基於機器學習模型與基因演算法優化對高性能水下寬帶吸音超材料的設計與優化

Abstract

本研究使用前人歸納出的多種吸音機制，提取複數機制結合後的模型特徵，並進行基因演算法優化，使用目標函數對初始結果進行轉換之後帶入進化/變異算法，找出優化的特徵組合為何，但由於基因驗算法優化後的結果為一個數值，無法得知真實的吸音表現，所以建立由機器學習所建構而成的替代模型，以便快速了解優化後的特徵組合具有何種吸音表現，最後藉由此套方法開發出在2 kHz ~ 20 kHz 範圍中，皆具有高達0.9以上吸音係數的高性能水下吸音超材料，同時解決了在吸音超材料中無法兼顧高低頻吸音的問題，達到寬帶吸音。

This study utilizes various absorption mechanisms identified by previous research to extract combined features of multiple mechanisms. The extracted features are then optimized using a genetic algorithm, where an objective function is applied to transform the initial results and input them into an evolutionary/mutation algorithm. The goal is to identify the optimized combination of features. However, since the results of the genetic algorithm optimization are numerical values, the actual absorption performance cannot be determined. Therefore, an alternative model constructed by machine learning is established to quickly understand the absorption performance of the optimized feature combination. Finally, using this approach, a high-performance underwater absorption metamaterial is developed with absorption coefficients exceeding 0.9 across the range of 2 kHz to 20 kHz. This solution simultaneously addresses the challenge of achieving broadband absorption in metamaterials, which typically struggle with absorbing both high and low frequencies.