CAJ | 학술논문

针对捷联惯导初始对准中 UKF 滤波中噪声的统计特性与实际不符时,滤波精度严重降低甚至发散的问题,提出一种基于凸线性组合支持向量机的初始对准方法.将测试样本对分为四组,分别用三组训练第一层和一组训练第二层的支持向量机,第一层为几组支持向量机的并行计算,第二层是把第一层单个支持向量机以凸线性组合的形式进行信息融合,构成凸线性组合支持向量机,从而实现捷联惯导系统的初始对准.最后通过UKF滤波、SVM、CLC-SVM进行仿真对比,结果表明CLC-SVM较单一SVM性能提高,实时性比UKF滤波提高一个数量级,泛化能力增强.
침대첩련관도초시대준중 UKF 려파중조성적통계특성여실제불부시,려파정도엄중강저심지발산적문제,제출일충기우철선성조합지지향량궤적초시대준방법.장측시양본대분위사조,분별용삼조훈련제일층화일조훈련제이층적지지향량궤,제일층위궤조지지향량궤적병행계산,제이층시파제일층단개지지향량궤이철선성조합적형식진행신식융합,구성철선성조합지지향량궤,종이실현첩련관도계통적초시대준.최후통과UKF려파、SVM、CLC-SVM진행방진대비,결과표명CLC-SVM교단일SVM성능제고,실시성비UKF려파제고일개수량급,범화능력증강.
The filtering precision would be severely decreased or even divergent when the noise statistical characteristics of UKF filter in SINS initial alignment does not conform to the actual one. To solve this problem, an initial alignment method based on support vector machine(SVM) is proposed. The test samples are split into four groups, in which three groups are trained for the SVMs in the first layer, and the last group is trained for the SVMs in the second layer. The parallel computing is trained for several groups of support vector machines in first layer, and the information of various single-SVMs in the first layer are trained to be fused by convex linear combination. In this way the initial alignment of SINS is realized. The results from the simulation contrast among UKF filter, SVM, CLC-SVM shows that the performance of CLC-SVM has improved compared with that of single SVM, and its real-time performance increases one order of magnitude compared with that of UKF filtering. Meanwhile, its generalization ability is enhanced.