基于代价敏感集成学习的战场态势评估模型  

Battlefield Situation Assessment Model Based on Cost-sensitive Ensemble Learning

作　　者：霍士伟 田八林 郭圣明[3] 唐宇波[3] HUO Shiwei;TIAN Balin;GUO Shengming;TANG Yubo(Graduate School,National Defense University,Beijing 100091;School of Information and Communication,National University of Defense Technology,Xi'an 710106;School of Joint Operations,National Defense University,Beijing 100091)

机构地区：[1]国防大学研究生院,北京100091 [2]国防科技大学信息通信学院,西安710106 [3]国防大学联合作战学院,北京100091

出　　处：《舰船电子工程》2021年第12期75-78,165,共5页Ship Electronic Engineering

基　　金：装备预研项目(编号:41401050201)资助。

摘　　要：针对误分代价不平衡条件下基于机器学习的战场态势评估误分代价较高问题,提出基于代价敏感集成学习的战场态势评估模型。以具有良好非线性建模能力的BP神经网络模型为基础分类器,通过AdaCost代价敏感集成学习方法综合考虑误分类代价对BP神经网络进行集成训练,使模型具有代价敏感特性。实验表明,在误分代价不平衡条件下,相对于单独的BP神经网络模型来说,所提模型在评估准确率和误分总代价方面都有较大优势。Aiming at the problem that the misclassification cost of battlefield situation assessment based on machine learning is high under the condition of unbalanced misclassification cost,a battlefield situation assessment model based on cost-sensitive en⁃semble learning is proposed.Based on the BP neural network model with good nonlinear modeling ability,the cost-sensitive ensem⁃ble learning method of AdaCost is used to train the BP neural network to make the model cost sensitive.The experimental results show that the proposed model has great advantages in the evaluation accuracy and the total cost of misclassification compared with the single BP neural network model under the condition of unbalanced misclassification cost.

关 键 词：态势评估 误分代价 代价敏感集成学习 神经网络 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]