引用本文
  •    [点击复制]
  •    [点击复制]
【打印本页】 【在线阅读全文】【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 23次   下载 12 本文二维码信息
码上扫一扫!
基于SVR的导航传感器自适应野值检测方法
戴海发,卞鸿巍,马恒,王荣颖
0
(海军工程大学电气工程学院, 武汉 430033)
摘要:
针对Lagrange插值法无法处理连续野值的问题,提出了一种基于改进支持向量回归(Support Vector Regression, SVR)的导航传感器自适应野值检测方法。该方法结合了支持向量回归利用小样本数据就能够准确建模和3σ准则计算简易的优点,利用支持向量回归在线建立舰船的运动模型对测量值进行实时预测,并利用3σ准则自适应地计算阈值,然后通过比较阈值与预测残差来判别测量值是否为野值点。该方法可以自动地学习舰船的运动趋势,建立舰船的真实运动模型,而且不受连续野值点的影响,能够在没有其他传感器辅助的条件下完成野值检测。海试实测数据表明,提出的方法对离散和连续的野值点均具有较好的检测效果,同时可以更好地估计传感器的真实测量值。
关键词:  导航传感器  野值剔除  Lagrange插值法  支持向量回归  自适应
DOI:
基金项目:国家自然科学基金 (41876222)
Adaptive Navigation Sensor Outlier Detection Method Based on SVR
DAI Hai-fa,BIAN Hong-wei,MA Heng,WANG Rong-ying
(Collage of Electrical Engineering, Naval University of Engineering, Wuhan 430033, China)
Abstract:
Aiming at the problem that Lagrange interpolation can't deal with continuous outliers, this paper proposes an adaptive outlier detection method for navigation sensors based on improved support vector regression (SVR). This method combines the advantages of support vector regression with small sample data for accurate modeling and 3σ for simple calculation. The support vector regression is used to establish the ship's motion model online to predict the measured values in real time, and adaptively using 3σ criteria. The threshold is calculated, and then the measured value is compared to the predicted residual to determine whether the measured value is an outlier. The method can automatically learn the movement trend of the ship, establish the real motion model of the ship, without being affected by the continuous outlier, and can perform the outlier detection without other sensors assistance. The sea-tested data shows that the proposed method has a good detection effect on both discrete and continuous outliers, and at the same time obtains a better estimate of the true measured value of the sensor.
Key words:  Navigation sensor  Outlier elimination  Lagrange interpolation  Support vector regression  Adaptive

用微信扫一扫

用微信扫一扫