(中南大學(xué) 地球科學(xué)與信息物理學(xué)院 有色金屬成礦預(yù)測教育部重點實驗室，長沙 410083)
摘 要: 比較了幾種大地電磁阻抗估計方法的性能。為了給阻抗估計結(jié)果提供對照標(biāo)準(zhǔn)，首先介紹大地電磁模擬數(shù)據(jù)的生成方法，并從含噪數(shù)據(jù)中提取典型的噪聲時間序列；然后介紹最小二乘估計、M回歸估計、有界影響估計和重復(fù)中值估計這4種回歸方法的基本原理，并用一組標(biāo)準(zhǔn)數(shù)據(jù)測試了它們的性能；最后采用4種估計方法處理不同程度加噪的模擬數(shù)據(jù)，與模擬結(jié)果進(jìn)行比較。結(jié)果表明：重復(fù)中值估計的穩(wěn)健性最優(yōu)；當(dāng)磁道存在噪聲時，有界影響估計給出了明顯優(yōu)于M回歸估計的結(jié)果，也能滿足要求。因此，在實測數(shù)據(jù)處理中，推薦使用重復(fù)中值估計方法與有界影響估計方法。
關(guān)鍵字: 大地電磁；穩(wěn)健阻抗估計；M回歸估計；有界影響估計；重復(fù)中值估計
（Key Laboratory of Metallogenic Prediction of Nonferrous Metals, Ministry of Education,
School of Geoscience and Info-Physics, Central South University, Changsha 410083, China）
Abstract:The performances of different kinds of methods for magnetotelluric impedance estimation were compared. The method of generating magnetotelluric simulated data was introduced, the typical noise series was extracted from noisy data. The basic principles of four regression methods of least square estimator, M-regression estimator, bounded influence estimator and repeated median estimator were introduced. These methods were tested using a set of standard data. At last, simulated data added by noise at different levels were processed by the four different estimators and compared with the simulated results. The results show that the repeated median estimator is the most robust. When the noise exists in the magnetic channel, the results estimated by the bounded influence method are acceptable, which are more stable than those estimated by the M-regression method. So, in the magnetotelluric data processing, the repeated median method and bounded influence method are recommended.
Key words: magnetotelluric; robust estimation; M-regression estimation; bounded influence estimation; repeated median estimation
