Hybrid Method for Optimizing the Structure of Power Generation Capacities Using An Evolutionary Algorithm and An MILP Solver

Abstract

A hybrid method is proposed for solving large-scale mixed-integer linear programming (MILP) problems that arise in the optimization of the generation capacity structure of electric power systems combining conventional and renewable energy sources. The novelty of the proposed approach lies in combining the decomposition of the original generation expansion planning problem into investment-level search and operational-level evaluation with evolutionary exploration of the discrete space of generation capacity decisions and exact parallel solution of operational MILP subproblems on high-performance computing (HPC) resources. This combination enables detailed operational evaluation without replacing the operational model with a surrogate approximation. Numerical experiments using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and the Solving Constraint Integer Programs (SCIP) solver showed that the global optimum was achieved in 98% of runs for a population size of 80 and that, compared with the single-threaded SCIP baseline, the 128-thread configuration reduced the average runtime for successful runs by 3.77×. At the same time, scaling with respect to computation time and memory usage exhibited nearly linear behavior. A comparison with the alternative Evolutionary Centers Algorithm demonstrated the superiority of CMA-ES in convergence reliability. The obtained results indicate that the proposed method is well-suited for deployment in HPC environments and can serve as an effective tool for strategic planning of electric power system development. Future research should focus on extending the applicability of the approach to multi-period planning problems and analyzing the impact of algorithm parameters on convergence and solution accuracy.

Received: 30 August 2025 | Revised: 24 March 2026 | Accepted: 19 May 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.20214170.

Author Contribution Statement

Sergii Saukh: Conceptualization, Methodology, Validation, Formal analysis, Resources, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Taras Puchko: Methodology, Software, Investigation, Data curation, Writing – review & editing, Visualization.