摘要
[目的]为使舰艇能在短时间内正确预判空中来袭目标的意图,提出应用异质集成学习器解决该模糊不确定性分类问题。[方法]首先选取极限学习机、决策树、Skohonen神经网络和学习矢量化(LVQ)神经网络4种子学习器,使用集成学习结合策略构建异质集成学习器;然后利用该集成学习器训练测试训练集100次,得到该分类实验平均准确率和计算时间。为提高准确率,进行了集成修剪,剔除"劣质"的LVQ神经网络,重新构建效率更高的异质集成学习器,其实验结果具有极高的精度,但计算耗时长。为此,提出对Skohonen神经网络子分类器做"线下训练、线上调用"的改进。[结果]仿真实验表明,从探测到空中目标到预判出各来袭目标意图总用时为4.972 s,预判精度为99.93%,很好地满足了精度和实时性要求。[结论]该研究为作战决策提供了一种新颖而有效的方法,同时也为小样本分类识别问题提供了一种较好的实现途径。
[Objectives]This paper proposes a heterogeneous integrated learner to solve the problem of fuzzy uncertainty classification in order to judge the target intention of air attack in a short time.[Methods]First,a limit learning machine,decision tree,Skohonen neural network and LVQ neural network are selected to construct the heterogeneous integrated learner using the integrated learning strategy. Next,the training program is trained 100 times using the integrated learner to obtain the classification experiment average accuracy and calculation time. In order to improve the accuracy,integrated pruning is carried out to eliminate the "poor quality" LVQ neural network,and a more efficient heterogeneous integrated learner is reconstructed. The experimental results are extremely accurate but the calculation is time-consuming. In this paper,the Skohonen neural network sub-classifier is proposed as an "offline training and online call".[Results]Simulation experiments show that the time consumed from detecting the air targets to prejudging the intention of each incoming target is 4.972 s with an accuracy of 99.93%,which is excellent for meeting accuracy and real-time requirements.[Conclusions]This study provides a new and effective method for air defense decision-making. The method used in this paper also provides a better way of realizing the classification problem of small samples.
出处
《中国舰船研究》
CSCD
北大核心
2018年第1期133-139,共7页
Chinese Journal of Ship Research