Shale gas early-stage production dynamic prediction method, device and equipment and storage medium

Abstract

The embodiment of the invention provides a shale gas early-stage production dynamic prediction method and device, equipment and a storage medium, and the method comprises the steps: collecting target data, and preprocessing the target data, the target data comprising geological parameters, engineering parameters and production data related to a to-be-predicted shale gas well; establishing a tense fusion transformation TFT model; respectively inputting static information and dynamic variables in the target data into the TFT model to obtain a production dynamic prediction result; and analyzing productivity main control factors according to the production dynamic prediction result. According to the embodiment of the invention, under the condition that the shale gas well lacks historical data in the early stage, the accurate prediction model of the production dynamics such as the early yield and the pressure is established by fusing the production time sequence data and the geological engineering static information. The method is suitable for the initial stage of shale gas well commissioning, and important reference can be provided for new well productivity prediction, allocation production optimization, development system adjustment and the like.

Description

Technical Field

[0001] The present application relates to the field of geological exploration technology, and specifically to a method, device, equipment and storage medium for dynamic prediction of early production of shale gas. Background Art

[0002] In the process of shale gas reservoir development, accurate prediction of gas well production dynamics is of great guiding significance for production measures adjustment and rational resource allocation. Unlike conventional oil and gas reservoirs, shale gas reservoir development is affected by the complex flow system of underground fracture network, multi-stage hydraulic fracturing and other process flows, and different production allocation strategies. The data of shale gas well production and pressure fluctuate violently and the decline law is complex, which often makes prediction very difficult. Especially in the early stage of development and production, it is more difficult to effectively predict the production dynamics of gas wells under the condition of limited production history data.

[0003] Common production performance prediction methods include analytical methods, numerical simulations, and decline curve analysis. Among them, the analytical method establishes a seepage model based on the flow mechanism, and predicts shale gas production performance by establishing a chart or theoretical model. It is suitable for the initial stage of production when there is a lack of production data. However, this method is limited by a series of ideal assumptions and simplifications, and it is difficult to cope with the complex situations in actual production. Numerical simulation can achieve an accurate description of the seepage process by establishing a fine gas reservoir model to more comprehensively characterize the reservoir transformation characteristics, occurrence and migration mechanism, fluid-solid coupling effect, etc., so as to effectively predict production performance. However, this method is computationally expensive and places very high demands on the quality and integrity of geological data. The simulation results often have large uncertainties. The decline curve analysis method predicts production performance by establishing a decline trend model suitable for a single well based on a large amount of historical production data, targeting the characteristics of shale gas with high initial production, fast decline in the later period, and long gas production cycle. This method is easy to implement and can describe the changing trend of the long-term production process. It is suitable for areas where production wells have entered the decline period. However, this method does not link production dynamic changes with static geological engineering information, so the physical mechanism behind the method is weak and there is great uncertainty in predicting early production dynamics due to the lack of sufficient historical data support.

[0004] In recent years, production dynamics prediction methods based on machine learning and deep learning are gradually being widely used. This type of data-driven method can characterize complex gas well production dynamics due to the strong nonlinear fitting ability of the model, and performs well in mid- and late-stage predictions. Although the method has the potential to integrate dynamic data and static information for modeling, most current applications fail to effectively utilize geological engineering information when predicting shale gas production dynamics, so the method is not effective in the early stages of gas well production.

[0005] It should be pointed out that the information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or suggestion in any form that the information constitutes prior art already known to those skilled in the art. Summary of the invention

[0006] In view of this, the present application provides a shale gas early production dynamic prediction method, device, equipment and storage medium, so as to solve the problem of poor prediction effect in the early stage of gas well production in the prior art.

[0007] In a first aspect, an embodiment of the present application provides a method for predicting the dynamics of early shale gas production, comprising:

[0008] Collecting target data and preprocessing the target data, wherein the target data includes geological parameters, engineering parameters and production data related to the shale gas well to be predicted;

[0009] Establishing a temporal fusion transformation TFT model, wherein the TFT model is used to characterize the mapping relationship between the geological parameters, the engineering parameters and the production data;

[0010] Inputting the static information and dynamic variables in the target data into the TFT model respectively to obtain production dynamic prediction results;

[0011] The main controlling factors of production capacity are analyzed based on the production dynamics forecast results.

[0012] In a possible implementation, the geological parameters include: static parameters characterizing the geological characteristics or geological conditions of the shale gas reservoir, such as the depth of the gas layer, formation pressure, porosity and gas content;

[0013] The engineering parameters include: static parameters characterizing shale gas reservoir development engineering or fracturing design, such as gas test section length, number of sections, number of clusters, sand addition intensity, fluid intensity, etc.;

[0014] The production data include: dynamic parameters characterizing shale gas production, such as well opening days, oil pressure, casing pressure, daily gas production, daily liquid production, etc.

[0015] In a possible implementation manner, when the target data is the geological parameter and the engineering parameter, the preprocessing of the target data includes:

[0016] Eliminate missing data in the geological parameters and the engineering parameters;

[0017] For each type of data in the geological parameters and engineering parameters, outlier data points are removed based on data distribution and outlier analysis algorithms;

[0018] The correlation coefficient between any two parameters among the geological parameters and engineering parameters is calculated, and repeated parameters with strong correlation are eliminated.

[0019] In a possible implementation manner, when the target data is the production data, preprocessing the target data includes:

[0020] Eliminate or fill in missing and abnormal data in the production data.

[0021] In a possible implementation, the TFT model includes:

[0022] A variable selection network VSN, which is composed of a gated residual network GRN and is used to remove irrelevant features and unnecessary nonlinear transformations;

[0023] A dual-input structure is used to embed static discrete data and then superimpose it with continuous features. After passing through the static covariate encoder SCE, four different vectors are generated and flow in the following three directions: after superimposing with dynamic features, they enter the static enrichment layer, enter the LSTM encoder-decoder, and directly enter the VSN.

[0024] The time series processing module is used to use the Seq2Seq layer based on the long short-term memory LSTM encoder-decoder, input historical unknown data into the LSTM encoder, and input future known data into the LSTM decoder. The output unified time series features are used as the input of the next layer of decoder; the multi-head attention mechanism in the Transformer algorithm is modified and improved so that the multi-head shares the weight matrix W V , and sum and average the long results;

[0025] Output a loss function, where the loss function is the average of all quantile losses of the predicted values ​​for each time period;

[0026] The hyperparameter search module is used to reduce the training and testing errors of the model through hyperparameter selection, prevent underfitting and overfitting, and ultimately determine the optimal model framework.

[0027] In one possible implementation, the deep learning hyperparameters of the hyperparameter search module include a learning rate, a number of iterations, a number of hidden layers, a number of neurons, a number of attention heads, and / or a regularization multiplier.

[0028] In a possible implementation, analyzing the main control factors of production capacity according to the production dynamic prediction result includes:

[0029] Based on the prediction results and accuracy of each input feature in different ranges, analyze the feature distribution when the prediction effect is good or bad, so as to determine the space for exploring model improvement and the impact of features on the prediction results;

[0030] Analyze the impact of each static and dynamic feature on production dynamics by ranking the features by importance;

[0031] Analyze the time steps that have a critical impact on prediction based on attention weights;

[0032] The influence of each feature on the predicted value is analyzed through the partial dependence diagram, and a regularity diagram is drawn.

[0033] In a second aspect, an embodiment of the present application provides a shale gas early production dynamic prediction device, comprising:

[0034] A data collection and processing module, used to collect target data and pre-process the target data, wherein the target data includes geological parameters, engineering parameters and production data related to the shale gas well to be predicted;

[0035] A model building module, used to build a temporal fusion transformation TFT model, wherein the TFT model is used to characterize the mapping relationship between the geological parameters, the engineering parameters and the production data;

[0036] A data input module, used to input the static information and dynamic variables in the target data into the TFT model respectively to obtain a production dynamic prediction result;

[0037] The analysis module is used to analyze the main control factors of production capacity according to the production dynamic prediction results.

[0038] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0039] processor;

[0040] Memory;

[0041] And a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the electronic device executes any one of the methods described in the first aspect.

[0042] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described in the first aspect.

[0043] In the absence of historical data in the early stage of shale gas wells, the embodiment of the present application establishes an accurate prediction model for early production dynamics such as production output and pressure by integrating production time series data and geological engineering static information. The embodiment of the present application is applicable to the initial stage of shale gas well production, and can provide important reference for new well capacity prediction, production optimization, and development system adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 A schematic diagram of a process for predicting the dynamics of early shale gas production provided in an embodiment of the present application;

[0046] Figure 2A A schematic diagram of casing pressure prediction for a shale gas well in a certain block 30 days after it is put into production, provided in an embodiment of the present application;

[0047] Figure 2B A schematic diagram of casing pressure prediction for another shale gas well in a certain block 30 days after it is put into production, provided in an embodiment of the present application;

[0048] Figure 3A A schematic diagram of static parameter importance ranking provided in an embodiment of the present application;

[0049] Figure 3B A schematic diagram of historical data importance ranking provided in an embodiment of the present application;

[0050] Figure 3C A schematic diagram of ranking the importance of future known data provided in an embodiment of the present application;

[0051] Figure 4 A partial dependency graph provided for an embodiment of the present application;

[0052] Figure 5 The structural block diagram of a shale gas early production dynamic prediction device provided in this application is

[0053] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0055] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0056] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0057] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0058] In view of the problem of poor prediction effect in the early stage of gas well production in the prior art, the embodiment of the present application provides a production dynamic prediction method suitable for the early stage of shale gas well production. By integrating multi-well production time series data and geological engineering static information, a deep learning model with comprehensive multi-dimensional data body information is established, so as to efficiently and accurately predict the production dynamic change trend of shale gas well early production, pressure, etc., and provide important reference for new well capacity prediction, production optimization, development system adjustment, etc. The specific implementation method is described in detail below.

[0059] See also Figure 1 , is a schematic diagram of a process flow of a shale gas early production dynamic prediction method provided in an embodiment of the present application. Figure 1 As shown, the process mainly includes the following steps.

[0060] Step S101: collecting target data and preprocessing the target data, wherein the target data includes geological parameters, engineering parameters and production data related to the shale gas well to be predicted.

[0061] Exemplarily, geological parameters include: static parameters characterizing the geological characteristics or geological conditions of shale gas reservoirs, such as medium depth of gas layer, formation pressure, porosity and gas content; engineering parameters include: static parameters characterizing shale gas reservoir development engineering or fracturing design, such as test gas section length, number of sections, number of clusters, sand addition intensity, liquid use intensity; production data include: dynamic parameters characterizing shale gas production, such as well opening days, oil pressure, casing pressure, daily gas production, daily liquid production, etc.

[0062] In order to meet the requirements of the model, after obtaining the target data, the target data needs to be preprocessed accordingly. Among them, the preprocessing method of geological parameters and engineering parameters is: eliminate missing data in geological parameters and engineering parameters; for each type of data in geological parameters and engineering parameters, eliminate outlier data points based on data distribution and outlier analysis algorithms; calculate the correlation coefficient between any two parameters in geological parameters and engineering parameters, and eliminate repeated parameters with strong correlation. The preprocessing method of production data is: eliminate or fill in missing and abnormal data in production data.

[0063] Step S102: Establish a temporal fusion transformation TFT model, where the TFT model is used to characterize the mapping relationship between geological parameters, engineering parameters and production data.

[0064] In the embodiment of the present application, the TFT model specifically includes the following modules or units, which are introduced in detail below.

[0065] Variable Selection Network VSN:

[0066] VSN consists of a gated residual network (GRN) to remove irrelevant features and unnecessary nonlinear transformations. The expression of GRN is shown in Formula 1.

[0067] Formula 1:

[0068] GRN ω(a，c) =LayerNorm(a+GLU ω (η1)),

[0069] η1=W 1，ω η2+b 1，ω

[0070] η2=ELU(W 2，ω a+W 3，ω c+b 2，ω )

[0071] Among them, a is the original input, W and b are the learned model parameters, c is the context vector, LayerNorm is layer normalization, and ELU is the elu activation function.

[0072] GRN is composed of gated linear units (GLUs), which are similar to the gating mechanism of LSTM and are used to select features that are helpful for the training process. The expression of GLU is shown in Formula 2.

[0073] Formula 2:

[0074] GLU ω (η1)=σ(W 4，ω η1+b 4，ω )⊙(W 5，ω η1+b 5，ω )

[0075] Among them, W and b are the learned model parameters, σ is the sigmoid activation function, and ⊙ represents the multiplication of matrix elements.

[0076] VSN uses GRN for individual features and concatenates all features, generates feature weights using the Softmax function, and finally generates a weighted sum of the outputs of a single GRN and selects the relevant features at each time step.

[0077] Dual input structure:

[0078] After static discrete data is embedded, it is superimposed with continuous features and passes through the static covariate encoder SCE to generate four different vectors, which flow in the following three directions: (1) after being superimposed with dynamic features, it enters the static enrichment layer; (2) enters the LSTM encoder-decoder; (3) directly enters the VSN.

[0079] Timing processing module:

[0080] To learn the short-term correlation of time series data, a Seq2Seq layer based on long short-term memory LSTM encoder-decoder is used. Historical unknown data is input into the LSTM encoder, and future known data is input into the LSTM decoder. The output unified time series features are used as the input of the next layer of decoder. To learn long-term correlation, the multi-head attention mechanism in the Transformer algorithm is modified and improved so that the multi-heads share the weight matrix W. V , and sum and average the multi-head results to improve the interpretability of the algorithm while calculating long-term correlation;

[0081] Output loss function:

[0082] The average of all quantile losses of the predicted values ​​for each time period is expressed as shown in Formula 3.

[0083] Formula 3:

[0084]

[0085] where y t ∈Ω is the M samples of the training or test set, y pred is the predicted value, q∈(2%, 10%, 25%, 50%, 75%, 90%, 98%) is the quantile prediction, which can be used for uncertainty analysis, τ m is the total duration of the forecast.

[0086] Hyperparameter search module:

[0087] Through the selection of hyperparameters, the training and testing errors of the model are reduced, underfitting and overfitting are prevented, and the optimal model framework is finally determined. Among them, deep learning hyperparameters include learning rate, number of iterations, number of hidden layers, number of neurons, number of attention heads and / or regularization multipliers, etc.

[0088] Step S103: input the static information and dynamic variables in the target data into the TFT model respectively to obtain the production dynamic prediction results.

[0089] Static information input: Static information does not change over time (such as the geological parameters and engineering parameters mentioned above), and can be divided into two categories: continuous variables and discrete variables. Among them, continuous variables include geological and engineering parameters that characterize the characteristics of a single well, such as the depth of the gas layer, porosity, and gas test length. Discrete variables include the block, well type, etc. Some discrete variables need to be preprocessed by one-hot encoding according to data attributes.

[0090] Dynamic variable input: Dynamic variables (such as the production data mentioned above) are divided into two categories: known variables and unknown variables. Among them, known variables include data that can be determined in advance during the prediction stage, such as the number of days to open the well, the daily production under the fixed production system, and the bottom hole pressure under the fixed pressure system. Unknown variables represent the production dynamic data that need to be predicted, including the casing pressure and oil pressure under the fixed production system, and the daily production under the open production system.

[0091] Step S104: Analyze the main control factors of production capacity according to the production dynamics forecast results.

[0092] In a possible implementation, analyzing the main control factors of production capacity according to the production dynamic forecast results may include one or more of the following aspects:

[0093] (1) Based on the prediction results and accuracy of each input feature in different ranges, analyze the feature distribution when the prediction effect is good or bad, so as to determine the space for exploring model improvement and the impact of features on the prediction results;

[0094] (2) Analyze the impact of each static and dynamic feature on production dynamics by ranking the features by importance;

[0095] (3) Analyze the time steps that have a critical impact on prediction based on attention weights;

[0096] (4) Analyze the impact of each feature on the predicted value through partial dependence diagram and draw a regularity diagram.

[0097] The embodiment of the present application can establish an accurate prediction model for early production dynamics such as output and pressure by integrating production time series data and geological engineering static information when there is a lack of historical data in the early stage of shale gas wells. The embodiment of the present application is suitable for the initial stage of shale gas well production, and can provide important reference for new well capacity prediction, production optimization, and development system adjustment.

[0098] The embodiment of the present application is based on advanced multi-dimensional data fusion and deep learning modeling technology. Compared with the traditional production dynamic prediction method, the temporal fusion transformation model established by the present invention can effectively integrate dynamic and static information, significantly reduce the uncertainty of early production dynamic prediction, and have higher prediction accuracy. In addition, the embodiment of the present application automatically selects the main control parameters in each time period to establish a model, and combines the attention mechanism to learn the correlation between long-term time series data. The model has higher transparency and stronger interpretability, which can provide more theoretical guidance for production practice.

[0099] For ease of understanding, a shale gas block is taken as an example to illustrate the shale gas early production dynamic prediction method provided in the embodiment of the present application.

[0100] Specifically, taking a shale gas block as an example, based on the exploration, development and production data of the block, a daily gas production forecast is carried out during the initial production of the gas well through casing production.

[0101] According to the implementation process of the embodiment of this application, first collect the static information of each shale gas production well in the block, including geological parameters such as gas layer depth and TOC, engineering parameters such as gas test length, liquid intensity, sand addition intensity, and production data such as open flow, technically recoverable reserves, and cumulative production. Eliminate missing values ​​and outliers, and remove duplicate data after correlation analysis, and finally sort out the data of 70 shale gas wells, each of which contains 10 static parameters, some of which are shown in Table 1.

[0102] Table 1:

[0103]

[0104] The production time series data of gas wells are collected. Each gas well includes seven kinds of dynamic data, including the number of days of opening, daily oil pressure, casing pressure, daily gas production, daily liquid production, cumulative gas production, and cumulative liquid production. Some of the data are shown in Table 2.

[0105] Table 2:

[0106]

[0107] In the initial stage of production, the prediction target was selected as daily gas production. By studying the production history of the past 30 days and combining the main controlling geological and engineering parameters, as well as the oil and casing pressure data under the fixed pressure production system, the daily gas production changes in the next 30 days were predicted.

[0108] In this TFT model, static continuous variables include the number of single-stage clusters, fluid intensity, sand addition intensity, casing pressure and other parameters that have a greater impact on production capacity. Dynamic known variables are the number of days a single well is put into production and the oil pressure casing pressure under the constant pressure production system. Dynamic unknown variables are the daily gas production of a single well. Through model hyperparameter search and model training, the prediction accuracy of the test set is as follows: Figure 2A and 2B As shown, Figure 2A and 2B They respectively represent the casing pressure prediction of two shale gas wells in a certain block 30 days after they were put into production. The results show that the invention can effectively integrate dynamic and static data of different dimensions, accurately capture the overall law of dynamic changes in production in the early stage of production, and provide important basis for medium- and long-term capacity prediction, production optimization, and development system adjustment.

[0109] In terms of model interpretability, Figures 3A-3C The importance ranking of input features is shown, including static features, historical dynamic features, and future known features that have the greatest impact on daily gas production. The results of the main control factor analysis show that the single well number as a category feature has the greatest impact on daily gas production, followed by the representation of historical daily gas production processed by the encoder. The casing pressure, sand addition intensity, and fluid intensity have a certain impact on daily gas production, and the single-stage cluster number has a small impact.

[0110] Figure 4 The results of analyzing the effect of casing pressure on daily gas production based on the partial dependence diagram (PDP) are presented, which shows that the daily gas production gradually increases with the decrease of casing pressure, and also includes the uncertainty range.

[0111] Corresponding to the above-mentioned embodiments, the present application also provides a shale gas early production dynamic prediction device.

[0112] See also Figure 5 , which is a structural block diagram of a shale gas early production dynamic prediction device provided in this application. Figure 5 As shown, it mainly includes the following modules.

[0113] The data collection and processing module 501 is used to collect target data and pre-process the target data, wherein the target data includes geological parameters, engineering parameters and production data related to the shale gas well to be predicted;

[0114] A model building module 502 is used to build a temporal fusion transformation TFT model, wherein the TFT model is used to characterize the mapping relationship between the geological parameters, the engineering parameters and the production data;

[0115] A data input module 503, used to input the static information and dynamic variables in the target data into the TFT model to obtain a production dynamic prediction result;

[0116] The analysis module 504 is used to analyze the main control factors of production capacity according to the production dynamic prediction results.

[0117] It should be pointed out that the specific contents involved in the embodiments of the present application can be found in the description of the above method embodiments, and for the sake of brevity, they will not be repeated here.

[0118] Corresponding to the above embodiment, an embodiment of the present application further provides an electronic device.

[0119] See also Figure 6 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device 600 may include: a processor 601, a memory 602 and a communication unit 603. These components communicate via one or more buses. Those skilled in the art will appreciate that the electronic device structure shown in the figure does not limit the embodiments of the present application, and it may be a bus structure or a star structure, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0120] The communication unit 603 is used to establish a communication channel so that the electronic device can communicate with other devices.

[0121] The processor 601 is the control center of the electronic device. It uses various interfaces and lines to connect various parts of the entire electronic device. It runs or executes software programs and / or modules stored in the memory 602, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 601 can include only a central processing unit (CPU). In the embodiment of the present application, the CPU can be a single computing core or multiple computing cores.

[0122] Memory 602 is used to store execution instructions of processor 601. Memory 602 can be implemented by any type of volatile or non-volatile storage device or a combination of them, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0123] When the execution instructions in the memory 602 are executed by the processor 601 , the electronic device 600 is enabled to execute part or all of the steps in the above method embodiment.

[0124] Corresponding to the above embodiment, the embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium may store a program, wherein when the program is running, the device where the computer-readable storage medium is located may be controlled to perform some or all of the steps in the above method embodiment. In a specific implementation, the computer-readable storage medium may be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0125] Corresponding to the above-mentioned embodiment, the embodiment of the present application further provides a computer program product, which includes executable instructions. When the executable instructions are executed on a computer, the computer executes some or all of the steps in the above-mentioned method embodiment.

[0126] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0127] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] 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.

[0129] In several embodiments provided in the present application, any function can be stored in a computer-readable storage medium if it is implemented in the form of a software functional unit and sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a 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 application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., and other media that can store program codes.

[0130] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting the dynamics of early shale gas production, characterized in that: include: Collecting target data and preprocessing the target data, wherein the target data includes geological parameters, engineering parameters and production data related to the shale gas well to be predicted; Establishing a temporal fusion transformation TFT model, wherein the TFT model is used to characterize the mapping relationship between the geological parameters, the engineering parameters and the production data; Inputting the static information and dynamic variables in the target data into the TFT model respectively to obtain production dynamic prediction results; The main controlling factors of production capacity are analyzed based on the production dynamics forecast results.

2. The method according to claim 1, characterized in that: The geological parameters include: static parameters characterizing the geological characteristics or geological conditions of shale gas reservoirs; The engineering parameters include: static parameters characterizing shale gas reservoir development engineering or fracturing design; The production data include: dynamic parameters characterizing shale gas production.

3. The method according to claim 1, characterized in that When the target data is the geological parameter and the engineering parameter, the preprocessing of the target data includes: Eliminate missing data in the geological parameters and the engineering parameters; For each type of data in the geological parameters and engineering parameters, outlier data points are removed based on data distribution and outlier analysis algorithms; The correlation coefficient between any two parameters among the geological parameters and engineering parameters is calculated, and repeated parameters with strong correlation are eliminated.

4. The method according to claim 1, characterized in that When the target data is the production data, the preprocessing of the target data includes: Eliminate or fill in missing and abnormal data in the production data.

5. The method according to claim 1, characterized in that The TFT model includes: A variable selection network VSN, which is composed of a gated residual network GRN and is used to remove irrelevant features and unnecessary nonlinear transformations; A dual-input structure is used to embed static discrete data and then superimpose it with continuous features. After passing through the static covariate encoder SCE, four different vectors are generated and flow in the following three directions: after superimposing with dynamic features, they enter the static enrichment layer, enter the LSTM encoder-decoder, and directly enter the VSN. The time series processing module is used to use the Seq2Seq layer based on the long short-term memory LSTM encoder-decoder, input historical unknown data into the LSTM encoder, and input future known data into the LSTM decoder. The output unified time series features are used as the input of the next layer of decoder; the multi-head attention mechanism in the Transformer algorithm is modified and improved so that the multi-head shares the weight matrix W V , and sum and average the long results; Output a loss function, where the loss function is the average of all quantile losses of the predicted values ​​for each time period; The hyperparameter search module is used to reduce the training and testing errors of the model through hyperparameter selection, prevent underfitting and overfitting, and ultimately determine the optimal model framework.

6. The method according to claim 5, characterized in that The deep learning hyperparameters of the hyperparameter search module include learning rate, number of iterations, number of hidden layers, number of neurons, number of attention heads and / or regularization multipliers.

7. The method according to claim 1, characterized in that The analyzing the main control factors of production capacity according to the production dynamic prediction results includes: Based on the prediction results and accuracy of each input feature in different ranges, analyze the feature distribution when the prediction effect is good or bad, so as to determine the space for exploring model improvement and the impact of features on the prediction results; Analyze the impact of each static and dynamic feature on production dynamics by ranking the features by importance; Analyze the time steps that have a critical impact on prediction based on attention weights; The influence of each feature on the predicted value is analyzed through the partial dependence diagram, and a regularity diagram is drawn.

8. A shale gas early production dynamic prediction device, characterized in that: include: A data collection and processing module, used to collect target data and pre-process the target data, wherein the target data includes geological parameters, engineering parameters and production data related to the shale gas well to be predicted; A model building module, used to build a temporal fusion transformation TFT model, wherein the TFT model is used to characterize the mapping relationship between the geological parameters, the engineering parameters and the production data; A data input module, used to input the static information and dynamic variables in the target data into the TFT model respectively to obtain a production dynamic prediction result; The analysis module is used to analyze the main control factors of production capacity according to the production dynamic prediction results.

9. An electronic device, characterized in that: include: processor; Memory; And a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

Images