基于图网络的集群运动预测研究

集群动力学是软物质领域的研究热点和前沿视角，集群运动的同步机制存在着丰富的潜在规律和应用价值。该文构建了一种基于加权集群动力学的图网络模型，该模型从粒子的位置、运动方向特征以及邻居的影响中学习集群运动的演化机制，可以实现对集群运动过程的长期预测。实验结果表明，图网络模型可以对集群运动过程中的序参量进行预测，涵盖不同的噪声和视野半径，预测效果较好。模型构建后，无需进行复杂的动力学模拟和计算，就可以得到不同条件参数下集群运动系统序参量的值，从而快速量化集群运动的同步程度，节省时间成本，对集群的智能控制具有重要意义。

Research on Collective Movement Prediction Based on Graph Network

Collective dynamics is a research hotspot and frontier perspective in the field of soft matter. The synchronization mechanism of collective movement has rich potential laws and application values. This paper constructs a graph network model based on weighted collective dynamics, which learns the evolution mechanism of collective movement from the position of particles, the movement direction and the influence of neighbors, and can realize long-term prediction of the evolution movement process of the collective movement. Results show that the graph network model can predict the order parameters of the movement process, covering different noises and field of view radius, and the prediction effect is better. After the model is constructed, the value of the order parameter of the system can be obtained without complex dynamic simulation and calculation, so as to quickly quantify the synchronization degree of the collective movement, which can save time and cost, and have important significance for the intelligent control of the collective.