened: the city has grown a massive, digital brain. This "brain" is made of computing power—the invisible engine that runs our hospitals, banks, and communication networks. Just like a human brain needs oxygen to think, this digital brain needs electricity to work. The tricky part is that they are now locked in a dance: the digital brain needs electricity to run, but the rest of the city needs the digital brain to function. If the power goes out, the brain stops thinking, and the city freezes. If the brain gets too much power but the rest of the city doesn't, the city still can't function. This paper explores how to keep this delicate dance going when a disaster hits and the power supply starts to shrink.

The researchers, Minqiao Zheng and Zejun Yang from Shanghai University, decided to build a new way of looking at this problem. They call their creation the "Electricity-Computing Interdependency System" (ECIS). Think of it as a smart traffic controller for a city in crisis. In the past, when a disaster struck and power was scarce, city planners used a simple "priority list." They would say, "Hospitals get power first, then banks, then schools." If a Data Center (the giant warehouse full of computers that provides the city's digital brain) was on that list, it was treated just like a regular building that needed lights and air conditioning.

The authors argue that this old way of thinking is broken. They point out that a Data Center isn't just a building that uses power; it's a factory that makes something else: computing power. It takes electricity and turns it into the digital fuel that hospitals and banks need to operate. If you treat the Data Center like a regular building, you might give it too much power when there isn't enough to go around, causing the rest of the city to starve. Or, you might give it too little, leaving the city's digital brain asleep even though the lights are on.

To fix this, the team created a computer simulation to test a new strategy. They modeled a city with different types of buildings: life-saving places like hospitals (Level 1), public services like water systems and universities (Level 2), and regular areas like shops and homes (Level 3). They then simulated a disaster where the available electricity dropped to as low as 20% of normal.

They tested two approaches. The first was the old "Priority List" method. They tried giving the Data Center a high priority (making it a VIP) and a low priority (making it a regular guest). The results were messy. When the Data Center was a VIP and got all the power at 30% supply, the Data Center was happy, but the hospitals and banks had no power and couldn't function, so the whole city's score was zero. When the Data Center was a regular guest, the hospitals got power, but they couldn't do their jobs because they needed the Data Center's computing power to run their systems. The city was still stuck.

Then, they tried their new "Electrical-Computing Interdependency" strategy. This was like giving the traffic controller a superpower: the ability to see the whole picture. Instead of a fixed list, this new strategy asked, "How much power does the Data Center need right now to help the most critical buildings?"

The simulation showed that this new approach was a game-changer. When power was extremely low (20-40%), the strategy wisely decided to temporarily "sacrifice" the Data Center, giving power directly to the hospitals and banks so they could keep their basic lights on. As the power supply slowly recovered to 50-80%, the strategy started feeding the Data Center more power, knowing that the Data Center could now generate enough computing power to help the other buildings work better.

The results were impressive. In these simulations, the new strategy improved the city's overall ability to bounce back (resilience) by more than 20% compared to the old methods. Specifically, the new strategy achieved a resilience score of 0.553, while the best of the old strategies only managed 0.449 or 0.433. The researchers also found that if you ignored the connections between buildings (like how a bank depends on the Data Center), you would overestimate how well the city was doing. By including these connections, the model showed a 9.5% improvement in resilience scores.

In short, the paper suggests that in our AI-driven world, we can't just look at electricity anymore. We have to understand that electricity and computing power are a team. By letting a smart system decide how to split the power between the "brain" (the Data Center) and the "body" (the rest of the city) based on the current situation, we can keep our cities running much better when the lights flicker. It's not a magic fix for every disaster, but in these computer models, it shows a clear path to a more resilient future.

Technical Summary: Resilience Enhancement of Electricity-Computing Interdependency System against Contingency

Problem Statement
Modern urban infrastructures, including hospitals, financial institutions, and water systems, rely heavily on stable computing services provided by data centers (DCs). However, existing resilience models for post-disaster recovery typically view DCs merely as electrical loads with fixed priority levels. This perspective overlooks the critical interdependency where DCs convert electrical power into computing power to support other urban functions. Consequently, traditional "electrical power on, function on" models fail to capture the cascading failures that occur when computing power supply is disrupted, even if electrical power is partially restored. Furthermore, current resource scheduling studies often assume sufficient electrical resources, neglecting the constraints imposed by extreme disasters on computing power availability. There is a lack of a unified framework to describe the coordination among computing power systems, electrical power systems, and urban infrastructures under extreme conditions.

Methodology
To address these gaps, the authors propose an Electricity-Computing Interdependency System (ECIS) framework and a collaborative optimization strategy.

System Modeling:
- Functional Level: The system's overall functional level ( $F_{sys}$ ) is defined as a weighted sum of individual facility functional levels, varying with the electrical power supply ratio ( $k$ ).
- DC Modeling: The DC is modeled as a facility that transforms electrical power into computing power. A piecewise function ( $g(P_{dc})$ ) maps electrical power input to computing power output, capturing startup thresholds, ramping stages, and saturation.
- Facility Functionality: The basic functional level ( $B_i$ ) of an infrastructure is a weighted sum of its electrical power satisfaction and computing power satisfaction. Weights ( $\alpha_i, \beta_i$ ) reflect the specific dependence of each facility type (e.g., financial institutions have high $\beta_i$ ).
- Dependency Propagation: An iterative model based on a dependency weight matrix ( $A$ ) simulates how the failure of one facility affects others. The actual functional level ( $F_i$ ) is the product of the facility's basic function and its external dependency satisfaction level, updated until convergence.
Optimization Strategy:
- The objective is to maximize overall system resilience ( $R$ ), defined as the area under the system functional level curve across recovery stages.
- The problem is formulated as a nonlinear constrained optimization problem, solved using a Genetic Algorithm.
- Two strategies are compared:
  - Strategy A (Baseline): A predefined priority list restoration strategy where the DC is treated as a conventional load with a fixed weight.
  - Strategy B (Proposed): A collaborative optimization strategy that dynamically adjusts the DC's electrical power quota based on system states, explicitly considering the indirect benefits of computing power supply and infrastructure interdependencies.

Key Contributions

ECIS Framework: The paper introduces a framework characterizing the coupling relationships among computing power, electrical power, and infrastructure interdependencies in post-disaster scenarios.
Collaborative Optimization Strategy: A new strategy is developed that integrates computing power supply and infrastructure dependencies into electrical power allocation. Unlike traditional methods, it treats the DC not just as a load but as a producer of a critical resource (computing power) for the broader system.

Results
Case studies simulating electrical power supply ratios from 20% to 100% reveal the following:

Limitations of Fixed Priorities (Strategy A): Assigning a fixed high priority to the DC leads to system functional collapse during severe power shortages (20–40% supply) because other critical facilities starve of power. Conversely, assigning a low priority causes computing power shortages that degrade the functionality of high-dependency facilities (e.g., financial institutions) during moderate recovery (50–80% supply).
Performance of Proposed Strategy (Strategy B): Strategy B dynamically adjusts the DC's electrical power satisfaction level. It sacrifices the DC during severe shortages to sustain basic urban functions and increases DC power allocation as supply recovers to leverage computing power benefits.
Quantitative Improvements:
- Strategy B achieves a system resilience value of 0.553, which is 23.2% higher than Strategy A with a low DC weight and 27.7% higher than Strategy A with a high DC weight.
- Considering infrastructure interdependencies within Strategy B further improves resilience by 9.5% compared to a version of the strategy that ignores these dependencies.
- Overall, the proposed strategy improves system resilience by more than 20% during power shortages compared to traditional predefined priority list strategies.

Significance
The paper claims that in the context of the AI revolution and digital infrastructure development, computing power has become a critical resource for urban resilience. Assessing post-disaster recovery solely from an electrical power perspective is no longer sufficient. By modeling the DC as a converter of electrical to computing power and accounting for interdependencies, the proposed ECIS framework provides a more accurate and effective approach for optimizing resource allocation during extreme events. The study highlights that ignoring these interdependencies leads to an overestimation of urban functional levels and suboptimal resilience planning. Future work is noted to involve integrating real distribution network topologies and multi-data-center coordination.

More like this