| 引用本文: | 杨任农,张振兴,房育寰,等.深度置信网络在导弹攻击区分类中的应用.[J].国防科技大学学报,2019,41(2):98-106.[点击复制] |
| YANG Rennong,ZHANG Zhenxing,FANG Yuhuan,et al.Application of deep belief network in classification of missile launch envelopes[J].Journal of National University of Defense Technology,2019,41(2):98-106[点击复制] |
|
| |
|
|
| 本文已被：浏览 6739次 下载 5904次 |
| 深度置信网络在导弹攻击区分类中的应用 |
| 杨任农, 张振兴, 房育寰, 左家亮, 张彬超 |
|
(空军工程大学 空管领航学院， 陕西 西安 710038)
|
| 摘要: |
| 针对传统导弹攻击区解算方法忽略双方态势变化等问题，提出运用深度置信网络的导弹攻击区分类模型。根据导弹命中情况与目标机动间的关系，将导弹攻击区划分为五类。通过分析影响导弹攻击结果的态势参数，构建导弹攻击结果预测模型。在实验部分，结合重构误差和测试错误率确定深度置信网络的网络结构，通过逐层提取数据法分析模型参数特征并且讨论微调数据的采样方式。使用反向传播神经网络和支持向量机进行分类有效性对比实验。实验结果表明：深度置信网络运行速度和预测准确度明显优于其他两种方法，满足实时性和准确性要求，所提方法具有良好的应用价值。 |
| 关键词: 导弹攻击区分类 深度置信网络 特征提取 微调数据采样 |
| DOI：10.11887/j.cn.201902015 |
| 投稿日期：2017-12-31 |
| 基金项目:航空科学基金资助项目（20155196022）；国家自然科学基金青年科学基金资助项目（71501184）；陕西省自然科学基金资助项目（2016JQ6050） |
|
| Application of deep belief network in classification of missile launch envelopes |
| YANG Rennong, ZHANG Zhenxing, FANG Yuhuan, ZUO Jialiang, ZHANG Binchao |
|
(Air Traffic Control and Navigation College, Air Force Engineering University, Xi′an 710038, China)
|
| Abstract: |
| Aiming at the problems such as neglecting the dynamic changes in the conventional missile attacking envelop solution, a missile attack envelop classification model based on deep belief network was proposed. The missile attack envelop was divided into five parts according to the relationship between missile hits and target maneuvers. By analyzing the situation parameters which affect the attack result of air-to-air missile, a missile attack prediction model was constructed. In the experiments, the reconstruction error and the test error rate were used to determine the network structure. Through extracting data layer by layer, the features of parameters were analyzed and the approaches of fine-tuning data sampling were discussed. Back propagation network and support vector machine were selected for classification comparison experiments. The results show that the deep belief network performs better than the other two algorithms in speed and prediction accuracy and the presented method is of great practical value. |
| Keywords: missile attack envelop classification deep belief network feature extraction fine-tuning data sampling |
|
|
|
|
|
