摘要
雷达技术人员在制定目标跟踪策略时,采用机器学习技术完成同型号飞行目标跟踪情况复盘,在实际使用中仍面临数据过拟合、调参耗时费力、实施过程难以规范统一等问题。基于不同自动化机器学习框架,实现雷达跟踪电平预测,从模型搭建耗时、推理耗时、预测平均绝对误差、均方根误差等方面比较自动化机器学习框架和人工设计的传统模型的优劣。实验结果表明:基于自动机器学习的雷达电平预测方法比传统机器学习方案建模速度更快,预测结果与传统机器学习方案接近,电平预测偏差小于2.90 dB,均方根误差小于3.88 dB,自动化机器学习框架对电平预测技术的快速落地应用具有重要价值。
Radar technicians typically formulate tracking strategies by applying machine learning models to analyze the tracking records of the same type of flying targets in the past.However,due to the scarcity of data samples machine learning faces problems such as data overfitting,time-consuming and laborious parameter tuning,and difficulty in standardizing and unifying the implementation processes,there remains a gap for its application.In this paper,the radar automatic gain control(AGC)level prediction based on automatic machine learning(AutoML)framework is studied.The advantages and disadvantages of the automatic machine learning framework and traditional models were compared in terms of model construction time,inference time,mean absolute error,and root mean square error.The experimental results show that the proposed radar AGC level prediction method is faster than the traditional machine learning framework,and the prediction results are comparable to the traditional machine learning scheme.The mean absolute error is less than 2.90 dB and the root mean square error is less than 3.88 dB.The automatic machine learning framework holds significant value for the rapid application of radar AGC level prediction.
作者
曹一文
李伟宗
张润杰
孟艳峰
CAO Yiwen;LI Weizong;ZHANG Runjie;MENG Yanfeng(Taiyuan Satellite Launch Center,Taiyuan Shanxi 030017,China)
出处
《航天工程大学学报》
2025年第5期74-79,共6页
Journal of Space Engineering University
关键词
自动化机器学习框架
电平预测
数据挖掘
测量控制
脉冲雷达
automatic machine learning(AutoML)framework
radar automatic gain control(AGC)level prediction
data mining
measurement and control
mono-pulse radar
