Graph-based reinforcement learning for discrete cross-section optimization of planar steel frames

A combined method of graph embedding (GE) and reinforcement learning (RL) is developed for discrete cross-section optimization planar steel frames, in which the section size each member selected from a prescribed list standard sections. The RL agent aims to minimize total structural volume under various practical constraints. GE extracting features data with irregular connectivity. While most existing methods aim at node features, an improved formulation edges associated members this study. Owing proposed operations, capable grasping property columns beams considering their connectivity frame arbitrary as feature vectors same size. Using vectors, trained estimate accurate return action take proper actions on reduce or increase using algorithm. applicability versatile because frames different numbers nodes can be used both training application phases. In numerical examples, agents outperform particle swarm benchmark terms computational cost design quality cross-sectional changes; successfully assign reasonable cross-sections geometry, connectivity, support load conditions frames.