Natural gas hydrate stratum collapse early warning method and system

Abstract

The invention discloses a natural gas hydrate stratum collapse early warning method and system, and the method comprises the steps: collecting the stratum parameters of a natural gas hydrate stratum according to sound waves, generating a stratum pressure image of the natural gas hydrate stratum, and carrying out the collapse prediction of the natural gas hydrate stratum according to the stratum pressure image. Meanwhile, the first training data are used as training data to train a convolutional neural network model, a plurality of different formation collapse prediction models are generated, the second image data are predicted according to the formation collapse prediction models to generate pseudo labels of the second image data, the sample data are further expanded, and the prediction accuracy of the formation collapse prediction models is improved. Therefore, the integrated formation collapse prediction model is trained according to the training data, and the precision and accuracy of the model for predicting the formation collapse of the natural gas hydrate are improved.

Description

Technical Field

[0001] The invention relates to the technical field of formation collapse, and in particular to a natural gas hydrate formation collapse early warning method and system. Background Art

[0002] Natural gas hydrates are ice-like, cage-shaped clean compounds formed by the mixing of natural gas and water under medium-high pressure and low temperature conditions. They are a new type of clean energy with huge potential reserves. Natural gas hydrates are widely distributed in the continental permafrost, slope areas of islands, uplifts of active and passive continental margins, polar continental shelves, and deep-water environments of oceans and some inland lakes. One of the difficulties in the exploitation of natural gas hydrates is the stability of the reservoir. The exploitation of natural gas hydrates is a sensitive system. Slight changes in reservoir temperature, pressure, and volume will lead to the decomposition and formation of hydrates. The exploitation of hydrates is a multiphase transformation process. In the process of transforming from solid phase to gas phase and liquid phase, changes in reservoir properties, pressure, temperature, etc. can easily cause the decomposition of hydrates and cause formation collapse.

[0003] In the existing natural gas hydrate mining process, the possibility of natural gas hydrate formation collapse is generally reduced to a certain extent by improving the stability of the well wall during mining. However, this kind of mining method is costly and cannot predict the collapse of natural gas hydrate formations. It is impossible to completely avoid the collapse of natural gas hydrate formations, which affects the mining efficiency. Summary of the invention

[0004] The present invention provides a natural gas hydrate formation collapse early warning method and system to solve the problem that the existing natural gas hydrate exploitation cannot predict the natural gas hydrate formation collapse, cannot completely avoid the natural gas hydrate formation collapse, and affects the exploitation efficiency.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a natural gas hydrate formation collapse early warning method, comprising: Acquiring acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generating a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves; Randomly dividing the time-frequency atlas into first image data and second image data, and calculating and annotating formation parameters in the first image data to generate first training data; Constructing a convolutional neural network model, randomly and equally dividing the first training data into a plurality of groups of first sub-training data, and respectively training the convolutional neural network model according to the plurality of groups of first sub-training data to obtain a plurality of formation collapse prediction models; Inputting the second image data into the plurality of stratum collapse prediction models respectively, obtaining prediction results of each stratum collapse prediction model, annotating the second image data according to the prediction results to generate second training data, and integrating the first training data and the second training data to generate a training set; The multiple formation collapse prediction models are integrated, and the integrated formation collapse prediction models are trained according to the training set to output an optimal formation collapse prediction model, and collapse prediction and early warning of the natural gas hydrate formation are performed according to the optimal formation collapse prediction model.

[0006] The present invention collects formation parameters of a natural gas hydrate formation according to acoustic waves, thereby generating a formation pressure image of the natural gas hydrate formation, so as to predict the collapse of the natural gas hydrate formation according to the formation pressure image. At the same time, the first training data is used as training data to train a convolutional neural network model, generate a number of different formation collapse prediction models, and predict the second image data according to the formation collapse prediction model to generate a pseudo label for the second image data, further expand the sample data to make the training data more diversified, and thus train the integrated formation collapse prediction model according to the training data to improve the model prediction precision and prediction accuracy.

[0007] Furthermore, the acoustic wave time difference curves of different natural gas hydrate formations are collected according to the first transmitted acoustic wave and the second transmitted acoustic wave, and a time-frequency atlas is generated according to Fourier transform and the acoustic wave time difference curves, specifically: Transmitting a first transmitted sound wave and a second transmitted sound wave to a natural gas hydrate formation, and recording arrival times of a first reflected sound wave and a second reflected sound wave; Calculating the longitudinal wave velocity and the transverse wave velocity according to the arrival time of the first reflected sound wave and the second reflected sound wave, and constructing a sound wave time difference curve; Performing short-time Fourier transform on the acoustic wave time difference curve to generate a deep frequency graph of the acoustic wave time difference curve and obtain a time-frequency atlas.

[0008] Furthermore, the calculation and annotation of the formation parameters in the first image data to generate the first training data is specifically as follows: Calculating rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data; the rock mechanical characteristic parameters include rock cohesion, rock internal friction angle, uniaxial compressive strength, rock tensile strength, dynamic Young's modulus and dynamic Poisson's ratio; the formation pressure data includes formation collapse pressure and formation fracture pressure; The collapse probability of each sample data is predicted and labeled according to the formation collapse pressure and formation fracture pressure of each sample data to generate the first training data.

[0009] Furthermore, the plurality of formation collapse prediction models are integrated, and the integrated formation collapse prediction model is trained according to the training set to output the optimal formation collapse prediction model, specifically: Randomly extracting sample data from the first training data and inputting it into the integrated formation collapse prediction model to obtain prediction results, and calculating the prediction accuracy based on the prediction results and labels to set the weights of each formation collapse prediction model; The integrated formation collapse prediction model is iteratively trained according to the training set, and sample data in the training set is randomly selected to train the integrated formation collapse prediction model until the preset conditions are met, the optimal formation collapse prediction model is output, and the formation collapse prediction model is evaluated according to the prediction accuracy and recall rate.

[0010] Furthermore, the collapse prediction and early warning of the natural gas hydrate formation are performed according to the optimal formation collapse prediction model, specifically: Transmitting a first transmitting sound wave and a second transmitting sound wave to the natural gas hydrate formation, calculating formation pressure data of the natural gas hydrate formation according to an acoustic wave time difference method, and calculating a first prediction result according to the formation pressure data and reference pressure data; Constructing an acoustic wave time difference curve according to the first transmitted acoustic wave and the second transmitted acoustic wave, converting the curve into a time-frequency diagram, and inputting the time-frequency diagram into the formation collapse prediction model to obtain a second prediction result; A weighted average is performed on the first prediction result and the second prediction result to obtain a final prediction result, and a graded collapse warning is performed according to the final prediction result.

[0011] In a second aspect, the present invention provides a natural gas hydrate formation collapse early warning system, including: a data acquisition module, a data annotation module, a data expansion module, a model training module and a prediction module; The data acquisition module is used to collect acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generate a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves; The data annotation module is used to randomly divide the time-frequency atlas into first image data and second image data, and calculate and annotate formation parameters in the first image data to generate first training data; The data expansion module is used to construct a convolutional neural network model, randomly and equally divide the first training data into a plurality of groups of first sub-training data, and respectively train the convolutional neural network model according to the plurality of groups of first sub-training data to obtain a plurality of formation collapse prediction models; The model training module is used to input the second image data into the plurality of stratum collapse prediction models respectively, obtain prediction results of each stratum collapse prediction model, annotate the second image data according to the prediction results, generate second training data, and integrate the first training data and the second training data to generate a training set; The prediction module is used to integrate the multiple formation collapse prediction models, train the integrated formation collapse prediction model according to the training set, output the optimal formation collapse prediction model, and perform collapse prediction and early warning for the natural gas hydrate formation according to the optimal formation collapse prediction model.

[0012] Furthermore, the data acquisition module is specifically used for: Transmitting a first transmitted sound wave and a second transmitted sound wave to a natural gas hydrate formation, and recording arrival times of a first reflected sound wave and a second reflected sound wave; Calculating the longitudinal wave velocity and the transverse wave velocity according to the arrival time of the first reflected sound wave and the second reflected sound wave, and constructing a sound wave time difference curve; Performing short-time Fourier transform on the acoustic wave time difference curve to generate a deep frequency graph of the acoustic wave time difference curve and obtain a time-frequency atlas.

[0013] Furthermore, the data annotation module is specifically used to: Calculating rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data; the rock mechanical characteristic parameters include rock cohesion, rock internal friction angle, uniaxial compressive strength, rock tensile strength, dynamic Young's modulus and dynamic Poisson's ratio; the formation pressure data includes formation collapse pressure and formation fracture pressure; The collapse probability of each sample data is predicted and labeled according to the formation collapse pressure and formation fracture pressure of each sample data to generate the first training data.

[0014] Furthermore, the prediction module is specifically used for: Randomly extracting sample data from the first training data and inputting it into the integrated formation collapse prediction model to obtain prediction results, and calculating the prediction accuracy based on the prediction results and labels to set the weights of each formation collapse prediction model; The integrated formation collapse prediction model is iteratively trained according to the training set, and sample data in the training set is randomly selected to train the integrated formation collapse prediction model until the preset conditions are met, the optimal formation collapse prediction model is output, and the formation collapse prediction model is evaluated according to the prediction accuracy and recall rate.

[0015] Furthermore, the prediction module is specifically used for: Transmitting a first transmitting sound wave and a second transmitting sound wave to the natural gas hydrate formation, calculating formation pressure data of the natural gas hydrate formation according to an acoustic wave time difference method, and calculating a first prediction result according to the formation pressure data and reference pressure data; Constructing an acoustic wave time difference curve according to the first transmitted acoustic wave and the second transmitted acoustic wave, converting the curve into a time-frequency diagram, and inputting the time-frequency diagram into the formation collapse prediction model to obtain a second prediction result; A weighted average is performed on the first prediction result and the second prediction result to obtain a final prediction result, and a graded collapse warning is performed according to the final prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a natural gas hydrate formation collapse early warning method provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a natural gas hydrate formation collapse early warning system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Embodiment 1 Please refer to Figure 1 , Figure 1 A flow chart of a natural gas hydrate formation collapse early warning method provided by an embodiment of the present invention includes steps 101 to 105, which are specifically as follows: Step 101: collecting acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generating a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves; In this embodiment, the acoustic wave time difference curves of different natural gas hydrate formations are collected according to the first transmitted acoustic wave and the second transmitted acoustic wave, and a time-frequency atlas is generated according to Fourier transform and the acoustic wave time difference curves, specifically: Transmitting a first transmitted sound wave and a second transmitted sound wave to a natural gas hydrate formation, and recording arrival times of a first reflected sound wave and a second reflected sound wave; Calculating the longitudinal wave velocity and the transverse wave velocity according to the arrival time of the first reflected sound wave and the second reflected sound wave, and constructing a sound wave time difference curve; Performing short-time Fourier transform on the acoustic wave time difference curve to generate a deep frequency graph of the acoustic wave time difference curve and obtain a time-frequency atlas.

[0019] In this embodiment, the gas hydrate formation types include undrilled gas hydrate formations, drilled but unexploited gas hydrate formations, exploited gas hydrate formations, and depleted gas hydrate formations. By transmitting acoustic waves to different gas hydrate formations to collect formation parameters of different gas hydrate formations, the sample data is more diversified.

[0020] In this embodiment, a first transmitting sound wave and a second transmitting sound wave are transmitted to a natural gas hydrate formation according to a signal transmitter, and the time when the first reflected sound wave and the second reflected sound wave arrive at a signal receiver is recorded, wherein the arrival time of the first reflected sound wave is a first time, and the arrival time of the second reflected sound wave is a second time, the longitudinal wave velocity is calculated according to the first time and the second time, and the shear wave velocity is calculated according to the longitudinal wave velocity, thereby constructing a sound wave time difference curve of the natural gas hydrate formation according to the longitudinal wave velocity and the shear wave velocity.

[0021] In this embodiment, the acoustic wave time difference curve is subjected to short-time Fourier transform to generate a deep frequency graph of the formation pressure curve and obtain a time-frequency graph set. The short-time Fourier transform is specifically as follows: (1); in, is the acoustic time difference curve, The center is located in the depth of the well The window function at the formation, is the frequency domain parameter of Fourier transform, The depth domain parameters of the acoustic wave time difference curve, is the acoustic logging depth.

[0022] In this embodiment, the time-frequency atlas is divided into a first atlas, a second atlas, a third atlas and a fourth atlas according to the type of natural gas hydrate formation. The first atlas is a time-frequency atlas of undrilled natural gas hydrate formations, the second atlas is a time-frequency atlas of drilled but unexploited natural gas hydrate formations, the third atlas is a time-frequency atlas of exploited natural gas hydrate formations, and the fourth atlas is a time-frequency atlas of depleted natural gas hydrate formations.

[0023] Step 102: randomly dividing the time-frequency atlas into first image data and second image data, and calculating and annotating formation parameters in the first image data to generate first training data; In this embodiment, the formation parameters in the first image data are calculated and annotated to generate the first training data, specifically: Calculating rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data; the rock mechanical characteristic parameters include rock cohesion, rock internal friction angle, uniaxial compressive strength, rock tensile strength, dynamic Young's modulus and dynamic Poisson's ratio; the formation pressure data includes formation collapse pressure and formation fracture pressure; The collapse probability of each sample data is predicted and labeled according to the formation collapse pressure and formation fracture pressure of each sample data to generate the first training data.

[0024] In this embodiment, the time-frequency atlas is randomly and equally divided into first image data and second image data, and sample data in the first image data is selected for calculation and labeling.

[0025] In this embodiment, the rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data are calculated, so as to calculate the rock mechanical characteristic parameters of the natural gas hydrate formation according to the longitudinal wave velocity and the shear wave velocity.

[0026] In this embodiment, the rock mechanical characteristic parameters include the dynamic Young's modulus and dynamic Poisson's ratio of the rock, specifically:

[0027] (2)

[0028] (3)

[0029] in, is the dynamic Young's modulus of rock, is the dynamic Poisson’s ratio, is the rock density, is the longitudinal wave velocity, is the shear wave velocity.

[0030] In this embodiment, formation pressure data is calculated according to the acoustic wave time difference curve, and the formation pressure data includes overlying formation pressure, formation pore pressure and formation fracture pressure.

[0031] In this embodiment, the natural gas hydrate formation is stratified into lithologic sections according to the acoustic time difference curve, and the overlying formation pressure of the natural gas hydrate formation is calculated, the formation pore pressure of the natural gas hydrate formation is calculated according to the equal depth method, and the formation fracture pressure is calculated according to the overlying formation pressure and the formation pore pressure. Specifically: (4); in, is the formation fracture pressure, is the overlying formation pressure, is the formation pore pressure, is the dynamic Poisson's ratio.

[0032] In this embodiment, the formation collapse pressure is calculated according to the rock mechanical characteristic parameters, specifically: (5); in, is the equivalent drilling fluid density of the collapse pressure; is the logging depth; , is the internal friction angle of rock; C is the cohesion of rock; is the effective stress coefficient.

[0033] In this embodiment, the equivalent drilling fluid density of the collapse pressure is expanded according to the Mohr-Coulomb criterion, the stable drilling fluid density is calculated, and the collapse probability of each sample data is predicted and marked according to the stable drilling fluid density.

[0034] In this embodiment, the sample data is labeled according to the stable drilling fluid density. The sample data labels include easy to collapse and stable. When the drilling fluid density is greater than the stable drilling fluid density, the sample data label is stable. When the drilling fluid density is less than the stable drilling fluid density, the sample data label is stable. After the labeling is completed, the first training data is generated.

[0035] Step 103: constructing a convolutional neural network model, randomly and equally dividing the first training data into a plurality of groups of first sub-training data, and respectively training the convolutional neural network model according to the plurality of groups of first sub-training data to obtain a plurality of formation collapse prediction models; In this embodiment, the convolutional neural network model includes an input layer, three convolutional layers, three BN layers, a global pooling layer and a fully connected layer.

[0036] In this embodiment, the labeled first training data is randomly and equally divided into several groups of first sub-training data, and the divided first sub-training data is adjusted so that each first sub-training data contains sample data labeled as easy to collapse and labeled as stable. The convolutional neural network model is trained according to each first sub-training data to obtain several formation collapse prediction models.

[0037] Step 104: inputting the second image data into the plurality of stratum collapse prediction models respectively, obtaining prediction results of the respective stratum collapse prediction models, annotating the second image data according to the prediction results to generate second training data, and integrating the first training data and the second training data to generate a training set; In this embodiment, the second image data is divided into several groups of image data, and the image data of each group is respectively input into several formation collapse prediction models to obtain the prediction results of each sample data in the second image data. The second image data is annotated based on the prediction results as labels of the second image data to generate second training data, and the first training data and the second training data are integrated to generate a training set.

[0038] Step 105: Integrate the multiple formation collapse prediction models, and train the integrated formation collapse prediction model according to the training set, output the optimal formation collapse prediction model, and perform collapse prediction and early warning for the natural gas hydrate formation according to the optimal formation collapse prediction model.

[0039] In this embodiment, the plurality of formation collapse prediction models are integrated, and the integrated formation collapse prediction model is trained according to the training set to output the optimal formation collapse prediction model, specifically: Randomly extracting sample data from the first training data and inputting it into the integrated formation collapse prediction model to obtain prediction results, and calculating the prediction accuracy based on the prediction results and labels to set the weights of each formation collapse prediction model; The integrated formation collapse prediction model is iteratively trained according to the training set, and sample data in the training set is randomly selected to train the integrated formation collapse prediction model until the preset conditions are met, the optimal formation collapse prediction model is output, and the formation collapse prediction model is evaluated according to the prediction accuracy and recall rate.

[0040] In this embodiment, the several formation collapse prediction models are integrated. Specifically, a plurality of output branches are constructed according to the several formation collapse prediction models, and sample data are randomly selected from the first training data and input into the integrated formation collapse prediction model to obtain prediction results, and the weights of each formation collapse prediction model are set according to the prediction results and the label calculation prediction accuracy, and the weights of each output branch are set according to the weights of each formation collapse prediction model.

[0041] In this embodiment, the output of the integrated formation collapse prediction model is: (6); in, For the The output of the output branch, that is, An unintegrated ground collapse prediction model, For the is the weight of the unintegrated formation collapse prediction model, and N is the number of unintegrated formation collapse prediction models.

[0042] In this embodiment, sample data in the training set are randomly selected to train the integrated formation collapse prediction model until the prediction accuracy of the integrated formation collapse prediction model reaches a preset threshold, the iteration is stopped, and the optimal formation collapse prediction model is output.

[0043] In this embodiment, after obtaining the optimal formation collapse prediction model, it also includes testing the optimal formation collapse prediction model according to the first atlas, the second atlas, the third atlas and the fourth atlas, and the prediction accuracy of the optimal formation collapse prediction model for each formation type.

[0044] In this embodiment, the collapse prediction and early warning of the natural gas hydrate formation are performed according to the optimal formation collapse prediction model, specifically: Transmitting a first transmitting sound wave and a second transmitting sound wave to the natural gas hydrate formation, calculating formation pressure data of the natural gas hydrate formation according to an acoustic wave time difference method, and calculating a first prediction result according to the formation pressure data and reference pressure data; Constructing an acoustic wave time difference curve according to the first transmitted acoustic wave and the second transmitted acoustic wave, converting the curve into a time-frequency diagram, and inputting the time-frequency diagram into the formation collapse prediction model to obtain a second prediction result; A weighted average is performed on the first prediction result and the second prediction result to obtain a final prediction result, and a graded collapse warning is performed according to the final prediction result.

[0045] In this embodiment, before transmitting the first transmitted sound wave and the second transmitted sound wave to the natural gas hydrate formation, it is necessary to confirm the formation type of the natural gas hydrate formation to be detected, and set the prediction result weight of the optimal formation collapse prediction model according to the prediction accuracy of the optimal formation collapse prediction model for each formation type.

[0046] In this embodiment, after obtaining the first prediction result and the second prediction result, a weighted average is performed according to the weights of their respective prediction results to obtain a final prediction result, and a classified and graded warning is performed according to the final prediction result and the formation type of the natural gas hydrate formation.

[0047] Please refer to Figure 2 , Figure 2 A schematic diagram of a structure of a natural gas hydrate formation collapse early warning system provided by an embodiment of the present invention, comprising a data acquisition module 201, a data annotation module 202, a data expansion module 203, a model training module 204 and a prediction module 205; The data acquisition module 201 is used to collect acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generate a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves; The data annotation module 202 is used to randomly divide the time-frequency atlas into first image data and second image data, and calculate and annotate formation parameters in the first image data to generate first training data; The data expansion module 203 is used to construct a convolutional neural network model, randomly and equally divide the first training data into a plurality of groups of first sub-training data, and respectively train the convolutional neural network model according to the plurality of groups of first sub-training data to obtain a plurality of formation collapse prediction models; The model training module 204 is used to input the second image data into the plurality of stratum collapse prediction models respectively, obtain prediction results of each stratum collapse prediction model, annotate the second image data according to the prediction results, generate second training data, and integrate the first training data and the second training data to generate a training set; The prediction module 205 is used to integrate the multiple formation collapse prediction models, train the integrated formation collapse prediction model according to the training set, output the optimal formation collapse prediction model, and perform collapse prediction and early warning for the natural gas hydrate formation according to the optimal formation collapse prediction model.

[0048] In this embodiment, the data acquisition module is specifically used for: Transmitting a first transmitted sound wave and a second transmitted sound wave to a natural gas hydrate formation, and recording arrival times of a first reflected sound wave and a second reflected sound wave; Calculating the longitudinal wave velocity and the transverse wave velocity according to the arrival time of the first reflected sound wave and the second reflected sound wave, and constructing a sound wave time difference curve; Performing short-time Fourier transform on the acoustic wave time difference curve to generate a deep frequency graph of the acoustic wave time difference curve and obtain a time-frequency atlas.

[0049] In this embodiment, the data annotation module is specifically used to: Calculating rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data; the rock mechanical characteristic parameters include rock cohesion, rock internal friction angle, uniaxial compressive strength, rock tensile strength, dynamic Young's modulus and dynamic Poisson's ratio; the formation pressure data includes formation collapse pressure and formation fracture pressure; The collapse probability of each sample data is predicted and labeled according to the formation collapse pressure and formation fracture pressure of each sample data to generate the first training data.

[0050] In this embodiment, the prediction module is specifically used for: Randomly extracting sample data from the first training data and inputting it into the integrated formation collapse prediction model to obtain prediction results, and calculating the prediction accuracy based on the prediction results and labels to set the weights of each formation collapse prediction model; The integrated formation collapse prediction model is iteratively trained according to the training set, and sample data in the training set is randomly selected to train the integrated formation collapse prediction model until the preset conditions are met, the optimal formation collapse prediction model is output, and the formation collapse prediction model is evaluated according to the prediction accuracy and recall rate.

[0051] In this embodiment, the prediction module is specifically used for: Transmitting a first transmitting sound wave and a second transmitting sound wave to the natural gas hydrate formation, calculating formation pressure data of the natural gas hydrate formation according to an acoustic wave time difference method, and calculating a first prediction result according to the formation pressure data and reference pressure data; Constructing an acoustic wave time difference curve according to the first transmitted acoustic wave and the second transmitted acoustic wave, converting the curve into a time-frequency diagram, and inputting the time-frequency diagram into the formation collapse prediction model to obtain a second prediction result; A weighted average is performed on the first prediction result and the second prediction result to obtain a final prediction result, and a graded collapse warning is performed according to the final prediction result.

[0052] In this embodiment, the formation parameters of the natural gas hydrate formation are collected according to the acoustic wave, thereby generating a formation pressure image of the natural gas hydrate formation, so as to predict the collapse of the natural gas hydrate formation according to the formation pressure image. At the same time, the first training data is used as training data to train the convolutional neural network model, generate several different formation collapse prediction models, and predict the second image data according to the formation collapse prediction model to generate pseudo labels for the second image data, further expand the sample data to make the training data more diversified, so as to train the integrated formation collapse prediction model according to the training data, and improve the model prediction precision and prediction accuracy.

[0053] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A natural gas hydrate formation collapse early warning method, characterized in that: include: Acquiring acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generating a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves; Randomly dividing the time-frequency atlas into first image data and second image data, and calculating and annotating formation parameters in the first image data to generate first training data; Constructing a convolutional neural network model, randomly and equally dividing the first training data into a plurality of groups of first sub-training data, and respectively training the convolutional neural network model according to the plurality of groups of first sub-training data to obtain a plurality of formation collapse prediction models; Inputting the second image data into the plurality of stratum collapse prediction models respectively, obtaining prediction results of each stratum collapse prediction model, annotating the second image data according to the prediction results to generate second training data, and integrating the first training data and the second training data to generate a training set; The multiple formation collapse prediction models are integrated, and the integrated formation collapse prediction models are trained according to the training set to output an optimal formation collapse prediction model, and collapse prediction and early warning of the natural gas hydrate formation are performed according to the optimal formation collapse prediction model.

2. The natural gas hydrate formation collapse early warning method according to claim 1, characterized in that: The method of collecting acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generating a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves, is specifically as follows: Transmitting a first transmitted sound wave and a second transmitted sound wave to a natural gas hydrate formation, and recording arrival times of a first reflected sound wave and a second reflected sound wave; Calculating the longitudinal wave velocity and the transverse wave velocity according to the arrival time of the first reflected sound wave and the second reflected sound wave, and constructing a sound wave time difference curve; Performing short-time Fourier transform on the acoustic wave time difference curve to generate a deep frequency graph of the acoustic wave time difference curve and obtain a time-frequency atlas.

3. The natural gas hydrate formation collapse early warning method according to claim 2, characterized in that: The calculating and labeling of the formation parameters in the first image data to generate the first training data is specifically as follows: Calculating rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data; the rock mechanical characteristic parameters include rock cohesion, rock internal friction angle, uniaxial compressive strength, rock tensile strength, dynamic Young's modulus and dynamic Poisson's ratio; the formation pressure data includes formation collapse pressure and formation fracture pressure; The collapse probability of each sample data is predicted and labeled according to the formation collapse pressure and formation fracture pressure of each sample data to generate the first training data.

4. The natural gas hydrate formation collapse early warning method according to claim 3, characterized in that: The integrating the plurality of formation collapse prediction models, training the integrated formation collapse prediction model according to the training set, and outputting the optimal formation collapse prediction model is specifically: Randomly extracting sample data from the first training data and inputting it into the integrated formation collapse prediction model to obtain prediction results, and calculating the prediction accuracy based on the prediction results and labels to set the weights of each formation collapse prediction model; The integrated formation collapse prediction model is iteratively trained according to the training set, and sample data in the training set is randomly selected to train the integrated formation collapse prediction model until the preset conditions are met, the optimal formation collapse prediction model is output, and the formation collapse prediction model is evaluated according to the prediction accuracy and recall rate.

5. The natural gas hydrate formation collapse early warning method according to claim 4, characterized in that: The collapse prediction and early warning of the natural gas hydrate formation according to the optimal formation collapse prediction model is specifically as follows: Transmitting a first transmitting sound wave and a second transmitting sound wave to the natural gas hydrate formation, calculating formation pressure data of the natural gas hydrate formation according to an acoustic wave time difference method, and calculating a first prediction result according to the formation pressure data and reference pressure data; Constructing an acoustic wave time difference curve according to the first transmitted acoustic wave and the second transmitted acoustic wave, converting the curve into a time-frequency diagram, and inputting the time-frequency diagram into the formation collapse prediction model to obtain a second prediction result; A weighted average is performed on the first prediction result and the second prediction result to obtain a final prediction result, and a graded collapse warning is performed according to the final prediction result.

6. A natural gas hydrate formation collapse early warning system, characterized in that: include: Data collection module, data annotation module, data expansion module, model training module and prediction module; The data acquisition module is used to collect acoustic wave time difference curves of different natural gas hydrate formations according to the first transmitted acoustic wave and the second transmitted acoustic wave, and generate a time-frequency atlas according to Fourier transform and the acoustic wave time difference curves; The data annotation module is used to randomly divide the time-frequency atlas into first image data and second image data, and calculate and annotate formation parameters in the first image data to generate first training data; The data expansion module is used to construct a convolutional neural network model, randomly and equally divide the first training data into a plurality of groups of first sub-training data, and respectively train the convolutional neural network model according to the plurality of groups of first sub-training data to obtain a plurality of formation collapse prediction models; The model training module is used to input the second image data into the plurality of stratum collapse prediction models respectively, obtain prediction results of each stratum collapse prediction model, annotate the second image data according to the prediction results, generate second training data, and integrate the first training data and the second training data to generate a training set; The prediction module is used to integrate the multiple formation collapse prediction models, train the integrated formation collapse prediction model according to the training set, output the optimal formation collapse prediction model, and perform collapse prediction and early warning for the natural gas hydrate formation according to the optimal formation collapse prediction model.

7. The natural gas hydrate formation collapse early warning system according to claim 6, characterized in that: The data acquisition module is specifically used for: Transmitting a first transmitted sound wave and a second transmitted sound wave to a natural gas hydrate formation, and recording arrival times of a first reflected sound wave and a second reflected sound wave; Calculating the longitudinal wave velocity and the transverse wave velocity according to the arrival time of the first reflected sound wave and the second reflected sound wave, and constructing a sound wave time difference curve; Performing short-time Fourier transform on the acoustic wave time difference curve to generate a deep frequency graph of the acoustic wave time difference curve and obtain a time-frequency atlas.

8. The natural gas hydrate formation collapse early warning system according to claim 7, characterized in that: The data annotation module is specifically used for: Calculating rock mechanical characteristic parameters and formation pressure data of each sample data in the first image data; the rock mechanical characteristic parameters include rock cohesion, rock internal friction angle, uniaxial compressive strength, rock tensile strength, dynamic Young's modulus and dynamic Poisson's ratio; the formation pressure data includes formation collapse pressure and formation fracture pressure; The collapse probability of each sample data is predicted and labeled according to the formation collapse pressure and formation fracture pressure of each sample data to generate the first training data.

9. The natural gas hydrate formation collapse early warning system according to claim 8, characterized in that: The prediction module is specifically used for: Randomly extracting sample data from the first training data and inputting it into the integrated formation collapse prediction model to obtain prediction results, and calculating the prediction accuracy based on the prediction results and labels to set the weights of each formation collapse prediction model; The integrated formation collapse prediction model is iteratively trained according to the training set, and sample data in the training set is randomly selected to train the integrated formation collapse prediction model until the preset conditions are met, the optimal formation collapse prediction model is output, and the formation collapse prediction model is evaluated according to the prediction accuracy and recall rate.

10. The natural gas hydrate formation collapse early warning system according to claim 9, characterized in that: The prediction module is specifically used for: Transmitting a first transmitting sound wave and a second transmitting sound wave to the natural gas hydrate formation, calculating formation pressure data of the natural gas hydrate formation according to an acoustic wave time difference method, and calculating a first prediction result according to the formation pressure data and reference pressure data; Constructing an acoustic wave time difference curve according to the first transmitted acoustic wave and the second transmitted acoustic wave, converting the curve into a time-frequency diagram, and inputting the time-frequency diagram into the formation collapse prediction model to obtain a second prediction result; A weighted average is performed on the first prediction result and the second prediction result to obtain a final prediction result, and a graded collapse warning is performed according to the final prediction result.

Images