To analyze and predict decision-making in complex multi-party crisis game scenarios, a novel strategy preference fitting method that integrates agent-based modeling with the “Genetic-DTW (genetic-dynamic time warping)” algorithm is proposed. By fitting historical time-series data from intelligent agents in multi-party crisis game, this study simulates the decision-making processes of international actors during crises and predicts system state transitions. Combining expert knowledge with machine learning, our model achieves promising results in capturing strategy preferences of actors in multi-party games, obtaining an average DTW distance of 9.35 in historical case state sequence prediction tasks. This approach provides an innovative research path for understanding multi-party crisis decision-making. The proposed modeling and calibration methods can be widely applied to other complex multi-party game scenarios, including multilateral multi-round negotiations, multilateral economic sanctions, and regional conflicts.