報(bào)告題目:A High-Fidelity PDEs-Embedded Reinforcement Learning Framework for Shape Optimization of Airfoils
報(bào) 告 人:胡光輝 (澳門(mén)大學(xué) 教授、博士生導(dǎo)師)
報(bào)告時(shí)間:1月18日 15:30-17:30
報(bào)告地點(diǎn):明理樓C302B
報(bào)告人簡(jiǎn)介:
胡光輝本科及碩士畢業(yè)于四川大學(xué)數(shù)學(xué)學(xué)院,博士畢業(yè)于香港浸會(huì)大學(xué)數(shù)學(xué)系,后在密歇根州立大學(xué)數(shù)學(xué)系從事博士后研究工作,目前為澳門(mén)大學(xué)科技學(xué)院數(shù)學(xué)系教授,博士生導(dǎo)師,研究領(lǐng)域?yàn)槠⒎址匠虜?shù)值方法的設(shè)計(jì)分析、算法的高性能實(shí)現(xiàn)、及在計(jì)算物理問(wèn)題中的應(yīng)用,發(fā)表研究論文60余篇,主持國(guó)自然優(yōu)青(港澳)、面上、澳門(mén)科技發(fā)展基金等多個(gè)項(xiàng)目,擔(dān)任中國(guó)數(shù)學(xué)會(huì)計(jì)算數(shù)學(xué)會(huì)常務(wù)理事, 為SCI雜志Commun. Comput. Phys.及Advan. Appl. Math. Mech.編委。
報(bào)告內(nèi)容摘要:
In this work, we present a novel framework that integrates a high-fidelity PDE solver directly into a reinforcement learning (RL) loop for aerodynamic shape optimization. Our approach introduces a task-specific reward function designed to recover continuous optimization targets and leverages steady Euler equations as the environment for the RL agent. The proposed method successfully optimizes an airfoil parameterized by 132 design variables within O(10^3) simulations, which is rarely attainable by gradient-free methods.
主辦單位:理學(xué)院
人工智能研究院
科學(xué)技術(shù)發(fā)展研究院