基于机器学习的导弹干扰试验效果评估实证研究

针对复杂电磁环境下导弹干扰试验影响因素众多,难以量化,试验数据采集困难以及实验数据中普遍存在类不平衡等问题,基于机器学习创建导弹试验干扰效果评估模型,采用随机森林、支持向量机、朴素贝叶斯、多层感知机等常见模型对导弹试验干扰效果进行评估。特别针对小数据样本中的类不平衡问题提出 2阶段分类模型,采用过采样方式解决类不平衡问题并采用随机森林进行分类。基于开源的导弹干扰效果评估数据,通过实证研究说明,基于过采样的随机森林模型在干扰效果评估问题中具有较强的泛化能力和鲁棒性,在 AUC指标上,该模型比多层感知机模型在中位数上最多提高 60%,建议在后续的试验中采用该模型进行导弹干扰效果评估。

Empirical Research on the Evaluation of Missile Jamming Test Effect Based on Machine Learning

In view of the problem that there are numerous influencing factors of missile jamming test in complex electro?magnetic environment, which are difficult to quantify, the data is difficult to collect and there is class imbalance problemin dataset, the evaluation model of missile jamming test effect based on machine learning is created . Models such as ran?dom forest, support vector machine, Naive Bayes and multilayer perceptron are used to evaluate the missile jamming testeffect. A two-stage classification model is proposed for the problem of class imbalance in small data samples particularly.Oversampling is used to solve the problem of class imbalance and random forest is used for classification. Based on theopen-source data of missile jamming test effect evaluation, the empirical research result shows that the random forest mod?el based on oversampling has stronger generalization ability and robustness in missle jamming effect evaluation. In termsof AUC indicator, the random model is better than the multilayer perceptron mode, its median is increased by up to 60%. Itis recommended to use this model in subsequent tests to evaluate missile jamming effects.