Inverse Q filtering seismic processing system and method based on artificial intelligence

Abstract

The invention provides an inverse Q filtering seismic processing system based on artificial intelligence, and the system comprises a historical Q calculation module which calculates a depth domain Q value sequence through a Lee empirical formula, and converts the depth domain Q value sequence into a corresponding frequency band depth domain Q value sequence in combination with a seismic wave dominant frequency; the time-depth conversion module is used for performing convolution on the seismic wavelets and the reflection coefficients to generate synthetic seismic records, and converting a depth domain Q value sequence into a time domain Q value sequence by using a time-depth relationship; the model construction module is used for training a convolutional neural network to generate a Q prediction model by taking the amplitude, the frequency, the phase and the time domain Q value of historical seismic waves as training data; and the inverse Q filtering processing module inputs the multiband parameters of the current seismic wave to the Q prediction model, outputs a time domain Q prediction value, and performs multiband compensation on amplitude attenuation, frequency loss and phase distortion of the seismic wave by adopting an inverse Q filtering formula. According to the method, the defect of insufficient precision of a traditional Q compensation method is avoided, and the calculation efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data processing, and particularly relates to an anti-Q filtering seismic processing system and method based on artificial intelligence. Background Art

[0002] Q compensation in earthquakes is a technology for processing seismic data, aiming to compensate for problems such as energy attenuation and phase distortion caused by factors such as medium absorption during the propagation of seismic waves, so as to improve the quality of seismic data.

[0003] Traditional Q compensation methods mainly include the following: one is Q compensation based on empirical formulas. This method relies on existing empirical formulas and is relatively simple to operate, but the accuracy is limited; the other is Q compensation based on inversion. Its accuracy is relatively high and can more accurately reflect the actual situation. However, the calculation process is complex, and the requirements for data quality and computing resources are also high.

[0004] Although the traditional inversion-based Q compensation method has high accuracy, its calculation process is complex, and the requirements for data quality and computing resources are extremely high, which limits its wide application in practice. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the prior art and provide an anti-Q filtering seismic processing system based on artificial intelligence, including:

[0006] A historical quality factor Q calculation module, configured to calculate a sequence of historical full-band depth-domain quality factor Q values of well logging through the Lee's empirical formula, and convert the sequence of historical full-band depth-domain quality factor Q values of well logging into a sequence of depth-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave beside the well logging according to the main frequency of the historical seismic wave beside the well logging;

[0007] A time-depth conversion module, configured to perform a convolution operation on all seismic wavelets and the reflection coefficient sequence corresponding to the historical seismic wave beside the well logging to obtain a synthetic seismic record, adjust the corresponding relationship between the seismic wave propagation time and depth according to the synthetic seismic record, and convert the sequence of depth-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave beside the well logging into a sequence of time-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave beside the well logging according to the adjusted corresponding relationship between the seismic wave propagation time and depth;

[0008] A model construction module, configured to use the amplitudes, frequencies, phases of all frequency bands of the historical seismic wave beside the well logging and the sequence of time-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave beside the well logging as training data to train a designed convolutional neural network to obtain a quality factor Q prediction model;

[0009] The inverse Q filtering processing module is used to input the amplitudes, frequencies, and phases of all frequency bands of the current seismic wave beside the well logging into the quality factor Q prediction model to obtain the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave, and compensate for the amplitude attenuation, frequency loss, and phase distortion of the current seismic wave beside the well logging according to the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave by using the inverse Q filtering formula.

[0010] Further, in the historical quality factor Q calculation module, the specific method for calculating the sequence of historical full-frequency band depth-domain quality factor Q values of the well logging by using the Lee's empirical formula is as follows:

[0011]

[0012] where, V p,历,i is the historical longitudinal wave velocity of the seismic wave at different depth segments i of the well logging, Q 历,i is the historical full-frequency band depth-domain quality factor at different depth segments i of the well logging, p is the longitudinal wave, and i is the depth segment.

[0013] Further, in the historical quality factor Q calculation module, the specific method for converting the sequence of historical full-frequency band depth-domain quality factor Q values of the well logging into the sequence of depth-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave beside the well logging according to the main frequency of the historical seismic wave beside the well logging is as follows:

[0014]

[0015] where, Q 历,i,j is the depth-domain quality factor corresponding to the main frequency band j of the historical seismic wave of the well logging, N is the coefficient for converting the historical full-frequency band depth-domain quality factor Q value of the well logging into the depth-domain quality factor Q value corresponding to the main frequency band of the historical seismic wave of the well logging, and the value of N is determined by the main frequency of the historical seismic wave.

[0016] Further, in the time-depth conversion module, the specific method for performing convolution operation on all seismic wavelets and the corresponding reflection coefficient sequence of the historical seismic wave beside the well logging to obtain the synthetic seismic record is as follows:

[0017] The convolution operation formula is as follows:

[0018]

[0019] d t = w t × r t + n t

[0020] where, d t is the synthetic seismic record, w t is the seismic wavelet in the historical seismic wave, r tis the reflection coefficient corresponding to historical seismic waves, n t is noise, Z p,历,t is the P-wave impedance of historical seismic waves at different time periods t, where t is the time period during which the historical seismic waves propagate;

[0021] Adjust the time axis of the synthetic seismic record d t so that it corresponds to the depth axis of the P-wave velocity V of historical seismic waves at different depth segments i of the well logging, and obtain the synthetic seismic record d with the time-depth relationship of seismic waves matched well p,历,i 。 t

[0022] Furthermore, for the P-wave velocity V of historical seismic waves at different depth segments i of the well logging p,历,i and the historical rock density ρ at different depth segments i 历,i perform an initial time-depth relationship conversion according to the corresponding historical acoustic wave time difference of the well logging, and obtain the P-wave velocity V of historical seismic waves at different time periods t after the initial time-depth relationship conversion p,历,t and the historical rock density ρ at different time periods t 历,t , and calculate the historical P-wave impedance Z at different time periods t according to the following formula p,历,t : Z p,历,t =V p,历,t ×ρ 历,t 。

[0023] Furthermore, in the time-depth conversion module, the specific method for converting the depth-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging into the time-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging according to the adjusted corresponding relationship between the seismic wave propagation time and depth is as follows:

[0024] Generate the time-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging by using linear interpolation or resampling on the depth-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging according to the adjusted corresponding relationship between the seismic wave propagation time and depth.

[0025] Furthermore, in the model construction module, the specific method for training the designed convolutional neural network with the amplitudes, frequencies, phases of all frequency bands of historical seismic waves beside the well logging and the time-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging as training data to obtain the quality factor Q prediction model is as follows:

[0026] ​The designed convolutional neural network is as follows: The amplitudes, frequencies, and phases of all frequency bands of historical seismic waves are sent to each channel of the input layer. Each channel of the input layer corresponds to a frequency band, and each channel has data on amplitude, frequency, and phase. In the first convolutional layer, the convolutional kernel traverses each channel of the input layer to extract the local spatial features of the seismic wave. The local spatial features include the fluctuation pattern of amplitude, the change trend of frequency, and the stability of phase. Subsequent convolutional layers extract the remaining spatial features of the seismic wave. The remaining spatial features include the correlation between amplitude, frequency, and phase, the interaction between all frequency bands, and the change pattern of the seismic wave in the time series. In the pooling layer, the average pooling method is used for the feature map matrix output by the convolutional layer. The fully connected layer maps the features in the pooled feature map to the predicted value of the quality factor Q, and the output layer is used to output the predicted value of the quality factor Q that maps the features in the pooled feature map.

[0027] The amplitudes, frequencies, phases of all frequency bands of historical seismic waves and the time-domain quality factor Q value sequence of the well logging corresponding to the main frequency band of the historical seismic wave are used as training data to be input into the designed convolutional neural network. After training, a quality factor Q prediction model is obtained.

[0028] Furthermore, in the inverse Q filtering processing module, the specific method for compensating the amplitude attenuation, frequency loss, and phase distortion of the current seismic wave beside the well logging by using the inverse Q filtering formula according to the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave is as follows:

[0029]

[0030] where d(w) is the spectrum of the current seismic wave, d comp (w) is the spectrum after compensating the spectrum of the current seismic wave, Q 预测 is the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave, w is the angular frequency of the current seismic wave, and τ is the propagation time of the current seismic wave.

[0031] An inverse Q filtering seismic processing method based on artificial intelligence includes:

[0032] Calculating the historical full-frequency band depth-domain quality factor Q value sequence of the well logging through the Lee's empirical formula, and converting the historical full-frequency band depth-domain quality factor Q value sequence of the well logging into the depth-domain quality factor Q value sequence of the well logging corresponding to the main frequency band of the historical seismic wave according to the main frequency of the historical seismic wave beside the well logging;

[0033] Convolve all the seismic wavelets in the historical seismic wave beside the well log with the reflection coefficient sequence corresponding to the historical seismic wave to obtain a synthetic seismic record. Adjust the corresponding relationship between the seismic wave propagation time and depth according to the synthetic seismic record. Convert the Q-value sequence of the depth domain corresponding to the main frequency band of the historical seismic wave of the well log into the Q-value sequence of the time domain corresponding to the main frequency band of the historical seismic wave of the well log according to the adjusted corresponding relationship between the seismic wave propagation time and depth.

[0034] Use the amplitudes, frequencies, phases of all frequency bands of the historical seismic wave beside the well log and the Q-value sequence of the time domain corresponding to the main frequency band of the historical seismic wave of the well log as training data to train the designed convolutional neural network to obtain a Q-factor prediction model.

[0035] Input the amplitudes, frequencies, and phases of all frequency bands of the current seismic wave beside the well log into the Q-factor prediction model to obtain the predicted Q-factor value of the time domain corresponding to the main frequency of the current seismic wave. Compensate for the amplitude attenuation, frequency loss, and phase distortion of the current seismic wave beside the well log according to the predicted Q-factor value of the time domain corresponding to the main frequency of the current seismic wave using the inverse Q-filtering formula.

[0036] A computer program product includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the above-mentioned artificial intelligence-based inverse Q-filtering seismic processing method is implemented.

[0037] The beneficial effects of the present invention are as follows:

[0038] 1. The present invention cleverly combines the advantages of the inverse Q-filtering technology with the artificial intelligence algorithm (convolutional neural network), not only avoiding the defect of insufficient accuracy of the traditional Q-compensation method, but also greatly improving the calculation efficiency, and at the same time significantly reducing the dependence on data quality and computing resources, thus providing a more efficient, reliable and easy-to-implement solution for seismic data processing.

[0039] 2. The design of the input layer is based on the frequency band dimension of the historical seismic wave, and each frequency band corresponds to an input channel. Such a design can make full use of the information of the seismic wave in different frequency bands. Each channel contains three-dimensional data: amplitude, frequency, and phase, and these dimensions jointly describe the characteristics of the seismic wave in this frequency band. Through the first convolutional layer, the network can automatically learn the basic characteristics of the seismic wave in different frequency bands, laying a foundation for subsequent feature combination and high-level feature extraction. The subsequent convolutional layers can gradually construct more advanced and discriminative feature representations by combining and transforming the features of the previous layer, which is crucial for improving the prediction accuracy of the Q-factor. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a block diagram of the artificial intelligence-based inverse Q-filtering seismic processing system of the present invention.

[0041] Figure 2 This is the flowchart of the inverse Q-filtering seismic processing based on artificial intelligence of the present invention.

[0042] Figure 3 This is the Q-body profile.

[0043] Figure 4 This is the comparison chart of the cross-well seismic profiles before and after the inverse Q-filtering seismic processing. Detailed implementation manners

[0044] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] Embodiment 1

[0046] Reference Figure 1 , an inverse Q-filtering seismic processing system based on artificial intelligence, comprising:

[0047] A historical quality factor Q calculation module, configured to calculate a sequence of historical full-band depth-domain quality factor Q values of well logging through the Lee empirical formula, and convert the sequence of historical full-band depth-domain quality factor Q values of well logging into a sequence of depth-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave beside the well logging according to the main frequency of the historical seismic wave beside the well logging;

[0048] A time-depth conversion module, configured to perform a convolution operation on all seismic wavelets and the reflection coefficient sequence corresponding to the historical seismic wave in the historical seismic wave beside the well logging to obtain a synthetic seismic record, adjust the corresponding relationship between the seismic wave propagation time and depth according to the synthetic seismic record, and convert the sequence of depth-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave of the well logging into a sequence of time-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave according to the adjusted corresponding relationship between the seismic wave propagation time and depth;

[0049] A model construction module, configured to use the amplitudes, frequencies, phases of all frequency bands of the historical seismic wave beside the well logging and the sequence of time-domain quality factor Q values corresponding to the main frequency band of the historical seismic wave of the well logging as training data to train the designed convolutional neural network to obtain a quality factor Q prediction model;

[0050] An inverse Q-filtering processing module, configured to input the amplitudes, frequencies, and phases of all frequency bands of the current seismic wave beside the well logging into the quality factor Q prediction model to obtain a predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave, and compensate for the amplitude attenuation, frequency loss, and phase distortion of the current seismic wave beside the well logging according to the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave by using the inverse Q-filtering formula.

[0051] This technical solution skillfully combines the advantages of inverse Q filtering technology with those of artificial intelligence algorithms (convolutional neural networks), not only avoiding the defect of insufficient accuracy in traditional Q compensation methods, but also significantly improving the calculation efficiency, while significantly reducing the dependence on data quality and computing resources, thus providing a more efficient, reliable and easy-to-implement solution for seismic data processing.

[0052] (1) In the historical quality factor Q calculation module, the specific method for calculating the historical full-band depth-domain quality factor Q value sequence of well logging through the Lee empirical formula is as follows:

[0053]

[0054] Among them, V p,历,i is the historical P-wave velocity of seismic waves at different depth segments i of well logging, Q 历,i is the historical full-band depth-domain quality factor at different depth segments i of well logging, p is the P-wave, and i is the depth segment.

[0055] The P-wave velocity of seismic waves (unit: m / s) can be calculated by a special acoustic logging instrument from the acoustic wave travel time (s / m) measured by well logging. Specifically, the P-wave velocity of seismic waves is the reciprocal of the acoustic wave travel time. Calculating the Q value through the P-wave velocity has a high credibility near the well point, so this method is used to calculate the Q value.

[0056] (2) In the historical quality factor Q calculation module, the specific method for converting the historical full-band depth-domain quality factor Q value sequence of well logging into the depth-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves beside the well logging is as follows:

[0057]

[0058] Among them, Q 历,i,j is the depth-domain quality factor corresponding to the main frequency band j of historical seismic waves of well logging, N is the coefficient for converting the historical full-band depth-domain quality factor Q value sequence of well logging into the depth-domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves, and the value of N is determined by the main frequency of historical seismic waves.

[0059] Beside the well logging refers to the seismic trace beside the well logging. The historical seismic wave data beside the well logging is the seismic wave data of the trace closest to the well logging data, generally within 50m. Seismic waves have different frequency components, but the frequency of seismic waves lacks high frequencies. It is necessary to convert the frequency of the Q value to be consistent with that of the seismic wave (perform filtering) in order to perform learning in subsequent artificial intelligence, which is very important for the prediction of subsequent convolutional neural networks.

[0060] (3) In the time-depth conversion module, the specific method for convolving all seismic wavelets in the historical seismic waves beside the well logging and the reflection coefficient sequence corresponding to the historical seismic waves to obtain a synthetic seismic record is as follows:

[0061] For the historical seismic wave P-wave velocity V of different depth segments i of the well logging p,历,i and the historical rock density ρ of different depth segments i 历,i According to the corresponding historical acoustic wave travel time of the well logging, perform an initial time-depth relationship conversion to obtain the historical seismic wave P-wave velocity V of different time periods t after the initial time-depth relationship conversion p,历,t and the historical rock density ρ of different time periods t 历,t , calculate the historical P-wave impedance Z of different time periods t according to the following formula p,历,t : Z p,历,t = V p,历,t × ρ 历,t .

[0062] The convolution operation formula is as follows:

[0063]

[0064] d t = w t × r t + n t

[0065] where d t is the synthetic seismic record, w t is the seismic wavelet in the historical seismic waves, r t is the reflection coefficient corresponding to the historical seismic waves, n t is the noise, Z p,历,t is the historical seismic wave P-wave impedance of different time periods t, and t is the time period for the historical seismic wave to propagate;

[0066] Adjust the time axis of the synthetic seismic record d t to make it correspond to the depth axis of the historical seismic wave P-wave velocity V of different depth segments i of the well logging p,历,i to obtain a synthetic seismic record d t with the seismic wave time and depth relationship well matched. Generate the time-domain quality factor Q value sequence of the well logging corresponding to the main frequency band of the historical seismic waves by using the linear interpolation or resampling method according to the corresponding relationship between the adjusted seismic wave propagation time and depth of the well logging.

[0067] Since the relationship between depth and time can be initially obtained based on the P-wave velocity and acoustic wave travel time, so first through the acoustic wave travel time for V p,历,i and ρ 历,iPerform the initial time-depth relationship conversion, then obtain the synthetic record through convolution calculation, and finely adjust the time-depth relationship through the synthetic record. At this time, using the finely adjusted time-depth relationship to convert the full-seismic-band historical quality factor Q value sequence in the depth domain into the full-seismic-band historical quality factor Q value sequence in the time domain can greatly improve the data quality of the quality factor Q.

[0068] (4) In the model construction module, the amplitudes, frequencies, phases of all frequency bands of the historical seismic waves beside the well logs and the time-domain quality factor Q value sequence of the well logs corresponding to the main frequency band of the historical seismic waves are used as training data to train the designed convolutional neural network. The specific method for obtaining the quality factor Q prediction model is as follows:

[0069] The designed convolutional neural network is as follows: The amplitudes, frequencies, and phases of all frequency bands of the historical seismic waves are sent to each channel of the input layer. Each channel of the input layer corresponds to a frequency band, and each channel has data of amplitude, frequency, and phase; in the first convolutional layer, the convolutional kernel traverses each channel of the input layer to extract the local spatial features of the seismic waves. The local spatial features include the fluctuation pattern of the amplitude, the change trend of the frequency, and the stability of the phase. The subsequent convolutional layers extract the remaining spatial features of the seismic waves. The remaining spatial features include the correlation between the amplitude, frequency, and phase, the interaction between all frequency bands, and the change pattern of the seismic waves in the time series; in the pooling layer, the average pooling method is used for the feature map matrix output by the convolutional layer; the fully connected layer maps the features in the pooled feature map to the predicted value of the quality factor Q, and the output layer is used to output the predicted value of the quality factor Q that maps the features in the pooled feature map.

[0070] The amplitudes, frequencies, phases of all frequency bands of the historical seismic waves and the time-domain quality factor Q value sequence of the well logs corresponding to the main frequency band of the historical seismic waves are used as training data and input into the designed convolutional neural network. After training, the quality factor Q prediction model is obtained.

[0071] During the training process, the amplitudes, frequencies, phases of the historical seismic waves of different frequency bands and the historical quality factor Q value sequence corresponding to the seismic frequency bands are used as training data and input into the designed convolutional neural network. The parameters of the designed convolutional neural network are optimized through a loss function (such as the mean square error loss function) to enable the model to accurately predict the value of the quality factor Q. The backpropagation algorithm and the gradient descent optimization algorithm can be used during the training process to update the weights and bias terms of the network.

[0072] The design of the input layer is based on the frequency band dimension of historical seismic waves. Each frequency band corresponds to an input channel. Such a design can make full use of the information of seismic waves in different frequency bands. Each channel contains three-dimensional data: amplitude, frequency, and phase. These dimensions jointly describe the characteristics of seismic waves in that frequency band. Through the first convolutional layer, the network can automatically learn the basic characteristics of seismic waves in different frequency bands, laying a foundation for subsequent feature combination and high-level feature extraction. Subsequent convolutional layers can gradually construct more advanced and discriminative feature representations by combining and transforming the features of the previous layer, which is crucial for improving the prediction accuracy of the quality factor Q. The pooling layer can reduce the computational complexity, reduce the risk of overfitting, and improve the generalization ability of the model by reducing the dimension of the feature map.

[0073] The convolutional neural network can make full use of the amplitude, frequency, and phase information of seismic waves in different frequency bands, and construct advanced and discriminative feature representations through layer-by-layer feature extraction and dimensionality reduction. These feature representations can accurately reflect the characteristics of seismic waves and provide strong support for the prediction of the quality factor Q.

[0074] (5) In the inverse Q filtering processing module, the specific method for compensating the amplitude attenuation, frequency loss, and phase distortion of the current seismic wave beside the well logging by using the inverse Q filtering formula according to the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave is as follows:

[0075]

[0076] where d(w) is the spectrum of the current seismic wave, and d comp (w) is the spectrum after compensating the spectrum of the current seismic wave, Q 预测 is the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave, w is the angular frequency of the current seismic wave, and τ is the propagation time of the current seismic wave.

[0077] After inputting the amplitude, frequency, and phase of the seismic waves in all current frequency bands into the quality factor Q prediction model to obtain the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave, plotting the predicted values of the time-domain quality factor Q for all frequency bands, what is obtained is the Q value distribution map (Q body profile map) on the section of the formation, as Figure 3 shown.

[0078] Inverse Q filtering processing is used to compensate the observed seismic waves to restore the high-frequency components attenuated due to the earth absorption effect. In seismic exploration, the earth's absorption of seismic waves will cause the high-frequency components to attenuate faster, resulting in a reduction in the resolution of the observed seismic waves. Through inverse Q filtering processing, the observed seismic waves can be compensated by using the known Q value and the propagation time of the seismic wave, thereby improving the resolution of seismic data, refer to Figure 4 .

[0079] Example 2

[0080] Reference Figure 2 , an anti-Q filtering seismic processing method based on artificial intelligence, comprising:

[0081] Calculating the historical full-band depth-domain quality factor Q value sequence of well logging through the Li's empirical formula, and converting the historical full-band depth-domain quality factor Q value sequence of well logging into the depth-domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic wave beside the well logging according to the main frequency of the historical seismic wave beside the well logging;

[0082] Performing convolution operation on all seismic wavelets and the reflection coefficient sequence corresponding to the historical seismic wave in the historical seismic wave beside the well logging to obtain a synthetic seismic record, adjusting the corresponding relationship between the seismic wave propagation time and depth according to the synthetic seismic record, and converting the depth-domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic wave of the well logging into the time-domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic wave according to the adjusted corresponding relationship between the seismic wave propagation time and depth;

[0083] Using the amplitudes, frequencies, phases of all frequency bands of the historical seismic wave beside the well logging and the time-domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic wave of the well logging as training data to train the designed convolutional neural network to obtain a quality factor Q prediction model;

[0084] Inputting the amplitudes, frequencies, phases of all frequency bands of the current seismic wave beside the well logging into the quality factor Q prediction model to obtain the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave, and compensating for the amplitude attenuation, frequency loss, and phase distortion of the current seismic wave beside the well logging according to the predicted value of the time-domain quality factor Q corresponding to the main frequency of the current seismic wave by using the anti-Q filtering formula.

[0085] Example 3

[0086] A computer program product, comprising computer programs / instructions, which when executed by a processor implement the anti-Q filtering seismic processing method based on artificial intelligence in Example 2.

[0087] The content not described in detail in this specification belongs to the prior art well known to those skilled in the art. Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in multiple blocks.

[0089] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in multiple blocks.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in multiple blocks.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that after reading the present invention, various changes, modifications, or equivalent replacements can still be made to the specific implementation manners of the invention, but these changes, modifications, or equivalent replacements are all within the scope of the claims of the invention awaiting approval.

Claims

1. An artificial intelligence-based inverse Q-filter seismic processing system, characterized in that: include: The historical quality factor Q calculation module is used to calculate the historical full-band depth domain quality factor Q value sequence of the well logging by using the Li empirical formula, and convert the historical full-band depth domain quality factor Q value sequence of the well logging into the depth domain quality factor Q value sequence of the well logging corresponding to the main frequency band of the historical seismic wave according to the main frequency of the historical seismic wave next to the well logging; A time-depth conversion module is used to perform convolution operations on all seismic wavelets in the historical seismic waves near the well logging and the reflection coefficient sequence corresponding to the historical seismic waves to obtain a synthetic seismic record, adjust the corresponding relationship between the seismic wave propagation time and the depth according to the synthetic seismic record, and convert the depth domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic waves of the well logging into the time domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic waves of the well logging according to the adjusted corresponding relationship between the seismic wave propagation time and the depth; The model building module is used to train the designed convolutional neural network by using the amplitude, frequency, phase of all frequency bands of historical seismic waves beside the well logging and the time domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic waves of the well logging as training data to obtain the quality factor Q prediction model; The inverse Q filtering processing module is used to input the amplitude, frequency and phase of all frequency bands of the current seismic wave beside the well logging into the quality factor Q prediction model to obtain the time domain quality factor Q prediction value corresponding to the main frequency of the current seismic wave, and use the inverse Q filtering formula to compensate for the amplitude attenuation, frequency loss and phase distortion of the current seismic wave beside the well logging according to the time domain quality factor Q prediction value corresponding to the main frequency of the current seismic wave.

2. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 1, characterized in that: In the historical quality factor Q calculation module, the specific method for calculating the historical full-band depth domain quality factor Q value sequence of well logging by using the Li empirical formula is: Among them, V p,历,i is the historical seismic wave P-wave velocity at different depths i of the well, Q 历,i is the historical full-band depth domain quality factor of different logging depth segments i, p is the longitudinal wave, and i is the depth segment.

3. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 2, characterized in that: In the historical quality factor Q calculation module, the specific method of converting the historical full-band depth domain quality factor Q value sequence of the well logging into the depth domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic wave of the well logging according to the main frequency of the historical seismic wave next to the well logging is: Among them, Q 历,i,j is the depth domain quality factor of the logging corresponding to the main frequency band j of the historical seismic wave, N is the coefficient of converting the historical full-band depth domain quality factor Q value of the logging into the depth domain quality factor Q value corresponding to the main frequency band of the historical seismic wave, and the N value is determined by the main frequency of the historical seismic wave.

4. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 3, characterized in that: In the time-depth conversion module, the specific method for performing convolution operation on all seismic wavelets in the historical seismic waves near the well logging and the reflection coefficient sequence corresponding to the historical seismic waves to obtain the synthetic seismic record is: The convolution operation formula is as follows: d t =w t ×r t +n t Among them, d t is the synthetic seismic record, w t is the seismic wavelet in the historical seismic wave, r t is the reflection coefficient corresponding to the historical seismic wave, n t is noise, Z p,历,t is the longitudinal wave impedance of historical seismic waves at different time periods t, where t is the time period of historical seismic wave propagation; Adjust the synthetic seismic record d t The time axis is connected with the historical seismic wave P-wave velocity V of different depth segments i of the well logging. p,历,i The depth axis corresponds to the time axis of the seismic wave, and the synthetic seismic record d is obtained. t .

5. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 4, characterized in that: The historical seismic wave longitudinal wave velocity V of different depth segments i of the well logging p,历,i and the historical rock density ρ at different depths i 历,i The initial time-depth relationship is converted according to the corresponding historical acoustic wave time difference of the well logging, and the historical seismic wave longitudinal wave velocity V of different time periods t after the initial time-depth relationship conversion is obtained. p,历,t and the historical rock density ρ at different time periods t 历,t , the historical longitudinal wave impedance Z of different time periods t is calculated according to the following formula p,历,t :Z p,历,t =V p,历,t ×ρ 历,t .

6. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 5, characterized in that: In the time-depth conversion module, the specific method of converting the depth domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of well logging into the time domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of well logging according to the adjusted correspondence between the seismic wave propagation time and the depth is: The depth domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of well logging is generated by linear interpolation or resampling method according to the correspondence between the adjusted seismic wave propagation time and depth to generate the time domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of well logging.

7. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 1, characterized in that: In the model building module, the amplitude, frequency, phase of all frequency bands of historical seismic waves beside the well logging and the time domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging are used as training data to train the designed convolutional neural network. The specific method for obtaining the quality factor Q prediction model is as follows: The designed convolutional neural network is as follows: the amplitude, frequency and phase of all frequency bands of historical seismic waves are transmitted to each channel of the input layer, each channel of the input layer corresponds to a frequency band, and each channel has amplitude, frequency and phase data; in the first convolution layer, the convolution kernel traverses each channel of the input layer to extract the local spatial features of the seismic wave, which include the fluctuation pattern of the amplitude, the change trend of the frequency and the stability of the phase, and the subsequent convolution layer extracts the remaining spatial features of the seismic wave, which include the correlation between the amplitude, frequency and phase, the interaction between all frequency bands and the change pattern of the seismic wave in the time series; in the pooling layer, the average pooling method is used for the feature map matrix output by the convolution layer; the fully connected layer maps the features in the pooled feature map to the predicted value of the quality factor Q, and the output layer is used to output the predicted value of the quality factor Q mapped with the features in the pooled feature map; The amplitude, frequency, phase of all frequency bands of historical seismic waves and the time domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of well logging are input into the designed convolutional neural network as training data, and the quality factor Q prediction model is obtained after training.

8. The artificial intelligence-based inverse Q-filter seismic processing system according to claim 1, characterized in that: In the inverse Q filtering processing module, the specific method of using the inverse Q filtering formula to compensate for the amplitude attenuation, frequency loss and phase distortion of the current seismic wave beside the well logging according to the predicted value of the time domain quality factor Q corresponding to the main frequency of the current seismic wave is: Among them, d(w) is the spectrum of the current seismic wave, d comp (w) is the spectrum after compensating the spectrum of the current seismic wave, Q 预测 is the predicted value of the time domain quality factor Q corresponding to the main frequency of the current seismic wave, w is the angular frequency of the current seismic wave, and τ is the propagation time of the current seismic wave.

9. An artificial intelligence-based inverse Q-filter seismic processing method, characterized in that: include: The historical full-band depth domain quality factor Q value sequence of the well logging is calculated by Lee's empirical formula, and the historical full-band depth domain quality factor Q value sequence of the well logging is converted into the depth domain quality factor Q value sequence of the well logging corresponding to the main frequency band of the historical seismic wave according to the main frequency of the historical seismic wave next to the well logging; Convolution operation is performed on all seismic wavelets in the historical seismic waves near the well logging and the reflection coefficient sequence corresponding to the historical seismic waves to obtain a synthetic seismic record, the corresponding relationship between the seismic wave propagation time and the depth is adjusted according to the synthetic seismic record, and the depth domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic waves of the well logging is converted into the time domain quality factor Q value sequence corresponding to the main frequency band of the historical seismic waves of the well logging according to the adjusted corresponding relationship between the seismic wave propagation time and the depth; The amplitude, frequency, phase of all frequency bands of historical seismic waves beside the well logging and the time domain quality factor Q value sequence corresponding to the main frequency band of historical seismic waves of the well logging are used as training data to train the designed convolutional neural network and obtain the quality factor Q prediction model; The amplitude, frequency and phase of all frequency bands of the current seismic wave beside the logging well are input into the quality factor Q prediction model to obtain the time domain quality factor Q prediction value corresponding to the main frequency of the current seismic wave. According to the time domain quality factor Q prediction value corresponding to the main frequency of the current seismic wave, the inverse Q filtering formula is used to compensate for the amplitude attenuation, frequency loss and phase distortion of the current seismic wave beside the logging well.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the artificial intelligence-based inverse Q filtering seismic processing method according to claim 9 is implemented.

Images