Evolutionary Algorithms Using Pairwise Ranking Machine Learning Models for Expensive Optimization Problems

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K17826
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61040:Soft computing-related
• Research Institution: Saitama University (2023); Tokyo Metropolitan University (2021-2022)
• Principal Investigator: Harada Tomohiro 埼玉大学, 理工学研究科, 准教授 (40755518)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 進化計算 / 機械学習 / 最適化

Outline of Final Research Achievements

This research proposed a new binary classification surrogate model called ELDR for evolutionary algorithms (EAs). By applying the EA using ELDR to constrained and unconstrained optimization problems, this research demonstrated superior performance compared to existing methods, with particularly significant performance improvements for high-dimensional problems. This research applied ELDR to a real-world optimization problem and showed that it enables better designs with fewer evaluations than conventional methods. This research achieved a wide range of results, including the proposal of ELDR, the establishment and validation of the effectiveness of EAs using ELDR, and the application to real-world problems.

Free Research Field

計算知能

Academic Significance and Societal Importance of the Research Achievements

本研究は，進化的アルゴリズムによる効率的な解探索のために，少ないデータ数で高精度に評価値を推定可能な新しい二値分類型サロゲートモデルELDRを提案し，その有効性を検証した．実世界の多くの最適化問題は，解候補の評価にシミュレーションや複雑な数値計算を用いるため評価コストが高く，最適解の獲得までに莫大な計算時間を要する．本研究の研究成果によって，このような実世界の高コストな最適化問題に対して，従来より少ない評価回数で高品質な解を効率的に獲得できることが期待される．