A knowledge‐enhanced deep reinforcement learning‐based shape optimizer for aerodynamic mitigation of wind‐sensitive structures

Structural shape optimization plays an important role in the design of wind‐sensitive structures. The numerical evaluation of aerodynamic performance for each shape search and update during the optimization process typically involves significant computational costs. Accordingly, an effective shape optimization algorithm is needed. In this study, the reinforcement learning (RL) method with deep neural network (DNN)‐based policy is utilized for the first time as a shape optimization scheme for aerodynamic mitigation of wind‐sensitive structures. In addition, “tacit” domain knowledge is leveraged to enhance the training efficiency. Both the specific direct‐domain knowledge and general cross‐domain knowledge are incorporated into the deep RL‐based aerodynamic shape optimizer via the transfer‐learning and meta‐learning techniques, respectively, to reduce the required datasets for learning an effective RL policy. Numerical examples for aerodynamic shape optimization of a tall building are used to demonstrate that the proposed knowledge‐enhanced deep RL‐based shape optimizer outperforms both gradient‐based and gradient‐free optimization algorithms.