Optimizing the relationship between training load and physical health of volleyball players using GOT (graph of thoughts) graph-structure modeling

Abstract The relationship between volleyball players'training load and physical health is dynamic and individualized. However, current methods lack interpretable and adaptive modeling. Existing methods struggle to achieve interpretable pathway analysis based on Granger causality and dynamic personalized optimization under multivariate coupling. To address this issue, this study proposes a GOT (Graph of Thoughts)-based graph-structure model approach to support personalized training optimization. First, the athletes' daily training load and physiological response data are collected to construct a dynamic training-health relationship graph. Then, a thinking node generation mechanism based on the collaboration of expert rules and LLM (Large Language Model) is designed to form a multi-path reasoning network. Then, MCTS (Monte Carlo Tree Search) is used to search for the optimal decision path, dynamically generate load adjustment suggestions, and output an explainable reasoning chain. Experimental results statistically demonstrate that the model effectively improves individual adaptation of training load. The experimental group's Individualized Load Adaptation Score (ILAS, measuring load-physiology alignment) increased significantly from an initial 0.42 to 0.78. The normalized contribution of all CHI (Comprehensive Health Index,) indicators reached 2.42, significantly exceeding the control group's 0.93. The Granger causality coefficients for different paths at all time lags were significantly lower than those in the control group. The model also demonstrates good coaching understandability and human–computer synergy consistency (overall Decision Consistency Index: 0.69). This study provides a new, interpretable, and adaptive paradigm for intelligent sports training.