Abstract
As power grids integrate more renewable energy sources, coordination between Transmission System Operators (TSOs) and Distribution System Operators (DSOs) becomes critical. However, legal and competitive barriers often prevent the sharing of sensitive grid data. This paper introduces a geographically distributed framework that solves the AC Optimal Power Flow (OPF) problem while maintaining strict data privacy. By combining the ALADIN optimization algorithm with the eCoSim co-simulation environment, the proposed method allows multiple entities to optimize the grid collaboratively without exposing internal network topologies. Testing on a geographically distributed network shows that the framework achieves centralized-level accuracy with a manageable increase in runtime caused by communication latency.
Introduction
The transition toward sustainable energy requires more sophisticated grid management. Traditionally, AC Optimal Power Flow (OPF)—the mathematical determination of the most efficient power routing within physical constraints—is solved centrally. This requires a single entity to possess comprehensive data for the entire grid. In practice, TSOs and DSOs are often prohibited from sharing detailed models or customer data.
Existing distributed optimization methods frequently suffer from two major drawbacks: a lack of numerical convergence guarantees for non-convex problems like AC OPF, and a failure to account for real-world communication hurdles. This research addresses these gaps by implementing a framework that is both mathematically robust and designed for actual distributed IT infrastructures.
Methods
The framework integrates two primary components to bridge the gap between theory and practice: the ALADIN (Augmented Lagrangian based Alternating Direction Inexact Newton method) algorithm and the eCoSim (energy system Co-Simulation) environment.
ALADIN Optimization
Unlike the common ADMM algorithm, ALADIN is specifically designed for non-convex problems. It provides convergence guarantees and achieves a locally quadratic convergence rate, ensuring high-accuracy solutions once the algorithm nears the optimum.
eCoSim and eASiMOV
The eCoSim module, part of the eASiMOV framework, serves as a communication wrapper. It enables different simulation modules, representing distinct grid regions, to exchange data over a network using TCP protocols and shared storage.
The Distributed Workflow
The optimization process follows a four-step cycle:
- Local Step: Each client (DSO or TSO) solves its own local optimization problem using private data.
- Coordination Step: Clients send only "senseless" derivatives and dual variables to a central coordinator, preserving internal privacy.
- Global Step: The coordinator solves a simplified coupled problem to align regions at their borders (tie-lines) and sends updates back.
- Communication: Data exchange occurs via a secure network, ensuring no entity sees another's "black box" grid model.
Experimental Setup
The framework was evaluated using an Integrated Transmission and Distribution (ITD) system:
- TSO Model: A 57-bus transmission system (PGLib).
- DSO Models: Two 33-bus distribution systems (MATPOWER).
- Integration: Merged using the rapidPF toolbox. The system was tested across five computers in a 15km radius on the KIT network to simulate real-world geographic distribution.
Results and Discussion
The study compared several configurations, from centralized solvers to geographically distributed setups.
Accuracy and Convergence
The distributed approach successfully matched the centralized solution's cost (approximately 34,210.55 units) with a negligible optimality gap of $2.46 \times 10^{-8}$. The ALADIN algorithm demonstrated high efficiency, reaching this accuracy in just five iterations.
The Privacy-Performance Trade-off
The primary finding is the quantification of the "cost of privacy." While a centralized solver on a single machine takes ~0.42 seconds, the geographically distributed setup via VPN takes ~18.15 seconds. This 25x slowdown is primarily attributed to network latency and communication overhead rather than mathematical complexity. For grid operators, this delay is often considered a justified trade-off to meet legal privacy requirements.
Conclusion
This research proves that AC OPF can be solved accurately in a privacy-preserving, distributed manner across real-world networks. The framework's "black-box" compatibility allows different operators to use their preferred internal software, facilitating flexible collaboration. Future work will focus on reducing communication bottlenecks—potentially through direct data exchange—and testing the scalability of the coordination step for larger systems involving thousands of buses and numerous DSOs.
Summary updated