稀疏LS-SVM算法在海底趋势面模型构建中的应用  

Application of Sparse LS-SVM Algorithm in the Construction of Trend Surface Model

作　　者：李宏武 刘玉红 黄贤源[2,3] 翟国君 陆秀平[2] 黄辰虎[2] 范龙[2] LI Hongwu;LIU Yuhong;HUANG Xianyuan;ZHAI Guojun;LU Xiuping;HUANG Chenhu;FAN Long(The First Military Representative Office of the Navy,Tianjin 300061,China;92859 Troops,Tianjin 300061,China;Department of Military Oceanography and Hydrography,Dalian Naval Academy,Dalian 116018,China)

机构地区：[1]海军装备部第一军事代表室,天津300061 [2]92859部队,天津300061 [3]海军大连舰艇学院军事海洋与测绘系,辽宁大连116018

出　　处：《海洋测绘》2020年第1期53-57,共5页Hydrographic Surveying and Charting

基　　金：国家自然科学基金(41706111)。

摘　　要：利用最小二乘向量机(LS-SVM)算法构造海底趋势面的过程中,由于算法解缺乏稀疏性,使得异常测深训练样本对最终构造的函数模型也产生影响。为了解决该问题,在对留一样本交叉检核法研究的基础上提出了LS-SVM稀疏算法,由于留一样本交叉检核法求解的残差序列可以有效地表示函数预测值偏离实际水深的程度,因此利用该原则重新修剪后的样本数据不仅使算法具有稀疏特性,而且构造的函数模型更合理。为了检验算法的有效性,选取实测的多波束测深数据进行验证,计算结果表明留一样本交叉检核法能够合理地筛选出对函数模型构造贡献程度大的测深训练样本,使得构造的函数模型更合理。In the process of constructing the seabed trend surface by using LS-SVM algorithm.Due to the lack of sparsity in algorithm,the abnormal sounding training samples have an impact on the final constructed function model,so it is necessary to screen the sounding training samples before constructing the function model,that is,to study the sparsity of the LS-SVM algorithm.The residual sequence solved by the method of least one out cross validation can effectively express the deviation of the predictive value of the function from the actual depth of water.The Lagrange multiplier reconstructed based on this principle can better reflect the contribution of the sounding data to the structure of the function model.The sample data after pruning not only make the algorithm sparse,but also construct the function model more reasonable.In order to verify the validity of least one out cross validation method,the measured data of multi-beam sounding are selected.The results show that the least one out cross validation method can reasonably screen out the sounding training samples with large contribution to the function model,making the constructed function model more reasonable.

关 键 词：最小二乘支持向量机 留一样本交叉检核法 稀疏性 拉格朗日乘子 海底趋势面构建 

分 类 号：P229[天文地球—大地测量学与测量工程]