Transmission Topology Optimization using accelerated MapElites

Abstract: Transmission Topology Optimization has great potential to improve efficiency and flexibility of grid operations through non-costly switching actions, but previous approaches struggle with runtime performance and scalability. In this work, we present an optimization approach that leverages GPU acceleration to speed up computations. In a genetic algorithm setting, topologies are randomly mutated and evaluated in parallel for multiple optimization criteria. Combined with a fully GPU-native DC loadflow solver, there is no CPU-GPU data transfer required in the DC optimization loop. Using a variant of the illumination algorithm MapElites, we efficiently generate a set of diverse candidate solutions on the pareto front. Together with an importing and AC validation step, we present an end-to-end optimization solution that runs in under 15 minutes. The approach is currently under evaluation by operational planning operators in two European TSOs. We furthermore open-source our code at github.com/eliagroup/ToOp.

• The paper introduces a GPU-accelerated MapElites method that processes up to 768,064 candidate topologies in minutes.
• It combines DC optimization with rigorous AC validation to ensure operational feasibility under stringent N–1 security standards.
• Results show significant congestion relief and suggest strong potential for industrial-scale deployment in modern power systems.

Transmission Topology Optimization using Accelerated MapElites: An Expert Review

Problem Context and Motivation

Transmission Topology Optimization (TTO) is a significant paradigm in modern power system operations for mitigating congestion and improving grid flexibility without incurring operational costs. The ever-increasing spatial mismatch between renewable generation and load centers, particularly in countries like Germany, exacerbates transmission bottlenecks and necessitates costly redispatch interventions. Grid expansion is fundamentally slow; thus, TTO via non-costly switching actions (e.g., branch disconnections, busbar splits, and asset reassignments) provides an efficacious alternative. Nevertheless, practical deployment is hampered by the combinatorial intractability of the problem (NP-hardness), real-time operational constraints, and the need for compliance with multi-dimensional operational security criteria.

Methodological Advances

End-to-End GPU Pipeline

The paper introduces a fully GPU-accelerated workflow, comprising:

• Importer: Preprocessing steps for CGMES grid model loading, action space enumeration using Tarjan's algorithm for bridge detection, and PTDF matrix computation for DC flow approximation.
• DC-Optimizer: An accelerated MapElites variant implemented entirely in GPU memory, leveraging the QDax framework for quality-diversity optimization and extensive batch-parallel evaluation of tens of thousands of candidate solutions per second.
• AC-Validator: Post-selection AC power flow validation utilizing powsybl and polars frameworks with CPU parallelization, worst-k contingency case filtering, and pruning heuristics to ensure operational feasibility under N-1 security.

Quality-Diversity Optimization

Advancing beyond traditional genetic algorithmic frameworks, the authors deploy MapElites to systematically illuminate the Pareto front across operational intervention descriptors. Descriptor dimensions (number of splits, disconnections, and reassignments) are used to bin candidate solutions, allowing diverse high-performing behaviors to coexist. This is a salient move in real-world TTO, where trade-offs between aggressive topology actions and operational risk cannot be adequately encapsulated by single-objective optimization.

Mutation and crossover mechanisms are rigorously implemented with feasibility constraints to ensure only valid topologies are perpetuated. Fitness heuristics combine DC-acquired overload metrics and critical branch counts, augmented by "do-not-make-it-worse" constraints for busbar outage behavior.

Numerical Performance and Results

Scalability and Efficiency

The approach efficiently traverses the exponential search space, evaluating up to $768{,}064$ topologies in three minutes for one grid scenario, with end-to-end optimization (including AC validation) completed within the 15-minute operational requirement. The majority of the runtime is absorbed by preprocessing (Importing stage), with hardware-accelerated DC optimization and selective AC validation introducing manageable bottlenecks. Notably, the DC optimization alone achieves throughput orders of magnitude higher than legacy approaches.

Solution Quality

Congestion relief is substantial. For the TSO 1 benchmark, a pre-optimization overload energy of $1269\,\mathrm{MW}$ was reduced only via topologies with at least two splits or disconnections, highlighting the non-additive nature of topology actions. In TSO 2, the best topology achieved $905\,\mathrm{MW}$ overload (down from $2637\,\mathrm{MW}$), with practical preference for simpler solutions due to operational risk considerations. AC validation of DC-proposed candidates revealed a relatively low acceptance rate (16.9%), primarily due to convergence issues (59.1%) and DC–AC modeling mismatches.

Evolution Dynamics

Rapid improvement in fitness is observed in early optimization stages, with later stagnation; the initial seeding of the repertoire is suboptimal, suggesting future methodological refinements in initialization strategies.

Contradictory and Bold Claims

• Industrialization Claim: The authors assert this is the first full-scale industrial deployment of TTO at any European TSO using a GPU-accelerated quality-diversity search, supported by ongoing operational evaluation.
• Non-Additivity Observation: The paper demonstrates empirically that certain topology actions are only effective in combination, contradicting assumptions implicit in linear optimization models and underscoring the necessity for multi-action search strategies.

Practical and Theoretical Implications

The study delivers a viable operational tool for rapid, non-costly congestion relief in transmission grids, making it feasible for integration with day-ahead planning and intra-day operational workflows. By providing a spectrum of candidate topologies across the Pareto front, system operators can exercise informed risk-benefit trade-offs, moving away from manual, experience-driven interventions.

Theoretically, the fusion of hardware-accelerated quality-diversity search with operational grid constraints exemplifies a new paradigm for solving NP-hard power system problems. The AC validation bottleneck exposes fundamental DC–AC modeling discrepancies, highlighting the need for improved surrogate modeling (e.g., DC+) and feedback mechanisms between optimization stages.

Future Directions

Prospects for enhancement include:

• GPU-Accelerated AC Validation: Leveraging recent developments in GPU-based AC loadflow could significantly reduce runtime overhead [47].
• Integrated Redispatch Optimization: Joint optimization of topology and redispatch could improve feasibility and alleviate non-convergence, potentially leading to significant cost savings.
• Feedback from AC to DC Stages: Introducing tighter DC+ approximations or iterative feedback could mitigate DC–AC mismatch-induced rejection rates [48].
• Extended Non-Costly Action Set: Inclusion of PST/HVDC setpoint adjustments and maintenance cancellations would broaden the operational impact.
• Automated Recommender Integration: Seamless integration with human-in-the-loop decision systems is necessary to respect operational timings and workflow constraints.

Conclusion

The paper delivers a rigorous, scalable, and interpretable solution for large-scale transmission topology optimization, underpinned by GPU-accelerated MapElites and AC validation. Empirical results confirm substantial reductions in congestion, illumination of diverse operationally feasible topologies, and compliance with stringent runtime requirements. While DC–AC bottlenecks and convergence limitations remain, the methodology provides a foundation for industrial-scale deployment and further advances in power system optimization algorithms. The openly released code base accelerates the transfer of research findings into operational practice and future methodological extensions.

Citation: "Transmission Topology Optimization using accelerated MapElites" (2605.10128)