针对动物领导关系的可变寻正时滞传递熵  

Variable positive-lag transfer entropy for animal leadership

作　　者：周立方 万亚平[1,2] ZHOU Li-fang;WAN Ya-ping(School of Computer Science,University of South China,Hengyang 421001,China;Hunan Provincial Base for Scientific and Technological Innovation Cooperation,China Hunan Provincial Science and Technology Department,Hengyang 421001,China;Research Center of Intelligent Manufacturing,Aerosun Corporation,Nanjing 211100,China)

机构地区：[1]南华大学计算机学院,湖南衡阳421001 [2]湖南省科技厅湖南省医疗大数据国际科技合作基地,湖南衡阳421001 [3]航天晨光股份有限公司智能制造研究中心,江苏南京211100

出　　处：《计算机工程与设计》2024年第9期2785-2795,共11页Computer Engineering and Design

基　　金：国防科技创新特区基金项目(17-163-15-XJ-002-002-04);湖南省2020年度创新型省份建设专项抗击新冠肺炎疫情应急专题基金项目(2020SK3010);十三五核能开发基金项目。

摘　　要：为改进可变时滞传递熵(VL传递熵),提出一种可变寻正时滞传递熵(VPL传递熵)。根据提出的时滞寻正设想,重新确定用来矫正DTW时滞序列的最佳固定时滞,用经矫正后的最佳时滞序列对时序进行“扭曲”,进行因果推断。在两个公开的动物运动数据集中进行领导关系因果推断,实验结果表明,VPL传递熵因果推断的准确度比VL传递熵提升了75%至100%。所提方法在动物群体领导关系发现上优于VL传递熵。To improve the variable-lag transfer entropy(VL transfer entropy),the variable positive-lag transfer entropy(VPL transfer entropy)was proposed.The optimal fixed time delay was redefined for correcting DTW time delay sequences based on the proposed idea of time delay correction.The corrected optimal time delay sequence was used to distort the time series,followed by the causal inference.The causal inference of leadership relationships was conducted in two publicly available animal movement datasets.Experimental results show that the accuracy of causal inference for VPL transfer entropy is improved by 75%to 100%compared to that of VL transfer entropy.The proposed method outperforms VL transfer entropy in discovering animal population leadership relationships.

关 键 词：因果推断 领导机制 时间序列 视野感知模型 动态时间规整算法 传递熵 可变时滞传递熵 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]