Construction method of logging data processing pre-training large model and related equipment

Abstract

The invention relates to the technical field of petroleum and natural gas exploration and development, and discloses a construction method and related equipment for a logging data processing pre-training large model, and the method comprises the steps: carrying out the preprocessing of original logging data, dividing the preprocessed logging data according to a logging interpretation task, and obtaining the logging curve data of different layers; building a logging monomer large model based on a Transform algorithm, and designing a loss function; logging curve data of different layer sections are input into the logging single body large model for training, loss values of the logging curve data of the different layer sections are obtained through calculation according to a loss function in the network training process, and parameter updating is conducted on the loss values of the logging curve data of the different layer sections through a stochastic gradient descent optimization algorithm; according to the method, the well logging task reasoning efficiency and the generalization ability under various intelligent scenes are greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil and gas exploration and development, in particular to a method for constructing a well logging data processing pre-training large model and related equipment. BACKGROUND

[0002] With the rapid development of society and economy, the consumption of oil by countries is increasing, but as a non-renewable resource, the reserves of oil will only become less and less, and even exhausted. Compared with developing new oil fields to alleviate the problem of oil resource shortage, deep development of the remaining oil and gas resources in the existing developed oil fields and promotion of high and stable production of the existing oil fields are more cost-effective and more practical. For a long time, the well logging interpretation task mainly adopts manual processing method, that is, the well logging interpreter makes subjective judgment according to the shape transformation characteristics of the well logging curve and his own experience. Manual well logging interpretation not only wastes manpower and material resources and is low in efficiency, but also is affected by subjective factors and is limited by the technical level of the well logging interpreter. Especially for the interpreter who is not familiar with the geological conditions and lacks experience, the division result is quite different and the error rate is high.

[0003] Since 2012, artificial intelligence technology dominated by deep learning has developed rapidly, and well logging interpretation technology based on deep learning has made great breakthroughs. One of the current research difficulties is how to make the well logging artificial intelligence model maintain relatively good and stable generalization performance and interpretation accuracy in different oil field blocks. Therefore, the present application automatically learns the characteristics of conventional well logging curve data in a self-supervised manner to form an upstream general well logging curve feature extraction model. Based on the upstream basic model, various downstream task corresponding intelligent models are applied, such as intelligent geological layering, reservoir division, fluid property identification and reservoir parameter prediction models, to find a well logging intelligent interpretation large model with better generalization performance and greater scalability. SUMMARY

[0004] The present application aims to provide a method for constructing a well logging data processing pre-training large model and related equipment to solve the technical problems of poor generalization performance of intelligent algorithms and single model knowledge in the scenarios of geological layering, fluid property identification and physical property parameter prediction caused by different block well logging data characteristics.

[0005] The present application is achieved by the following technical solutions:

[0006] In a first aspect, the present application provides a method for constructing a well logging data processing pre-training large model, comprising:

[0007] The original well logging data is preprocessed, and the preprocessed well logging data is divided according to the well logging interpretation task to obtain well logging curve data of different layers;

[0008] The well logging monomer large model based on the Transformer algorithm is built, and a loss function is designed.

[0009] The well logging curve data of different intervals is input into the well logging monomer large model for training. In the network training process, the loss values of the well logging curve data of different intervals are calculated according to the loss function. The loss values of the well logging curve data of different intervals are updated by the parameter through the stochastic gradient descent optimization algorithm, and the construction of the well logging data processing pre-training large model is completed.

[0010] Preferably, the specific process of preprocessing the original well logging data is as follows:

[0011] The abnormal values in the original well logging data are removed, the values exceeding the reasonable range are removed, and the measurement curve is filtered to remove sharp values.

[0012] The original well logging data after removing the abnormal values is standardized, and the standardization formula is as follows:

[0013]

[0014] Wherein, x i represents the normalized well logging data, x i represents the original well logging data, x max and x min represent the maximum and minimum values of a curve in a well, respectively.

[0015] Preferably, the well logging interpretation task includes a geological layering task, a physical property parameter prediction task, and a fluid property identification task. The geological layering task and the physical property parameter prediction task input the well logging curve features in the interval as a sample. The fluid property identification task inputs the well logging curve features of each layer of the geological layering result as a sample.

[0016] Preferably, in the well logging monomer large model based on the Transformer, the data sample is input to make the network learn the key features of the well logging curve in a self-supervised manner, and form a basic public feature skeleton model. The well logging monomer large model includes an encoder and a decoder of self-attention mechanism, and the input of the decoder is set as the output of the encoder. The feature relationship between the well logging data sequences is learned through a multi-head self-attention module. The specific formula is as follows:

[0017] b m = concatenate(b m，1 , b m，2 ,..., b m，h )W m

[0018] Wherein, b m，1The feature output by the first attention head module, b m The comprehensive feature output obtained by performing dot product operation on the features output by the plurality of attention modules and W m The weight matrix, to obtain the feature space of the final base model m.

[0019] Further, in the pre-training logging monomer large model, a plurality of simple logging curve feature extraction modules are connected in series to form a decoding encoder by learning the feature relationship between logging data sequences through a multi-head self-attention module, to perform feature extraction on the logging curve.

[0020] Further, in the logging monomer large model of the pre-training algorithm, a plurality of downstream algorithm models are constructed, based on high-dimensional curve features, to form a parallel structure or a single model structure, to complete the interpretation task.

[0021] Preferably, in the geological layering and physical parameter prediction of the logging interpretation task, the prediction network outputs the layering identification result of each sampling point, and the mean square error of the identification result of each layer is calculated, to perform back propagation and model parameter updating;

[0022] In the fluid property identification task of the logging interpretation task, the logging curve feature data in one of the layer intervals is input, and the prediction network is used to output the fluid property classification result of the layer.

[0023] In a second aspect, the present application provides a logging data processing pre-training large model construction system, comprising

[0024] The data processing module is used for pre-processing the original logging data, and dividing the pre-processed logging data according to the logging interpretation task to obtain logging curve data of different layer sections;

[0025] The model building module is used for building a logging monomer large model based on the Transformer algorithm, and designing a loss function;

[0026] The model training module is used for inputting the logging curve data of different layer sections into the logging monomer large model for training, and calculating the loss value of the logging curve data of different layer sections according to the loss function during the training network process, and updating the parameters of the logging curve data of different layer sections through the stochastic gradient descent optimization algorithm, to complete the construction of the logging data processing pre-training large model.

[0027] In a third aspect, the present application provides a mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the logging data processing pre-training large model construction method as described above.

[0028] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements the steps of the method for constructing a well data processing pre-training large model.

[0029] Compared with the prior art, the present application has the following beneficial technical effects:

[0030] The present application provides a method for constructing a well data processing pre-training large model. The original well logging data is preprocessed, and the preprocessed well logging data is divided according to the well logging interpretation task to obtain well logging curve data of different intervals, so that the well logging data meets different well logging interpretation tasks and is divided. The model is fine-tuned using a new data set, the model development process is simplified, the generalization performance of the well logging interpretation task is improved, and higher precision recognition and regression results are achieved compared with traditional single machine learning and deep learning algorithms. This model has good expansibility, supports the addition of downstream models and output detection networks, so that it can be more easily fine-tuned in different scenarios, such as geological layering. For each task, learning can be based on existing knowledge features, greatly improving the well logging task inference efficiency and generalization ability in various intelligent scenarios, improving the work efficiency of interpreters, and reducing the cost loss in the generation process.

[0031] Further, for different well logging interpretation tasks, the data input required by the well logging pre-training large model is different. For geological layering or physical property parameter prediction tasks, the well logging curve features in an interval can be input as a sample. In the algorithm running and model training process, the data length needs to be limited to relieve the hardware burden. By setting the window length and step length, the well logging data of the whole well section can be equally divided to form a usable data set. In the fluid property identification task, the well logging curve features of each layer of the geological layering result are input as a sample unit, which facilitates improving the well logging task inference efficiency and generalization ability in various intelligent scenarios.

[0032] Further, the parallel structure of the model in the present application can realize the well logging interpretation multi-task learning model. Compared with the training of multiple models respectively, on the one hand, the multi-task model can greatly save the training time and reduce the memory occupation, and on the other hand, the same well logging feature expression space can be used to efficiently perform the well logging interpretation task. Since the downstream multiple parallel network structures are relatively simple, they are responsible for up-sampling the high-dimensional feature space and outputting fluid property identification, reservoir parameter prediction and other tasks, so during actual training, only batch well logging data needs to be input, and based on the existing model parameters, the model is fine-tuned, so that good model generalization effect can be realized. On the other hand, the model in the present application can also be connected with a multi-level network to realize more complex well logging interpretation tasks, for example, the reservoir division model and the fluid property identification model are connected in series to form a downstream model connected to the Backbone model, thereby realizing a multi-level fluid property identification model based on the Transformer. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The flow chart of the method for constructing the well logging data processing pre-training large model in the present application is shown in the figure.

[0034] Figure 2 The architecture diagram of the well logging monomer large model in the present application is shown in the figure.

[0035] Figure 3 The schematic diagram of the Transformer model in the present application is shown in the figure.

[0036] Figure 4 The schematic diagram of the multi-head self-attention module in the present application is shown in the figure.

[0037] Figure 5 The schematic diagram of the fine-tuning stage of the well logging monomer large model in the present application is shown in the figure.

[0038] Figure 6 The schematic diagram of the geological layering effect based on the well logging pre-training monomer large model in the present application is shown in the figure.

[0039] Figure 7 The schematic diagram of the construction system structure of the well logging data processing pre-training large model in the present application is shown in the figure.

[0040] In the figure: 1-data processing module; 2-model building module; 3-model training module. DETAILED DESCRIPTION

[0041] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort should belong to the protection scope of the present application.

[0042] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device including a series of steps or units does not necessarily have to include only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product, or device.

[0043] The present application will be described in further detail below with reference to the drawings:

[0044] The present application aims to provide a well logging data processing pre-training large model construction method and related equipment, to solve the technical problems of poor generalization performance of intelligent algorithms and single model knowledge in the scenarios of geological layering, fluid property identification, and physical property parameter prediction caused by different block well logging data characteristics.

[0045] Embodiment 1

[0046] Referring to Figure 1 In an embodiment of the present application, there is provided

[0047] Step 1, the original well logging data is preprocessed, and the well logging data after preprocessing is divided according to the well logging interpretation task to obtain well logging curve data of different layers;

[0048] Specifically, the original data is preprocessed, including removing abnormal values such as 0, -9999, and other values outside the reasonable range, and performing simple filtering processing on the measurement curve to remove sharp values generated by machine noise, etc. Secondly, in order to ensure the features used for unified training, the data needs to be standardized, and the standardization formula is as follows:

[0049]

[0050] Wherein, x iX represents normalized logging data i X represents original logging data max X represents normalized logging data min X represents normalized logging data

[0051] Specifically, the conventional data is still full well section data after preprocessing. For different logging interpretation tasks, the data input required by the pre-trained large model is different. For the geological layering or physical property parameter prediction task, the logging curve features in an interval can be used as a sample input. In the algorithm running and model training process, the data length needs to be limited to relieve the hardware burden. By setting the window length and step length, the full well section logging data can be equally divided to form a usable data set. In the present application, a sliding window with a length of 256 and a step length of 256 is used to divide the data, and a large amount of logging curve data of different intervals can be obtained; in the fluid property identification task, the logging curve features of each layer of the geological layering result are used as a sample unit for input. The logging interpretation application large model in the present application is composed of a feature extraction basic model based on Transformer and a network for different downstream logging interpretation tasks. The feature extraction basic model needs to learn the features of the curve itself without additional labels. In the present application, part of the training data is covered, and the covered area data is used as a label. The public feature extraction model uses logging curve data containing a mask for self-supervised feature learning. Other downstream logging interpretation models use complete logging curve data as input during training, and the labels are obtained from the labels corresponding to each task.

[0052] Step 2, according to Figure 2 and Figure 3 , a logging monomer large model based on the Transformer algorithm is built, and a loss function is designed;

[0053] Specifically, the logging data processing large model used in the present embodiment mainly includes two processes, i.e. by inputting a large amount of data samples, the Transformer network learns the key features of the logging curve in a self-supervised manner to form a basic public feature backbone model; the second stage is to adapt intelligent learning models in different intelligent scenarios, such as geological layering models, etc., to complete a number of downstream tasks, as shown in Figure 5 . Among them, the Transformer in the first stage is composed of an encoder and a decoder containing a self-attention mechanism. The input of the decoder in the present application is set to the output of the encoder, and the Multi-head Self-Attention learns the feature relationship between the logging data sequences, as shown in Figure 4 .

[0054] In the Transformer-based encoder-decoder of this invention, the self-attention module uses Q and K to represent the correlation coefficients of each input logging curve sequence, and V to represent the original vector of each logging curve sequence. These three vectors are used to output features and corresponding weight matrices W from the previous input sequence. q W k and W v The dot product of Q, K, and V is obtained, and the formulas for calculating them are as follows:

[0055] q m =W q a m

[0056] K m =W k a m

[0057] V m =W v a m

[0058] Where a m To output the features of the logging curve sequence m, this method generates Q, K, and V feature vectors for multiple input sequences. Based on these three vectors, a feature space a is output that comprehensively considers the features of multiple input sequences. m , used to express well logging curve sequence b m The calculation formula is as follows:

[0059]

[0060] Where a m,i This represents the dot product of model i in the correlation feature space of the logging curve sequence m. It is a m,i The intermediate result after the Softmax function, b m for and corresponding v i The output characteristics of the base model m, which integrates the results of multiple other well logging curve sequences, are obtained by performing a dot product and summing the results. m By integrating all output feature information, the pre-trained network uses this module to further learn the weights of all logging curve sequences. Building upon this, this invention applies a multi-head attention mechanism module, comprehensively considering the output b obtained from multiple attention modules. m The final feature is output as shown in the following formula:

[0061] b m =concatenate(b m,1 ,b m,2 ,…,b m,h W m

[0062] wherein b m,1 is the feature output of the first attention head module, b m is the comprehensive feature output obtained by performing dot product operation on the output of the plurality of attention modules and W m is a weight matrix, to obtain a feature space of the final log curve sequence m.

[0063] Specifically, the feature space b m is obtained by the MSA network, and the application is based on the MSA network to connect a plurality of simple log curve feature extraction modules (Curve Feature Block, CF Block) to form a decoding encoder to perform feature extraction on the log curve. Assuming that the input of the previous layer in the network is z l-1 , after the CF module, the output z l-1 of the next layer can be obtained, and the specific process is shown in the following formula:

[0064]

[0065] The CF module is composed of MSA and MLP, and the MLP in the application is stacked by three fully connected neural networks (Fully Connected Network). Each output in the module integrates the input of the module, the output of the MSA and the output of the MLP, so that better feature extraction effect is achieved, and the calculation amount is reduced through the MLP layer.

[0066] Specifically, the embodiment proposes to implement intelligent logging interpretation by matching downstream models used in different logging interpretation tasks. Specifically, a feature extraction network based on a Transformer is used to learn logging data features and has the ability of curve reconstruction. In the face of various logging interpretation scenarios, the encoder part of this network can be used as a basic Backbone model to receive logging data input and output high-dimensional curve features. The present application constructs various downstream algorithm models in the logging intelligent interpretation model process, forms a parallel structure or a single model structure based on high-dimensional curve features, and completes the interpretation task. The parallel structure based on the Backbone model in the present application can realize a multitask learning model (Multitask Learning Model) for logging interpretation. This multitask model can greatly save training time and reduce memory occupation on the one hand, and can efficiently perform logging interpretation tasks by using the same logging feature expression space on the other hand. Since the multiple parallel network structures in the downstream are relatively simple, they are responsible for upsampling the high-dimensional feature space and outputting fluid property identification, reservoir parameter prediction and other tasks, so in actual training, only batch logging data needs to be input, and the model can be fine-tuned based on the existing Backbone model parameters to achieve good model generalization effect. On the other hand, the Backbone model in the present application can also be matched with a multi-level network to realize more complex logging interpretation tasks, for example, a reservoir division model and a fluid property identification model are connected in series to form a downstream model matched to the Backbone model, realizing a multi-level fluid property identification model based on a Transformer.

[0067] Step 3: Input the logging curve data of different layers into the logging monomer large model for training. In the training network process, the loss values of the logging curve data of different layers are calculated according to the loss function, and the loss values of the logging curve data of different layers are updated by the stochastic gradient descent optimization algorithm to complete the construction of the logging data processing pre-training large model.

[0068] Specifically, in the training process, for geologic layering and physical property parameter prediction, the prediction network of the present application outputs the layering results of each sampling point, and the present application calculates the mean square error of the identification results of each layer for back propagation and model parameter updating to ensure that the classification loss values of layers are not related; for fluid property identification, the present application inputs the logging curve feature data in a certain layer interval, and outputs the fluid property classification results of the layer through the prediction network. The fluid property identification and geologic layering both use a Softmax function as a loss function, as shown in the following formula:

[0069]

[0070] wherein Loss gLoss represents the loss function of geological layering, Loss f y represents the loss function of fluid property identification, y i represents the i-th component of the one-hot encoding of the geological layering label corresponding to a sampling point, indicating whether the depth of the sampling point is located in the geological layer i, p i is the prediction result of the geological layering, and C is the number of horizon classification.

[0071] In the physical property parameter prediction, the mean square error is used as the loss function:

[0072]

[0073] where Loss p represents the loss function of physical property parameter prediction, m is the number of samples, h(x i ) is the output of the final activation function, i is the sample index, y i represents the physical property parameter label corresponding to the sample data with index i.

[0074] Specifically, in the training network process in the embodiment, the loss value is calculated, and the parameter is updated through the stochastic gradient descent (SGD) optimization algorithm. In the training process of the pre-trained model in the application, after the network output prediction result, the mean square error is calculated with the label, and the parameter update of the prediction network is performed through the back propagation algorithm, and the generalization performance of the model is continuously improved. Finally, a usable pre-trained model module is obtained which integrates multiple logging knowledge and features.

[0075] Step 4: Apply the trained logging pre-trained large model to the test set to perform geological layering, fluid property identification or physical property parameter prediction, and evaluate the model performance, wherein the geological layering effect based on the logging pre-trained large model is as shown in Figure 6 .

[0076] Specifically, the logging data processing single large model based on the Transformer is trained to have good generalization performance and scalability in logging interpretation tasks. For different interpretation tasks, the intelligent logging interpretation process is realized through the organic combination of the logging data public feature extraction network and multiple downstream task networks. In the subsequent model development process, logging personnel can introduce new downstream base models, fine-tune the network with a small amount of data, enrich the algorithmic capabilities of the logging single large model of the application, meet the logging task requirements, and further improve the generalization ability.

[0077] In summary, this invention provides a method for constructing a pre-trained large-scale model for well data processing. It preprocesses the raw well logging data and then divides the preprocessed data according to the well logging interpretation task, obtaining well logging curve data for different layers. This ensures that the well logging data meets different well logging interpretation tasks and is divided accordingly. The model is then fine-tuned using a new dataset, simplifying the model development process and improving the generalization performance for well logging interpretation tasks. It achieves higher accuracy in recognition and regression results compared to traditional single-unit machine learning and deep learning algorithms. This model has good scalability, supporting the addition of downstream models and output detection networks, making it easier to fine-tune the model in different scenarios, such as geological stratification. For each task, learning can be based on existing knowledge features, greatly improving the inference efficiency and generalization ability of well logging tasks in various intelligent scenarios, increasing the work efficiency of interpreters, and reducing cost losses during the generation process.

[0078] Example 2

[0079] according to Figure 7 As shown, the present invention also provides a system for constructing a pre-trained large model for well logging data processing, comprising:

[0080] Data processing module 1 is used to preprocess the raw logging data and divide the preprocessed logging data according to the logging interpretation task to obtain logging curve data of different layers.

[0081] Model building module 2 is used to build a large-scale well logging model based on the Transformer algorithm and design the loss function;

[0082] Model training module 3 is used to input well logging curve data from different layers into the large well logging model for training. During the network training process, the loss value of the well logging curve data from different layers is calculated according to the loss function. The loss value of the well logging curve data from different layers is used to update the parameters through the stochastic gradient descent optimization algorithm, thus completing the construction of the pre-trained large model for well logging data processing.

[0083] Example 3

[0084] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, such as a program for building a pre-trained large model for well logging data processing.

[0085] When the processor executes the computer program, it implements the steps of the above-described method for constructing a pre-trained large model for well logging data processing, for example:

[0086] The raw logging data is preprocessed, and the preprocessed logging data is divided according to the logging interpretation task to obtain logging curve data of different layers;

[0087] Build a large-scale well logging model based on the Transformer algorithm and design the loss function;

[0088] Well logging curve data from different layers are input into a large-scale well logging model for training. During the training process, the loss value of the well logging curve data from different layers is calculated based on the loss function. The loss value of the well logging curve data from different layers is used to update the parameters through a stochastic gradient descent optimization algorithm, thus completing the construction of the pre-trained large-scale well logging data processing model.

[0089] Alternatively, when the processor executes the computer program, it implements the functions of each module in the above system, for example:

[0090] Data processing module 1 is used to preprocess the raw logging data and divide the preprocessed logging data according to the logging interpretation task to obtain logging curve data of different layers.

[0091] Model building module 2 is used to build a large-scale well logging model based on the Transformer algorithm and design the loss function;

[0092] Model training module 3 is used to input well logging curve data from different layers into the large well logging model for training. During the network training process, the loss value of the well logging curve data from different layers is calculated according to the loss function. The loss value of the well logging curve data from different layers is used to update the parameters through the stochastic gradient descent optimization algorithm, thus completing the construction of the pre-trained large model for well logging data processing.

[0093] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the mobile terminal.

[0094] For example, the computer program can be divided into a data processing module 1, a model building module 2, and a model training module 3;

[0095] Data processing module 1 is used to preprocess the raw logging data and divide the preprocessed logging data according to the logging interpretation task to obtain logging curve data of different layers.

[0096] Model building module 2 is used to build a large-scale well logging model based on the Transformer algorithm and design the loss function;

[0097] Model training module 3 is used to input well logging curve data from different layers into the large well logging model for training. During the network training process, the loss value of the well logging curve data from different layers is calculated according to the loss function. The loss value of the well logging curve data from different layers is used to update the parameters through the stochastic gradient descent optimization algorithm, thus completing the construction of the pre-trained large model for well logging data processing.

[0098] The mobile terminal can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The mobile terminal may include, but is not limited to, a processor and memory.

[0099] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the mobile terminal, connecting various parts of the mobile terminal via various interfaces and lines.

[0100] The memory can be used to store the computer program and / or module. The processor implements various functions of the mobile terminal by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0101] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function (such as sound playback or image playback). The data storage area may store data created based on the use of the phone (such as audio data or a phonebook). Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMediaCards (SMC), Secure Digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0102] Example 4

[0103] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for constructing a pre-trained large model for well logging data processing.

[0104] If the modules / units integrated in the mobile terminal are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0105] Based on this understanding, all or part of the processes in the above method can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-described method for constructing a pre-trained large model for well logging data processing. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate form.

[0106] The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0107] It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a pre-trained large model for processing well logging data, characterized in that, Comprise: The original logging data is preprocessed, and the preprocessed logging data is divided according to the logging interpretation task to obtain logging curve data of different intervals; A logging monomer large model based on the Transformer algorithm is built, and a loss function is designed; The logging curve data of different intervals is input into the logging monomer large model for training, and the loss value of the logging curve data of different intervals is calculated according to the loss function during the training network process. The loss value of the logging curve data of different intervals is updated by the stochastic gradient descent optimization algorithm, and the construction of the logging data processing pre-training large model is completed.

2. The method of claim 1, wherein the method comprises: The specific process of preprocessing the original logging data is as follows: The abnormal values in the original logging data are removed, the values exceeding the reasonable range are removed, and the sharp values are filtered; The original logging data after removing the abnormal values is standardized, and the standardization formula is as follows: where x i represents the normalized log data, X i represents the original log data, X max and X min represent the maximum and minimum values of a curve in a well, respectively.

3. The method of claim 1, wherein the method comprises: training the pre-trained large model using the well logging data and the well logging data label; and outputting the pre-trained large model. The logging interpretation task includes geological layering task, physical property parameter prediction task and fluid property identification task, wherein the geological layering task and the physical property parameter prediction task input the logging curve features in the interval as a sample; the fluid property identification task inputs the logging curve features of each layer of the geological layering result as a sample.

4. The method of claim 1, wherein the method comprises: In the logging monomer large model based on the Transformer, the network learns the key features of the logging curve in a self-supervised manner by inputting data samples to form a basic common feature skeleton model. The logging monomer large model includes an encoder and a decoder with self-attention mechanism, and the input of the decoder is set to the output of the encoder. The feature relationship between the logging data sequences is learned through a multi-head self-attention module, and the specific formula is as follows: b m = concatenate(b m,1 , b m,2 , …, b m,h )W m Among them, b m,1 b is the feature output by the first attention head module m To concatenate the outputs of multiple attention modules and W m The comprehensive feature output obtained by performing a dot product operation on the weight matrix yields the feature space of the final base model m.

5. The method of claim 4, wherein the method further comprises: training the pre-trained large model using the well logging data and the well logging data label; and outputting the pre-trained large model. In the pre-training logging monomer large model, the multi-head self-attention module is used to learn the feature relationship between the logging data sequences, and a plurality of simple logging curve feature extraction modules are connected in series to form a decoding encoder for feature extraction of the logging curve.

6. The method of claim 4, wherein the method further comprises: In the pre-training algorithm logging monomer large model, a plurality of downstream algorithm models are constructed based on high-dimensional curve features to form a parallel structure or a single model structure to complete the interpretation task.

7. The method of claim 1, wherein the method further comprises: training the pre-trained large model using the well logging data and the well logging data label; and outputting the pre-trained large model. In the geological layering and physical property parameter prediction of the logging interpretation task, the prediction network outputs the layering identification result of each sampling point, and calculates the mean square error of the identification result of each layer for back propagation and model parameter update; In the fluid property identification task of the logging interpretation task, the logging curve feature data in one layer interval is input, and the fluid property classification result of the layer is output through the prediction network. 8.A system for constructing a pre-trained large model for processing well logging data, characterized in that, Comprise A data processing module (1) for preprocessing the original logging data, and dividing the preprocessed logging data according to the logging interpretation task to obtain logging curve data of different intervals; A model building module (2) for building a logging monomer large model based on the Transformer algorithm, and designing a loss function; The model training module (3) is used for inputting the logging curve data of different layer sections into a logging single-body large model for training, and loss values of the logging curve data of different layer sections are calculated according to a loss function in the network training process, the loss values of the logging curve data of different layer sections are updated in parameters through a stochastic gradient descent optimization algorithm, and the construction of the logging data processing pre-training large model is completed.

9. A mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The computer program is executed by the processor to realize the steps of the construction method of the logging data processing pre-training large model according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to realize the steps of the construction method of the logging data processing pre-training large model according to any one of claims 1-7.

Images