Acoustic logging time difference determination method and device and electronic equipment

Abstract

The invention relates to the technical field of acoustic logging, and discloses an acoustic logging time difference determination method and device and electronic equipment, and the acoustic logging time difference determination method comprises the steps: obtaining a first acoustic wave train signal and a second acoustic wave train signal of the same sound source; determining the time difference of arrival between the first sound wave train signal and the second sound wave train signal through a preset encoder according to the first sound wave train signal and the second sound wave train signal; wherein the preset encoder is obtained by training according to a preset self-supervised learning strategy. According to the acoustic logging time difference determination method and device and the electronic equipment provided by the invention, the arrival time difference is determined by using the preset encoder, so that the acoustic logging time difference determination speed can be effectively improved; moreover, the preset encoder is obtained through the training of the preset self-supervised learning strategy, training can be completed on the sample data without labels, the training process of the preset encoder is more efficient, the feature extraction precision of the preset encoder is improved, and the calculation precision of the time difference of arrival is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic logging, and particularly relates to a method, a device, and an electronic device for determining acoustic logging time difference. Background Art

[0002] When sound waves propagate in different media, their acoustic characteristics such as speed, amplitude, and frequency change differently. Since geological factors such as the lithology, structural plane conditions, weathering degree, stress state, and water content of rock masses can directly cause changes in the sound (ultrasonic) wave speed, amplitude, and frequency, the geological conditions of rocks (rock masses) can be understood by the sound (ultrasonic) wave speed, frequency, and amplitude received by the receiver, and some mechanical parameters of the rocks (rock masses) (such as Poisson's ratio, dynamic elastic modulus, compressive strength, elastic resistance coefficient, etc.) and some other engineering geological property indicators (such as weathering coefficient, fracture coefficient, anisotropy coefficient, etc.) can be obtained. Acoustic logging is a logging method that uses these acoustic properties of rocks to study the geological profile of a well. By studying the propagation characteristics of sound waves in downhole rock formations and media, the geological characteristics of the rock formations can be understood and the cementing quality can be judged.

[0003] Currently, the most commonly used method for extracting acoustic logging time difference is the time difference - time correlation analysis method (abbreviated as the STC method). However, the STC method has problems such as a large number of calculations, which affects the speed of time difference extraction, and it is easy to fall into local maximum points during peak searching, which affects the accuracy of time difference extraction. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method, a device, and an electronic device for determining acoustic logging time difference.

[0005] The present invention provides a method for determining acoustic logging time difference, the method comprising:

[0006] Obtaining a first acoustic wave train signal and a second acoustic wave train signal of the same sound source;

[0007] Determining the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal according to the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder;

[0008] Wherein, the preset encoder is trained according to a preset self - supervised learning strategy.

[0009] Further, the preset self - supervised learning strategy includes:

[0010] Obtaining an acoustic wave train sample signal;

[0011] Dividing the acoustic wave train sample signal into multiple sub - sample signals on average, and randomly dividing the sub - sample signals into a first group and a second group;

[0012] Extract the features of each sub-sample signal in the first group and the second group respectively through a preset transformation network to obtain the true feature set corresponding to the acoustic wave train sample signal, and predict the features of each missing sub-sample signal in the first group and the second group respectively to obtain the predicted feature set corresponding to the acoustic wave train sample signal;

[0013] Determine the loss value according to the true feature set and the predicted feature set;

[0014] Update the parameters of the preset transformation network according to the loss value, and continue to train the preset transformation network until the loss value is less than the preset loss threshold to obtain the trained preset transformation network;

[0015] Use the trained preset transformation network as the preset encoder.

[0016] Further, determining the loss value according to the true feature set and the predicted feature set includes:

[0017] Determine the first distance value between the true feature and the predicted feature according to the true feature set and the predicted feature set;

[0018] Decode the features in the true feature set and the predicted feature set respectively through a preset decoder to obtain the first acoustic wave train decoded signal and the second acoustic wave train decoded signal;

[0019] Determine the second distance value between the first acoustic wave train decoded signal and the second acoustic wave train decoded signal;

[0020] Determine the third distance value between the acoustic wave train sample signal and the first acoustic wave train decoded signal;

[0021] Determine the fourth distance value between the acoustic wave train sample signal and the second acoustic wave train decoded signal;

[0022] Determine the loss value according to the first distance value, the second distance value, the third distance value and the fourth distance value.

[0023] Further, determine the loss value L through the following formula 总 :

[0024] L 总 =γK 1 +(1 - γ)K 2 +K 3 +K 4

[0025] Where K 1 is the first distance value, K 2 is the second distance value, K 3 is the third distance value, K 4 is the fourth distance value, and γ is a constant parameter.

[0026] Further, the preset self-supervised strategy further includes:

[0027] Updating the parameters of the preset decoder according to the loss value, so that when determining the loss value based on the real feature set and the predicted feature set next time, the features in the real feature set and the predicted feature set are decoded respectively by the preset decoder with updated parameters.

[0028] Further, determining the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder according to the first acoustic wave train signal and the second acoustic wave train signal includes:

[0029] Intercepting a part of the first acoustic wave train signal as the first sub-acoustic wave train signal according to the preset intercepting strategy;

[0030] Extracting features from the first sub-acoustic wave train signal through a preset encoder to obtain a first feature set;

[0031] Slidingly segmenting the second acoustic wave train signal with a preset sliding window at a preset step length to obtain a plurality of second sub-acoustic wave train signals;

[0032] Extracting features from each second sub-acoustic wave train signal through a preset encoder to obtain a second feature set corresponding to each second sub-acoustic wave train signal respectively;

[0033] Calculating the distance values between the first feature set and each second feature set respectively, and taking the second feature set corresponding to the minimum distance value as the target feature set;

[0034] Taking the time difference between the second sub-acoustic wave train signal corresponding to the target feature set and the first sub-acoustic wave train signal as the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal.

[0035] The present invention also provides an acoustic logging time difference determination device, and the device includes:

[0036] An acquisition module, configured to acquire a first acoustic wave train signal and a second acoustic wave train signal of the same sound source;

[0037] A determination module, connected to the acquisition module, configured to determine the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder according to the first acoustic wave train signal and the second acoustic wave train signal; wherein, the preset encoder is trained according to a preset self-supervised learning strategy.

[0038] Further, the determination module includes:

[0039] An intercepting unit, configured to intercept a part of the first acoustic wave train signal as the first sub-acoustic wave train signal according to the preset intercepting strategy;

[0040] A sliding segmentation unit, used for performing sliding segmentation on the second sound wave train signal by using a sliding window of a preset size and a preset step length to obtain a plurality of second sub-sound wave train signals;

[0041] An extraction unit is connected to the interception unit and the sliding segmentation unit, respectively, and is used to extract features from the first sub-sound wave train signal through a preset encoder to obtain a first feature set; and is used to extract features from each second sub-sound wave train signal through a preset encoder to obtain a second feature set corresponding to each second sub-sound wave train signal;

[0042] a target feature set determination unit, connected to the extraction unit, for respectively calculating the distance value between the first feature set and each second feature set, and taking the second feature set corresponding to the minimum distance value as the target feature set;

[0043] The arrival time difference determination unit is connected to the target feature set determination unit and is used to use the time difference between the second sub-sound wave train signal and the first sub-sound wave train signal corresponding to the target feature set as the arrival time difference between the first sound wave train signal and the second sound wave train signal.

[0044] The present invention also provides a computer-readable storage medium, wherein the computer program stored in the computer-readable storage medium implements the steps of the above method when executed by one or more processors.

[0045] The present invention also provides an electronic device, comprising a memory and one or more processors, wherein a computer program is stored in the memory, and the memory and the one or more processors are communicatively connected to each other, and when the computer program is executed by the one or more processors, the steps of the above method are executed.

[0046] The method, device and electronic device for determining the acoustic logging time difference provided by the present invention have at least the following beneficial effects:

[0047] (1) When determining the acoustic logging time difference, the arrival time difference is determined by using a preset encoder, which can effectively improve the speed of determining the acoustic logging time difference.

[0048] (2) The preset encoder is an encoder that is pre-trained according to a preset self-supervised learning strategy. By adopting the preset self-supervised learning strategy, the training can be completed on unlabeled sample data to obtain the preset encoder, making the preset encoder training process more efficient.

[0049] (3) When training the preset encoder, the loss value is determined by obtaining the first distance value, the second distance value, the third distance value, and the fourth distance value based on the real feature set and the predicted feature set, thereby improving the feature extraction accuracy of the preset encoder and further improving the calculation accuracy of the arrival time difference. Brief Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0051] Figure 1 It is a flowchart of the steps of the acoustic logging time difference determination method provided in an embodiment of the present invention;

[0052] Figure 2 It is a flowchart of the steps of the preset self-supervised learning strategy method provided in an embodiment of the present invention;

[0053] Figure 3 It is a flowchart of the steps of the loss value determination method provided in an embodiment of the present invention;

[0054] Figure 4 It is a flowchart of the steps of the arrival time difference determination method provided in an embodiment of the present invention;

[0055] Figure 5 It is a schematic diagram of the preset encoder training process provided in an embodiment of the present invention;

[0056] Figure 6 It is a schematic diagram of the structure of the acoustic logging time difference determination device provided in an embodiment of the present invention;

[0057] Figure 7 It is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention;

[0058] Reference Signs:

[0059] Figure 6 Among them: 601 - acquisition module, 602 - determination module;

[0060] Figure 7 Among them: 700 - electronic device, 701 - processor, 702 - communication bus, 703 - user interface, 704 - communication interface, 705 - memory. Detailed Embodiments

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention. In the claims and the specification of the present invention, the use of "first", "second", "third", "fourth" is only for the purpose of easy distinction and has no special meaning, and is not intended to limit the present invention.

[0062] In an embodiment of the present invention, a method for determining the acoustic logging time difference is provided. As Figure 1 shown, the method includes the following steps:

[0063] Step S101: Obtain a first acoustic wave train signal and a second acoustic wave train signal of the same sound source.

[0064] Step S102: Determine the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder according to the first acoustic wave train signal and the second acoustic wave train signal.

[0065] Among them, the first acoustic wave train signal and the second acoustic wave train signal are acoustic signals from the same sound source received by two receiving transducers respectively.

[0066] The preset encoder is trained according to a preset self-supervised learning strategy.

[0067] In this embodiment, when determining the acoustic logging time difference, determining the arrival time difference according to the preset encoder can effectively improve the speed of determining the acoustic logging time difference.

[0068] In some embodiments, the method for determining the acoustic logging time difference further includes training a preset encoder in advance according to a preset self-supervised learning strategy. By using the preset self-supervised learning strategy, training can be completed on unlabeled sample data to obtain the preset encoder, making the training process of the preset encoder more efficient.

[0069] Specifically, as Figure 2 shown, the preset self-supervised learning strategy includes:

[0070] Step S201: Obtain an acoustic wave train sample signal;

[0071] Step S202: Divide the acoustic wave train sample signal into multiple sub-sample signals on average, and randomly divide the sub-sample signals into a first group and a second group.

[0072] Step S203: Extract the features of each sub-sample signal in the first group and the second group respectively through a preset transformation network to obtain the true feature set corresponding to the acoustic wave train sample signal, and predict the features of each missing sub-sample signal in the first group and the second group respectively to obtain the predicted feature set corresponding to the acoustic wave train sample signal.

[0073] Step S204: Determine the loss value according to the true feature set and the predicted feature set.

[0074] Step S205: Update the parameters of the preset transformation network according to the loss value, and continue to train the preset transformation network until the loss value is less than the preset loss threshold to obtain a trained preset transformation network.

[0075] Step S206: Use the trained preset transformation network as the preset encoder.

[0076] Among them, the number of sub-sample signals and the preset loss threshold can be set by those skilled in the art according to actual needs, and the present invention does not make any limitations thereto. Preferably, the number of sub-sample signals is a positive integer power of 2. For example, the number of sub-sample signals is 8 or 16.

[0077] The specific structure of the preset transformation network can also be set by those skilled in the art according to actual needs, and the present invention does not make any restrictions thereto. In one implementation, the preset transformation network includes a plurality of cascaded transformer models, and the number of transformer models is equal to the number of sub-sample signals.

[0078] It should be understood that before training the preset transformation network according to the preset self-supervised learning strategy, it is also necessary to initialize the parameters of the preset transformation network first.

[0079] For example, after obtaining the acoustic wave train sample signal f(t), the acoustic wave train sample signal f(t) is evenly divided into n sub-sample signals and randomly divided into the first group V 1 , the second group V 2 .

[0080] The first group V 1 includes n1 sub-sample signals, and the second group V 2 includes n2 sub-sample signals, where n1 + n2 = n.

[0081] Use a mask to represent each missing sub-sample signal in the first group V 1 , and input the sub-sample signals in the first group V 1 and the sub-sample signals missing in the first group V 1 represented by the mask into the preset transformation network in the corresponding time order, and the first group V 1 can be output.The features (true features) corresponding to each sub-sample signal in and the first group of V 1 The features (predicted features) corresponding to each missing sub-sample signal in.

[0082] Similarly, use a mask to represent each missing sub-sample signal in the second group of V 2 Input the sub-sample signals in the second group of V 2 and the sub-sample signals represented by the mask in the second group of V 2 into the preset transformation network in sequence according to the corresponding time order for the missing sub-sample signals, and then the features (true features) corresponding to each sub-sample signal in the second group of V 2 and the features (predicted features) corresponding to the missing sub-sample signals in the second group of V 2 can be output.

[0083] The true feature set includes the features (true features) corresponding to each sub-sample signal in the first group of V 1 and the features (true features) corresponding to each sub-sample signal in the second group of V 2 The true feature set includes the features (true features) corresponding to each sub-sample signal in the first group of V

[0084] The predicted feature set contains the features (predicted features) corresponding to each missing sub-sample signal in the first group of V 1 and the features (predicted features) corresponding to each missing sub-sample signal in the second group of V 2 The predicted feature set contains the features (predicted features) corresponding to each missing sub-sample signal in the first group of V

[0085] It should be understood that the missing sub-sample signals in the first group of V 1 are the sub-sample signals in the second group of V 2 and the missing sub-sample signals in the second group of V 2 are the sub-sample signals in the first group of V 1 In some embodiments, as shown in

[0086] Step S204 includes the following steps: Figure 3

[0087] Step S2041: Determine the first distance value between the true features and the predicted features according to the true feature set and the predicted feature set.

[0088] Specifically, the features (true features) in the true feature set are arranged in sequence according to the time order and denoted as Z 1 , Z 2 , Z 3 , ……, Z n ; the features (predicted features) in the predicted feature set are arranged in sequence according to the time order and denoted as where Z i and are the true features and predicted features corresponding to the i-th sub-sample signal, where j = 1, 2, 3, ……, n.

[0089] In this step, each pair is compared Calculate Then determine the first distance value K according to the following formula 1 :

[0090]

[0091] Step S2042: Decode the features in the true feature set and the predicted feature set respectively through a preset decoder to obtain a first acoustic wave train decoding signal and a second acoustic wave train decoding signal.

[0092] Among them, the structure of the preset decoder and the structure of the preset transformation network can be the same structure. The input of the preset decoder is a feature, and the output is an acoustic wave decoding signal. In one implementation, when the preset transformation network includes multiple cascaded transformer models, the preset decoder also includes multiple cascaded transformer models, and the number of transformer models in the preset transformation network is the same as the number of transformer models in the preset decoder.

[0093] Similarly, before training the preset transformation network according to the preset self-supervised learning strategy, it is also necessary to initialize the parameters of the preset encoder first.

[0094] Step S2043: Determine the second distance value between the first acoustic wave train decoding signal and the second acoustic wave train decoding signal.

[0095] Specifically, determine the second distance value K through the following formula 2 :

[0096]

[0097] Where is the first acoustic wave train decoding signal, is the second acoustic wave train decoding signal.

[0098] Step S2044: Determine the third distance value between the acoustic wave train sample signal and the first acoustic wave train decoding signal.

[0099] Specifically, determine the third distance value K through the following formula 3 :

[0100]

[0101] Step S2045: Determine the distance value between the acoustic wave train sample signal and the second acoustic wave train decoding signal as the fourth distance value.

[0102] Specifically, the fourth distance value K is determined by the following formula 4 :

[0103]

[0104] Step S2046: Determine the loss value according to the first distance value, the second distance value, the third distance value, and the fourth distance value.

[0105] Specifically, the loss value L is determined by the following formula 总 :

[0106] L 总 =γK 1 +(1 - γ)K 2 +K 3 +K 4

[0107] where γ is a constant parameter, 0 < γ < 1.

[0108] Furthermore, in some embodiments, the preset self-supervised learning strategy further includes:

[0109] Update the parameters of the preset decoder according to the loss value, so that when determining the loss value based on the real feature set and the predicted feature set next time, the features in the real feature set and the predicted feature set are decoded by the preset decoder with updated parameters respectively. That is, when the loss value is determined, in addition to updating the parameters of the preset transformation network according to the loss value, the parameters of the preset decoder are also updated.

[0110] For the preset self-supervised learning strategy provided by the present invention, when training to obtain the preset encoder, the loss value is determined by the first distance value, the second distance value, the third distance value, and the fourth distance value obtained from the real feature set and the predicted feature set, which improves the feature extraction accuracy of the preset encoder, and further improves the calculation accuracy of the time difference of arrival.

[0111] In some embodiments, as Figure 4 shown, step S102 includes the following steps:

[0112] Step S1021: Intercept a part of the signal from the first acoustic wave train signal as the first sub-acoustic wave train signal according to the preset interception strategy.

[0113] Specifically, the preset interception strategy may include:

[0114] Preset a preset interception length in advance, and intercept from the preset position of the first acoustic wave train signal according to the preset interception length to obtain the first sub-acoustic wave train signal with a length equal to the preset interception length. Among them, the preset interception length can be set by those skilled in the art according to actual needs, and the present invention does not limit this.

[0115] The preset position is: any position that satisfies the distance from the preset position to the end position of the first acoustic wave train signal is greater than or equal to the preset interception length.

[0116] Step S1022: Extract features from the first sub-acoustic wave train signal through a preset encoder to obtain a first feature set.

[0117] Similarly, when extracting features from the first sub-acoustic wave train signal through a preset encoder, first evenly divide the first sub-acoustic wave train signal into multiple sub-signals, and sequentially input them into the preset encoder according to the time sequence. The preset encoder outputs the features corresponding to each sub-signal to obtain a first feature set. Similarly, the number of sub-signals can be set by those skilled in the art according to actual needs, and the present invention does not limit this. Preferably, the number of sub-signals is a positive integer power of 2. By way of example, the number of sub-signals is 8 or 16.

[0118] Step S1023: Slide and segment the second acoustic wave train signal with a preset sliding window at a preset step length to obtain a plurality of second sub-acoustic wave train signals.

[0119] Specifically, in this step, the preset size is the same as the signal length size of the first sub-acoustic wave train signal. The preset step length is the same as the sub-signal length size. The signal length of the obtained second sub-acoustic wave train signal is the same as the signal length of the first sub-acoustic wave train signal.

[0120] Step S1024: Extract features from each second sub-acoustic wave train signal through a preset encoder to obtain a second feature set corresponding to each second sub-acoustic wave train signal respectively.

[0121] Similarly, when extracting features from the second sub-acoustic wave train signal through a preset encoder, first evenly divide the second sub-acoustic wave train signal into multiple sub-signals, and sequentially input them into the preset encoder according to the time sequence. The preset encoder outputs the features corresponding to each sub-signal to obtain a second feature set.

[0122] Among them, the number of sub-signals of the second sub-acoustic wave train signal is equal to the number of sub-signals of the first sub-acoustic wave train signal.

[0123] Step S1025: Calculate the distance values between the first feature set and each second feature set respectively, and use the second feature set corresponding to the minimum distance value as the target feature set;

[0124] Among them, the features in the first feature set are arranged in sequence according to the time sequence and denoted as F 1 ,F 2 ,F 3 ,……,F n ;

[0125] For any second feature set, arrange the features in the second feature set in chronological order and denote them as

[0126] Compare each pair of Calculate Then, according to the following formula, determine the distance value K between the first feature set and the second feature set 5 :

[0127]

[0128] After calculating the distance values between the first feature set and each second feature set respectively, the second feature set corresponding to the minimum distance value can be found.

[0129] Step S1026: Use the time difference between the second sub-acoustic wave train signal corresponding to the target feature set and the first sub-acoustic wave train signal as the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal.

[0130] Specifically, when calculating the time difference between the second sub-acoustic wave train signal corresponding to the target feature set and the first sub-acoustic wave train signal, the start time of the second sub-acoustic wave train signal corresponding to the target feature set can be obtained, and the start time of the first sub-acoustic wave train signal can be obtained. The difference between the two start times is the time difference.

[0131] To make it easier for those skilled in the art to understand the training process of the preset encoder in the present invention, in combination with Figure 5 The training process is further described as follows:

[0132] As shown in Figure 5, after obtaining the acoustic wave train sample signal f(t), the acoustic wave train sample signal f(t) is evenly divided into 4 sub-sample signals and randomly divided into the first group V 1 , the second group V 2 .

[0133] The first group V 1 includes the first and third sub-sample signals, and the second group V 2 includes the second and fourth sub-sample signals.

[0134] Use a mask to represent each missing sub-sample signal in the first group V 1 (that is, the second and fourth sub-sample signals). Arrange the sub-sample signals in the first group V 1 and the sub-sample signals missing in the first group V represented by the mask 1 in chronological order and input them into the preset transformation network in sequence, and the features Z corresponding to each sub-sample signal in the first group V 1 can be output.1 (True features corresponding to the first sub-sample signal), Z 3 (True features corresponding to the third sub-sample signal) and the features corresponding to the missing sub-sample signals in the first group V 1 (Predicted features corresponding to the second sub-sample signal), (Predicted features corresponding to the fourth sub-sample signal).

[0135] Similarly, use a mask to represent each missing sub-sample signal (i.e., the first and third sub-sample signals) in the second group V. Input the sub-sample signals (the second and fourth sub-sample signals) in the second group V and the sub-sample signals (the first and third sub-sample signals) represented by the mask in the second group V into the preset transformation network in the corresponding time order, and then the features Z corresponding to each sub-sample signal in the second group V can be output 2 2 2 2 2 (True features corresponding to the second sub-sample signal), Z 4 (True features corresponding to the fourth sub-sample signal) and the features corresponding to the missing sub-sample signals in the second group V 2 (Predicted features corresponding to the first sub-sample signal), (Predicted features corresponding to the third sub-sample signal).

[0136] The true feature set includes Z 1 , Z 2 , Z 3 , Z 4 ; The predicted feature set includes Subsequently, the loss value can be determined according to the true feature set and the predicted feature set.

[0137] Furthermore, compare each pair Calculate

[0138] Decode the features of the true feature set through the preset encoder to obtain the first acoustic wave train decoding signal Decode the features of the predicted feature set to obtain the second acoustic wave train decoding signal Calculate:

[0139]

[0140] Then determine the loss value L according to the following formula 总 :

[0141] ​​​​​​

[0142] Obtain the loss value L 总 After that, if the loss value is greater than or equal to the preset loss threshold, update the parameters of the preset transformation network and the parameters of the preset decoder according to the loss value, and continue to train the preset transformation network until the loss value is less than the preset loss threshold.

[0143] Specifically, in one implementation, the preset transformation network includes multiple cascaded transformer models, the preset decoder includes multiple cascaded transformer models, and the number of transformer models in the preset transformation network is the same as the number of transformer models in the preset decoder. More specifically, the preset transformation network is composed of 4 cascaded transformer models. Similarly, the corresponding preset decoder is also composed of 4 cascaded transformer models.

[0144] Taking the example of extracting the features of each sub-sample signal in the first group through the preset transformation network and predicting the features of the missing sub-sample signals in the first group, first, input the first sub-sample signal in the first group into the first transformer model, and the first transformer model outputs the true feature Z corresponding to the first sub-sample signal 1 ; then input the second sub-sample signal represented by the mask (the missing sub-sample signal in the first group) and Z 1 into the second transformer model, and the second transformer model outputs the predicted feature of the second sub-sample signal And so on. Starting from the second transformer model, each transformer model has two inputs. One input is the corresponding sub-sample signal (where the missing sub-sample signal is represented by a mask), and the other input is the output of the previous transformer model. Eventually, the features of each sub-sample signal in the first group can be extracted and the features of the missing sub-sample signals in the first group can be predicted.

[0145] Similarly, the process of decoding using the preset decoder is similar to the above encoding process. Taking the example of decoding the features in the true feature set, input Z in the true feature set 1 into the first transformer model in the preset decoder, and output the decoded sub-signal corresponding to Z 1 ; then input the decoded sub-signal corresponding to Z 1 and Z 2 as the input of the second transformer model in the preset decoder, and output Z 2The corresponding decoded sub-signals, and so on. Starting from the second transformer model in the preset decoder, each transformer model has two inputs. One input is the corresponding real feature, and the other input is the output of the previous transformer model. Finally, the decoded sub-signals corresponding to each real feature in the real feature set can be decoded. By splicing the decoded sub-signals corresponding to each real feature in the real feature set in chronological order, the first acoustic wave train decoded signal can be obtained. The process of using the preset decoder to decode the features in the predicted feature set is the same as the process of using the preset decoder to decode the features in the real feature set, which will not be elaborated here. Similarly, the decoded sub-signals corresponding to each predicted feature in the predicted feature set can be finally decoded. By splicing the decoded sub-signals corresponding to each predicted feature in the predicted feature set in chronological order, the second acoustic wave train decoded signal can be obtained.

[0146] In another embodiment of the present invention, as Figure 6 shown, there is also provided an acoustic logging time difference determination device, which includes:

[0147] An acquisition module 601, configured to acquire a first acoustic wave train signal and a second acoustic wave train signal of the same sound source;

[0148] A determination module 602, electrically connected to the acquisition module 601, and configured to determine the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder according to the first acoustic wave train signal and the second acoustic wave train signal; wherein, the preset encoder is trained according to a preset self-supervised learning strategy.

[0149] In some embodiments, the preset self-supervised learning strategy includes:

[0150] Acquire an acoustic wave train sample signal;

[0151] Average the acoustic wave train sample signal into multiple sub-sample signals, and randomly divide the sub-sample signals into a first group and a second group;

[0152] Extract the features of each sub-sample signal in the first group and the second group respectively through a preset transformation network to obtain a real feature set corresponding to the acoustic wave train sample signal, and predict the features of the missing sub-sample signals in the first group and the second group respectively to obtain a predicted feature set corresponding to the acoustic wave train sample signal;

[0153] Determine a loss value according to the real feature set and the predicted feature set;

[0154] Update the parameters of the preset transformation network according to the loss value, and continue to train the preset transformation network until the loss value is less than a preset loss threshold to obtain a trained preset transformation network;

[0155] Use the trained preset transformation network as the preset encoder.

[0156] In some embodiments, determining the loss value according to the real feature set and the predicted feature set includes:

[0157] Determine the first distance value between the real feature and the predicted feature according to the real feature set and the predicted feature set;

[0158] Decode the features in the real feature set and the predicted feature set respectively through the preset decoder to obtain the first acoustic wave train decoding signal and the second acoustic wave train decoding signal;

[0159] Determine the second distance value between the first acoustic wave train decoding signal and the second acoustic wave train decoding signal;

[0160] Determine the third distance value between the acoustic wave train sample signal and the first acoustic wave train decoding signal;

[0161] Determine the fourth distance value between the acoustic wave train sample signal and the second acoustic wave train decoding signal;

[0162] Determine the loss value according to the first distance value, the second distance value, the third distance value, and the fourth distance value.

[0163] In some embodiments, the loss value L is determined by the following formula 总 :

[0164] L 总 =γK 1 +(1 - γ)K 2 +K 3 +K 4

[0165] Wherein, K 1 is the first distance value, K 2 is the second distance value, K 3 is the third distance value, K 4 is the fourth distance value, and γ is a constant parameter, 0 < γ < 1.

[0166] In some embodiments, the preset self-supervised learning strategy further includes:

[0167] Update the parameters of the preset decoder according to the loss value, so that when determining the loss value according to the real feature set and the predicted feature set next time, decode the features in the real feature set and the predicted feature set respectively through the preset decoder with updated parameters.

[0168] In another embodiment of the present invention, the determining module includes:

[0169] An interception unit, configured to intercept a part of the signal from the first acoustic wave train signal according to a preset interception strategy as the first sub-acoustic wave train signal;

[0170] A sliding segmentation unit, configured to perform sliding segmentation on the second acoustic wave train signal through a sliding window of a preset size with a preset step length to obtain a plurality of second sub-acoustic wave train signals;

[0171] An extraction unit, connected to the interception unit and the sliding segmentation unit respectively, configured to perform feature extraction on the first sub-acoustic wave train signal through a preset encoder to obtain a first feature set; and, configured to perform feature extraction on each second sub-acoustic wave train signal through the preset encoder to obtain a second feature set corresponding to each second sub-acoustic wave train signal;

[0172] A target feature set determination unit, connected to the extraction unit, configured to calculate the distance values between the first feature set and each second feature set respectively, and use the second feature set corresponding to the minimum distance value as the target feature set;

[0173] An arrival time difference determination unit, connected to the target feature set determination unit, configured to use the time difference between the second sub-acoustic wave train signal corresponding to the target feature set and the first sub-acoustic wave train signal as the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal.

[0174] In another embodiment of the present invention, there is also provided a computer program product, which includes a computer program or instruction. When the computer program or instruction is executed by a processor, all or part of the steps of the method in the above method embodiment are implemented, and this embodiment will not be repeated here.

[0175] Further, the computer program product may include one or more computer-executable components configured to execute the embodiment when the program is running; the computer program product may also include a computer program tangibly contained on a computer-readable medium, and the computer program includes program codes for executing any method in the implementation manner of the present invention. In such an implementation manner, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium.

[0176] In another embodiment of the present invention, there is also provided a computer-readable storage medium. The computer program stored in the computer-readable storage medium, when executed by one or more processors, implements all or part of the steps of the method in the above method embodiment, and this embodiment will not be repeated here.

[0177] In another embodiment of the present invention, there is also provided an electronic device 700, Figure 7 which is a schematic structural diagram of the electronic device provided in the embodiment of the present invention, as Figure 7As shown in the figure, the electronic device 700 includes: at least one processor 701, at least one communication bus 702, a user interface 703, at least one external communication interface 704, and a memory 705. Among them, the communication bus 702 is configured to implement connection communication between these components. Among them, the user interface 703 may include a display screen, and the external communication interface 704 may include a standard wired interface and a wireless interface. A computer program is stored on the memory 705, and the memory 705 and one or more processors 701 are communicatively connected to each other. When the computer program is executed by one or more processors, the processor 701 is configured to execute the computer program stored in the memory to implement all or part of the steps of the method in the above method embodiments, and this embodiment will not be repeated here.

[0178] The method, device, and electronic device for determining the acoustic logging time difference provided by the present invention can effectively improve the speed of determining the acoustic logging time difference by determining the arrival time difference through a preset encoder when determining the acoustic logging time difference; and the preset encoder is an encoder obtained by training in advance according to a preset self-supervised learning strategy. By using the preset self-supervised learning strategy, training can be completed on unlabeled sample data to obtain the preset encoder, making the training process of the preset encoder more efficient. In addition, when training the preset encoder in the present invention, the loss value is determined by a first distance value, a second distance value, a third distance value, and a fourth distance value obtained according to a true feature set and a predicted feature set, improving the feature extraction accuracy of the preset encoder, and further improving the calculation accuracy of the arrival time difference.

[0179] Each embodiment in the present invention is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized.

[0180] The protection scope of the present invention is not limited to the above embodiments. Obviously, those skilled in the art can make various changes and deformations to the present invention without departing from the scope and spirit of the present invention. If these changes and deformations fall within the scope of the claims of the present invention and their equivalent technologies, the intention of the present invention also includes these changes and deformations.

Claims

1. A method for determining the acoustic logging time difference, characterized in that, the method includes: obtaining a first acoustic wave train signal and a second acoustic wave train signal of the same sound source; determining the arrival time difference between the first acoustic wave train signal and the second acoustic wave train signal according to the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder; wherein, the preset encoder is obtained by training according to a preset self-supervised learning strategy.

2. The method for determining the acoustic logging time difference according to claim 1, characterized in that, the preset self-supervised learning strategy includes: obtaining an acoustic wave train sample signal; averaging the acoustic wave train sample signal into multiple sub-sample signals, and randomly dividing the sub-sample signals into a first group and a second group; extracting the features of each sub-sample signal in the first group and the second group respectively through a preset transformation network to obtain a true feature set corresponding to the acoustic wave train sample signal, and predicting the features of each missing sub-sample signal in the first group and the second group respectively to obtain a predicted feature set corresponding to the acoustic wave train sample signal; determining a loss value according to the true feature set and the predicted feature set; updating the parameters of the preset transformation network according to the loss value, and continuing to train the preset transformation network until the loss value is less than a preset loss threshold to obtain a trained preset transformation network; using the trained preset transformation network as the preset encoder.

3. The method for determining the acoustic logging time difference according to claim 2, characterized in that, determining the loss value according to the true feature set and the predicted feature set includes: determining a first distance value between the true feature and the predicted feature according to the true feature set and the predicted feature set; decoding the features in the true feature set and the predicted feature set respectively through a preset decoder to obtain a first acoustic wave train decoding signal and a second acoustic wave train decoding signal; determining a second distance value between the first acoustic wave train decoding signal and the second acoustic wave train decoding signal; determining a third distance value between the acoustic wave train sample signal and the first acoustic wave train decoding signal; determining a fourth distance value between the acoustic wave train sample signal and the second acoustic wave train decoding signal; determining the loss value according to the first distance value, the second distance value, the third distance value and the fourth distance value.

4. The method for determining the acoustic logging time difference according to claim 3, characterized in that, The loss value L is determined by the following formula 总 :[[-]] l 总 = γK 1 +(1 - γ)K 2 +K 3 +K 4 where K 0 is the first distance value, K 2 is the second distance value, K 3 is the third distance value, K 4 is the fourth distance value, and γ is a constant.

5. The method for determining the acoustic logging time difference according to claim 3, characterized in that, the preset self-supervised learning strategy further includes: updating the parameters of the preset decoder according to the loss value, so that when determining the loss value according to the true feature set and the predicted feature set next time, decoding the features in the true feature set and the predicted feature set respectively through the preset decoder with updated parameters.

6. The method for determining the acoustic logging time difference according to claim 1, characterized in that, Determining the time difference of arrival between the first acoustic wave train signal and the second acoustic wave train signal according to the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder includes: Intercepting a part of the signal from the first acoustic wave train signal according to a preset intercepting strategy to be used as a first sub-acoustic wave train signal; Performing feature extraction on the first sub-acoustic wave train signal through the preset encoder to obtain a first feature set; Performing sliding segmentation on the second acoustic wave train signal through a sliding window with a preset size at a preset step length to obtain a plurality of second sub-acoustic wave train signals; Performing feature extraction on each second sub-acoustic wave train signal through the preset encoder to respectively obtain a second feature set corresponding to each second sub-acoustic wave train signal; Calculating the distance value between the first feature set and each second feature set respectively, and taking the second feature set corresponding to the minimum distance value as the target feature set; Taking the time difference between the second sub-acoustic wave train signal corresponding to the target feature set and the first sub-acoustic wave train signal as the time difference of arrival between the first acoustic wave train signal and the second acoustic wave train signal.

7. An acoustic logging time difference determination device Characterized in that The device includes: An acquisition module, configured to acquire a first acoustic wave train signal and a second acoustic wave train signal of the same sound source; A determination module, connected to the acquisition module, configured to determine the time difference of arrival between the first acoustic wave train signal and the second acoustic wave train signal according to the first acoustic wave train signal and the second acoustic wave train signal through a preset encoder; wherein, the preset encoder is trained according to a preset self-supervised learning strategy.

8. The acoustic logging time difference determination device according to claim 7 Characterized in that The determination module includes: An intercepting unit, configured to intercept a part of the signal from the first acoustic wave train signal according to a preset intercepting strategy to be used as a first sub-acoustic wave train signal; A sliding segmentation unit, configured to perform sliding segmentation on the second acoustic wave train signal through a sliding window with a preset size at a preset step length to obtain a plurality of second sub-acoustic wave train signals; An extraction unit, respectively connected to the intercepting unit and the sliding segmentation unit, configured to perform feature extraction on the first sub-acoustic wave train signal through the preset encoder to obtain a first feature set; and configured to perform feature extraction on each second sub-acoustic wave train signal through the preset encoder to respectively obtain a second feature set corresponding to each second sub-acoustic wave train signal; A target feature set determination unit, connected to the extraction unit, configured to calculate the distance value between the first feature set and each second feature set respectively, and taking the second feature set corresponding to the minimum distance value as the target feature set; A time difference of arrival determination unit, connected to the target feature set determination unit, configured to take the time difference between the second sub-acoustic wave train signal corresponding to the target feature set and the first sub-acoustic wave train signal as the time difference of arrival between the first acoustic wave train signal and the second acoustic wave train signal.

9. A computer-readable storage medium Characterized in that The computer program stored in the computer-readable storage medium, when executed by one or more processors, implements the steps of the method according to any one of claims 1 to 6.

10. An electronic device, characterized in that it includes a memory and one or more processors, a computer program is stored on the memory, the memory and the one or more processors are communicatively connected to each other, and when the computer program is executed by the one or more processors, the steps of the method according to any one of claims 1 to 6 are executed.

Images