A distributed knowledge method for multi-agent power flow analysis based on consensus algorithms

Abstract

The paper introduces a novel gradient tracking-based algorithm for solving the power flow problem in a fully distributed manner, using the AB algorithm. The motivation for this work stems from the limitations of centralized approaches, which can be overcome with distributed implementations. Notably, the proposed distributed algorithm eliminates the need for a central monitoring facility, allowing all calculations, input data, and network intelligence to remain within individual buses (agents), thus removing single points of failure and preserving data privacy. The paper presents how this can be achieved by reformulating the power flow study as a purely distributed optimization problem, and then applying the AB algorithm, which can effectively converge even when only partial system information is available. To enhance the performance of the proposed algorithm, two significant modifications—cost function whitening and momentum—are introduced as an additional contribution, which enables faster convergence (in fewer than 20 iterations) while maintaining accuracy comparable to traditional centralized power flow algorithms. The effectiveness of the proposed framework is validated through tests on IEEE 14- and 300-bus systems, demonstrating its practical applicability and robustness. The paper also examines some extreme operating scenarios, such as instances when communication is lost with parts of the network, or when uncertainty exists in grid parameters.

Keywords

• Distributed power flow
• AB Algorithm
• Gradient-based model
• Single point of failure