基于序列趋势和集合距离的UAV态势相似性度量方法

在无人机（UAV）自主作战与智能决策的过程中，无人机获取的新态势与历史态势的相似性度量是态势评估与作战决策的重要环节，而现有的相似性度量方法主要处理离散时刻态势，采用的欧式距离等方法对数据敏感，不符合作战态势特性，且处理效率低下，针对该问题，提出基于序列趋势和集合距离的UAV态势相似性度量方法。该方法首先选取UAV作战态势要素并以时间序列形式表征数据；然后，使用经验模态分解方法提取历史态势与新态势的序列趋势以度量序列趋势的相似性；最后，对趋势相似的每条态势序列进行自组织映射聚类，得到若干聚类中心构成集合，利用最优子模式分配距离度量集合间的距离，提取集合距离较小的部分获得与新态势相似的历史经验态势。通过公用数据集的分类效果比对实验以及作战仿真态势的相似性度量实验，表明该方法能够有效度量两序列之间的相似性程度，度量效果好，分类实验中分类精度较传统方法最高提高18%，且方法简便，具有一定的实用价值。

Measurement of UAV situation similarity based on sequence trend and set distance

In the process of autonomous combat and intelligent decision-making of Unmanned Aerial Vehicles (UAVs), the similarity measure of the new situation and historical situation acquired by drones is an important link in situation assessment and operational decision-making. However, the existing similarity measurement methods mainly deal with discrete moments. The Euclidean distance method used in the existing methods shows low processing efficiency and is data-sensitive that does not fit the characteristics of the combat posture. To solve this problem, a multi-granularity UAV situation similarity measure method is proposed. This method first selects the UAV combat posture elements and characterizes the data in time series. Then, the empirical modality decomposition method is used to extract sequence trends of historical experience and new situation to measure their similarity. Finally, the Self-Organization Map (SOM) clustering is performed for each situation sequence with similar trend, gaining sets formed by a number of cluster centers. Using the optimal subpattern assignment distance to measure the distance between the sets, we extract the part with a small set distance to obtain a historical experience similar to the new situation. The comparison of the classification effect of common datasets and the similarity measurement experiment of the combat simulation situation shows that the proposed method can effectively measure the degree of similarity between two sequences with a good measurement effect. The classification accuracy in the classification experiment is 18% higher than the traditional method. The method is simple and has certain practical values.