[Paper] Exact and Evolutionary Algorithms for Sequential Multi-Objective Transmission Topology Planning

Source: arXiv - 2605.03753v1

Overview

The paper tackles day‑ahead transmission topology planning – the process TSOs use to decide which lines to switch on or off to keep the grid secure while minimizing operational hassle. By framing the task as a sequential, multi‑objective optimization problem, the authors deliver two complementary solution methods: an exact enumeration technique (the block algorithm) and a tailored evolutionary heuristic (a NSGA‑III‑based approach). Their work shows that, even on a real‑world high‑voltage network, the exact method can deliver a full Pareto front in minutes, providing a practical decision‑support tool and a benchmark for future heuristics.

Key Contributions

• Formal multi‑objective model with four realistic TSO criteria:
  1. Worst‑case line loading under N‑1 security,
  2. Topological depth (how far the network deviates from the reference topology),
  3. Number of switching actions,
  4. Cumulative time spent in non‑reference topologies over a 24‑h horizon.
• Block algorithm – an exact enumeration method that exploits the temporal block structure of feasible switching schedules, yielding polynomial‑time growth in the number of evaluations for fixed depth and switch‑count limits.
• Problem‑specific NSGA‑III heuristic – includes structure‑guided initialization and custom variation operators that respect the topology‑planning constraints.
• Empirical validation on TenneT’s Dutch high‑voltage grid, demonstrating:
  • Full Pareto front computed in < 3 min for a heavily congested day.
  • Evolutionary algorithm converges close to the exact front but does not fully recover it.
• Benchmark dataset – the exact front serves as ground truth for future heuristic, learning‑based, or reinforcement‑learning approaches.

Methodology

1. Sequential Decision Model

   • The 24‑hour planning horizon is split into blocks (contiguous time intervals) where the network topology stays unchanged.
   • Within each block, a binary vector indicates which lines are switched. The sequence of blocks forms a feasible schedule.

2. Exact Block Algorithm

   • Enumerates all possible block partitions respecting user‑defined limits on depth and switch count.
   • For each partition, it solves a single‑block sub‑problem (a deterministic OPF under N‑1 security) to evaluate the four objectives.
   • Because the number of partitions grows polynomially with the horizon when depth/switch limits are fixed, the algorithm remains tractable for realistic day‑ahead horizons.

3. Evolutionary Heuristic (NSGA‑III)

   • Initialization: seeds the population with schedules derived from historical TSO actions and from random block structures.
   • Variation Operators: custom crossover and mutation that add, remove, or shift block boundaries while preserving feasibility (e.g., no simultaneous contradictory switches).
   • Selection: standard NSGA‑III non‑dominated sorting with reference points aligned to the four objectives.

4. Evaluation

   • Both methods call a security‑constrained optimal power flow (SC‑OPF) solver for each block to obtain worst‑case line loadings.
   • The authors use real load and generation forecasts from TenneT for a selected congested day.

Results & Findings

The exact method confirms that, for the test day, the trade‑off surface is relatively smooth, and the heuristic quickly homes in on the most relevant regions. However, the heuristic struggles to capture the extreme low‑switch, high‑security solutions that the exact algorithm identifies.

Practical Implications

• Decision‑support for TSOs: The block algorithm can be integrated into existing Energy Management Systems (EMS) to provide operators with a complete set of optimal switching schedules, enabling transparent trade‑off analysis between security, operational effort, and system stability.
• Benchmark for AI/ML research: The exact Pareto front serves as a gold‑standard dataset for training reinforcement‑learning agents or surrogate models that aim to approximate the multi‑objective landscape faster.
• Reduced computational burden: Because the exact method’s evaluation count is polynomial under realistic bounds, it can be run daily without requiring high‑performance clusters, making it viable for real‑time or near‑real‑time planning.
• Customizable objectives: The framework is flexible; TSOs can replace or augment the four objectives (e.g., adding emission targets or market‑price considerations) while retaining the same algorithmic backbone.
• Risk‑aware operation: By explicitly optimizing worst‑case N‑1 line loadings, the approach aligns with regulatory security standards, potentially lowering the need for costly post‑hoc remedial actions.

Limitations & Future Work

• Scalability to larger networks: While the block algorithm is polynomial for fixed depth/switch limits, the constants can become large for very high‑dimensional grids or when many switching actions are allowed.
• Heuristic gaps: The NSGA‑III variant, though fast, does not guarantee coverage of the entire Pareto front, especially in extreme regions; further refinement of variation operators or hybrid exact‑heuristic schemes could close this gap.
• Dynamic uncertainties: The current model assumes deterministic forecasts for load and generation; incorporating stochastic or robust optimization to handle forecast errors is an open avenue.
• Real‑time adaptation: Extending the method to handle intra‑day re‑optimization (e.g., after unforeseen outages) would increase its operational relevance.

Overall, the paper delivers a solid blend of theory, algorithmic innovation, and real‑world validation, offering a practical toolkit for modern transmission system operators and a benchmark for the next generation of AI‑driven grid optimization methods.

Authors

• Job Groeneveld
• Miguel Muñoz
• Jan Viebahn
• Alessandro Zocca

Paper Information

• arXiv ID: 2605.03753v1
• Categories: math.OC, cs.NE, eess.SY
• Published: May 5, 2026
• PDF: Download PDF