MGA Compass

Package for Discovery and Interactive Exploration of Near-Optimal Solutions


Sina Hajikazemi

Prof. Florian Steinke

Technical University of Darmstadt

28.09.2026

OpenMod Workshop, Freiburg, 28-29 September 2026

Near Optimal Solutions

Geometry


Objective

Determine the optimal capacity and operational schedule of these facilities at the minimum cost.

Question?

What are the other options if we allow the solution to be ε-suboptimal?

Electricity Hydrogen (H₂) Solar PV Wind Load Shedding Demand Battery Storage Electrolysis Turbine H₂ Storage
PV Capacity Wind Turbine Capacity Feasible solutions Near-optimal feasible solutions cost ≤ (1+ε) cost* -c

Network schematic based on PyPSA's single-node capacity expansion example.

Near Optimal Solutions

Motivation


Use cases

  • Preferable alternatives in political, social, or environmental terms.[1]
  • Robust solutions to weather variability[2] and cost uncertainty[3].
  • Participatory modeling that lets stakeholders explore trade-offs near the optimum.[4]
PV Capacity Wind Turbine Capacity Near-optimal feasible solutions cost ≤ (1+ε) cost* -c

[1] Neumann, Fabian, and Tom Brown. "The near-optimal feasible space of a renewable power system model." Electric Power Systems Research 190 (2021): 106690.

[2] Grochowicz, Aleksander, et al. "Intersecting near-optimal spaces: European power systems with more resilience to weather variability." Energy Economics 118 (2023): 106496.

[3] Neumann, Fabian, and Tom Brown. "Broad ranges of investment configurations for renewable power systems, robust to cost uncertainty and near-optimality." iScience 26.5 (2023).

[4] Vågerö, Oskar, et al. "Exploring near-optimal energy systems with stakeholders: a novel approach for participatory modelling." arXiv preprint arXiv:2501.05280 (2025).

Challenges

 


Discover

  • Unknown 𝒵
  • Inner approximation ℐ
  • Outer approximation 𝒪
𝒵 𝒪 ℐ 𝑑 zI zO 𝑑 zO 𝑑 ≤ 𝛿 ? terminate

Our method builds on ORACLE[1] that approximates 𝒵 with two polytopes.

However, it solves a costly bilevel problem to find 𝑑.

We propose G-ORACLE that replaces the bilevel problem with a geometric solution.

Runtime per iteration on a log scale: ORACLE rises to about 300 seconds within 50 iterations G-ORACLE stays below about 30 seconds per iteration through more than 300 iterations

Explore

PV Capacity Wind TurbineCapacity cost ≤ (1+ε) cost* X Y X Y

We propose MGA-Compass[2], an interactive real-time platform:

  • User defines a question
  • Platform solves the Navigation problem (≈ 2 s)
  • User traverses between the endpoints (instantaneous)

[1] Turan, Evren Mert, Stefano Moret, and André Bardow. "ORACLE: A rigorous metric and method to explore all near-optimal designs for energy systems." Computers & Chemical Engineering 210 (2026): 109630.

[2] Hajikazemi, Sina, and Florian Steinke. "Interactive exploration of near-optimal solutions of energy planning models." 22nd International Conference on the European Energy Market (EEM) (2026): 1–6.

Thank You For Your Attention!

QR code linking to the code repository

Code

QR code linking to the live demo

Live Demo