中国惯性技术学报
中國慣性技術學報
중국관성기술학보
JOURNAL OF CHINESE INERTIAL TECHNOLOGY
2009年
5期
572-576
,共5页
张湘平%祝俊淞%吴文启%吴美平
張湘平%祝俊淞%吳文啟%吳美平
장상평%축준송%오문계%오미평
非线性辨识%最小概率%直接加权优化%带宽选择%激光陀螺%温度误差模型
非線性辨識%最小概率%直接加權優化%帶寬選擇%激光陀螺%溫度誤差模型
비선성변식%최소개솔%직접가권우화%대관선택%격광타라%온도오차모형
nonlinear identification%minimal probability%direct weight optimization%bandwidth selection%laser gyro%temperature error model
在原有研究的基础上,针对实验数据观测点疏密分布均匀或不均匀的工程实际情况,分别运用全局准则和局部准则,研究最小概率DWO非线性辨识方法中的带宽选择关键问题,提出了校正AIC准则和LCV准则两种不同的带宽选择方法,并将这些方法应用于四频差动激光陀螺的温度误差模型辨识中,比较和验证了这些方法的正确性和适应性.研究结果表明:①对于"分布均匀"的情况,宜采用校正AIC准则;②对于"分布不均匀"的情况,宜采用LCV准则;③形成了自动带宽选择算法.总之,这些方法为解决"带宽选择"问题提供了有效途径,从而进一步提高了最小概率DWO方法的工程应用价值.
在原有研究的基礎上,針對實驗數據觀測點疏密分佈均勻或不均勻的工程實際情況,分彆運用全跼準則和跼部準則,研究最小概率DWO非線性辨識方法中的帶寬選擇關鍵問題,提齣瞭校正AIC準則和LCV準則兩種不同的帶寬選擇方法,併將這些方法應用于四頻差動激光陀螺的溫度誤差模型辨識中,比較和驗證瞭這些方法的正確性和適應性.研究結果錶明:①對于"分佈均勻"的情況,宜採用校正AIC準則;②對于"分佈不均勻"的情況,宜採用LCV準則;③形成瞭自動帶寬選擇算法.總之,這些方法為解決"帶寬選擇"問題提供瞭有效途徑,從而進一步提高瞭最小概率DWO方法的工程應用價值.
재원유연구적기출상,침대실험수거관측점소밀분포균균혹불균균적공정실제정황,분별운용전국준칙화국부준칙,연구최소개솔DWO비선성변식방법중적대관선택관건문제,제출료교정AIC준칙화LCV준칙량충불동적대관선택방법,병장저사방법응용우사빈차동격광타라적온도오차모형변식중,비교화험증료저사방법적정학성화괄응성.연구결과표명:①대우"분포균균"적정황,의채용교정AIC준칙;②대우"분포불균균"적정황,의채용LCV준칙;③형성료자동대관선택산법.총지,저사방법위해결"대관선택"문제제공료유효도경,종이진일보제고료최소개솔DWO방법적공정응용개치.
Based on the former researches, the bandwidth selection of minimal probability DWO(direct weight optimization) is investigated using global and local criterions respectively in view of the engineering practice situations (the observed points of test data are in uniform or non-uniform distribution). Two different bandwidth selection ways including corrected AIC (Akaike Information Criteria) and LCV (Local Cross Validation) criteria are presented and applied to identify the temperature error model of four-frequency differential laser gyro to compare and validate their correctness and adaptability. The research result shows: (1)In case of "uniform distribution", the corrected AIC should be adopted; (2)In case of "non-uniform distribution", the LCV criteria should be adopted; (3)Automatic bandwidth selection algorithms are developed. In a word, these methods are effective ways to overcome the problem of "bandwidth selection" and are able to advance the engineering significance of minimal probability DWO.