中国惯性技术学报
中國慣性技術學報
중국관성기술학보
JOURNAL OF CHINESE INERTIAL TECHNOLOGY
2014年
3期
409-415
,共7页
惯性测量组件%在线最小二乘支持向量机%动态数据窗%鲁棒性%故障预测
慣性測量組件%在線最小二乘支持嚮量機%動態數據窗%魯棒性%故障預測
관성측량조건%재선최소이승지지향량궤%동태수거창%로봉성%고장예측
inertial measurement units%online least squares support vector machine%dynamic data window%robustness%fault prediction
为提高惯导系统工作的可靠性和导航性能,对其惯性测量组件的故障模式和检测模型进行了分析。针对最小二乘支持向量机(LS-SVM)回归算法做了两点改进,具体方法是先对输入样本观察窗平移更新的每个样本数据进行异常点滤波判断并用牛顿插值法进行处理,接着通过对在线LS-SVM回归过程的研究,提出了一种递推求解的快速算法,将惯性测量组件的输出量、舵偏角改变量并辅以环境因素作为观测样本序列,应用该算法来提高模型检测的准确性和时效性。最后对惯性测量组件无故障和出现卡滞、恒偏差时的故障模式进行了仿真实验,结果表明,与应用LS-SVM、SVM和BP神经网络算法相比,提出的惯性测量组件故障在线检测方法具有较强的鲁棒性和较快的速度。
為提高慣導繫統工作的可靠性和導航性能,對其慣性測量組件的故障模式和檢測模型進行瞭分析。針對最小二乘支持嚮量機(LS-SVM)迴歸算法做瞭兩點改進,具體方法是先對輸入樣本觀察窗平移更新的每箇樣本數據進行異常點濾波判斷併用牛頓插值法進行處理,接著通過對在線LS-SVM迴歸過程的研究,提齣瞭一種遞推求解的快速算法,將慣性測量組件的輸齣量、舵偏角改變量併輔以環境因素作為觀測樣本序列,應用該算法來提高模型檢測的準確性和時效性。最後對慣性測量組件無故障和齣現卡滯、恆偏差時的故障模式進行瞭倣真實驗,結果錶明,與應用LS-SVM、SVM和BP神經網絡算法相比,提齣的慣性測量組件故障在線檢測方法具有較彊的魯棒性和較快的速度。
위제고관도계통공작적가고성화도항성능,대기관성측량조건적고장모식화검측모형진행료분석。침대최소이승지지향량궤(LS-SVM)회귀산법주료량점개진,구체방법시선대수입양본관찰창평이경신적매개양본수거진행이상점려파판단병용우돈삽치법진행처리,접착통과대재선LS-SVM회귀과정적연구,제출료일충체추구해적쾌속산법,장관성측량조건적수출량、타편각개변량병보이배경인소작위관측양본서렬,응용해산법래제고모형검측적준학성화시효성。최후대관성측량조건무고장화출현잡체、항편차시적고장모식진행료방진실험,결과표명,여응용LS-SVM、SVM화BP신경망락산법상비,제출적관성측량조건고장재선검측방법구유교강적로봉성화교쾌적속도。
In order to improve the reliability and navigation performance of the inertial navigation system, the fault mode and test model were analyzed. Two ameliorations were made for the method of the online least squares support vector machine(LS-SVM): 1) The singularity was found out and disposed with Newton interpolation method among the sample data which was shifted and updated in the observation window. 2) A recursive solution method was put forward based on the process regression analysis of online LS-SVM, and the inertial measurement units outputs complement with elevator angle variation and environmental factors were chosen as the observed sample sequence. Then the proposed method was used to improve the accuracy and timeliness of the online test model for the inertial navigation system. Finally, the simulation was made when the inertial navigation system has no fault and has lock fault or constant bias fault. The results show that, compared with SVM, LS-SVM, and BP neural network modeling, the proposed method has higher learning speed and robustness performance.