中国有色金属学报
中國有色金屬學報
중국유색금속학보
THE CHINESE JOURNAL OF NONFERROUS METALS
2014年
10期
2601-2608
,共8页
深度加权%密度约束%聚焦反演%混合正则化%边界识别
深度加權%密度約束%聚焦反縯%混閤正則化%邊界識彆
심도가권%밀도약속%취초반연%혼합정칙화%변계식별
depth matrix%bound constraint%focusing inversion%hyper-parameter regularization%edge-preserving
在实际地下地质构造是一类多尺度的构造(如断层和褶皱等),而传统的正则化方法多基于最小光滑策略,其反演密度模型一般不易辨识以上构造。为此在分裂Bregman迭代正则化框架下引入混合正则化方法以充分利用非光滑反演和小波多尺度反演算子的特性,引入与衰减系数无关的深度加权矩阵以更好地描述深部异常;针对非光滑反演中异常幅值易于超出现实及理论异常范围,引入密度成像中的约束以确保反演具有物理意义。通过设置两类模型,对比多类正则化反演方法。反演结果显示:混合正则化反演能有效地勾勒异常边界;在处理埋深不同的异常源时,相对于聚焦反演出现的过度聚焦现象而导致的反演深度描述不准确、异常歪斜,混合正则化反演的聚焦效应相对较弱、但深度描述准确。这表明本研究反演确实可行、有效,且具有更强的适应性。
在實際地下地質構造是一類多呎度的構造(如斷層和褶皺等),而傳統的正則化方法多基于最小光滑策略,其反縯密度模型一般不易辨識以上構造。為此在分裂Bregman迭代正則化框架下引入混閤正則化方法以充分利用非光滑反縯和小波多呎度反縯算子的特性,引入與衰減繫數無關的深度加權矩陣以更好地描述深部異常;針對非光滑反縯中異常幅值易于超齣現實及理論異常範圍,引入密度成像中的約束以確保反縯具有物理意義。通過設置兩類模型,對比多類正則化反縯方法。反縯結果顯示:混閤正則化反縯能有效地勾勒異常邊界;在處理埋深不同的異常源時,相對于聚焦反縯齣現的過度聚焦現象而導緻的反縯深度描述不準確、異常歪斜,混閤正則化反縯的聚焦效應相對較弱、但深度描述準確。這錶明本研究反縯確實可行、有效,且具有更彊的適應性。
재실제지하지질구조시일류다척도적구조(여단층화습추등),이전통적정칙화방법다기우최소광활책략,기반연밀도모형일반불역변식이상구조。위차재분렬Bregman질대정칙화광가하인입혼합정칙화방법이충분이용비광활반연화소파다척도반연산자적특성,인입여쇠감계수무관적심도가권구진이경호지묘술심부이상;침대비광활반연중이상폭치역우초출현실급이론이상범위,인입밀도성상중적약속이학보반연구유물리의의。통과설치량류모형,대비다류정칙화반연방법。반연결과현시:혼합정칙화반연능유효지구륵이상변계;재처리매심불동적이상원시,상대우취초반연출현적과도취초현상이도치적반연심도묘술불준학、이상왜사,혼합정칙화반연적취초효응상대교약、단심도묘술준학。저표명본연구반연학실가행、유효,차구유경강적괄응성。
Traditional regularization inversions based on minimum smooth cannot distinguish the discontinuity of the underground subsurface (such as faults and folds, etc.). Hyper-parameter regularization method was introduced under split Bregman iterative regularization framework for taking advantage of edge-preserving inversion and wavelet multiscale operator. A new depth matrix was created based on sensitive matrix for giving an exact description of deep density abnormity. Physical bounds were imposed in each inverse iteration to obtain meaningful solutions and to avoid insignificant abnormality in non-smooth inversion. The smoothness inversion, Marquardt inversion, Occam inversion and focusing inversion with two synthetic models were compared. The inversion results show that the hyper-parameter regularization inversion can preserve edge effectively and has relatively weak focusing effect when treating the designed model with different depth sources. Meanwhile, this method can avoid inaccurate depth description and obliqueness caused by over-regularization in focusing inversion. Moreover, the inversion proposed in this study is feasible, effective, and has better adaptability.