基于滑窗式置信度检测和改进SVM的量测数据预处理方法

针对无源观测平台量测数据高度的非线性特征和各种随机噪声对滤波器估计精度所造成的严重影响，提出了一种对量测数据进行预处理的算法。首先，利用量测数据的置信度对量测数据进行滑窗式检测，初步判别无效数据，然后对支持向量机(support vector machine，SVM)的核函数进行改进，利用数据自身的特征信息对核函数进行加权处理，再用改进的SVM对数据进行再分类处理，从而最终确定无效数据。利用所提算法对实测数据进行预处理，结果说明所提算方法能有效剔除各类噪声对测量数据的影响，具有工程应用价值。

Measurement data pre-processing method based on the improvement of theSVM and the confidence level detection in sliding window type

Aiming at the nonlinear characteristics of passive observation platforms’ measurements and serious impact on the filter estimate accuracy caused by the random noise, a measurement data pre-processing algorithm is proposed. First of all, the confidence level of the measured data is used to detect the measurements in sliding window type and the invalid data are discriminated initially, then the support vector machine (SVM) is used whose kernel function has been weighted with data’s characteristic to re classify the measurements, in order to determine the invalid data ultimately. Finally a test based on the actual measurements is given， whose result shows that the proposed algorithm can effectively eliminate the impact of various types of noise on the measurement data, and the proposed algorithm has some engineering application value.