Grid congestion is today relieved mainly by redispatch, one of the main drivers of grid operation cost. Busbar switching allows transmission system operators (TSOs) to reconfigure substation connectivity and redistribute power flows without redispatching generation, but systematically optimizing busbar configurations across large transmission grids remains an open problem, particularly for integration into operational TSO processes.

Existing solution methods, including mathematical optimization and exact enumeration approaches, predominantly evaluate candidate topologies with a linearized DC power flow, sacrificing voltage accuracy and reactive power modeling. This approximation is most consequential in voltage-sensitive, high reactive-power operating conditions. Genetic algorithms (GAs) are not subject to this limitation, as they handle discrete switching variables natively and place no restriction on the fitness function, allowing evaluation by a full AC power flow.

We investigate GA setups for AC-exact transmission topology optimization (TTO), in which every candidate topology is evaluated with an AC power flow. TTO is formulated as a multi-time-step problem, so that temporal coupling between switching actions is part of the search itself. On top of a baseline AC-exact GA, we examine a distance-biased mutation scheme that redistributes mutation probability toward switches electrically close to congested lines, and elitism and skip-bad-individual rules that reduce the number of power flow evaluations. To address the computational cost of AC evaluation at scale, we further examine three mixed-fidelity acceleration schemes that combine cheap DC pre-selection with AC power flow validation: a one-shot switch from DC to AC evaluation, a periodic DC/AC switching cadence, and an AC archive that tracks top-ranked, AC-verified individuals alongside a DC-evaluated population.

The setups are benchmarked on the IEEE case118 and case2869pegase grid models, expanded into double-busbar substations, over multi-time-step congestion windows constructed to test both stationary and moving congestion patterns. Setups are compared under an equal wall-clock budget, using feasibility rate, number of switching actions, worst-case loading and voltage margins, and runtime and power-flow-evaluation counts as KPIs. This provides a first systematic evaluation of AC-exact GA search for busbar-based congestion management across multiple time steps and realistic grid sizes, and practical guidance on the trade-off between DC-accelerated throughput and AC-exact solution quality for TSOs seeking to deploy automated topology optimization at scale.