Scalability Analysis of Convex Relaxation Methods for Branch Flow AC Optimal Power Flow

Today's power grid is in a transitional stage to cater to the needs of energy efficiency, climate change, and environmental targets. In the process of designing the future power grid, one of the most fundamental models to be utilized is AC optimal power flow (AC-OPF). Since the feasible space of AC-OPF is non-convex, the optimization models developed using it often result in multiple local minima. To avoid such computational challenges in solving optimization models, various relaxation methods have been developed in the past. In the literature, these relaxation methods are mainly tested on specific networks. However, the scalability of relaxation techniques on branch-flow-based AC-OPF is yet to be explored. In this context, this paper compares the performance of different relaxation methods with the well-established MATPOWER AC-OPF solver in terms of the mean square error (MSE), maximum squared error, minimum and maximum values of voltage magnitude, and the average simulation time. In addition, the scalability of these models is tested on various radial and mesh networks with nodes ranging from 33 to 6655 nodes and 9 to 6515 nodes, respectively. In this manner, the trade-off between computational complexity and solution accuracy is demonstrated and analyzed in depth. This provides an enhanced understanding of the suitability and efficiency of the compared relaxation methods, helping, in turn, the efficiency of optimization models for varying sizes and types (i.e., radial or meshed) of networks.