Design optimization of post–beam–panel mass timber frame systems using Monte Carlo Tree Search and policy neural network

Mass timber construction offers significant sustainability benefits, but its adoption is limited by the high cost of engineered timber components. This study develops a Monte Carlo Tree Search with policy neural network (MCTS+NN) optimization framework to minimize the material cost of post-beam-panel mass timber systems under gravity loads. The proposed framework formulates the structural design process as a Markov Decision Process and integrates a policy neural network to guide the MCTS search. Its performance is evaluated using three building design scenarios together with a parametric study of the MCTS + NN algorithmic hyperparameters, namely neural-network training frequency and exploration rate. The results indicate that more frequent neural-network training combined with lower exploration rates provides the best optimization performance. Under the reported experimental conditions and equivalent computational effort, the proposed MCTS + NN framework achieved higher objective values than conventional MCTS, demonstrating the potential of the proposed methodology for structural design optimization.