Chromatographic inversion method and device, electronic equipment and storage medium

Abstract

The invention provides a tomography inversion method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring shot point space coordinates, multiple detection point space coordinates and shot detection point travel time data; according to the shot point space coordinates and the geophone space coordinates, a partial differential equation solving space is created, and partial differential equation space coordinates are obtained through discretization; determining a travel time data error of the shot and detection point travel time data under the constructed deep learning neural network and a partial differential equation error of the shot and detection point travel time data in the partial differential equation solving space; an augmented Lagrange framework is adopted, the travel time data error and the partial differential equation error are fused, and a target velocity field prediction neural network is constructed; and inputting the partial differential equation space coordinates into the target velocity field prediction neural network to obtain a tomography velocity model. Therefore, simultaneous prediction and constraint of the travel time field and the velocity field are realized, and the velocity model is updated.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic exploration data processing, and in particular to a tomography inversion method, device, electronic equipment and storage medium. Background Art

[0002] Shallow surface velocity modeling is an important step in the industrial seismic exploration data processing process. Tomographic inversion can not only provide a velocity model for static correction calculations, but also provide an initial velocity model for reverse time migration (RTM) and full waveform inversion.

[0003] At present, conventional tomographic inversion is usually based on travel time forward modeling operators under equally spaced fixed grids or variable grids. In the conventional tomographic inversion process, an initial velocity model is first given, and then the travel time forward modeling operator is used to calculate the predicted travel time under the velocity model. Next, the predicted travel time is compared with the observed travel time, and the gradient of the travel time is calculated. The velocity model is iteratively adjusted according to the gradient information so that the predicted travel time gradually approaches the observed travel time. The iterative process will continue until a certain iterative convergence criterion is met, and finally a velocity model that fits the observed data is obtained.

[0004] With the rise of neural networks, there is also a speed modeling technology based on data-driven design. The corresponding travel time curve is calculated through the conventional eikonal equation algorithm to generate various speed models, and then the speed model and travel time curve are used as sample labels for neural network training. This data-driven approach not only has high requirements for the training set, but also is not flexible enough for the observation system. In other words, when the observation system changes, it is uncertain whether the neural network can infer a reasonable result. Summary of the invention

[0005] The present invention provides a tomographic inversion method, device, electronic equipment and storage medium, which are used to shorten the time required for tomographic inversion while ensuring the accuracy of underground medium velocity model.

[0006] In a first aspect, the present invention provides a tomographic inversion method, comprising:

[0007] Obtain the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points;

[0008] According to the shot point spatial coordinates and the detection point spatial coordinates, a partial differential equation solution space is created, and the partial differential equation spatial coordinates are discretized;

[0009] Determine the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0010] Adopting an augmented Lagrangian framework, fusing the travel time data error and the partial differential equation error, and constructing a target velocity field prediction neural network;

[0011] The partial differential equation spatial coordinates are input into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0012] Optionally, an augmented Lagrangian framework is adopted to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network, including:

[0013] Combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model, constructing an augmented Lagrangian function under the augmented Lagrangian framework;

[0014] The augmented Lagrangian function is used as the objective function, an optimization algorithm is used, and network weights are updated during the training process of the deep learning neural network to obtain the target velocity field prediction neural network.

[0015] Optionally, the deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; determining the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space, including:

[0016] Inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data;

[0017] Inputting the partial differential equation spatial coordinates into the velocity field prediction neural network to obtain a predicted velocity field;

[0018] Based on the predicted travel time data and the predicted velocity field, the travel time data error and the partial differential equation error are determined.

[0019] Optionally, determining the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field includes:

[0020] Calculating the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space;

[0021] Based on a preset partial differential equation error calculation expression, the partial differential equation error is obtained in combination with the predicted velocity field and the predicted travel time data.

[0022] In a second aspect, the present invention provides a tomographic inversion device, comprising:

[0023] An acquisition module is used to acquire the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points;

[0024] A solution space creation module, used to create a partial differential equation solution space according to the shot point space coordinates and the detection point space coordinates, and discretize the partial differential equation space coordinates;

[0025] An error determination module, used to determine the travel time data error of the shot check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0026] A network construction module is used to adopt an augmented Lagrangian framework to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network;

[0027] The tomography module is used to input the partial differential equation space coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0028] Optionally, the network construction module includes:

[0029] A function construction submodule, for constructing an augmented Lagrangian function under the augmented Lagrangian framework by combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model;

[0030] The network construction submodule is used to use the augmented Lagrangian function as the objective function, use an optimization algorithm, update the network weights during the training process of the deep learning neural network, and obtain the target velocity field prediction neural network.

[0031] Optionally, the deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; the error determination module includes:

[0032] A travel time data prediction submodule, used for inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data;

[0033] A velocity field prediction submodule, used for inputting the partial differential equation space coordinates into the velocity field prediction neural network to obtain a predicted velocity field;

[0034] The error determination submodule is used to determine the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field.

[0035] Optionally, the error determination submodule includes:

[0036] A travel time field error calculation unit is used to calculate the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space;

[0037] The partial differential equation error calculation unit is used to obtain the partial differential equation error based on a preset partial differential equation error calculation expression in combination with the predicted velocity field and the predicted travel time data.

[0038] In a third aspect, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the method provided in the first aspect are performed.

[0039] In a fourth aspect, the present application provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps in the method provided in the first aspect are performed.

[0040] It can be seen from the above technical solutions that the present invention has the following advantages:

[0041] The present invention provides a tomographic inversion method, device, electronic device and storage medium, the method comprising: obtaining the spatial coordinates of the shot point, the spatial coordinates of multiple detection points and the travel time data of the shot detection point; creating a partial differential equation solution space according to the spatial coordinates of the shot point and the spatial coordinates of the detection point, and discretizing the partial differential equation space coordinates; determining the travel time data error of the shot detection point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space; using an augmented Lagrangian framework, fusing the travel time data error and the partial differential equation error, and constructing a target velocity field prediction neural network; inputting the partial differential equation space coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model. The target velocity field prediction neural network is constructed by fusion of the augmented Lagrangian framework, data error and partial differential equation error, realizing simultaneous prediction and constraint of travel time field and velocity field, and updating the velocity model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0043] Figure 1 A flowchart of a tomographic inversion method according to a first embodiment of the present invention;

[0044] Figure 2 A flowchart of a second embodiment of a tomographic inversion method of the present invention;

[0045] Figure 3 FIG. 1 is a flow chart of a third embodiment of a tomographic inversion method of the present invention;

[0046] Figure 4 A schematic diagram of model terrain undulations in a fourth embodiment of a tomographic inversion method of the present invention;

[0047] Figure 5 A schematic diagram of a longitudinal section of a real model of a fourth embodiment of a tomographic inversion method of the present invention;

[0048] Figure 6 It is a schematic diagram of tomographic velocity of a tomographic inversion method embodiment 4 of the present invention;

[0049] Figure 7 It is a schematic diagram of comparison between tomographic velocity and true velocity in a fourth embodiment of a tomographic inversion method of the present invention;

[0050] Figure 8 It is a structural block diagram of an embodiment of a tomographic inversion device of the present invention. DETAILED DESCRIPTION

[0051] The embodiments of the present invention provide a tomographic inversion method, device, electronic equipment and storage medium, which are used to shorten the time required for tomographic inversion while ensuring the accuracy of underground medium velocity model.

[0052] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below 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.

[0053] For example, see Figure 1 , Figure 1 The present invention is a flowchart of a tomographic inversion method according to Embodiment 1, comprising:

[0054] Step S101, obtaining the spatial coordinates of the shot point, the spatial coordinates of multiple detection points and the travel time data of the shot detection points;

[0055] Step S102, creating a partial differential equation solution space according to the shot point space coordinates and the detection point space coordinates, and discretizing the partial differential equation space coordinates;

[0056] Step S103, determining the travel time data error of the shot check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0057] Step S104, using an augmented Lagrangian framework to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network;

[0058] Step S105, inputting the partial differential equation spatial coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0059] In a tomographic inversion method provided in an embodiment of the present invention, the spatial coordinates of the shot point, the spatial coordinates of multiple detection points and the travel time data of the shot point are obtained; based on the spatial coordinates of the shot point and the spatial coordinates of the detection point, a partial differential equation solution space is created, and the spatial coordinates of the partial differential equation are discretized; the travel time data error of the shot point travel time data under the constructed deep learning neural network and the partial differential equation error in the partial differential equation solution space are determined; the travel time data error and the partial differential equation error are fused using an augmented Lagrangian framework to construct a target velocity field prediction neural network; the partial differential equation spatial coordinates are input into the target velocity field prediction neural network to obtain a tomographic velocity model. The target velocity field prediction neural network is constructed by fusion of the augmented Lagrangian framework, data error and partial differential equation error, to achieve simultaneous prediction and constraint of the travel time field and velocity field, and to update the velocity model.

[0060] For example 2, please refer to Figure 2 , Figure 2 The present invention is a flow chart of a second embodiment of a tomographic inversion method, comprising:

[0061] Step S201, obtaining the spatial coordinates of the shot point, the spatial coordinates of multiple detection points and the travel time data of the shot detection points;

[0062] Step S202, creating a partial differential equation solution space according to the shot point space coordinates and the detection point space coordinates, and discretizing the partial differential equation space coordinates;

[0063] Step S203, determining the travel time data error of the shot check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0064] Step S204, constructing an augmented Lagrangian function under the augmented Lagrangian framework by combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model;

[0065] Step S205, taking the augmented Lagrangian function as the objective function, using an optimization algorithm, updating the network weights during the training process of the deep learning neural network, and obtaining the target velocity field prediction neural network;

[0066] Step S206, inputting the partial differential equation spatial coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0067] A tomographic inversion method provided in an embodiment of the present invention introduces Lagrangian multipliers and augmented terms, combines data errors, partial differential equation errors, and boundary conditions of the partial differential equation model, constructs an augmented Lagrangian function, and uses this as the objective function to train a prediction neural network to obtain a target velocity field prediction neural network that satisfies both data constraints and physical constraints and can provide accurate tomographic inversion.

[0068] For example 3, please refer to Figure 3 , Figure 3 The present invention is a flowchart of a fourth embodiment of a tomographic inversion method, comprising:

[0069] Step S301, obtaining the spatial coordinates of the shot point, the spatial coordinates of multiple detection points and the travel time data of the shot detection points;

[0070] Step S302, creating a partial differential equation solution space according to the shot point space coordinates and the detection point space coordinates, and discretizing the partial differential equation space coordinates;

[0071] Step S303, inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data;

[0072] Step S304, inputting the partial differential equation space coordinates into the velocity field prediction neural network to obtain a predicted velocity field;

[0073] In the embodiment of the present invention, the prediction neural network adopts an independent dual network structure to fit the travel time field and the velocity field respectively, that is:

[0074]

[0075] in, is the spatial point coordinate, N t is the travel time field prediction neural network, which can be understood as a composite function; N vIt is a velocity field prediction neural network, which can also be understood as a composite function. The corresponding physical meaning is that for any spatial point, the coordinates are input, the travel time field prediction neural network can give a travel time data from the earthquake source to the current spatial point, and the velocity field prediction neural network can give the velocity value of the current spatial point. In summary, the intelligent tomographic inversion problem is to obtain the velocity field prediction neural network and the travel time field prediction neural network at the same time after neural network training.

[0076] Step S305, calculating the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space;

[0077] Step S306, based on a preset partial differential equation error calculation expression, combined with the predicted velocity field and the predicted travel time data, obtain the partial differential equation error;

[0078] Step S307, constructing an augmented Lagrangian function under the augmented Lagrangian framework by combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model;

[0079] The embodiment of the present invention adopts an augmented Lagrangian framework, in which the data error and the partial differential equation error are integrated. For a single source, the corresponding augmented Lagrangian function is:

[0080]

[0081] in, is the data error, is the partial differential equation error,

[0082] is the boundary condition of the partial differential equation. Specifically, N r is the number of receivers, is the value of the travel time field at the receiver location, is the observed travel time, N is the number of points in the discrete velocity field, T is the travel time field, v is the velocity field, if the travel time is at the source location, is zero.

[0083] Step S308, taking the augmented Lagrangian function as the objective function, using an optimization algorithm, updating the network weights during the training process of the deep learning neural network, and obtaining the target velocity field prediction neural network;

[0084] In an embodiment of the present invention, the travel time curve is roughly positively correlated with the shot detection distance. When the detector coincides with the shot point, the travel time is zero; the farther the detection point is from the shot point, the greater the travel time. In the process of solving intelligent tomography, the model may not be accurate enough at large offsets due to the imbalance of travel time data. Therefore, in order to speed up the convergence speed of intelligent tomography inversion, an analytical background travel time function can be given in advance, such as a function proportional to the distance of the shot detection point. When predicting the neural network training, the travel time field prediction neural network is pre-decomposed. The decomposition of the time variable can be achieved by multiplying T=T0τ or adding T=T0+τ to construct a travel time field prediction network about the intermediate variable τ to replace the travel time field prediction network about T. Thereby, the value range of the intermediate variable becomes smaller. For example, for the travel time data generated by the uniform model, the travel time data is proportional to the distance; after decomposition by product, the true value of the intermediate variable is 1, thereby balancing the contribution of each offset distance to the tomography inversion.

[0085] At the same time, in order to speed up the training of intelligent tomographic inversion, for a single shot set, the inversion range is automatically selected according to its detection point coverage range to avoid inputting the coordinates of the entire work area, thereby reducing the demand for memory and speeding up network training.

[0086] Furthermore, a seismic exploration area usually contains tens of thousands of shot sets. With limited computing resources, it is difficult to read in the travel times of all the shot sets at once. At this time, it is necessary to use incremental data processing methods to solve the problem of massive data, that is, input the data into the neural network in batches, and each neural network training is updated on the previous network model.

[0087] In addition, the embodiment of the present invention adopts progressive application of data components to solve the problem of slow neural network training under the condition that the travel time curve may appear, that is, the functional relationship between the first arrival travel time and the shot check point distance or the shot check distance is complex. That is to say, the observed data is first fitted with a polynomial, and the neural network is trained on the fitted data; after the training is completed, high-frequency data components are gradually added until the original first arrival travel time is added. This avoids the situation that when the actual geological model is complex and the travel time curve is complex and changeable, if the travel time curve is directly input into the prediction neural network, the high-frequency components of the travel time curve need to be fitted after multiple rounds of iterations.

[0088] Finally, in order to reduce unnecessary calculations, the embodiment of the present invention allows the user to specify the lower interface.

[0089] Step S309: input the partial differential equation spatial coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0090] A tomographic inversion method, by obtaining the spatial coordinates of the shot point, the spatial coordinates of multiple detection points and the travel time data of the shot detection point; according to the spatial coordinates of the shot point and the spatial coordinates of the detection point, a partial differential equation solution space is created, and the spatial coordinates of the partial differential equation are discretized; the travel time data error of the shot detection point travel time data under the constructed deep learning neural network and the partial differential equation error in the partial differential equation solution space are determined; the travel time data error and the partial differential equation error are fused by using an augmented Lagrangian framework to construct a target velocity field prediction neural network; the partial differential equation spatial coordinates are input into the target velocity field prediction neural network to obtain a tomographic velocity model. The target velocity field prediction neural network is constructed by fusion of the augmented Lagrangian framework, data error and partial differential equation error, so as to achieve simultaneous prediction and constraint of travel time field and velocity field, and update the velocity model.

[0091] To facilitate those skilled in the art to understand the beneficial effects of the embodiments of the present invention, the velocity model obtained by tomographic inversion in the fourth embodiment of the present invention and the real model are compared and illustrated below.

[0092] See also Figure 4-5 , Figure 4 Schematic diagram of model terrain fluctuations in a fourth embodiment of a tomographic inversion method of the present invention, Figure 5 This is a schematic diagram of a longitudinal section of a real model of a tomographic inversion method embodiment 4 of the present invention, based on Figure 4 and Figure 5 , using the target travel time field prediction neural network and the target velocity field prediction neural network to predict, we get Figure 6 The schematic diagram of the layer velocity is shown, and then the comparison method of interval display is obtained. Figure 7 From the schematic diagram of comparison between the tomographic velocity and the true velocity, it can be seen that the velocity model obtained by the tomographic inversion method of the embodiment of the present invention is relatively consistent with the true velocity model.

[0093] Example 4, please refer to Figure 8 , Figure 8 The structure diagram of a tomographic inversion device embodiment of the present invention includes:

[0094] An acquisition module 501 is used to acquire the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points;

[0095] A solution space creation module 502 is used to create a partial differential equation solution space according to the shot point space coordinates and the detection point space coordinates, and discretize the partial differential equation space coordinates;

[0096] An error determination module 503 is used to determine the travel time data error of the shot check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0097] A network construction module 504 is used to adopt an augmented Lagrangian framework to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network;

[0098] The tomographic module 505 is used to input the partial differential equation space coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0099] In an optional embodiment, the network construction module 504 includes:

[0100] A function construction submodule, for constructing an augmented Lagrangian function under the augmented Lagrangian framework by combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model;

[0101] The network construction submodule is used to use the augmented Lagrangian function as the objective function, use an optimization algorithm, update the network weights during the training process of the deep learning neural network, and obtain the target velocity field prediction neural network.

[0102] In an optional embodiment, the deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; the error determination module 503 includes:

[0103] A travel time data prediction submodule, used for inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data;

[0104] A velocity field prediction submodule, used for inputting the partial differential equation space coordinates into the velocity field prediction neural network to obtain a predicted velocity field;

[0105] The error determination submodule is used to determine the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field.

[0106] In an optional embodiment, the error determination submodule includes:

[0107] A travel time field error calculation unit is used to calculate the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space;

[0108] The partial differential equation error calculation unit is used to obtain the partial differential equation error based on a preset partial differential equation error calculation expression in combination with the predicted velocity field and the predicted travel time data.

[0109] Embodiment 5. The embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of a tomographic inversion method, including:

[0110] Obtain the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points;

[0111] According to the shot point spatial coordinates and the detection point spatial coordinates, a partial differential equation solution space is created, and the partial differential equation spatial coordinates are discretized;

[0112] Determine the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0113] Adopting an augmented Lagrangian framework, fusing the travel time data error and the partial differential equation error, and constructing a target velocity field prediction neural network;

[0114] The partial differential equation spatial coordinates are input into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0115] In an optional embodiment, an augmented Lagrangian framework is used to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network, including:

[0116] Combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model, constructing an augmented Lagrangian function under the augmented Lagrangian framework;

[0117] The augmented Lagrangian function is used as the objective function, an optimization algorithm is used, and network weights are updated during the training process of the deep learning neural network to obtain the target velocity field prediction neural network.

[0118] In an optional embodiment, the deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; determining the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space, including:

[0119] Inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data;

[0120] Inputting the partial differential equation spatial coordinates into the velocity field prediction neural network to obtain a predicted velocity field;

[0121] Based on the predicted travel time data and the predicted velocity field, the travel time data error and the partial differential equation error are determined.

[0122] In an optional embodiment, determining the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field includes:

[0123] Calculating the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space;

[0124] Based on a preset partial differential equation error calculation expression, the partial differential equation error is obtained in combination with the predicted velocity field and the predicted travel time data.

[0125] Embodiment 6. The embodiment of the present invention further provides a computer storage medium, on which a computer program is stored. When the computer program is executed by the processor, the steps of implementing a tomographic inversion method include:

[0126] Obtain the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points;

[0127] According to the shot point spatial coordinates and the detection point spatial coordinates, a partial differential equation solution space is created, and the partial differential equation spatial coordinates are discretized;

[0128] Determine the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space;

[0129] Adopting an augmented Lagrangian framework, fusing the travel time data error and the partial differential equation error, and constructing a target velocity field prediction neural network;

[0130] The partial differential equation spatial coordinates are input into the target velocity field prediction neural network to obtain a tomographic velocity model.

[0131] In an optional embodiment, an augmented Lagrangian framework is used to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network, including:

[0132] Combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model, constructing an augmented Lagrangian function under the augmented Lagrangian framework;

[0133] The augmented Lagrangian function is used as the objective function, an optimization algorithm is used, and network weights are updated during the training process of the deep learning neural network to obtain the target velocity field prediction neural network.

[0134] In an optional embodiment, the deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; determining the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space, including:

[0135] Inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data;

[0136] Inputting the partial differential equation spatial coordinates into the velocity field prediction neural network to obtain a predicted velocity field;

[0137] Based on the predicted travel time data and the predicted velocity field, the travel time data error and the partial differential equation error are determined.

[0138] In an optional embodiment, determining the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field includes:

[0139] Calculating the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space;

[0140] Based on a preset partial differential equation error calculation expression, the partial differential equation error is obtained in combination with the predicted velocity field and the predicted travel time data.

[0141] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0142] In the several embodiments provided in the present application, it should be understood that the methods, devices, electronic devices and storage media disclosed in the present invention can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0143] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0144] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0145] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a readable storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned readable storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0146] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A tomographic inversion method, characterized in that: include: Obtain the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points; According to the shot point spatial coordinates and the detection point spatial coordinates, a partial differential equation solution space is created, and the partial differential equation spatial coordinates are discretized; Determine the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space; Adopting an augmented Lagrangian framework, fusing the travel time data error and the partial differential equation error, and constructing a target velocity field prediction neural network; The partial differential equation spatial coordinates are input into the target velocity field prediction neural network to obtain a tomographic velocity model.

2. The tomographic inversion method according to claim 1, characterized in that: The augmented Lagrangian framework is adopted to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network, including: Combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model, constructing an augmented Lagrangian function under the augmented Lagrangian framework; The augmented Lagrangian function is used as the objective function, an optimization algorithm is used, and network weights are updated during the training process of the deep learning neural network to obtain the target velocity field prediction neural network.

3. The tomographic inversion method according to claim 1, characterized in that: The deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; determining the travel time data error of the gun check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space, including: Inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data; Inputting the partial differential equation spatial coordinates into the velocity field prediction neural network to obtain a predicted velocity field; Based on the predicted travel time data and the predicted velocity field, the travel time data error and the partial differential equation error are determined.

4. The tomographic inversion method according to claim 3, characterized in that: Determining the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field includes: Calculating the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space; Based on a preset partial differential equation error calculation expression, the partial differential equation error is obtained in combination with the predicted velocity field and the predicted travel time data.

5. A tomographic inversion device, characterized in that: include: An acquisition module is used to acquire the spatial coordinates of the shot point, the spatial coordinates of multiple detection points, and the travel time data of the shot detection points; A solution space creation module, used to create a partial differential equation solution space according to the shot point space coordinates and the detection point space coordinates, and discretize the partial differential equation space coordinates; An error determination module, used to determine the travel time data error of the shot check point travel time data under the constructed deep learning neural network, and the partial differential equation error in the partial differential equation solution space; A network construction module is used to adopt an augmented Lagrangian framework to fuse the travel time data error and the partial differential equation error to construct a target velocity field prediction neural network; The tomography module is used to input the partial differential equation space coordinates into the target velocity field prediction neural network to obtain a tomographic velocity model.

6. The tomographic inversion device according to claim 5, characterized in that: The network building module includes: A function construction submodule, for constructing an augmented Lagrangian function under the augmented Lagrangian framework by combining the travel time data error, the partial differential equation error, and the boundary conditions of the partial differential equation model; The network construction submodule is used to take the augmented Lagrangian function as the objective function, use an optimization algorithm, update the network weights during the training process of the deep learning neural network, and obtain the target velocity field prediction neural network.

7. The tomographic inversion device according to claim 5, characterized in that: The deep learning neural network is an independent dual network structure, including: a travel time field prediction neural network and a velocity field prediction neural network; the error determination module includes: A travel time data prediction submodule, used for inputting the shot point spatial coordinates, the detection point spatial coordinates and the shot detection point travel time data into the travel time field prediction neural network to obtain predicted travel time data; A velocity field prediction submodule, used for inputting the partial differential equation space coordinates into the velocity field prediction neural network to obtain a predicted velocity field; The error determination submodule is used to determine the travel time data error and the partial differential equation error based on the predicted travel time data and the predicted velocity field.

8. The tomographic inversion device according to claim 7, characterized in that: The error determination submodule comprises: A travel time field error calculation unit is used to calculate the travel time field error of each shot check point under the predicted travel time data and the corresponding travel time data of the shot check point in the partial differential equation solution space; The partial differential equation error calculation unit is used to obtain the partial differential equation error based on a preset partial differential equation error calculation expression in combination with the predicted velocity field and the predicted travel time data.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 4 is executed.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is executed.

Images