【作者】 高少武；

【导师】 贺振华；

【作者基本信息】 成都理工大学 ， 地球探测与信息技术， 2010， 博士

【摘要】 在地震数据采集过程中,如果地震测线上空有高压输电线通过,地震记录中就会出现一个50Hz左右的单频干扰。常见的单频干扰压制方法有两种:频率域压制法和陷频滤波法。频率域压制法是相对于振幅的,压制量很难取准。压制不足,在地震记录中会存在很强的剩余干扰;压制过量,会伤害有效信号。而且干扰频率不是整50Hz,计算窗口长度难以恰当选取,使得频率域压制法出现一些难以克服的困难。陷频滤波法会伤及单频附近有效信号的频率成分。为了减少伤害,往往会将压制频带取得很窄,这样对应的时间域算子就会很长,从而耗费很多的运算时间,导致计算效率低下,且会产生严重的边界效应。另外,时间域算子不可能无限长,这样陷频滤波法的压制频带必定有一个宽度,因此必然造成对有效信号频率成分的伤害。鉴于从浅层到深层,单频干扰振幅、频率和相位基本保持不变,这样可以在时间域采用余弦函数或者正弦函数与余弦函数的组合逼近单频干扰,并从地震记录中减去,达到消除干扰目的。余弦函数逼近的三个参数是:振幅、频率和时延。单频干扰与频率和时延呈非线性关系,可以采用各种算法进行估算。而单频干扰与振幅呈线性正比关系,可以采用公式直接计算。本文推导出余弦函数振幅计算公式,且提出了基于余弦函数振幅计算公式的频率时延估算方法。频率时延估算方法有三种:扫描法、最速下降法、自适应法。在余弦函数自适应法中,我们推导出时延参数计算公式,并应用公式直接计算时延参数。同时我们提出了基于这些频率时延算法的单频干扰识别与消除方法。正余弦函数逼近的三个参数是:正弦函数振幅、余弦函数振幅和频率。频率可以采用各种方法估算,而单频干扰与正弦函数振幅和余弦函数振幅呈线性正比关系,可以采用公式直接计算。本文推导出正弦函数振幅和余弦函数振幅计算公式,且提出了基于正弦函数振幅和余弦函数振幅计算公式的频率估算方法。频率估算方法有四种:扫描法、最速下降法、自适应法、线性调频谱法。本文首次提出并应用线性调频谱法频率估算算法,同时提出了基于这些频率算法的单频干扰识别与消除方法。本文首次提出并应用自相关法单频干扰识别与消除方法和自相关褶积法单频干扰识别与消除方法。在自相关法中,通过地震数据的自相关运算,直接估算单频干扰余弦函数,由余弦函数自适应减算法直接消除单频干扰。在自相关褶积法中,通过地震数据的自相关褶积运算,直接估算干扰正弦函数和余弦函数,由正余弦函数自适应减算法直接消除单频干扰。自相关法和自相关褶积法都不需要估算单频干扰振幅、频率和时延三个参数。本文对参数估算误差造成单频干扰估算误差进行了全面深入的分析。振幅估算误差对单频干扰影响很大,即使频率(和时延)参数估算准确,而仅仅由于振幅估算存在误差,也会产生较强的剩余单频干扰。频率(和时延)估算误差不但造成振幅计算误差和单频干扰误差,且会引起单频干扰在时间方向发生变化。为了准确估算振幅和单频干扰,频率估算精度至少应当是0.01Hz;而对于余弦函数逼近,时延估算精度为0.1。使用理论数据说明单频干扰参数估算误差对单频干扰的影响。在实际地震数据处理时,对地震数据需要做分析,以确定单频干扰初始频率。本文提出了自动识别包含单频干扰的地震数据道算法,可以大大提高单频干扰消除处理运算效率。另外还提出利用地震数据浅层初至时间段或深层时间段估算单频干扰,可以有效提高单频干扰估算精度。对于多个单频干扰,采用逐个估算频率(和时延),整体振幅标定的策略。在正余弦函数逼近中,单频干扰仅仅与频率呈非线性关系,因此在同类频率估算方法中,正余弦函数逼近方法计算效率要比余弦函数逼近高得多。这些单频干扰识别与消除方法最大的优点就是在时间域能够有效地消除单频干扰,而不伤害干扰频率附近有效信号,从而提高了单频频率分量附近数据的信噪比。合成数据和实际数据例子试算结果也表明这些方法是有效和可行的。实际数据处理表明,线性调频谱法运算效率高,节省时间,更加适用于海量地震数据处理需要,是消除单频干扰最有效的方法。更多还原

【Abstract】 During seismic data acquisition, there would be a strong monofrequency interference ( MFI ) around 50Hz in seismic records if a high-tension line crosses over the seismic lines. Traditional MFI suppression methods include the frequency domain suppression and the notched-frequency filtering. The former only suppresses on the amplitude spectrum of seismic data, nor on the phase spectrum. Therefore, the suppression amount is not easily controlled. If the suppression amount is too small, the strong residual MFI exists in seismic records. On the contrary, the effective signal could be injured if the suppression amount is too large. Additionally, since the interference frequency is not exactly at 50 Hz, the choice of computed window length is so difficult that the method itself is hard to implement effectively. Similarly, the notched-frequency filtering method could injure the frequency components of effective signals around the monofrequency. In order to decrease the injury, this kind of methods usually narrows the suppression frequency bands. Thus the corresponding time domain operator is so long that the computation time is also long. It results in the low computation efficiency and the severe boundary effect. Consider that the length of time domain operator would not be unlimited, the suppressed band must have a limited width. Hence, the frequency components of effective signals must be injured.In the seismic records, the amplitude, frequency and phase can be considered as basically invariable from the shallow to the deeper layer. Thus a MFI would be approximated by a cosine function or a combination of a cosine with a sine function in the time domain. Following that, the MFI is subtracted from the seismic record to eliminate the interference. For the cosine function approximation, there are three parameters such as amplitude, frequency and time-delay. The MFI has a linear relation with the amplitude, so it is directly computed according to a formula. The computational formula of the cosine function amplitude is first derived, and the frequency and time-delay estimation algorithms based on this formula are also proposed for the first time. At the same time, the MFI has nonlinear relations with the frequency and the time-delay, so these two parameters are estimated by a variety of algorithms. These estimation methods include scanning algorithm, steepest descent algorithm, and adaptive algorithm. For the cosine function adaptive algorithm, the computational formula of the time-delay is first derived, and the time-delay parameter is also computed according to it. Furthermore, we also propose and apply the identification and elimination methods of the cosine function MFI based on these frequency and time-delay estimation algorithms.In the sine-cosine function approximation, there are also three parameters: frequency, sine amplitude, and cosine amplitude. The sine and cosine amplitudes are directly computed according to their formulas for these two parameters are linearly proportional to the MFI. We first derive the computational formulas of the sine-cosine function amplitudes, and we also propose the frequency estimation algorithms for the first time, as well as the FMI identification and elimination methods based on these amplitude formulas. The frequency is estimated by a variety of algorithms. These methods include scanning algorithm, steepest descent algorithm, adaptive algorithm, and linear frequency-modulation spectrum algorithm ( LFMS ). Additionally, the LFMS algorithm is first proposed, too, as well as first applied to the frequency estimation algorithm and the method of identifying and eliminating the MFI. Furthermore, the identification and elimination methods of the sine-cosine function MFI based on these frequency estimations are first proposed, as well as first applied.We also propose two kinds of methods on identification and elimination of the MFI based on an autocorrelation and an autocorrelation-convolution algorithm. For the autocorrelation algorithm, the cosine function of a MFI is directly estimated by the autocorrelation operation of the seismic data. After that, the MFI is eliminated by the cosine function adaptive subtraction algorithm. Similarly, in the autocorrelation-convolution algorithm, by the autocorrelation and convolution operations of the seismic data, the sine and cosine functions of a MFI are directly estimated, and the MFI is eliminated by the sine-cosine function adaptive subtraction algorithm. However in these two algorithms, it is not necessary to estimate the three parameters of the MFI.A detailed and profound analysis is first performed about the FMI estimation error induced by three parameter estimation errors. The error of the amplitude estimation has a huge influence on the MFI. Even though the frequency and phase estimations are exact, it can bring a very strongly residual MFI in the seismic records. Additionally, the errors of the frequency and phase estimations can not only bring ones of the amplitude and MFI computations, but they can also make the MFI change in the time direction. In order to estimate the amplitude and MFI exactly and precisely, the accuracy of the frequency should not be less than 0.01Hz, and one of the time-delay should be 0.1 for the cosine function approximation. The synthetic data examples illustrate that the errors of the parameter estimations would truly affect the identification and elimination of the MFI.During the real seismic processing, the data analysis must be performed to confirm the initial frequency of the MFI. Using the method on an automatic identification of the MFI data trace, the operation efficiency eliminating the MFI could be greatly improved. At the same time, the estimated precision of three parameters and the MFI can be also enhanced by estimating the MFI in the shallow first-arrival or deep time zone. For multiple MFIs, the frequencies ( and time-delays ) are estimated one by one, and the amplitudes are computed by a system strategy.In the sine-cosine function approximation, the MFI only has a nonlinear relation with the frequency parameter. Therefore, the computational efficiency of this method is higher than that of the cosine function approximation within the same kinds of estimation methods. The distinguished merit of these methods is that the MFI can be effectively eliminated in the time domain. Meanwhile, the useful signals around the interference have not been injured. Consequently, the signal-to-noise ratio has been improved in the vicinity of monofrequency component. The synthetic and real data examples illustrate that the proposed methods are feasible and effective.The real data processing examples also illustrate that the computational time of the linear frequency-modulation spectrum method is shortest, and the operational efficiency is highest. So it adapts the application of the great capacity seismic data, and is the most effective method of eliminating the MFI.更多还原