Neural Network-Based Time-Lapse Seismic Prediction Method and System

Abstract

The present invention discloses a time-lapse seismic prediction method and system based on a neural network, including: establishing a non-linear mapping relationship between streamer data in a non-pipeline area and OBN data, training a neural network model, and when the model is stable, using the streamer data in the pipeline area to predict the OBN pipeline area data, and then putting the predicted data into the OBN data pipeline area to complete the reconstruction of the OBN data. The advantages of the present invention are: predicting and reconstructing the data in the pipeline area more accurately, and the time-lapse seismic based on the reconstructed seismic data can more effectively monitor the remaining gas reservoirs. Improve the recovery rate and achieve high-efficiency development.

Description

Technical Field

[0001] The present invention relates to the technical field of earthquake prediction, and particularly relates to a time-lapse seismic prediction method and system based on a U-Net neural network or a LeNet neural network. Background Art

[0002] In the western South China Sea, there are various types of producing gas fields, covering different development strata, different temperature-pressure systems, different drive types, different lithologic combinations, etc. After years of development, each gas field has successively entered the production decline period. Among them, due to the low-permeability reservoirs of some gas groups in the gas fields, with strong spatial heterogeneity, the reservoir connectivity is poor and the development is inefficient. In order to achieve efficient development of the gas fields, seismic monitoring of the gas reservoirs is required, and then the remaining gas prediction and adjustment and tapping potential are carried out for the gas reservoirs that have reached the middle and late stages or even the end stage of development. Time-lapse seismic, as an important and effective technology for monitoring the changes of oil and gas reservoirs during development and quantitatively predicting the distribution of remaining oil and gas, has become an important research direction and means for tapping potential and adjusting the gas fields.

[0003] Based on two sets of 3D seismic data collected at different times in the same work area, namely the 3D towed cable data collected early and the 3D OBN data collected recently. Since the geophones are deployed avoiding the submarine pipeline area during the OBN data collection, data gaps appear in the collected 3D OBN data. Time-lapse seismic, as an important technology for monitoring the changes of oil and gas reservoirs during development and quantitatively predicting the distribution of remaining oil and gas, requires two sets of data collected at different times to perform seismic monitoring on the oil and gas reservoirs. The pipeline part data with gaps in the 3D OBN seismic data in this work area contains a large number of main producing areas of oil and gas reservoirs and lacks a large amount of information. In order to effectively monitor the remaining oil and gas reservoirs using time-lapse technology, a data set is made using the 3D towed cable data collected early, and the U-Net neural network is used to predict the gaps in the recent 3D OBN data. The two sets of data are in the same work area and have the same underground structure, which can reflect a relatively complete non-linear mapping relationship. The existing neural network intelligent prediction technology has been applied in many aspects in the geophysical field. Using the neural network to learn the differences between the two sets of seismics and perform non-linear mapping can provide more complete monitoring data for time-lapse seismic. Technical solution of the prior art one

[0004] Chinese Invention Patent: A 5D Interpolation Method for Seismic Data Based on 5D-CNN, Patent No.: CN115184985A, Publication Date: October 14, 2022; discloses a 5D interpolation method for seismic data based on 5D-CNN, including: using low-dimensional convolutional cascades to build 5D convolutions and designing 5D convolutional layers; constructing a 5D-CNN network based on the 5D convolutional layer; constructing a training data set; constructing a loss function; inputting the training set Dtrain into the 5D-CNN network, minimizing the loss function through an optimization algorithm, training the 5D-CNN network parameters, and evaluating the network performance using the validation set Dval after each round of training; after training is completed, using the trained 5D-CNN network to perform interpolation processing on data from other work areas.

[0005] Disadvantages of the Prior Art I

[0006] This patent can only make a training set in a certain work area for model training and then apply the model to another work area for interpolation processing. The sample work area of the training set of this technology and the final application work area are not within the same work area range. The underground structure information in different work areas is different, and the underground information of the application work area cannot be learned in the model, requiring the model to have extremely strong generalization ability. Summary of the Invention

[0007] In view of the defects of the prior art, the present invention provides a time-lapse seismic prediction method and system based on a neural network. Conduct comprehensive interpretation and analysis of the time-lapse seismic differences of the gas reservoir, implement the distribution of remaining gas in the gas field, so as to guide the adjustment of the development plan, improve the recovery rate, and achieve efficient development.

[0008] In order to achieve the above invention purposes, the technical solutions adopted by the present invention are as follows:

[0009] A time-lapse seismic prediction method based on a neural network, including the following steps:

[0010] Step 1: Data matching and selection based on vector offset.

[0011] Collect streamer data and Ocean Bottom Node (OBN) data. To make the two sets of data reflect the same geological information, prestack streamer data and prestack OBN data are sorted into the CDP (Common Depth Point) domain, and data matching and selection based on vector offset are performed.

[0012] Step 2: Prestack consistency processing. To eliminate non-repeatable factors in the data and accurately obtain the time-lapse seismic response differences of the oil and gas field, global matching correction needs to be performed on the CDP gather after vector offset selection;

[0013] Step 3: Post-stack consistency processing: Using a statistical cross-equalization method, the energy, frequency, and phase differences of the stacked seismic data are corrected to effectively eliminate unwanted differences.

[0014] Step 4: Build a U-Net neural network or a LeNet neural network and select the Tanh function as the activation function.

[0015] Step 5: Create a sample set. The sample set includes features and labels. Features serve as network inputs, and labels serve as desired outputs. The training set includes an equal amount of feature data and label data. The streamer data will serve as the neural network input features, and the OBN data will serve as the desired output labels. Extract the streamer and OBN data into a 128*128 two-dimensional matrix, removing samples containing empty channels.

[0016] Step 6: Model training. Input the training set into the U-Net network, initialize the weight vector w and bias b, and calculate the error L between the node output and the expected output MSE , when L MSE Once converged, the training ends and the weights of each parameter are stored.

[0017] Step 7: Pipeline Data Reconstruction. Extract the streamer pipeline data into a 128x128 two-dimensional matrix, perform zero-meaning processing, and use it as model input. Input the streamer data into the model generated in Step 6 to obtain the expected output. After de-zero-meaning processing, place the data back into the OBN pipeline area to complete the OBN pipeline data reconstruction.

[0018] Furthermore, the data matching selection based on vector offset in step 1 is as follows:

[0019] Calculate the interception range of the gather, intercept the gather where the offset of the streamer and OBN data overlap, and use it as the basic data for subsequent processing. Count the maximum longitudinal offset of the OBN data (MaxOffsetY) to define the interception range for the streamer data. The calculation formula is as follows:

[0020] MaxOffsetY=(max{SY 1max ,RY 1max}-min{SY 1min ,RY 1min})÷2

[0021] Where SY1 is the longitudinal coordinate value of the OBN data shot point, and RY1 is the longitudinal coordinate value of the OBN data detection point.

[0022] MaxOffsetY is finally determined based on the distribution of the maximum longitudinal offset of OBN.

[0023] Statistically calculate the maximum lateral offset distance (MaxOffsetX) of the towed cable data to delimit the interception range for OBN data. The calculation formula is as follows:

[0024] MaxOffsetX = (max{SX 2max , RX 2max} - min{SX 2min , RX 2min}) ÷ 2

[0025] In the formula, SX1 is the lateral coordinate value of the shot point of the towed cable data, and RX1 is the lateral coordinate value of the geophone point of the towed cable data.

[0026] Finally determine MaxOffsetX according to the distribution of the maximum lateral offset distance of the towed cable.

[0027] Further, in step two, the global matching correction includes: noise suppression, ghost wave suppression, multiple wave suppression, wavelet shaping, consistent velocity analysis, and consistent prestack time migration.

[0028] Further, in step five, to avoid the problem of gradient explosion during the model training process, the sample set is processed to have zero mean. The calculation formula is:

[0029] X′ = X - μ

[0030] In the formula, X is the original data, and μ is the mean.

[0031] Further, in step six, calculate the error L between the output of the node and the expected output, specifically as follows:

[0032] Express the composition formula of the neuron as:

[0033]

[0034] where y represents the output of the neuron, f(·) represents the activation function, ω i represents the weight of the i-th input signal, x i represents the i-th input feature, and b represents the bias.

[0035] First, initialize the weights ω1, ω2, …, ω i and the bias b in the neural network, and then input the features x1, x2, …, x i into the neural network, and forward propagate to obtain the output y.

[0036] Then calculate the loss function L between the neuron output value and the expected output MSE , and the calculation formula is:

[0037]

[0038] Among them, n is the number of samples, and y i is the true value, and

[0039] MSE is the predicted value. Update and iterate the weight and bias parameters according to the loss function. When L

[0040] converges, the optimal solution of the model is obtained.

[0041] The present invention also discloses a time-lapse seismic prediction system, which can be used to implement the above time-lapse seismic prediction method. Specifically, it includes: a data matching and selection module, a pre-stack consistency processing module, a post-stack consistency processing module, a neural network module, a pipeline area data reconstruction module, and a seismic judgment module;

[0042] Data matching and selection module: Collect streamer data and OBN data, sort the pre-stack streamer data and pre-stack OBN data into the CDP domain, and perform data matching and selection based on vector offset.

[0043] Pre-stack consistency processing module: Perform global matching correction on the CDP gathers after vector offset selection;

[0044] Post-stack consistency processing module: Adopt a statistical-based mutual equalization processing method to correct the differences in energy, frequency, and phase of the stacked seismic data, etc., to eliminate unwanted differences.

[0045] Neural network module: Through the made training sample set, input the training set into the neural network, initialize the weight vector, calculate the error between the node output and the expected output. When the error converges, the training ends, and each parameter weight is stored. Input the streamer data into the training model to obtain the model output and perform anti-normalization.

[0045] Pipeline area data reconstruction module: After anti-normalization, put the data back into the OBN pipeline area for reconstruction.

[0046] Seismic judgment module: According to the reconstruction result, judge whether there is time-lapse seismic.

[0047] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above time-lapse seismic prediction method is implemented.

[0048] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the program is executed by the processor, the above time-lapse seismic prediction method is implemented.

[0049] Compared with the prior art, the advantages of the present invention are:

[0050] More accurately predict and reconstruct the data in the pipeline area. Time-lapse seismic based on the reconstructed seismic data can more effectively monitor the remaining gas reservoirs, improve the recovery rate, and achieve high-efficiency development. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a comparison diagram of the azimuth angles of the streamer and OBN data traces in an embodiment of the present invention.

[0052] Figure 2 It is a result diagram of statistically analyzing the vertical offset data of OBN in an embodiment of the present invention.

[0053] Figure 3 It is a result diagram of statistically analyzing the horizontal offset data of the streamer in an embodiment of the present invention.

[0054] Figure 4 It is a comparison diagram before and after data reconstruction in an embodiment of the present invention. a: Original OBN data, b: Reconstructed OBN data in the pipeline area. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the following provides further detailed descriptions of the present invention based on the drawings and by listing embodiments.

[0056] A time-lapse seismic prediction method based on a neural network includes the following steps:

[0057] Step 1: Data matching and extraction based on vector offsets.

[0058] Due to the influence of factors such as acquisition and processing on seismic data in different years, non-uniformity is inevitable. Signal differences caused by non-reservoir factors will lead to misinterpretation of time-lapse seismic information. Therefore, it is necessary to perform matching processing on time-lapse seismic data to eliminate the influence of non-reservoir factors, ensure that the signal differences reflect the current gas reservoir exploitation situation and the remaining gas distribution characteristics, and realize the study of the time-lapse changes of reservoir parameters. The streamer data is collected in the north-south direction, and the Ocean Bottom Node (OBN) data is collected in the east-west direction. To make the two sets of data reflect the same geological information, the pre-stack streamer data and pre-stack OBN data are sorted into the CDP (Common Depth Point) domain, and data matching and extraction based on vector offsets are performed.

[0059] Further, the data matching and extraction based on vector offsets in Step 1 are specifically as follows:

[0060] Calculate the trace gather truncation range. The comparison of the azimuth angles of the streamer and OBN data traces is as Figure 1, intercept the gather where the offset between the towing cable and the OBN data is roughly coincident as the basic data for subsequent processing. Statistically calculate the maximum offset in the Y direction (MaxOffsetY) of the OBN data to delimit the interception range for the towing cable data. The calculation formula is as follows:

[0061] MaxOffsetY = (max{SY 1max , RY 1max} - min{SY 1min , RY 1min ) ÷ 2

[0062] In the formula, SY1 is the longitudinal coordinate value of the shot point of the OBN data, and RY1 is the Y coordinate value of the geophone point of the OBN data.

[0063] The statistical result of the OBN data is as Figure 2 , and finally determine MaxOffsetY = 800m according to the distribution of the maximum longitudinal offset of the OBN.

[0064] Statistically calculate the maximum offset in the X direction (MaxOffsetX) of the towing cable data to delimit the interception range for the OBN data. The calculation formula is as follows:

[0065] MaxOffsetX = (max{SX 2max , RX 2max} - min{SX 2min , RX 2min ) ÷ 2

[0066] In the formula, SX1 is the X coordinate value of the shot point of the towing cable data, and RX1 is the X coordinate value of the geophone point of the towing cable data.

[0067] The statistical result of the towing cable data is as Figure 3 , and finally determine MaxOffsetX = 300m according to the distribution of the maximum lateral offset of the towing cable.

[0068] Step 2: Pre-stack consistency processing. To eliminate non-repeatable factors in the data and accurately obtain the time-lapse seismic response difference of the oil and gas field, it is necessary to perform global matching correction on the CDP gather after vector offset selection, including noise suppression, ghost wave suppression, multiple wave suppression, wavelet shaping, consistency velocity analysis, consistency pre-stack time migration, etc.

[0069] Step 3: Post-stack consistency processing. Adopt a statistical-based mutual equalization processing method to effectively eliminate unwanted differences by correcting the differences in energy, frequency, and phase of the stacked seismic data.

[0070] Step 4: Construct a U-Net neural network or a LeNet neural network. The U-Net neural network includes a pooling layer, three downsampling layers, three upsampling layers, and a single convolutional layer. Each sampling layer contains a convolutional layer, a BatchNorm layer, and an activation function. The LeNet neural network contains three convolutional layers, a pooling layer, and a fully connected layer. The Tanh function is selected as the activation function for the above networks. The expression of the Tanh function is:

[0071]

[0072] Step 5: Make a sample set. The sample set includes features and labels. The features are the network inputs, and the labels are the expected outputs. The training set includes an equal number of feature data and label data. The towed cable data will be used as the input (features) of the U-Net or LeNet network, and the OBN data is the expected output (label). The towed cable and OBN data are extracted into a 128*128 two-dimensional matrix, and the samples containing empty channels are removed. To avoid the problem of gradient explosion during model training, the sample set is zero-mean normalized, and the calculation formula is:

[0073] X′ = X - μ

[0074] where X is the original data and μ is the mean.

[0075] Step 6: Model training. Input the training set into the U-Net network, initialize the weight vector w and the bias b, calculate the error L between the node output and the expected output. When L converges, the training ends, and all parameter weights are stored.

[0076] The specific calculation method of Step 6 is as follows:

[0077] The basic unit in the neural network is the neuron. The composition formula of the neuron is expressed as:

[0078]

[0079] where y represents the neuron output, f(·) represents the activation function, ω i represents the weight of the i-th input signal, x i represents the i-th input feature, and b represents the bias.

[0080] First, initialize the weights ω1, ω2, …, ω i and the bias b in the neural network, and then input the features x1, x2, …, x i into the neural network, and forward propagate to obtain the output y.

[0081] Then calculate the loss function L MSE between the neuron output value and the expected output. The calculation formula is:

[0082]

[0083] where n is the number of samples, y i is the true value, is the predicted value.

[0084] Update and iterate the weight and bias parameters according to the loss function. When L MSE converges, the optimal solution of the model is obtained.

[0085] Step Seven: Reconstruct the data in the pipeline area. Extract the data in the pipeline area of the streamer number and intercept it as a 128*128 two-dimensional matrix. After zero-mean processing, it is used as the model input. Input the streamer data into the model obtained in Step Six to get the expected output. After inverse zero-mean processing, the data is put back into the OBN data pipeline area to complete the reconstruction of the OBN pipeline area data.

[0086] The streamer data is complete three-dimensional data, and the OBN data lacks the data in the pipeline area. In this embodiment, a non-linear mapping relationship between the non-pipeline area streamer data and the OBN data is established, and a neural network model is trained. When the model is stable, the streamer data in the pipeline area is used to predict the OBN pipeline area data, and then the predicted data is put into the OBN data pipeline area to complete the reconstruction of the OBN data. The comparison before and after data reconstruction is as Figure 4 shown.

[0087] In another embodiment of the present invention, a time-lapse seismic prediction system is provided. The system can be used to implement the above time-lapse seismic prediction method. Specifically, it includes: a data matching and selection module, a pre-stack consistency processing module, a post-stack consistency processing module, a neural network module, a pipeline area data reconstruction module, and a seismic judgment module;

[0088] Data matching and selection module: Collect streamer data and OBN data, sort the pre-stack streamer data and pre-stack OBN data into the CDP domain, and perform data matching and selection based on the vector offset.

[0089] Pre-stack consistency processing module: Perform global matching correction on the CDP gather after vector offset selection;

[0090] Post-stack consistency processing module: Adopt a statistical cross-equalization processing method to correct the differences in energy, frequency, and phase of the stacked seismic data, and eliminate the non-expected differences.

[0091] Neural network module: Through the made training sample set, input the training set into the neural network, initialize the weight vector, calculate the error between the node output and the expected output. When the error converges, the training ends, and each parameter weight is stored. Input the streamer data into the trained model to get the model output, and perform inverse normalization.

[0092] Pipeline area data reconstruction module: After denormalization, the data is put back into the OBN pipeline area for reconstruction.

[0093] Seismic judgment module: According to the reconstruction result, judge whether there is time-lapse seismicity.

[0094] In another embodiment of the present invention, a terminal device is provided. The terminal device includes a processor and a memory. The memory is used to store a computer program. 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 embodiment of the present invention can be used for the operations of the above time-lapse seismic prediction.

[0095] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is the memory device in the 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 the operating system of the terminal is stored in this storage space. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. 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.

[0096] One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps related to time-lapse seismic prediction in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to implement the corresponding steps of a time-lapse seismic prediction.

[0097] 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 storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0098] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one or more of the processes or a plurality of processes and / or blocks Figure 1 one or more of the blocks or a plurality of blocks.

[0099] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that realize the functions specified in Figure 1 one or more of the processes or a plurality of processes and / or blocks Figure 1 one or more of the blocks or a plurality of blocks.

[0100] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the implementation methods of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A time-lapse seismic prediction method based on a neural network, characterized in that, The following steps are involved: Step 1: Data matching and selection based on vector offset; To collect streamer data and Ocean Bottom Node (OBN) data, pre-stack streamer data and pre-stack OBN data are sorted into the CDP (Common Depth Point) domain to ensure that the two sets of data reflect the same geological information. Then, data matching and selection based on vector offset are performed. Step 2: Prestack consistency processing: To eliminate non-repeatability factors in the data and accurately obtain the time-lapse seismic response differences of the oil and gas fields, it is necessary to perform global matching correction on the CDP gathers after vector offset selection. Step 3: Post-stack consistency processing: Using a statistical cross-equalization method, the energy, frequency, and phase differences of the stacked seismic data are corrected to effectively eliminate unwanted differences. Step 4: Build a U-Net neural network or a LeNet neural network and select the Tanh function as the activation function; Step 5: Create a sample set. The sample set includes features and labels. Features are network inputs, and labels are expected outputs. The training set includes an equal amount of feature data and label data. The streamer data will serve as the input features of the neural network, and the OBN data will serve as the expected output labels. The streamer and OBN data will be extracted into a 128*128 two-dimensional matrix, and samples containing empty channels will be removed. Step 6: Model training; Input the training set into the U-Net network, initialize the weight vector w and the bias b, and calculate the error between the node output and the expected output When converges, the training ends, and all parameter weights are stored; Step 7: Reconstruct the pipeline area data; extract the streamer data from the pipeline area and truncate it into a 128*128 two-dimensional matrix, perform zero-mean processing on it and use it as the model input; input the streamer data into the model obtained in step 6 to obtain the expected output, perform anti-zero-mean processing on it and place the data back into the OBN data pipeline area to complete the OBN pipeline area data reconstruction.

2. The time-lapse seismic prediction method based on a neural network according to claim 1, wherein: The data matching and selection based on vector offset in step 1 are as follows: Calculate the intercept range of the trace gather, intercept the trace gather of the overlapping part of the streamer and the OBN data offset, and use it as the basic data for subsequent processing; count the maximum vertical offset of the OBN data , define the intercept range for the streamer data, and the calculation formula is as follows: wherein, is the vertical coordinate value of the OBN data shot point, is the vertical coordinate value of the OBN data geophone point; According to the distribution of the maximum longitudinal offset of OBN, MaxOffsetY is finally determined; Statistical maximum lateral offset of towed cable data , delimit the interception range for OBN data, and the calculation formula is as follows: Wherein, is the lateral coordinate value of the shot point of the streamer data, is the lateral coordinate value of the geophone point of the streamer data; MaxOffsetX is finally determined based on the distribution of the maximum lateral offset of the streamer.

3. A time-lapse seismic prediction method based on a neural network according to claim 2, characterized in that: In step 2, global matching correction includes: noise suppression, ghost suppression, multiple suppression, wavelet shaping, consistent velocity analysis and consistent prestack time migration.

4. The time-lapse seismic prediction method based on a neural network according to claim 3, characterized in that: In step 5, to avoid the problem of gradient explosion during model training, the sample set is zero-mean processed. The calculation formula is: Wherein, X is the original data, is the mean value.

5. A time-lapse seismic prediction method based on a neural network according to claim 4, characterized in that: Calculate the error between the output of the computing node and the expected output in Step 6 , which is specifically as follows: The neuron composition formula is expressed as: Among them represents the neuron output, represents the activation function, represents the i weight of the th input signal, i represents the th input feature, and b represents the bias; First, initialize the weights and biases in the neural network. Then, input the features into the neural network and perform forward propagation to obtain the output y; Then calculate the loss function between the neuron output value and the expected output , and the calculation formula is: Among them, is the number of samples, is the true value, is the predicted value; Update and iterate the weight and bias parameters according to the loss function. When converges, the optimal solution of the model is obtained.

6. A time-lapse seismic prediction system, characterized in that: The system can be used to implement the time-lapse seismic prediction method according to any one of claims 1 to 5, and specifically comprises: a data matching and selection module, a pre-stack consistency processing module, a post-stack consistency processing module, a neural network module, a pipeline area data reconstruction module, and an earthquake judgment module; Data matching and selection module: collects streamer data and OBN data, sorts pre-stack streamer data and pre-stack OBN data into the CDP domain, and performs data matching and selection based on vector offset; Pre-stack consistency processing module: performs global matching correction on the CDP gathers after vector offset selection; Post-stack consistency processing module: adopts a statistical cross-equalization processing method to correct the energy, frequency and phase differences of stacked seismic data to eliminate unexpected differences; Neural network module: By using the made training sample set, input the training set into the neural network, initialize the weight vector, calculate the error between the node output and the expected output. When the error converges, the training ends and store various parameter weights; input the towed cable data into the training model, obtain the model output, and perform anti-normalization; Pipeline area data reconstruction module: After anti-normalization, put the data back into the OBN pipeline area for reconstruction; Seismic judgment module: According to the reconstruction result, judge whether there is time-lapse seismicity.

7. A computer device, including 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 the time-lapse seismic prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the time-lapse seismic prediction method according to any one of claims 1 to 5.

Images