⚡ electrical engineering

Threshold Optimization and Dynamic Adaptation of Distributed Optimal Power Flow in 5G Networks

This paper experimentally evaluates ADMM-based Distributed Optimal Power Flow in a real 5G-enabled Smart Grid testbed, demonstrating that a proposed dynamic threshold adaptation mechanism significantly reduces convergence time compared to both baseline and static threshold approaches.

Original authors: Biswajit Kumar Dash, Garrett Thomas, Adedoyin Inaolaji, Filippo Malandra

Published 2026-06-29

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Biswajit Kumar Dash, Garrett Thomas, Adedoyin Inaolaji, Filippo Malandra

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, complex power grid as a giant orchestra. In the old days, one conductor (a central computer) stood on a podium, shouted instructions to every single musician, and waited for everyone to play in perfect sync. But as the grid grows with more solar panels and wind turbines (the musicians), this "one-conductor" approach gets too slow and fragile. If the conductor gets sick or the phone line breaks, the whole orchestra stops.

To fix this, the power grid is moving to a "distributed" approach. Instead of one conductor, the orchestra is split into five small sections. Each section has its own local leader (a Local Controller) who tells their musicians what to do. These leaders talk to each other to make sure the whole orchestra sounds good. They use a mathematical dance called ADMM (Alternating Direction Method of Multipliers) to figure out the perfect timing and volume for everyone.

However, there's a catch: these leaders talk to each other over 5G cell networks, just like your phone. And 5G networks are messy. Sometimes the signal is strong, sometimes it's weak, and sometimes the message takes a long time to arrive (latency).

The Problem: Waiting for the Slowest Friend

In a perfect world, the central coordinator would wait for every single local leader to send back their update before moving to the next step. But in the real world, if one leader is stuck in a bad signal area, the whole orchestra has to sit in silence waiting for them. This is like a group project where the teacher won't let anyone move forward until the slowest student finishes their part, even if everyone else is ready.

The Solution: The "Wait-Too-Long" Timer

The researchers built a real-life test lab called ExTODS using five small computers (Raspberry Pis) connected to a real commercial 5G network to simulate this power grid. They discovered that waiting forever is inefficient.

They introduced a simple rule: Set a Timer.

The Rule: The coordinator sets a "deadline" (a threshold). If a leader sends their update before the timer rings, great! If the timer rings and the update hasn't arrived, the coordinator says, "Okay, I'll use the last thing I heard from you" and moves on immediately.
The Analogy: Imagine a pizza delivery. If the pizza doesn't arrive in 30 minutes, you don't starve; you order a backup or eat what you have. You don't wait 2 hours for the perfect slice.

The Findings: Finding the Sweet Spot

The team tested different timer lengths to see what worked best:

Too Short (0.10 seconds): The timer rings before anyone is ready. The coordinator keeps using old, outdated information. The system gets confused and never finishes the job (it "doesn't converge").
Too Long (0.50 seconds): The coordinator waits too long, wasting time, even though the updates were ready much earlier.
Just Right (0.20 seconds): They found a "Goldilocks" timer. By using this specific time limit, they finished the optimization 7.75% faster than if they had waited for everyone every single time. They didn't lose accuracy; the power grid still ran perfectly, just faster.

The Upgrade: The "Smart" Timer

But 5G networks change. Sometimes you are in a tunnel; sometimes you are in a field. A fixed timer (like a stopwatch) can't handle these changes. If the network gets slow, the fixed timer might be too short. If the network gets fast, the fixed timer wastes time.

So, the researchers built a Dynamic (Smart) Timer.

How it works: The coordinator watches the network in real-time. If it notices the signals are getting weak and slow, it automatically extends the timer. If the network is super fast, it shortens the timer.
The Result: When they tested this against the fixed "Goldilocks" timer, the Smart Timer was 26.42% faster. It adapted to the bad signal conditions by giving the slow leaders a little more time, ensuring they could still send their updates without holding up the whole group.

The Bottom Line

This paper proves that for smart power grids to work in the real world, they can't just rely on perfect math. They need to be aware of the messy reality of 5G networks. By using a "smart timer" that knows when to wait and when to move on, the system becomes much faster and more reliable, ensuring your lights stay on even when the network gets a little shaky.

Technical Summary: Threshold Optimization and Dynamic Adaptation of Distributed Optimal Power Flow in 5G Networks

Problem Statement
The increasing integration of Distributed Energy Resources (DERs) into power distribution networks necessitates a shift from centralized to decentralized control architectures to address scalability, privacy, and resilience challenges. The Alternating Direction Method of Multipliers (ADMM) is a leading technique for distributed optimization in this context. However, the performance of ADMM is heavily dependent on the underlying communication network. In real-world deployments, particularly over cellular networks, communication delays and packet loss can cause deadlocks or significantly slow convergence if the coordinator waits indefinitely for all Local Controllers (LCs) to update. While previous studies have proposed delay-threshold mechanisms to mitigate this, they often rely on analytical delay models that fail to capture the stochastic variability of real-world networks. Furthermore, static thresholds cannot adapt to time-varying network conditions, leading to either unnecessary waiting times or the rejection of valid updates.

Methodology
This work utilizes ExTODS, a fully experimental testbed featuring commercial 5G connectivity and real-time control, to evaluate ADMM-based Distributed Optimal Power Flow (DOPF). The study focuses on the IEEE 123-bus unbalanced distribution feeder, partitioned into five distinct areas using spectral clustering. Each area is managed by a Local Controller implemented on a Raspberry Pi (RPi) equipped with a 5G module, communicating with a central coordinator via a commercial 5G network.

The methodology proceeds in three stages:

Experimental Characterization: The authors conduct comprehensive experiments to measure the impact of communication delays (Round-Trip Times, RTT) and local computation times on ADMM convergence. They evaluate the system under various static delay thresholds ($Thr$) ranging from 0.10s to 0.50s.
Static Threshold Optimization: The authors derive an empirical expression for an optimal static threshold ( $\hat{Thr}^*$ ) based on the maximum observed RTT and local optimization time of the slowest area. This model is validated against experimental data to minimize total convergence time without sacrificing solution accuracy.
Dynamic Adaptation: To address time-varying network conditions, a dynamic threshold tuning framework is proposed. This mechanism updates the threshold in real-time (every $B=5$ iterations) based on the 90th percentile of observed delays within a batch, allowing the system to adapt to fluctuating signal quality (e.g., induced by physical obstructions).

Key Contributions

Experimental Validation on Real 5G: The paper scales the ExTODS testbed to a five-area partition of the IEEE 123-bus feeder, providing a rare experimental evaluation of ADMM over a live commercial 5G network, rather than relying on simulations or analytical models.
Optimal Static Threshold Derivation: The authors experimentally derive a linear model ( $\hat{Thr}^* = 1.27 \cdot RTT_{max} + 0.92 \cdot T_{lo,max}$ ) to predict the optimal static threshold. This model achieves a high coefficient of determination ( $R^2 = 0.96$ ) on validation data, enabling the selection of a threshold that balances convergence speed and accuracy without extensive trial-and-error.
Dynamic Threshold Strategy: A novel dynamic threshold selection strategy is introduced that adapts to changing network conditions using measured RTTs and computation times. This approach is shown to consistently outperform static optimal thresholds in variable environments.

Results

Static Threshold Performance: The study identifies that thresholds below 0.20s prevent convergence due to excessive update rejections, while thresholds above 0.20s allow reliable convergence. The optimal static threshold ( $Thr^* = 0.20s$ ) reduces the total convergence time by 7.75% compared to a "no-threshold" baseline (where the coordinator waits for all updates), while maintaining solution accuracy within 4% of the centralized benchmark.
Dynamic Threshold Performance: Under time-varying network conditions (simulated by periodically obstructing the 5G signal), the dynamic threshold approach reduces the total convergence time by 26.42% compared to the optimal static threshold. It achieves this by increasing the threshold during periods of high latency to ensure timely updates are accepted, and reducing it when conditions improve.
Robustness: The dynamic approach significantly improves the Local Update (LU) success rate for larger, more complex areas (e.g., Area 5) during network degradation, ensuring the coordinator receives fresher data and maintains convergence speed without compromising the maximum residual error (which remains well below the $10^{-4}$ tolerance).

Significance and Claims
The paper claims that its primary significance lies in bridging the gap between theoretical distributed optimization and practical Smart Grid (SG) deployment. By moving beyond abstract delay models to a fully experimental platform with real 5G variability, the work demonstrates that adaptive, communication-aware control strategies are essential for real-world efficiency. The authors assert that while static thresholds offer improvements over naive waiting policies, dynamic adaptation is necessary to handle the inherent unpredictability of cellular networks, yielding substantial gains in convergence speed for time-sensitive grid applications. The study concludes that the proposed dynamic framework provides a robust mechanism for maintaining distributed optimization performance in the face of real-world communication impairments.