Computing Necessary Conditions for Near-Optimality in Capacity Expansion Planning Problems

Antoine Dubois, Damien ErnstarXiv 2021 · arXiv

Introduction

In large-scale energy planning, the standard approach is to solve for the cheapest way to build a carbon-neutral grid. While optimization models provide a precise plan, real-world implementation often faces political, social, or technical hurdles. Decision-makers frequently require a range of options rather than a single point estimate. They need to understand which components of a plan are non-negotiable and which are flexible. For instance, if a 5% cost increase is acceptable, is a controversial new transmission line still essential? This paper addresses the need to identify essential infrastructure across all nearly-optimal solutions.

Methods

The proposed framework centers on the concept of $ε$-optimality.

  1. Defining the Space: The "$ε$-optimal feasible space" ($Χ^ε$) encompasses all grid configurations with a total cost no higher than $(1 + ε)$ times the absolute minimum cost $f(x^*)$.
  2. Identifying Necessary Conditions: A condition is considered "necessary" if it is satisfied by every solution within $Χ^ε$.
  3. Non-Implied Conditions: The focus is on finding the tightest possible bounds (e.g., the maximum "minimum required capacity") to provide actionable information.
  4. Algorithmic Approach: For conditions involving sums of variables, such as total regional solar capacity, the necessary condition is found by minimizing that sum over the $ε$-optimal space.
Illustration of the ε-optimal feasible space Χε around the optimal solution x*.

Case Study: The European Grid

The methodology was applied to a carbon-neutral 2050 European electricity grid model using PyPSA-Eur. The model utilized the TYNDP18 reference grid, weather data from Renewables.ninja, and demand profiles from the Open Power System Data project.

Key Insights

The analysis reveals significant flexibility in certain infrastructure categories while highlighting others as strictly necessary.

  • The Transmission "Plateau": Total required transmission investment decreases sharply with slight suboptimality. Allowing for a 10% cost deviation reduces the "necessary" new transmission capacity by approximately half.
  • National vs. European Necessity: While the aggregate European grid requires expansion, individual lines are rarely strictly necessary. No single line must be expanded to stay within 2.5% of the optimal cost, indicating high spatial flexibility.
  • Optionality of Storage: While the cost-optimal solution includes 249 GW of Li-ion batteries, the necessary condition for battery capacity drops to zero at just 5% suboptimality. This suggests that while batteries are cost-effective, they can be replaced by other flexibility options at a relatively low cost premium.
  • Non-Negotiable Renewables: In contrast to storage, a massive amount of renewable generation remains necessary even at 20% suboptimality. Specifically, at least 600 GW of combined wind and solar is required, with offshore wind proving particularly resilient across the $ε$ range.
Necessary capacity requirements for various technologies as a function of the suboptimality coefficient ε.

Results Summary

  • Transmission: The optimal plan adds 146 TWkm of lines; only ~70 TWkm is strictly necessary at 10% suboptimality.
  • Renewables: The optimal plan calls for 862 GW of combined capacity; at least ~600 GW is required even at 20% suboptimality.
  • Regional Variation: France and Germany show the highest "necessary" transmission needs, whereas needs for Great Britain and Norway drop to zero at very low $ε$ levels.

Conclusion and Future Work

This framework provides a robust tool for identifying "no-regret" investments in energy planning. Future research will focus on "meta-necessary conditions" that account for uncertainties in technology costs and weather data, as well as expanding the framework to non-linear conditions and multi-objective optimization (e.g., land use or direct CO2 minimization).

Summary updated

BibTeX
@misc{dubois2021computing,
  author = {Antoine Dubois and Damien Ernst},
  title = {{Computing Necessary Conditions for Near-Optimality in Capacity Expansion Planning Problems}},
  year = {2021},
  eprint = {2109.14272},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2109.14272},
}
Figure