<p>Flyrock caused by mine blasting is a harmful phenomenon. Effectively controlling flyrock distance remains a persistent challenge in blasting operations. This paper proposes an intelligent framework, which integrates the theory-guided machine learning method and the improved slime mould algorithm, for flyrock distance prediction and blast parameter optimization. First, due to difficulties in collecting blasting data and the limited generalization ability of prediction models under data scarcity, a conditional tabular generative adversarial network CTGAN was used to augment the original dataset. Statistical analysis indicated that the synthetic data closely matched the real data in key statistical properties. Second, to overcome the limitations of conventional data-driven machine learning models particularly their low robustness and limited interpretability, two theory-guided approaches including a hybrid model integrating empirical theory with data-driven models, and a loss function informed by domain knowledge, were proposed. These approaches were used to validate the effectiveness of synthetic data in flyrock distance prediction. Subsequently, the two theory-guided machine learning models were integrated to exploit their complementary capabilities. The performance of the hybrid model was rigorously evaluated via four distinct regression metrics, and the most accurate predictor for flyrock distance was identified. Finally, the slime mould algorithm was enhanced according to the characteristics of blasting operations and was then applied to optimize blasting parameters. The optimized parameters led to a 23.05% reduction in flyrock distance compared to the unoptimized method. The results demonstrated the effectiveness of the proposed framework in mitigating flyrock hazards and suggest its potential applicability to related engineering problems.</p>