Intelligent Partitioning in Distributed Optimization of Electric Power Systems

Distributed optimization techniques in electric power systems have drawn increased attention as they provide a scalable way to handle the increasingly complex and large-scale optimization problems associated with the optimal operation of the system. However, little effort has been reported on how to optimally partition the overall optimization problem into subproblems, which significantly affects the efficiency and convergence speed of distributed methods. To address this issue, this paper focuses on how to determine the optimal partition for a given system and optimization problem, and quantify the improvement obtained with the optimal partition in terms of number of iterations and convergence time for solving the ac optimal power flow problem. The proposed approach is based on spectral clustering using a combination of the Hessian matrix of the optimization problem and the admittance matrix as the affinity matrix. Simulation results for the IEEE test systems with 14, 30, 57, 118, and 300 buses confirm the effectiveness of the proposed partitioning method, and the robustness of the performance of a certain partition with respect to the operating point of the system.