A shale gas horizontal well ground stress prediction method based on deep learning

Abstract

This invention relates to the field of geostress prediction technology, and discloses a deep learning-based method for predicting geostress in shale gas horizontal wells. The method includes the following steps: identifying a designated horizontal well in a target area, and acquiring geostress distribution data and elemental logging data from the designated horizontal well; processing the data based on the geostress distribution data and elemental logging data using uniform depth calibration sampling according to well depth to determine the elemental combination data corresponding to the geostress distribution data; inputting the elemental combination data corresponding to the geostress distribution data into a BP neural network to construct a comprehensive geostress learning model; and predicting the geostress distribution data of the target horizontal well using the comprehensive geostress learning model. This invention solves the problem of low accuracy in predicting geostress without logging data.

Description

Technical Field

[0001] This invention relates to the field of geostress prediction technology, and in particular to a deep learning-based method for predicting geostress in shale gas horizontal wells. Background Technology

[0002] Currently, shale gas production is primarily achieved through horizontal well technology, and large-scale hydraulic fracturing is a key technology driving significant increases in the production of these horizontal wells. In-situ stress is a crucial basis for shale gas fracturing and clustering; therefore, calculating in-situ stress is essential when designing fracturing schemes for shale gas horizontal wells.

[0003] Hydraulic fracturing is currently the only method for measuring deep crustal stress during in-situ stress detection. However, it is relatively expensive, and the lack of data optimization further increases costs, leading to a significant increase in the time and effort required for testing. To save costs, some shale gas platforms currently lack logging data for their horizontal well sections. Specifically, only a handful of horizontal wells on some platforms have logging data, while most lack it. Furthermore, for safety design reasons, even in wells with logging data, logging operations cannot be performed on the third horizontal section of some of these wells.

[0004] In general, while hydraulic fracturing has played a crucial role in geostress detection, it still has significant limitations. Firstly, it consumes a large amount of local water resources, and the fracturing fluid may contain chemical additives harmful to groundwater and soil. Secondly, microseismic events induced during fracturing may impact local residents and infrastructure. Thirdly, properly treating the wastewater generated after fracturing (containing salts, heavy metals, and organic matter) remains a technical and economic challenge. Fourthly, chemical additives and gases that may be released during fracturing (such as methane) may pose risks to human health.

[0005] In this context, accurately predicting the geostress of shale gas horizontal wells without logging data becomes a crucial and urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to solve the problem of low accuracy in predicting geostress in shale gas horizontal wells without well logging data, and to provide a deep learning-based method for predicting geostress in shale gas horizontal wells.

[0007] To achieve the above-mentioned objectives, the embodiments of the present invention provide the following technical solutions:

[0008] A deep learning-based method for predicting in-situ stress in shale gas horizontal wells includes the following steps:

[0009] Step 1: Identify the designated horizontal wells in the target area and obtain the geostress distribution data and elemental logging data of the designated horizontal wells;

[0010] Step 2: Based on the geostress distribution data and element logging data, perform data processing by uniformly calibrating and sampling according to well depth to determine the element combination data corresponding to the geostress distribution data;

[0011] Step 3: Input the element combination data corresponding to the geostress distribution data into the BP neural network to construct a geostress comprehensive learning model;

[0012] Step 4: Predict the geostress distribution data of the target horizontal well using the geostress integrated learning model.

[0013] Furthermore, step 1 specifically includes the following steps:

[0014] Step 1-1: Determine the target area for this study, identify the designated horizontal wells within the target area, perform testing and calibration on the designated horizontal wells, record the logging data corresponding to the designated horizontal wells, and further extract the geostress distribution data from the logging data.

[0015] Steps 1-2 involve performing multi-point elemental logging for the specified horizontal well, analyzing the chemical elemental composition of the rocks in the specified horizontal well to determine the lithology of the rocks in the specified horizontal well, and recording multi-point detection data for different lithologies; based on the multi-point detection data, selecting the corresponding elemental logging data from the original elemental logging data of the specified horizontal well.

[0016] Furthermore, step 2 specifically includes the following steps:

[0017] Step 2-1: Acquire depth verification signals using a high-precision white light length and speed measuring sensor;

[0018] Step 2-2: Acquire pulse depth signals using the Martin-Dick depth measurement system, and convert the pulse depth signals according to the depth correction coefficient of the logging truck to generate the logging truck depth signal;

[0019] Steps 2-3 involve inputting the depth verification signal and the logging truck depth signal into the logging depth verification system, calculating the average error at multiple time points, and then generating the correction value of the logging truck's depth correction coefficient. Based on the correction value of the depth correction coefficient, the depth correction coefficients of the element logging data and the geostress distribution data are further corrected, thereby determining the element combination data corresponding to the geostress distribution data.

[0020] Furthermore, step 3 specifically includes the following steps:

[0021] Step 3-1: Input the element combination data corresponding to the ground stress distribution data into the BP neural network. The BP neural network includes an input layer, a hidden layer and an output layer. The input of the neuron nodes in the input layer is the element combination data, and the output of the neuron nodes in the output layer is the ground stress distribution data.

[0022] Step 3-2: Initialize the particle population and the number of iterations;

[0023] Step 3-3: Calculate the fitness of each particle in the particle population at the current iteration number;

[0024] Steps 3-4: Update the positions of particles in the particle population based on the particle's fitness.

[0025] Step 3-5: Determine if the current iteration count has reached the maximum iteration count. If yes, take the updated position of the particle population as the global optimal position and execute step 3-6. If no, increment the current iteration count by 1 and return to step 3-3.

[0026] Steps 3-6: Obtain the optimal weights and thresholds of the BP neural network based on the global optimal location, generate the trained BP neural network, and obtain the geostress integrated learning model.

[0027] Further, in step 3-3, the expression for calculating the fitness γi of the i-th particle in the particle population at the current iteration number is as follows:

[0028]

[0029] Where αkl represents the weight of the k-th neuron node in the input layer and the l-th neuron node in the hidden layer; N represents the number of neurons in the hidden layer; βl represents the weight of the l-th neuron node in the hidden layer and the neuron node in the output layer; and δ represents the vibration vector calculated in the search space of the input data.

[0030] Furthermore, in steps 3-4, the step of updating the positions of particles in the particle population specifically includes:

[0031] Step 3-4-1: Calculate the updated velocity of the i-th particle at the current iteration number t.

[0032]

[0033] Where ωt represents the inertia weight at the current iteration number t; This represents the velocity of the i-th particle before it is updated at the current iteration number t; This represents the individual extreme value of the i-th particle at the current iteration number t; represents the extreme value of the particle population at the current iteration number t; c represents the particle's attractiveness; rand represents a random number;

[0034] Step 3-4-2: Calculate the candidate position of the i-th particle at the current iteration number t.

[0035]

[0036] in, This indicates the position of the i-th particle before the update at the current iteration number t;

[0037] Step 3-4-3: Calculate the updated position of the i-th particle at the current iteration number t.

[0038]

[0039] Where γi-1 represents the fitness of the i-th particle in the previous iteration number t-1.

[0040] Furthermore, in step 3-4-1, the expression for the inertia weight ωt at the current iteration number t is specifically as follows:

[0041] ωt=rand·(ωmax-ωmin)+ωmin;

[0042] Where ωmax represents the maximum value of the inertial weight; ωmin represents the minimum value of the inertial weight.

[0043] Furthermore, step 4 specifically includes the following steps:

[0044] Step 4-1: Obtain element logging data of the target horizontal well, select element types that match the geological characteristics of the current target area, and obtain element combination data of the target horizontal well.

[0045] Step 4-2: Input the element combination data of the target horizontal well into the geostress integrated learning model to obtain the geostress distribution data of the target horizontal well.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] (1) This invention uses the same sampling interval for the corresponding horizontal wells through multi-point detection, and performs multiple sampling tests at the same well depth in different horizontal wells, thereby enhancing the accuracy of the detection of geostress distribution data and element logging data.

[0048] (2) This invention uses data processing, which involves uniformly calibrating and sampling element logging data and geostress distribution data according to well depth, to determine the element combination data corresponding to the geostress distribution data, thereby further enhancing the accuracy of the prediction results and saving time and effort for the staff.

[0049] (3) The invention uses particle swarm optimization to optimize the weights and thresholds of the BP neural network and establishes a comprehensive learning model for geostress. During the model training phase, the speed of the particles is continuously optimized based on their fitness and attraction to better scan and search for particle update positions. This will not increase the computational load of the fitness value too much, help the algorithm escape local optima, improve the optimization ability of the BP neural network training and improve its convergence efficiency, and ensure that the search position is globally optimal. Subsequently, it can accurately predict the geostress of the target horizontal well, meet the geostress prediction requirements of shale gas horizontal wells with incomplete logging data or horizontal wells without logging data, and its prediction results have high reliability.

[0050] (4) The present invention constructs a geostress comprehensive learning model based on particle swarm optimization BP neural network. After high-precision training, it can be used to optimize the drilling path with decision-makers, guide the wellbore trajectory design, and provide the optimal solution of the current design parameters to reduce the uncertainty and risk in the drilling process. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0053] Figure 2 This is a graph showing the geostress distribution data predicted by the geostress integrated learning model in an embodiment of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0055] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance, or suggesting any such actual relationship or order between these entities or operations. Additionally, the terms "connected," "linked," etc., can refer to a direct connection between elements or an indirect connection via other elements.

[0056] Example 1:

[0057] This invention is achieved through the following technical solutions, such as... Figure 1 As shown, a deep learning-based method for predicting in-situ stress in shale gas horizontal wells includes the following steps:

[0058] Step 1: Determine the specified horizontal well in the target area and obtain the geostress distribution data and elemental logging data of the specified horizontal well.

[0059] First, the target area for this study was determined, and then specific horizontal wells, including shale gas horizontal wells, were identified within this area. Next, the specific horizontal wells were calibrated and their corresponding logging data were recorded. Finally, geostress distribution data was extracted from the logging data.

[0060] Multi-point elemental logging is performed on a designated horizontal well to analyze the chemical elemental composition of the rocks, thereby determining the lithology of the rocks and recording multi-point detection data for different lithologies. Based on the multi-point detection data, corresponding elemental logging data are selected from the original elemental logging data of the designated horizontal well.

[0061] Step 2: Based on the geostress distribution data and element logging data, perform data processing by uniformly calibrating and sampling according to well depth to determine the element combination data corresponding to the geostress distribution data.

[0062] Step 2 specifically includes the following steps:

[0063] Step 2-1: Acquire depth verification signals using a high-precision white light length and speed measuring sensor.

[0064] Step 2-2: Acquire pulse depth signals using the Martin-Dick depth measurement system, and convert the pulse depth signals according to the depth correction coefficient of the logging truck to generate the logging truck depth signal.

[0065] Steps 2-3 involve inputting the depth verification signal and the logging truck depth signal into the logging depth verification system, calculating the average error at multiple time points, and then generating the correction value of the logging truck's depth correction coefficient. Based on the correction value of the depth correction coefficient, the depth correction coefficients of the element logging data and the geostress distribution data are further corrected, thereby determining the element combination data corresponding to the geostress distribution data.

[0066] Step 3: Input the element combination data corresponding to the geostress distribution data into the BP neural network to construct a geostress comprehensive learning model.

[0067] Step 3 specifically includes the following steps:

[0068] Step 3-1: Input the element combination data corresponding to the ground stress distribution data into the BP neural network. The BP neural network includes an input layer, a hidden layer, and an output layer. The input of the neuron nodes in the input layer is the element combination data, and the output of the neuron nodes in the output layer is the ground stress distribution data.

[0069] Step 3-2: Initialize the particle population and the number of iterations.

[0070] Step 3-3: Calculate the fitness of each particle in the particle population at the current iteration number.

[0071] In step 3-3, the expression for calculating the fitness γi of the i-th particle in the particle population at the current iteration number is as follows:

[0072]

[0073] Where αkl represents the weight of the k-th neuron node in the input layer and the l-th neuron node in the hidden layer; N represents the number of neurons in the hidden layer; βl represents the weight of the l-th neuron node in the hidden layer and the neuron node in the output layer; and δ represents the vibration vector calculated in the search space of the input data.

[0074] Steps 3-4: Update the positions of particles in the particle population based on the particle's fitness.

[0075] In steps 3-4, the step of updating the position of particles in the particle population specifically includes:

[0076] Step 3-4-1: Calculate the updated velocity of the i-th particle at the current iteration number t.

[0077]

[0078] Where ωt represents the inertia weight at the current iteration number t; This represents the velocity of the i-th particle before it is updated at the current iteration number t; This represents the individual extreme value of the i-th particle at the current iteration number t; represents the extreme value of the particle population at the current iteration number t; c represents the particle's attractiveness; rand represents a random number.

[0079] The expression for the inertia weight ωt at the current iteration number t is as follows:

[0080] ωt=rand·(ωmax-ωmin)+ωmin;

[0081] Where ωmax represents the maximum value of the inertial weight; ωmin represents the minimum value of the inertial weight.

[0082] Step 3-4-2: Calculate the candidate position of the i-th particle at the current iteration number t.

[0083]

[0084] in, This indicates the position of the i-th particle before the update at the current iteration number t.

[0085] Step 3-4-3: Calculate the updated position of the i-th particle at the current iteration number t.

[0086]

[0087] Where γi-1 represents the fitness of the i-th particle in the previous iteration number t-1.

[0088] Step 3-5: Determine if the current iteration count has reached the maximum iteration count. If yes, take the updated position of the particle population as the global optimal position and execute step 3-6. If no, increment the current iteration count by 1 and return to step 3-3.

[0089] Steps 3-6: Obtain the optimal weights and thresholds of the BP neural network based on the global optimal location, generate the trained BP neural network, and obtain the geostress integrated learning model.

[0090] Step 4: Predict the geostress distribution data of the target horizontal well using the geostress integrated learning model.

[0091] Elemental logging data of the target horizontal well is acquired, and element types that match the geological characteristics of the current target area are selected to obtain elemental combination data of the target horizontal well. This elemental combination data is then input into the geostress comprehensive learning model to obtain the geostress distribution data of the target horizontal well.

[0092] In this embodiment, several sets of real geostress data were recorded at multiple points in the horizontal well section of the corresponding study area. After analyzing the detected geostress distribution data and elemental logging data, 333 sets of effective stress distribution data corresponding to elemental combinations were selected and imported into a BP neural network. The BP neural network was used to learn the relationships between the elemental combinations corresponding to the geostress distribution data to construct a corresponding comprehensive geostress learning model. The comprehensive geostress learning model constructed in this invention can accurately predict the geostress distribution data of the target horizontal well with a low overall error rate. It can meet the requirements for predicting the geostress distribution data of shale gas horizontal wells with incomplete logging data or horizontal wells without logging data, and its prediction results have high reliability. Figure 2 The figure shown is a graph of the geostress distribution data predicted by the geostress integrated learning model constructed in this invention. The horizontal axis represents the well depth in the horizontal section, and the vertical axis represents the geostress. The blue curve represents the actual geostress distribution data, and the orange curve represents the predicted geostress distribution data. It can be seen that the prediction results highly overlap with the actual data.

[0093] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A deep learning-based method for predicting in-situ stress in shale gas horizontal wells, characterized in that: Includes the following steps: Step 1: Identify the designated horizontal wells in the target area and obtain the geostress distribution data and elemental logging data of the designated horizontal wells; Step 2: Based on the geostress distribution data and element logging data, perform data processing by uniformly calibrating and sampling according to well depth to determine the element combination data corresponding to the geostress distribution data; Step 3: Input the element combination data corresponding to the geostress distribution data into the BP neural network to construct a geostress comprehensive learning model; Step 4: Predict the geostress distribution data of the target horizontal well using the geostress integrated learning model.

2. The method for predicting in-situ stress in shale gas horizontal wells based on deep learning according to claim 1, characterized in that: Step 1 specifically includes the following steps: Step 1-1: Determine the target area for this study, identify the designated horizontal wells within the target area, perform testing and calibration on the designated horizontal wells, record the logging data corresponding to the designated horizontal wells, and further extract the geostress distribution data from the logging data. Steps 1-2 involve performing multi-point elemental logging for the specified horizontal well, analyzing the chemical elemental composition of the rocks in the specified horizontal well to determine the lithology of the rocks in the specified horizontal well, and recording multi-point detection data for different lithologies; based on the multi-point detection data, selecting the corresponding elemental logging data from the original elemental logging data of the specified horizontal well.

3. The deep learning-based method for predicting in-situ stress in shale gas horizontal wells according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2-1: Acquire depth verification signals using a high-precision white light length and speed measuring sensor; Step 2-2: Acquire pulse depth signals using the Martin-Dick depth measurement system, and convert the pulse depth signals according to the depth correction coefficient of the logging truck to generate the logging truck depth signal; Steps 2-3 involve inputting the depth verification signal and the logging truck depth signal into the logging depth verification system, calculating the average error at multiple time points, and then generating the correction value of the logging truck's depth correction coefficient. Based on the correction value of the depth correction coefficient, the depth correction coefficients of the element logging data and the geostress distribution data are further corrected, thereby determining the element combination data corresponding to the geostress distribution data.

4. The deep learning-based method for predicting in-situ stress in shale gas horizontal wells according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 3-1: Input the element combination data corresponding to the ground stress distribution data into the BP neural network. The BP neural network includes an input layer, a hidden layer and an output layer. The input of the neuron nodes in the input layer is the element combination data, and the output of the neuron nodes in the output layer is the ground stress distribution data. Step 3-2: Initialize the particle population and the number of iterations; Step 3-3: Calculate the fitness of each particle in the particle population at the current iteration number; Steps 3-4: Update the positions of particles in the particle population based on the particle's fitness. Step 3-5: Determine if the current iteration count has reached the maximum iteration count. If yes, take the updated position of the particle population as the global optimal position and execute step 3-6. If no, increment the current iteration count by 1 and return to step 3-3. Steps 3-6: Obtain the optimal weights and thresholds of the BP neural network based on the global optimal location, generate the trained BP neural network, and obtain the geostress integrated learning model.

5. The deep learning-based method for predicting in-situ stress in shale gas horizontal wells according to claim 4, characterized in that: In step 3-3, the expression for calculating the fitness γi of the i-th particle in the particle population at the current iteration number is as follows: Where αkl represents the weight of the k-th neuron node in the input layer and the l-th neuron node in the hidden layer; N represents the number of neurons in the hidden layer; βl represents the weight of the l-th neuron node in the hidden layer and the neuron node in the output layer; and δ represents the vibration vector calculated in the search space of the input data.

6. The deep learning-based method for predicting in-situ stress in shale gas horizontal wells according to claim 5, characterized in that: In steps 3-4, the step of updating the position of particles in the particle population specifically includes: Step 3-4-1: Calculate the updated velocity of the i-th particle at the current iteration number t. Where ωt represents the inertia weight at the current iteration number t; This represents the velocity of the i-th particle before it is updated at the current iteration number t; This represents the individual extreme value of the i-th particle at the current iteration number t; represents the extreme value of the particle population at the current iteration number t; c represents the particle's attractiveness; rand represents a random number; Step 3-4-2: Calculate the candidate position of the i-th particle at the current iteration number t. in, This indicates the position of the i-th particle before the update at the current iteration number t; Step 3-4-3: Calculate the updated position of the i-th particle at the current iteration number t. Where γi-1 represents the fitness of the i-th particle in the previous iteration number t-1.

7. The deep learning-based method for predicting in-situ stress in shale gas horizontal wells according to claim 6, characterized in that: In step 3-4-1, the expression for the inertia weight ωt at the current iteration number t is as follows: ωt=rand·(ωmax-ωmin)+ωmin; Where ωmax represents the maximum value of the inertial weight; ωmin represents the minimum value of the inertial weight.

8. The deep learning-based method for predicting in-situ stress in shale gas horizontal wells according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 4-1: Obtain element logging data of the target horizontal well, select element types that match the geological characteristics of the current target area, and obtain element combination data of the target horizontal well. Step 4-2: Input the element combination data of the target horizontal well into the geostress integrated learning model to obtain the geostress distribution data of the target horizontal well.

Images