@article{Liu2026, 
author = {Wanli Liu and Yuzhou Chen and Xin Liu and Manqing Fu and Zhi Li and Shiping Tang},
title = {Genetic-dynamic time warping strategy preference fitting in multi-party crisis games},
year = {2026},
journal = {Journal of National University of Defense Technology},
volume = {48},
number = {5},
pages = {120-131},
keywords = {agent-based modeling, computer simulation, genetic algorithm, dynamic time warping, unconventional complex games, international conflict management},
url = {https://www.sciopen.com/article/10.11887/j.issn.1001-2486.24040006},
doi = {10.11887/j.issn.1001-2486.24040006},
abstract = {ObjectiveThe study aims to develop an innovative multi-agent crisis game simulation model that combines ABM (agent-based modeling) with the Genetic-DTW (genetic-dynamic time warping) strategy fitting method to address complex game and decision-making predictions in multi-party crisis. The model particularly seeks to overcome limitations of traditional game-theoretical models in settings with multiple agents, diverse strategies, and incomplete information.MethodsThe study combined historical data with machine learning, using ABM to construct intelligent agent behavior models. By introducing the Genetic-DTW strategy fitting method, agent preferences within the model are calibrated to allow dynamic strategy adjustment in crisis simulations. The model was trained and calibrated based on real-world international crisis scenarios to optimize agents’ strategies to approximate their historical counterparts within multi-party games.ResultsThe model we proposed demonstrates significant potential in simulating and predicting crisis dynamics through the integration of state transition models, behavior preference models, and issue preference matrices. Using the “Genetic-DTW” algorithm, we optimize behavioral and issue domain preference parameters on historical data. In the subsequent state sequence prediction test, our model achieves an average DTW distance of 9.35 in testing, outperforming the baseline model’s 14.60. Ablation experiments confirm the contributions of each component, with DTW distances decreasing from 29.84 (base model) to 12.40 (state transition), 10.09 (behavior preference), and 9.35 (full model). In tests, the model is able to capture major shifts in crisis intensity, reconstructing macro-level dynamics from micro-level behavior rules. These results highlight the model’s potential for understanding strategic interactions and predicting crisis trajectories.ConclusionsOur model innovatively fills a gap in existing research on multi-agent crisis game simulations, enhancing decision-making simulation and crisis prediction in complex crisis scenarios through a data-driven ABM modeling approach. This study not only provides a new solution for multi-agent crisis games but also introduces a fresh perspective for future research in international relations and conflict management. Future work will further expand the model’s applications to other scenarios, such as multilateral negotiations, economic sanctions, and regional ethnic conflicts.}
}