Seismic wave impedance inversion method and system based on cavity space pyramid pooling

Abstract

The invention discloses a seismic wave impedance inversion method and system based on cavity space pyramid pooling, and relates to the field of seismic inversion and oil and gas reservoir prediction, and the method comprises the following steps: S1, carrying out the preprocessing of seismic data, obtaining seismic inversion initial data, and generating a training set; s2, constructing a one-dimensional hole space pyramid pooling network wave impedance inversion model; s3, training the cavity space pyramid pooling network wave impedance inversion model by using the training set to obtain a trained wave impedance inversion model; and S4, inputting seismic data to be inverted into the trained wave impedance inversion model, and outputting an inversion result. Compared with a traditional convolutional neural network seismic inversion method, the inversion method provided by the invention not only has better continuity and noise resistance, but also can effectively reduce the dependence on marked data, and improves the feature capturing capability and the seismic wave impedance inversion profile precision while maintaining the resolution.

Description

Technical Field

[0001] The present invention belongs to the field of seismic inversion and oil and gas reservoir prediction, and particularly relates to a seismic wave impedance inversion method and system based on dilated spatial pyramid pooling. Background Technique

[0002] Seismic inversion is a computational method that uses seismic record data observed on the surface to achieve tomographic imaging of the strata and provides support for reservoir prediction based on the visualized imaging results. The wave impedance parameter is an important lithological parameter and has wide applications in non-uniform geological layer boundary identification, rock fracture calculation, etc. From the perspective of the signal processing process, seismic wave impedance inversion is actually to use a seismic inversion algorithm to eliminate the influence of the source signal, that is, the wavelet signal, from the seismic record data, and finally restore the wave impedance parameter distribution profile of the strata within the corresponding time window.

[0003] According to different calculation methods, wave impedance inversion can be roughly divided into the standard recursive method, the sparse spike method, the model method, etc. These methods all require establishing a strict mathematical model between the seismic wave propagation law and the elastic impedance. Therefore, they are collectively referred to as model-driven seismic inversion. Currently, model-driven seismic inversion generally adopts a linear convolution model. In theory, if the established mathematical model is accurate enough, the inversion calculation should be able to obtain an inversion result with the same resolution as the logging data. However, for complex geological structures, it is difficult to describe the mapping relationship between the source signal and the seismic record detected on the surface through a simple mathematical model. Traditional seismic wave impedance inversion based on the convolution model has the problem of multi-solution, and at the same time, the inversion accuracy is also affected by the number of wells and the well pattern distribution. Generally speaking, the more wells and the more concentrated the well distribution, the higher the reliability of the inversion result. However, as the exploration depth increases, the drilling cost rises sharply, the number of logging points decreases, and the signal-to-noise ratio of deep seismic data decreases, thus affecting the inversion accuracy and stability. Therefore, developing a non-idealized seismic mathematical model suitable for the current data characteristics has become the key to improving the inversion accuracy.

[0004] When traditional inversion models process seismic data, the ability to extract local features is limited, resulting in the loss of local information. Especially under complex geological conditions, it is difficult to accurately capture subtle geological changes, thereby affecting the information correlation between long-distance regions. Summary of the Invention

[0005] In order to solve the technical problems existing in the background technique, the present invention aims to provide a seismic wave impedance inversion method and system based on dilated spatial pyramid pooling.

[0006] In order to solve the technical problems, the technical solution of the present invention is:

[0007] A seismic wave impedance inversion method based on dilated spatial pyramid pooling, the method comprising:

[0008] Preprocess the seismic data, obtain the initial data for seismic inversion, and generate a training set; construct a one-dimensional dilated spatial pyramid pooling network wave impedance inversion model, and use the training set to train the inversion model to obtain a trained wave impedance inversion model; input the seismic data to be inverted into the trained wave impedance inversion model and output the inversion result.

[0009] Furthermore, the preprocessing of the seismic data includes:

[0010] Denoise and normalize the seismic data, extract the effective reflection band, and record the seismic data as S;

[0011] Obtain the wave impedance values of the target formation section according to the logging data, where the wave impedance at the single-channel logging position is denoted as Z;

[0012] Randomly extract 1.0%-2.0% of the total amount of the seismic data and the corresponding wave impedance data obtained to create an inversion training set.

[0013] Furthermore, the construction of the one-dimensional dilated spatial pyramid pooling network wave impedance inversion model includes:

[0014] Encoding module:

[0015] This module includes 4 downsampling layers, and each downsampling layer consists of two consecutive convolutional blocks and a 1×2 max pooling layer;

[0016] Convolutional block structure:

[0017] Each convolutional block consists of the following three parts: a 1×3 one-dimensional convolutional layer for feature extraction; a batch normalization layer BN to normalize the data by mean and variance to improve the model stability; a ReLU activation function for non-linear transformation to enhance the feature expression ability;

[0018] Max pooling layer:

[0019] Extract local features through the pooling window and reduce the data dimension; the formula is to take the maximum value within the window, the stride is stride, and the window size is kernel_size;

[0020] Dilated spatial pyramid pooling module:

[0021] Adopt a parallel sampling method to enhance the perception ability of different-scale features; include a 1×1 convolutional layer, 1×3 dilated convolutional layers with dilation rates of 6, 12, and 18 respectively, and an adaptive pooling layer;

[0022] 1×1 convolutional layer:

[0023] Reduce the dimension or adjust the number of channels through a 1×1 convolution to control the computational complexity;

[0024] Atrous Convolution Layer:

[0025] 1×3 convolution layers with different dilation rates (6, 12, 18) are adopted to expand the receptive field and capture multi-scale features;

[0026] Adaptive Pooling Layer:

[0027] Adaptive average pooling is adopted to maintain a fixed output size and improve the generalization ability of the model;

[0028] Feature Fusion and Channel Adjustment:

[0029] The results of atrous convolution, multi-scale pooling, and 1×1 convolution are concatenated to expand the number of channels; the number of channels is reduced through 1×1 convolution to meet the expected output dimension of the network;

[0030] Decoding Module:

[0031] The structure is symmetric to the encoding module, including 4 upsampling layers, and each upsampling layer consists of a 1×3 transposed convolution layer and two convolution blocks;

[0032] Transposed Convolution Layer:

[0033] Upsampling is performed through transposed convolution to restore the data spatial resolution. The calculation method is the weighted sum of the input signal and the transposed convolution kernel to generate the upsampled output signal.

[0034] Furthermore, the construction of the one-dimensional atrous spatial pyramid pooling network wave impedance inversion model specifically includes:

[0035] Encoding Module: It includes 4 downsampling layers, and the downsampling layer includes two consecutive convolution blocks and a 1×2 max pooling layer;

[0036] Each convolution block consists of a 1×3 one-dimensional convolution layer, a batch normalization layer, and a rectified linear unit. Its formula is expressed as follows:

[0037] z (n) =Conv1D(W (n) ,a (n-1) )+b

[0038]

[0039] In the formula, W (n) is the corresponding convolution kernel; a (n-1) is the input of the convolution layer; b is the bias term; z (n) is the output of the convolution layer; E[z (n) is the expected value of z (n) ; Var[z (n) is the variance of z (n)Variance; ∈ usually represents a small positive number; is the result after batch normalization; is the ReLU activation function; a (n) is the output of the rectified linear unit function;

[0040] The formula for defining the max pooling layer is expressed as follows:

[0041] a (n) (N i , C j , k) = max h＝0,...,kernel_size-1 a (n-1) (N i , C j , stride × k + h)

[0042] In the formula, a (n) (N i , C j , k) is the output of the pooling; a (n-1) (N i , C j , idx) is the input of the pooling; kernel_size is the size of the pooling window; stride is the step size of the pooling operation; k is the position of the pooling window on the input data; h is the position index in the pooling window, ranging from 0 to kernel_size - 1;

[0043] Build a dilated spatial pyramid pooling module to perform parallel sampling on the input, including a 1×1 convolutional layer, 1×3 convolutional layers with dilation rates of 6, 12, and 18 respectively, and an adaptive pooling layer;

[0044] Define the 1×1 convolutional layer, and its formula is expressed as follows:

[0045] Y0 = Conv1D(X, k1)

[0046] In the formula, Conv1D is a one-dimensional convolution, k1 is the convolution kernel; X is the input feature map, and Y0 is the output;

[0047] Define the 1×3 convolutional layers with dilation rates of 6, 12, and 18 respectively, and their formulas are expressed as follows:

[0048] Y r = Conv1D r (X, k r ) where r is the dilation rate

[0049] Y = [Conv1D6(X, k6), Conv1D 12 (X, k 12 ), Conv1D 18 (X, k18 )]

[0050] Wherein, r is the porosity; k r is the convolution kernel; Y r is the output; dilation rate is the number of dilated convolutional layers;

[0051] Define the adaptive pooling layer, and its formula is expressed as follows:

[0052] Y aap = AdaptiveAvgPool(X, output size )

[0053] Wherein, AdaptiveAvgPool is the adaptive pooling; X is the input; output size is the output size; Y aap is the output;

[0054] Concatenate all the results to expand the number of channels, and then reduce the number of channels to the expected value through convolution for output, and its formula is expressed as follows:

[0055] Y final = Conv1D1([Y0, Y6, Y 12 , Y 18 , Y aap )

[0056] Wherein, Y final is the final output; [Y0, Y6, Y 12 , Y 18 , Y aap is the result after concatenating all the outputs;

[0057] Decoding module: Corresponding to the encoding module, it includes 4 upsampling layers, and each upsampling layer consists of a 1×3 transposed convolution layer and two convolutional blocks. The formula of the transposed convolution layer is expressed as follows:

[0058]

[0059] Wherein, x i-k+1 is the element in the input signal x; w k is the k-th element of the transposed convolution kernel; y i is the i-th element in the output signal y.

[0060] Furthermore, use the training set to train the inversion model, including:

[0061] Input of training data, input the training set data into the wave impedance inversion model of the dilated spatial pyramid pooling network in batches;

[0062] Calculate the objective function, calculate the error between the model output and the true wave impedance, use the mean square error (MSE) as the objective function, and track the minimum objective function value during the training process. If the current error is lower than the historical minimum, update the model parameters;

[0063] Adam algorithm parameter optimization:

[0064] Initialize parameters: Set the first moment estimate s0 = 0 and the second moment estimate v0 = 0;

[0065] Calculate the exponentially weighted average: Calculate the exponentially weighted averages of the first and second moments based on the gradients;

[0066] Bias correction: Correct the first and second moment estimates to avoid excessive bias in the initial stage;

[0067] Update the model parameters: Use the corrected moment estimates to adjust the weights and continue to optimize until the objective function converges and the training is completed.

[0068] Furthermore, use the training set to train the inversion model, specifically including:

[0069] Input the training set into the wave impedance inversion model of the hollow spatial pyramid pooling network in batches;

[0070] Calculate the objective function between the network output and the true wave impedance, track the minimum objective function value in the current training process in each iteration. When the current objective function value is less than the previous minimum objective function value, update and save the model parameters. The formula of the objective function is as follows:

[0071]

[0072] In the formula, N is the total number of wave impedance sampling points; Z i is the i-th sampling data point corresponding to the current wave impedance data Z; m is the linearization change parameter of the wave impedance vector of, is the i-th wave impedance sampling data point;

[0073] Update the parameters of the model using the adaptive moment estimation algorithm. The calculation steps of the algorithm are as follows:

[0074] First, initialize the model parameters:

[0075] s0 = 0, v0 = 0

[0076] In the formula, s is the first moment estimate and v is the second moment estimate;

[0077] Calculate the momentum exponentially weighted average:

[0078]

[0079] In the formula, β1 is the first - order moment decay rate, β2 is the second - order moment decay rate, g t is the current gradient, and t is the number of iterations;

[0080] Since at the initial stage of training, s t and v t have small values, deviation correction is performed:

[0081]

[0082] In the formula, is the first - order moment estimate for deviation correction, the second - order moment estimate for deviation correction;

[0083] Update the parameters according to the corrected moment estimates:

[0084]

[0085] In the formula, θ is the weight parameter, α is the learning rate, and ∈ usually represents a small positive number. Stop training until the objective function value drops to the minimum.

[0086] Furthermore, inputting the seismic data to be inverted into the trained wave impedance inversion model includes:

[0087] Input the seismic data into the trained hollow spatial pyramid pooling inversion model, and the inversion process is:

[0088]

[0089] In the formula, D (i) is the sequence data of a one - dimensional seismic trace, f θ the trained inversion model, is the predicted wave impedance value. This process is input channel by channel, that is, each time a seismic trace is input, and the model inverses each seismic trace separately;

[0090] Integrate all the inverted wave impedance data in order to form the final wave impedance profile:

[0091]

[0092] In the formula, is the wave impedance profile obtained by combining the prediction results of all seismic traces.

[0093] A seismic wave impedance inversion system based on hollow spatial pyramid pooling, the system is applied to any of the above - mentioned methods, and the system includes:

[0094] Data pre - processing module: used for pre - processing seismic data, including denoising and normalization;

[0095] Model training module: used to train the wave impedance inversion model of the atrous spatial pyramid pooling network using seismic data to obtain a trained wave impedance inversion model;

[0096] Inversion objective function construction module: calculates the objective function value of each sample according to the difference between the wave impedance output by the model and the true wave impedance;

[0097] Objective function optimization module: updates and optimizes the objective function through the adaptive moment estimation algorithm, thereby adjusting the model parameters;

[0098] Inversion result judgment module: tracks the minimum objective function value during the current training process in each iteration. When the current objective function value is less than the previous minimum objective function value, the model parameters are updated and saved;

[0099] Result output module: after the training is completed, outputs the optimal inversion model for subsequent wave impedance inversion.

[0100] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements an earthquake wave impedance inversion method based on atrous spatial pyramid pooling described in any one of the above.

[0101] A computer-readable storage medium stores a computer program thereon. When the program is executed by a processor, it implements an earthquake wave impedance inversion method based on atrous spatial pyramid pooling described in any one of the above.

[0102] Compared with the prior art, the advantages of the present invention are as follows:

[0103] 1. The method can reduce the dependence on a large amount of labeled data, is especially suitable for the situation where actual logging data is scarce, and improves the adaptability of the model in the case of scarce data.

[0104] 2. By introducing the atrous spatial pyramid pooling layer, it can effectively capture multi-scale local features and solve the problem that traditional inversion models cannot fully extract local information when processing seismic data.

[0105] 3. By strengthening the extraction of detail features, the inversion result is more accurate. Especially in complex geological environments, it can more accurately reflect geological changes and improve the inversion accuracy.

[0106] 4. The method has good continuity, can smoothly represent geological changes, and avoids errors caused by model discontinuity. At the same time, its strong anti-noise ability makes the inversion result more robust and can effectively cope with the noise interference in seismic data. Through the hollow spatial pyramid pooling, the model improves the ability to capture multi-scale features on the basis of maintaining a high spatial resolution, so as to more precisely identify and extract geological features at different levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 It is a flowchart of the seismic wave impedance inversion method according to an embodiment of the present invention;

[0108] Figure 2 It is a signal processing flowchart according to an embodiment of the present invention;

[0109] Figure 3 It is an impedance profile of the impedance inversion result of the Marmousi2 model using a convolutional neural network;

[0110] Figure 4 It is an error profile of the impedance inversion result of the Marmousi2 model using a convolutional neural network;

[0111] Figure 5 It is an impedance profile of the impedance inversion result of the Marmousi2 model of the algorithm proposed by the present invention;

[0112] Figure 6 It is an error profile of the impedance inversion result of the Marmousi2 model of the algorithm proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0113] The following describes the specific embodiments of the present invention in conjunction with the embodiments:

[0114] It should be noted that the structures, ratios, sizes, etc. shown in this specification are only used to cooperate with the content disclosed in the specification for those skilled in this technology to understand and read, and are not used to limit the limited conditions under which the present invention can be implemented. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed by the present invention.

[0115] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope in which the present invention can be implemented. The change or adjustment of their relative relationships, without substantial change in the technical content, should also be regarded as the scope in which the present invention can be implemented.

[0116] Embodiment 1:

[0117] AsFigure 1 As shown in Figure 1 , a seismic wave impedance inversion method based on dilated spatial pyramid pooling includes the following steps:

[0118] S1: Preprocess seismic data to obtain initial data for seismic inversion and generate a training set;

[0119] S101: Denoise and normalize the seismic data, extract the effective reflection band, and denote the seismic data as S;

[0120] S102: Obtain the wave impedance values of the target formation section according to well logging data, where the wave impedance at the single-channel well logging position is denoted as Z;

[0121] S103: Randomly select 1%-2% (5-10 pairs of data) of the total amount of the seismic data and the corresponding wave impedance data obtained as the training set.

[0122] S2: Construct a one-dimensional dilated spatial pyramid pooling network wave impedance inversion model;

[0123] S201: Encoding module: It contains 4 downsampling layers, and each downsampling layer includes two consecutive convolutional blocks and a 1×2 max pooling layer;

[0124] S202: Each convolutional block consists of a 1×3 one-dimensional convolutional layer, a batch normalization layer, and a rectified linear unit function, and its formula is expressed as follows:

[0125] z (n) = Conv1D(W (n) , a (n-1) ) + b

[0126]

[0127] In the formula, W (n) is the corresponding convolution kernel; a (n-1) is the input of the convolutional layer; b is the bias term; z (n) is the output of the convolutional layer; E[z (n) is the expected value (mean) of z (n) ; Var[z (n) is the variance of z (n) ; ∈ usually represents a small positive number (to prevent the denominator from being zero); is the result after batch normalization processing; is the ReLU activation function; a (n) is the output of the rectified linear unit function;

[0128] S203: Define the formula of the max pooling layer as follows:

[0129] a (n) (N i,C j ,k) = max h＝0,...,kernel_size-1 a (n-1) (N i ,C j ,stride × k + h)

[0130] In the formula, a (n) (N i ,C j ,k) is the output of pooling; a (n-1) (N i ,C j ,idx) is the input of pooling; kernel_size is the size of the pooling window; stride is the stride of the pooling operation; k is the position of the pooling window on the input data; h is the position index in the pooling window, ranging from 0 to kernel_size - 1;

[0131] S204: Construct a dilated spatial pyramid pooling module to perform parallel sampling on the input, including a 1×1 convolutional layer, 1×3 convolutional layers with dilation rates of 6, 12, and 18 respectively, and an adaptive pooling layer;

[0132] S205: Define a 1×1 convolutional layer, and its formula is expressed as follows:

[0133] Y0 = Conv1D(X, k1)

[0134] In the formula, Conv1D is a one-dimensional convolution, k1 is the convolution kernel; X is the input feature map, and Y0 is the output;

[0135] S206: Define 1×3 convolutional layers with dilation rates of 6, 12, and 18 respectively, and their formula is expressed as follows:

[0136] Y r = Conv1D r (X, k r ) where r is the dilation rate

[0137] Y = [Conv1D6(X, k6), Conv1D 12 (X, k 12 ), Conv1D 18 (X, k 18 )]

[0138] In the formula, r is the dilation rate; k r is the convolution kernel; Y r is the output; dilation rate is the number of dilated convolutional layers (3 in this method);

[0139] S207: Define an adaptive pooling layer, and its formula is expressed as follows:

[0140] Y aap = AdaptiveAvgPool(X, output size )

[0141] Where AdaptiveAvgPool is adaptive pooling; X is the input; output size is the output size; Y aap is the output;

[0142] S208: Concatenate all the results to expand the number of channels, and then reduce the number of channels to the expected value through convolution for output. Its formula is as follows:

[0143] Y final = Conv1D1([Y0, Y6, Y 12 , Y 18 , Y aap )

[0144] Where Y final is the final output; [Y0, Y6, Y 12 , Y 18 , Y aap is the result after concatenating all the outputs;

[0145] S209: Decoding module: Corresponding to the encoding module, it contains 4 upsampling layers. Each upsampling layer consists of a 1×3 transposed convolution layer and two convolutional blocks. The formula of the transposed convolution layer is as follows:

[0146]

[0147] Where x i-k+1 is the element in the input signal x; w k is the k-th element of the transposed convolution kernel; y i is the i-th element in the output signal y.

[0148] S3: Use the training set to train the hole spatial pyramid pooling network wave impedance inversion model to obtain a trained wave impedance inversion model;

[0149] S301: Batch input the training set into the hole spatial pyramid pooling network wave impedance inversion model;

[0150] S302: Calculate the objective function between the network output and the true wave impedance, and track the minimum objective function value during the current training process in each iteration. When the current objective function value is less than the previous minimum objective function value, update and save the model parameters. Its objective function formula is as follows:

[0151]

[0152] Wherein, N is the total number of wave impedance sampling points; Z i is the i-th sampling data point corresponding to the current wave impedance data Z; m is the linearized change parameter of the wave impedance vector ; is the i-th wave impedance sampling data point;

[0153] S303: Update the parameters of the model using the adaptive moment estimation algorithm. The calculation steps of the algorithm are as follows:

[0154] S304: First, initialize the model parameters:

[0155] s0 = 0, v0 = 0

[0156] Wherein, s is the first moment estimation, and v is the second moment estimation.

[0157] S305: Calculate the momentum exponentially weighted average:

[0158]

[0159] Wherein, β1 is the first moment decay rate (usually set to 0.9), β2 is the second moment decay rate (usually set to 0.999), g t is the current gradient, and t is the number of iterations;

[0160] S306: Since in the initial stage of training, s t and v t have small values, so bias correction is performed:

[0161]

[0162] Wherein, is the first moment estimation of bias correction, the second moment estimation of bias correction;

[0163] S307: Update the parameters according to the corrected moment estimation:

[0164]

[0165] Wherein, θ is the weight parameter, α is the learning rate, and ∈ usually represents a small positive number. Stop training until the value of the objective function drops to the minimum.

[0166] S4: Input the seismic data to be inverted into the trained wave impedance inversion model and output the inversion result;

[0167] S401: Input the seismic data into the trained hollow spatial pyramid pooling inversion model. The inversion process is as follows:

[0168]

[0169] where D (i) is the sequence data of one-dimensional seismic traces, and f θ is the trained inversion model, is the predicted wave impedance value. This process inputs one trace at a time, that is, each seismic trace is input one by one, and the model performs inversion on each seismic trace separately.

[0170] S402: Integrate all the inverted wave impedance data in order to form the final wave impedance profile:

[0171]

[0172] where is the wave impedance profile obtained by combining the prediction results of all seismic traces.

[0173] As Figure 2 shown, a seismic wave impedance inversion signal processing flow using hollow spatial pyramid pooling includes the following steps:

[0174] Preprocess the seismic data to obtain the initial data for seismic inversion. Randomly extract 2% (10 pairs) of the seismic data and the corresponding generated wave impedance data to form a training set. The dimension of each pair of data is (1, 2800), that is, 2800 sampling points.

[0175] Set relevant hyperparameters before starting model training. The total number of training epochs is 1000, and the training batch size is 5. The initial learning rate is set to 0.001, and the learning rate is reduced by 10 times every 250 epochs. Select a weight decay of 0.0001 to limit the L2 norm of the weights.

[0176] Start training the model with the input data. In each iteration, track the minimum value of the objective function during the current training process. When the current objective function value is less than the previous minimum objective function value, update and save the model parameters.

[0177] After training is completed, output the trained model. Finally, use the trained model to perform wave impedance inversion on the entire seismic data.

[0178] To verify the effectiveness of the proposed algorithm, apply the algorithm to the impedance inversion calculation of the Marmousi2 model, and compare it with the inversion results of the convolutional neural network (CNN) algorithm. Figure 3 and Figure 4 are the cross-sectional diagrams of the impedance inversion results and the impedance error cross-sectional diagrams of the Marmousi2 model of the convolutional neural network (CNN) algorithm. Figure 5 and Figure 6The cross-sectional view of the impedance inversion result and the impedance error cross-sectional view of the Marmousi2 model for the proposed seismic wave impedance inversion method based on dilated spatial pyramid pooling. As can be seen from the figure, the inversion result of the proposed method has better stability and continuity, and the calculation error is significantly lower than that of the convolutional neural network algorithm, and the resolution is significantly improved in the complex structure area.

[0179] Example 2:

[0180] The present invention provides a seismic wave impedance inversion system using dilated spatial pyramid pooling, which can be used to implement the above-mentioned seismic wave impedance inversion method based on dilated spatial pyramid pooling. Specifically, it includes:

[0181] Data preprocessing module: used to preprocess seismic data, including denoising and normalization.

[0182] Model training module: connected to the data preprocessing module, used to train the dilated spatial pyramid pooling network wave impedance inversion model using seismic data to obtain a trained wave impedance inversion model.

[0183] Inversion objective function construction module: calculates the objective function value of each sample according to the difference between the wave impedance output by the model and the true wave impedance.

[0184] Objective function optimization module: updates and optimizes the objective function through the adaptive moment estimation algorithm, thereby adjusting the model parameters.

[0185] Inversion result judgment module: tracks the minimum objective function value in the current training process in each iteration. When the current objective function value is less than the previous minimum objective function value, the model parameters are updated and saved.

[0186] Result output module: outputs the optimal inversion model after training for subsequent wave impedance inversion.

[0187] Example 3:

[0188] This embodiment provides a terminal device, which includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function. The processor described in the embodiments of the present invention can be used for the operation of a seismic wave impedance inversion method based on dilated spatial pyramid pooling, including the following steps:

[0189] S1: Preprocess the seismic data to obtain initial seismic inversion data and generate a training set;

[0190] S2: Construct a one-dimensional dilated spatial pyramid pooling network wave impedance inversion model;

[0191] S3: Use the training set to train the dilated spatial pyramid pooling network wave impedance inversion model to obtain a trained wave impedance inversion model;

[0192] S4: Input the seismic data to be inverted into the trained wave impedance inversion model and output the inversion result.

[0193] Embodiment 4:

[0194] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a terminal device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, and this storage space stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by a processor are stored in this storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0195] One or more instructions stored in the computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps in the above embodiment regarding a seismic wave impedance inversion method based on dilated spatial pyramid pooling; one or more instructions in the computer-readable storage medium are loaded and executed by a processor to perform the following steps:

[0196] S1: Preprocess the seismic data, obtain the initial data for seismic inversion, and generate a training set;

[0197] S2: Construct a one-dimensional dilated spatial pyramid pooling network wave impedance inversion model;

[0198] S3: Use the training set to train the dilated spatial pyramid pooling network wave impedance inversion model to obtain a trained wave impedance inversion model;

[0199] S4: Input the seismic data to be inverted into the trained wave impedance inversion model and output the inversion result.

[0200] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. 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 memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0201] 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 flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0202] 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 operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0203] These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0204] The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the above embodiments. Without departing from the scope of knowledge possessed by those of ordinary skill in the art, various changes can be made without departing from the spirit of the present invention.

[0205] Many other changes and modifications can be made without departing from the spirit and scope of the present invention. It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A seismic wave impedance inversion method based on void space pyramid pooling, characterized in that: The method comprises: Preprocess the seismic data, obtain initial seismic inversion data, and generate a training set; construct a one-dimensional void space pyramid pooling network wave impedance inversion model, use the training set to train the inversion model, and obtain a trained wave impedance inversion model; input the seismic data to be inverted into the trained wave impedance inversion model, and output the inversion result.

2. The seismic wave impedance inversion method based on void space pyramid pooling according to claim 1, characterized in that: The preprocessing of seismic data includes: De-noise and normalize the seismic data to extract the effective reflection band, and the seismic data is recorded as S; Obtain the wave impedance value of the target formation section according to the logging data, wherein the wave impedance at the single-channel logging position is recorded as Z; 1.0%-2.0% of the total seismic data and the corresponding acquired wave impedance data were randomly selected to create an inversion training set.

3. The seismic wave impedance inversion method based on void space pyramid pooling according to claim 1, characterized in that: The one-dimensional void space pyramid pooling network wave impedance inversion model is constructed, including: Encoding module: The module includes 4 downsampling layers, each of which consists of two consecutive convolution blocks and a 1×2 maximum pooling layer; Convolutional block structure: Each convolution block consists of the following three parts: 1×3 one-dimensional convolution layer for feature extraction; batch normalization layer BN normalizes data by mean and variance to improve model stability; ReLU activation function performs nonlinear transformation to enhance feature expression capabilities; Max Pooling Layer: Extract local features through pooling windows and reduce data dimensions; the formula is expressed as taking the maximum value in the window, the step length is stride, and the window size is kernel_size; Atrous spatial pyramid pooling module: A parallel sampling method is used to enhance the perception of features of different scales; including a 1×1 convolution layer, a 1×3 dilated convolution layer with dilation rates of 6, 12, and 18, and an adaptive pooling layer; 1×1 convolutional layer: Reduce the dimension or adjust the number of channels through 1×1 convolution to control the computational complexity; Atrous convolutional layer: A 1×3 convolutional layer with different dilation rates (6, 12, 18) is used to expand the receptive field to capture multi-scale features; Adaptive pooling layer: Adaptive average pooling is used to maintain a fixed output size and improve the generalization ability of the model; Feature fusion and channel adjustment: Concatenate the results of dilated convolution, multi-scale pooling, and 1×1 convolution to increase the number of channels; reduce the number of channels through 1×1 convolution to make it consistent with the expected output dimension of the network; Decoding module: The structure is symmetrical with the encoding module, including 4 upsampling layers, each of which consists of a 1×3 deconvolution layer and two convolution blocks; Deconvolution layer: Upsampling is performed through deconvolution to restore the spatial resolution of the data. The calculation method is the weighted sum of the input signal and the deconvolution kernel to generate the upsampled output signal.

4. The seismic wave impedance inversion method based on void space pyramid pooling according to claim 1, characterized in that: The construction of a one-dimensional void space pyramid pooling network wave impedance inversion model specifically includes: Encoding module: includes 4 downsampling layers, which include two consecutive convolution blocks and a 1×2 maximum pooling layer; Each convolution block consists of a 1×3 one-dimensional convolution layer, a batch normalization layer, and a linear rectification function, and its formula is as follows: z (n) =Conv1D(W (n) ,a (n-1) )+b Where W (n) is the corresponding convolution kernel; a (n-1) is the input of the convolutional layer; b is the bias term; z (n) is the output of the convolutional layer; E[z (n) ] is z (n) The expected value of Var[z (n) ] is z (n) The variance of ;∈ usually represents a small positive number; This is the result after batch normalization; is the ReLU activation function; a (n) is the output of the linear rectification function; The formula for defining the maximum pooling layer is as follows: a (n) (N i ,C j ,k)=max h＝0,...,kernel_size-1 a (n-1) (N i ,C j ,stride×k+h) In the formula, a (n) (N i ,C j ,k) is the output of pooling; a (n-1) (N i ,C j ,idx) is the pooling input; kernel_size is the size of the pooling window; stride is the step size of the pooling operation; k is the position of the pooling window on the input data; h is the position index in the pooling window, ranging from 0 to kernel_size-1; Construct a dilated spatial pyramid pooling module to sample the input in parallel, including a 1×1 convolutional layer, a 1×3 convolutional layer with dilation rates of 6, 12, and 18, and an adaptive pooling layer; Define a 1×1 convolutional layer, whose formula is as follows: Y0=Conv1D(X,k1) In the formula, Conv1D is a one-dimensional convolution, k1 is the convolution kernel; X is the input feature map, and Y0 is the output; Define a 1×3 convolutional layer with expansion rates of 6, 12, and 18, respectively. The formula is as follows: Y r =Conv1D r (X,k r )where r is the dilation rate Y=[Conv1D6(X,k6),Conv1D 12 (X,k 12 ),Conv1D 18 (X,k 18 )] In the formula, r is the void ratio; k r is the convolution kernel; Y r is the output; dilation rate is the number of hole convolution layers; Define the adaptive pooling layer, whose formula is as follows: Y aap =AdaptiveAvgPool(X,output size ) Where AdaptiveAvgPool is adaptive pooling; X is input; output size is the output size; Y aap is the output; All the results are concatenated to expand the number of channels, and then the number of channels is reduced to the expected value output through convolution. The formula is as follows: AND final =Conv1D1([Y0,Y6,Y 12 ,AND 18 ,AND aap ]) Where Y final is the final output; [Y0,Y6,Y 12 ,Y 18 ,Y aap ] is the result of concatenating all outputs; Decoding module: Corresponding to the encoding module, it includes 4 upsampling layers. Each upsampling layer consists of a 1×3 deconvolution layer and two convolution blocks. The deconvolution layer formula is expressed as follows: In the formula, x i-k+1 is the element in the input signal x; w k is the kth element of the deconvolution kernel; y i is the i-th element in the output signal y.

5. The seismic wave impedance inversion method based on void space pyramid pooling according to claim 1, characterized in that: The inversion model is trained using the training set, including: Training data input: input the training set data in batches into the void space pyramid pooling network wave impedance inversion model; Calculate the objective function, calculate the error between the model output and the actual wave impedance, use the mean square error (MSE) as the objective function, and track the minimum objective function value during the training process. If the current error is lower than the historical minimum value, update the model parameters. Adam algorithm parameter optimization: Initialization parameters: Set the first-order moment estimate s0 = 0 and the second-order moment estimate v0 = 0; Calculate exponentially weighted averages: Calculate the exponentially weighted averages of the first and second moments based on the gradient; Bias correction: Correct the first-order and second-order moment estimates to avoid excessive bias in the initial stage; Update model parameters: Use the corrected moment estimates to adjust the weights and continue optimizing until the objective function converges and the training is complete.

6. The seismic wave impedance inversion method based on void space pyramid pooling according to claim 1, characterized in that: The inversion model is trained using the training set, including: The training set is batch-inputted into the void space pyramid pooling network wave impedance inversion model; Calculate the objective function between the network output and the real wave impedance, and track the minimum objective function value in the current training process in each iteration. When the current objective function value is less than the previous minimum objective function value, update and save the model parameters. The objective function formula is as follows: Where N is the total number of wave impedance sampling points; Z i is the i-th sampling data point corresponding to the current wave impedance data Z; m is the wave impedance vector The linear variation parameters, is the i-th wave impedance sampling data point; Adaptive moment estimation algorithm is used to update the parameters of the model. The algorithm calculation steps are as follows: First, initialize the model parameters: s0=0,v0=0 In the formula, s is the first-order moment estimate, v is the second-order moment estimate; Calculate the weighted average of the momentum index: Where β1 is the first-order moment decay rate, β2 is the second-order moment decay rate, g t is the current gradient, t is the number of iterations; Since in the early stage of training, s t and v t The value is too small, so a deviation correction is performed: In the formula, is the bias-corrected first-order moment estimate, Bias-corrected second-moment estimation; Update the parameters based on the corrected moment estimates: In the formula, θ is the weight parameter, α is the learning rate, and ∈ usually represents a small positive number until the objective function value drops to the minimum and the training stops.

7. The seismic wave impedance inversion method based on void space pyramid pooling according to claim 1, characterized in that: The step of inputting the seismic data to be inverted into the trained wave impedance inversion model comprises: The seismic data is input into the trained cavity space pyramid pooling inversion model. The inversion process is: Where D (i) is the sequence data of one-dimensional seismic trace, f θ The trained inversion model, To predict the wave impedance value, this process is input one by one, that is, each time a seismic trace is input, the model performs inversion on each seismic trace separately; All inverted impedance data are sequentially integrated into the final impedance profile: In the formula, The wave impedance profile is obtained by combining the prediction results of all seismic traces.

8. A seismic wave impedance inversion system based on void space pyramid pooling, characterized in that: The system is applied to the method described in any one of claims 1 to 7, and the system comprises: Data preprocessing module: used to preprocess seismic data, including denoising and normalization; Model training module: used to train the cavity space pyramid pooling network wave impedance inversion model using seismic data to obtain a trained wave impedance inversion model; Inversion objective function construction module: Calculate the objective function value of each sample based on the difference between the wave impedance output by the model and the real wave impedance; Objective function optimization module: updates and optimizes the objective function through the adaptive moment estimation algorithm, thereby adjusting the model parameters; Inversion result judgment module: in each iteration, the minimum objective function value in the current training process is tracked. When the current objective function value is less than the previous minimum objective function value, the model parameters are updated and saved. Result output module: After the training is completed, the optimal inversion model is output for subsequent wave impedance inversion.

9. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a seismic wave impedance inversion method based on void space pyramid pooling according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements a seismic wave impedance inversion method based on void space pyramid pooling according to any one of claims 1 to 7.

Images