鱼类集群运动的注意力模型研究

Research on the Attention Model of Fish Swarm Movement

自然界中的生物集群运动现象十分普遍,研究生物集群行为是人类控制复杂集群系统的关键。针对真实鱼群运动数据,借鉴注意力机制,设计出一种适用于鱼群运动分析的注意力模型。通过对自身和邻居的观测信息进行有效编码,构建注意力网络模型以获取适宜的注意力权重,利用注意力权重加权平均邻居的编码信息并进行解码,输出单体的运动决策。将该模型与真实鱼类的宏观运动特性进行比较,并采用多智能体仿真实验进行验证。实验结果表明,所提模型能有效驱动智能体模拟鱼群行为,为大规模实现多智能体集群运动奠定了基础。

The movement of biological clusters is very common in nature,and studying the behavior of biological clusters is the key for human beings to control complex systems of clusters.In view of real fish swarm motion data,we design an attention model suitable for fish swarm motion analysis by referring to the attention mechanism.By effectively encoding the observed information of self and neighbors,the attention network model is constructed to obtain the appropriate attention weight,and the coded information of neighbors is weighted and decoded by the attention weight,and the single motion decision is output.The macroscopic motion characteristics of the model were compared with those of real fish,and verified by multi-agent simulation experiments.The experimental results show that the proposed method is consistent with the experimental results of real fish,and can effectively drive agents to simulate fish swarm behavior,which lays a foundation for the large-scale realization of multi-agent swarm motion.