A method for shale gas well production prediction
The shale gas well production prediction method, which combines grey relational analysis and neural networks with the Arps model, solves the problems of computational complexity and insufficient accuracy in traditional methods, and achieves higher accuracy in shale gas well production prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST PETROLEUM UNIV
- Filing Date
- 2023-06-02
- Publication Date
- 2026-07-03
AI Technical Summary
Existing technologies for predicting shale gas well production suffer from computational complexity and difficulty in ensuring accuracy, especially given the numerous influencing factors during fracturing and the fact that traditional methods fail to adequately consider the impact of physical parameters.
The grey relational analysis method is used to identify influencing factors. Combined with the error backpropagation neural network and the long short-term memory neural network, and constrained by the Arps model, a shale gas production prediction model is established. The data mining and physical model are combined through machine learning methods.
It improves the accuracy and reliability of shale gas well production prediction, reduces generalization error, and provides more accurate production prediction results.
Smart Images
Figure CN116658155B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of shale gas reservoir development technology, specifically relating to a method for predicting shale gas well production. Background Technology
[0002] Shale gas, tight sandstone gas, and coalbed methane are unconventional natural gas sources, and my country currently attaches great importance to their development and utilization. Simultaneously, given my country's strong demand for natural gas, shale gas will inevitably become an important supplement to my country's natural gas resources. Shale gas, tight sandstone gas, and coalbed methane are unconventional natural gas sources, and the country currently attaches great importance to the development and utilization of shale gas. Predicting shale gas production after fracturing is an important means of determining its economic viability; however, due to the complexity of the fracturing process and numerous influencing factors, traditional production prediction methods are computationally intensive, labor-intensive, and lack the accuracy to guarantee.
[0003] Currently, there are two main methods for predicting shale gas well production: methods based on engineering experience models and data-based methods. Methods based on physical experience models are relatively mature. Their success lies in deriving a concise formula based on the physical parameters of a specific shale gas well under certain conditions, facilitating direct engineering application. However, predicting the parameters within the model remains challenging and often becomes the biggest obstacle to solving the problem. Data-based methods primarily employ machine learning techniques for conventional oil and gas production prediction, mainly using machine learning algorithms for in-depth data mining. However, research on the impact of physical parameters is still insufficient. Data-based methods have seen considerable success in inferring unknown data from known shale gas reservoir data. However, because they do not consider the well's own physical parameters, the prediction accuracy is significantly affected by the similarity between the predicted well and known wells in various physical properties.
[0004] Current methods for predicting shale gas production after fracturing rarely combine physical models with data mining. To address the main problems of existing methods, this paper designs a method that combines engineering experience models and data mining. This method fully utilizes the predictive power of machine learning for unknowns and introduces classic experience models for control. After training with measured data, it achieves good prediction accuracy and makes more accurate predictions of shale gas well production in China. Summary of the Invention
[0005] The present invention aims to solve the technical problems existing in the prior art and provide a method for predicting the production of shale gas wells.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting shale gas well production, comprising the following steps:
[0007] Step 1: Collect geological data, engineering data, and production data of shale gas wells to generate basic data;
[0008] Step 2: Preprocess the basic data to identify the main controlling factors affecting shale gas well production;
[0009] Step 3: Construct a sample set. Set the main control factor affecting production obtained in Step 2 as Xi. Use Xi as the feature vector of the gas well in the target block. Select the basic data of the target block wells as the sample dataset. Extract a certain proportion of the sample data in the sample dataset as the training set. Use the remaining sample data as the test set.
[0010] Step 4: Through learning and training, randomly generate training and test sets in the sample dataset, combine mathematical methods of backpropagation neural network and long short-term memory neural network, and use Arps model for constraints to form an improved BLA algorithm. Establish a shale gas production prediction model through the improved BLA algorithm.
[0011] Step 5: Using the shale gas production prediction model established in Step 4, predict the production of the fracturing section of the gas well in the target block.
[0012] A further technical solution is that, in step 1, the basic data includes the median vertical depth of the horizontal well, porosity, total organic carbon content, pressure coefficient, horizontal section length, fracturing section length, number of fracturing stimulation sections, average cluster spacing, fracturing fluid volume, proppant dosage, quartz sand volume, average sand ratio, construction discharge rate, average section length, proppant addition intensity, and fluid intensity. Among these, the fracturing fluid is slickwater, and the proppant is 70 / 140 mesh quartz sand and 40 / 70 mesh ceramsite.
[0013] A further technical solution is that, in step 2, the preprocessing of the basic data is carried out by using the grey relational analysis method to sort the basic data according to the degree of influence on gas well production, thereby obtaining the main controlling factors affecting production.
[0014] A further technical solution is that, in step 3, the target attribute of the sample dataset is to predict the production of the fracturing section of the gas well.
[0015] A further technical solution is that, in step 4, the improved BLA algorithm includes the following steps:
[0016] Step 4-1: First, use the backpropagation neural network method to train the experimental data. After the accuracy requirement is met, use an algorithm structure that meets the training error requirement to make new predictions. Select gradient descent as the training function, with a maximum training number of 1000 and a training accuracy requirement of 0.0001.
[0017] Step 4-2: Before transforming the output layer into a result using a Long Short-Term Memory (LSTM) artificial neural network method, the previous result is stored in the current computation process. To achieve this, a storage layer is added to the original output layer. As the final output, the previous result is also stored, with the previous storage element denoted as... The element in this storage layer is denoted as ,
[0018] =
[0019] The output result for this iteration is jointly provided by the current error backpropagation neural network result and the previous final result, that is:
[0020] Pt=( , , ..., )+
[0021] Where the matrix Also given from training data;
[0022] Step 4-3: During the data training phase, the Arps model is used with empirical model constraints. Before providing the final prediction result, based on the previous three results, the empirical model is used for reasonable extrapolation, and this extrapolation is compared with the actual predicted value. If the deviation exceeds the set deviation range, it is unacceptable, and the corresponding coefficients are readjusted. The final prediction result is then accepted as valid. The basic equation of the Arps model is:
[0023]
[0024] In the formula For output at any given time, This is the initial output. For decreasing exponents, This is the initial decrease rate.
[0025] The principle and beneficial effects of this invention:
[0026] 1. Compared with existing technologies, this invention collects geological data, engineering data and production data of shale gas wells, uses the grey relational analysis method to preprocess the data, identifies the main controlling factors affecting production, and establishes a nonlinear mathematical model, which can more accurately predict the production of shale gas wells.
[0027] 2. Compared with existing technologies, the shale gas production prediction model established in this invention deeply integrates the advantages of mathematical methods such as error backpropagation neural networks and long short-term memory neural networks, and uses the Arps model as an engineering experience model for constraint, which effectively reduces generalization error, avoids overfitting, and improves the model prediction accuracy, providing reliable production prediction for the large-scale development of shale gas.
[0028] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0029] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:
[0030] Figure 1 This is a flowchart of the shale gas well production prediction method of the present invention;
[0031] Figure 2 This is a flowchart of the improved BLA algorithm for the shale gas well production prediction method of this invention;
[0032] Figure 3 This is a comparison chart of test results for the improved BLA algorithm of the shale gas well production prediction method of this invention. Detailed Implementation
[0033] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0034] This application provides a method for predicting shale gas well production, characterized by the following steps:
[0035] Step 1: Collect geological data, engineering data, and production data of shale gas wells to generate basic data.
[0036] Step 2: Preprocess the basic data to identify the main controlling factors affecting shale gas well production.
[0037] Step 3: Construct a sample set. Set the main control factor affecting production obtained in Step 2 as Xi, use Xi as the feature vector of the gas well in the target block, select the basic data of the collected wells in the target block as the sample dataset, extract sample data from the sample dataset according to a certain proportion as the training set, and use the remaining sample data as the test set.
[0038] Step 4: Through learning and training, randomly generate training and test sets in the sample dataset, combine mathematical methods of backpropagation (BP) neural network and long short-term memory (LSTM) neural network, and use Arps model for constraints to form an improved BLA algorithm, and establish a shale gas production prediction model through the improved BLA algorithm.
[0039] The improved BLA algorithm includes the following steps:
[0040] Step 4-1: First, use the backpropagation neural network method to train the experimental data. After the accuracy requirement is met, use an algorithm structure that meets the training error requirement to make new predictions. Select gradient descent as the training function, with a maximum training number of 1000 and a training accuracy requirement of 0.0001.
[0041] Step 4-2: Before transforming the output layer into a result using a Long Short-Term Memory (LSTM) artificial neural network method, the previous result is stored in the current computation process. To achieve this, a storage layer is added to the original output layer. As the final output, the previous result is also stored, with the previous storage element denoted as... The element in this storage layer is denoted as ,
[0042] =
[0043] The output result for this iteration is jointly provided by the current error backpropagation neural network result and the previous final result, that is:
[0044] Pt=( , , ..., )+
[0045] Where the matrix Also given from training data;
[0046] Step 4-3: During the data training phase, the Arps model is used with empirical model constraints. Before providing the final prediction result, based on the previous three results, a reasonable extrapolation is performed using the empirical model, and the extrapolation is compared with the actual predicted value. If the deviation exceeds the set deviation range, it is unacceptable, and the corresponding coefficients are readjusted. The final prediction result is then accepted as valid. The basic equation of the Arps model is:
[0047]
[0048] In the formula For output at any given time, This is the initial output. For decreasing exponents, This is the initial decrease rate.
[0049] Step 5: Using the shale gas production prediction model established in Step 4, predict the production of the fracturing section of the gas well in the target block.
[0050] The basic principles of this invention will be further explained below with reference to specific examples:
[0051] Collect geological, engineering, and production data from shale gas wells:
[0052] Geological, engineering, and production data of shale gas wells in a certain block were collected to generate basic data. The factors affecting shale gas production in the basic data include: median vertical depth of horizontal wells (m), porosity (%), total organic carbon content (TOC, %), pressure coefficient, horizontal section length (m), fractured section length (m), number of fractured sections, average cluster spacing (m), fracturing fluid volume (m^3), proppant dosage (m^3), quartz sand usage (t), average sand ratio (%), operational displacement (m^3 / min), average section length (m), proppant intensity (t / m), and fluid intensity (m^3 / section), totaling 16 parameters. The fracturing fluid used was slickwater, and the proppant was 70 / 140 mesh quartz sand and 40 / 70 mesh ceramsite.
[0053] Perform data preprocessing:
[0054] The grey relational analysis method was used to preprocess the factors affecting shale gas production in the basic data. Specifically, the grey relational analysis method was used to rank the above 16 factors according to the strength of their influence on gas well production. The top 8 factors in terms of correlation strength were selected as the main control factors for analysis. The main control factors are shown in Table 1 below.
[0055] Table 1
[0056]
[0057] Construct the sample set:
[0058] Based on the main controlling factors affecting production identified by grey relational analysis, Xi is defined as the feature vector of the gas wells in the target block. The target attribute is the predicted production of the fracturing section of the gas well. The selected well baseline data is used as the sample dataset. A certain amount of sample data is extracted as the training set, and the remaining sample data is used as the test set.
[0059] First, the sample dataset is imported. Then, through training, training and test sets are randomly generated from the sample dataset at a 7:3 ratio. By calculating the correlation coefficient between physical parameters (input layer) and the target well, a BP neural network is constructed. Combined with an LSTM neural network, and incorporating the constraints of the empirical Arps model, an improved BLA algorithm is formed. This algorithm is used to establish a shale gas production prediction model, with the coefficient of determination and root mean square error selected as two evaluation metrics.
[0060] Two evaluation metrics were calculated for the predictions of the ordinary BP neural network method and the improved algorithm. The statistics of the evaluation metrics for different algorithms are shown in Table 2 below.
[0061] Table 2
[0062]
[0063] like Figure 3 As shown, the predicted values and actual test values of the improved BLA algorithm, the classic BP network method, and the actual values were compared (the horizontal axis represents the segment number, and the vertical axis represents the fracturing segment production, in meters). 3 ).
[0064] The results show that the improved algorithm significantly improves prediction performance, meeting accuracy requirements for both overall trend and individual point error values. The improved BLA method combines the advantages of empirical and neural network models, greatly enhancing the accuracy of the results.
[0065] In the description of this specification, references to terms such as "preferred embodiment," "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0066] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims
1. A method for predicting shale gas well production, characterized in that, Includes the following steps: Step 1: Collect geological data, engineering data, and production data of shale gas wells to generate basic data; Step 2: Preprocess the basic data to identify the main controlling factors affecting shale gas well production; Step 3: Construct a sample set. Set the main control factor affecting production obtained in Step 2 as Xi. Use Xi as the feature vector of the gas well in the target block. Select the basic data of the target block wells as the sample dataset. Extract a certain proportion of the sample data in the sample dataset as the training set. Use the remaining sample data as the test set. Step 4: Through learning and training, randomly generate training and test sets in the sample dataset, combine mathematical methods of backpropagation neural network and long short-term memory neural network, and use Arps model for constraints to form an improved BLA algorithm. Establish a shale gas production prediction model through the improved BLA algorithm. The improved BLA algorithm includes the following steps: Step 4-1: First, use the backpropagation neural network method to train the experimental data. After the accuracy requirement is met, use an algorithm structure that meets the training error requirement to make new predictions. Select gradient descent as the training function, with a maximum training number of 1000 and a training accuracy requirement of 0.0001. Step 4-2: Before transforming the output layer into a result using a Long Short-Term Memory (LSTM) artificial neural network method, the previous result is stored in the current computation process. To achieve this, a storage layer is added to the original output layer. As the final output, the previous result is also stored, with the previous storage element denoted as... The element in this storage layer is denoted as , = The output result for this iteration is jointly provided by the current error backpropagation neural network result and the previous final result, that is: Pt=( , ,…, )+ Where the matrix Also given from training data; Step 4-3: During the data training phase, the Arps model is used with empirical model constraints. Before providing the final prediction result, based on the previous three results, the empirical model is used for reasonable extrapolation, and this extrapolation is compared with the actual predicted value. If the deviation exceeds the set deviation range, it is unacceptable, and the corresponding coefficients are readjusted. The final prediction result is then accepted as valid. The basic equation of the Arps model is: In the formula For output at any given time, This is the initial output. For decreasing exponents, This is the initial decrease rate; Step 5: Using the shale gas production prediction model established in Step 4, predict the production of the gas fracturing section in the target block.
2. The method for predicting shale gas well production as described in claim 1, characterized in that, In step 1, the basic data include the median vertical depth of the horizontal well, porosity, total organic carbon content, pressure coefficient, horizontal section length, fractured section length, number of fractured sections, average cluster spacing, fracturing fluid volume, proppant dosage, quartz sand volume, average sand ratio, construction flow rate, average section length, proppant strength, and fluid strength. Among these, the fracturing fluid is slickwater, and the proppant is 70 / 140 mesh quartz sand and 40 / 70 mesh ceramsite.
3. The method for predicting shale gas well production as described in claim 1, characterized in that, In step 2, the specific method for preprocessing the basic data is to use the grey relational analysis method to sort the basic data according to the strength of their influence on gas well production, thereby deriving the main controlling factors affecting production.
4. The method for predicting shale gas well production as described in claim 1, characterized in that, In step 3, the target attribute of the sample dataset is to predict the production of the fracturing section of the gas well.