Shale gas production prediction method based on fracturing operation parameter clustering
By reducing and clustering the construction fracturing parameters of shale gas wells, combined with the Transformer model, the accuracy and resource consumption balance of shale gas well production prediction in the existing technology are solved, and more efficient shale gas well production prediction is achieved .
Patent Information
- Application Number
- CN202410693140.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing shale gas output prediction methods are difficult to effectively deal with complex fracturing construction parameters diversity and high data volumes, resulting in a difficult balance between accuracy and resource consumption in shale gas well development and production.
By reducing the dimensionality and clustering of the construction fracturing parameters of shale gas wells, they are divided into multiple well groups, and the respective yield prediction models are trained using the Transformer model to predict gas well production.
It improves the accuracy of shale gas well production forecasting, can capture complex nonlinear data relationships more effectively, and reduces the consumption of computing resources.
Smart Images

Figure CN118485182B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of oil and gas development, and in particular to a shale gas production prediction method based on clustering of fracturing construction parameters. Background Art
[0002] As an important unconventional natural gas resource, shale gas is stored in shale layers formed during hundreds of millions of years of geological changes. China has huge potential for shale gas resources, especially in the Sichuan Basin and its neighboring areas. The complexity and diversity of shale gas geological conditions in southern Sichuan directly determine the pertinence and diversity of fracturing optimization technology. Fracturing technology is an important factor in increasing the production of shale gas wells. It can create cracks in low-permeability shales so that gas can flow to the wellhead. However, the construction fracturing parameters of each gas well, such as the number of perforation clusters in the main section, the length of the fracturing section, the intensity of fluid use, and the intensity of sand addition, will be different, which will directly affect the formation of the fracture network and thus affect the production of shale gas. Therefore, predicting shale gas production based on fracturing construction parameters has become an important step in guiding the rational development and production of shale gas wells.
[0003] There are two main traditional shale gas production prediction methods: 1. Decline curve. This method cannot fully consider the diversity and complexity of fracturing construction parameters, and it is difficult to process large amounts of data. 2. Numerical simulation requires complete geological development data, and the simulation of complex geological models usually consumes a lot of computing resources and time. These methods usually cannot achieve ideal results when dealing with shale gas, which has a high degree of heterogeneity and complex address structure. It can be seen that there is an urgent need for a shale gas production prediction method that can achieve a good balance between accuracy and resource consumption to guide the development and production of shale gas wells. Summary of the invention
[0004] In view of this, the purpose of the present invention is to provide a shale gas production prediction method based on fracturing construction parameter clustering, which obtains multiple groups of well groups by clustering the gas well data after dimensionality reduction, and trains the production prediction models corresponding to different well groups to predict the production of shale gas wells.
[0005] In order to solve at least one of the above technical problems, the technical solution provided by the present invention is:
[0006] A shale gas production prediction method based on fracturing construction parameter clustering is provided, which mainly includes the following steps:
[0007] Step S1: collecting construction fracturing parameters, production parameters and daily gas production of gas wells in the area to be predicted, and preprocessing the data;
[0008] Step S2: performing dimension reduction processing on the construction fracturing parameters, and clustering the reduced dimension data to divide the gas wells in the area to be predicted into multiple groups of wells;
[0009] Step S3: Calculate the Pearson correlation coefficient between the daily gas production of each well group and the construction fracturing parameters, and obtain the first few groups of construction fracturing parameters of each well group after sorting the absolute values of the Pearson correlation coefficients from large to small;
[0010] Step S4: using the first few groups of construction fracturing parameters of each well group and the production parameters and gas production data of the gas wells in the well group as variables to input into the Transformer model, and training to obtain the production prediction model corresponding to each well group;
[0011] Step S5: Determine the well group to which the gas well to be predicted belongs, and select the production prediction model corresponding to the well group to perform prediction on the gas well to be predicted.
[0012] The technical effects of the present invention are:
[0013] The shale gas production prediction model based on fracturing construction parameter clustering proposed in the present invention combines complex fracturing engineering data with deep learning technology, which can accurately capture complex nonlinear data relationships and effectively improve the accuracy of shale gas well production prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0015] Figure 1 It is a structural diagram of the Transformer yield prediction model in the present invention;
[0016] Figure 2 This is a schematic diagram of the well group division after the principal component analysis and dimensionality reduction in the present invention;
[0017] Figure 3 This is a schematic diagram of the construction fracturing parameters most relevant to the daily gas production of a well group in the present invention;
[0018] Figure 4 This is a schematic diagram of the construction fracturing parameters most relevant to the 2-day gas production of the well group in the present invention;
[0019] Figure 5 This is a schematic diagram of the construction fracturing parameters most relevant to the 3-day gas production of the well group in the present invention;
[0020] Figure 6 This is a comparison chart of the gas production prediction effects of different models in the present invention;
[0021] Figure 7 This is a comparison chart of the accuracy of the three prediction models of Transformer, LSTM, and GRU in the present invention. DETAILED DESCRIPTION
[0022] The present invention will be further described in detail below in conjunction with the embodiments and drawings.
[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention.
[0024] The shale gas production prediction method based on fracturing construction parameter clustering includes the following steps:
[0025] Step S1: collect construction fracturing parameters, production parameters and daily gas production of gas wells in the area to be predicted, and pre-process the data.
[0026] The selection principles of construction fracturing parameters are the key factors that can affect the fracturing effect during the fracturing process, including geological conditions, engineering design, selection of fracturing fluid and proppant, fracturing effect, etc.
[0027] Production parameters mainly include conventional parameters required for prediction by conventional prediction models, such as casing pressure, production time, daily water production, etc.
[0028] There may be a large number of missing values in the collected construction fracturing parameters, which need to be filled. Among them, the random forest interpolation method can be used for data preprocessing.
[0029] Step S2: Perform dimensionality reduction processing on the construction fracturing parameters, and cluster the reduced-dimensional data to divide the gas wells in the area to be predicted into multiple well groups.
[0030] The collected construction fracturing data often has a high data dimension, and there is a strong correlation between the construction data. In order to reduce the complexity of subsequent calculations, it is first necessary to reduce the dimension of the construction fracturing parameters. In order to retain as much information as possible when reducing the data dimension, the data dimension reduction method used in the present invention is principal component analysis (PCA). The dimension reduction process is to compress the multi-dimensional information into two principal components, wherein the first principal component is the eigenvector corresponding to the maximum eigenvalue of the data covariance matrix, and the second principal component is the eigenvector corresponding to the second largest eigenvalue of the data covariance matrix. The first principal component captures the most information in the original data, and it contains the part of information with the largest change in all construction fracturing parameters. The second principal component captures the most information except the first principal component, and it contains the part of information with the largest change in the construction fracturing parameters except the first principal component, so that the multi-dimensional information in the original data is compressed to the two principal components.
[0031] Afterwards, clustering is performed to provide a grouping basis for subsequent model training. The clustering method used in the present invention is the K-Means algorithm, wherein an increase in the K value will lead to an increase in the number of well groups obtained by division and further increase the complexity of the subsequent calculation model. Considering the complexity of the subsequent model calculation, the K value needs to be selected based on the actual situation on site to balance the complexity of the model and the prediction effect.
[0032] Step S3: Calculate the Pearson correlation coefficient between the daily gas production of each well group and the construction fracturing parameters, and obtain the first few groups of construction fracturing parameters of each well group after sorting the absolute values of the Pearson correlation coefficients from large to small.
[0033] It is known from the technology in the art that in the process of building a model, it is very important to select appropriate input variables. The above-mentioned groups of construction fracturing parameters have different degrees of correlation with the daily gas production of the gas well as the prediction target. Therefore, the present invention adopts the method of calculating the Pearson correlation coefficient to screen out the parameters with the highest correlation with the gas production for building a prediction model. The specific method is: the construction fracturing parameters and daily gas production of all gas wells in a well group are taken as a set, and the Pearson correlation coefficient between the daily gas production set and the construction fracturing parameter set is respectively calculated to obtain multiple groups of Pearson correlation coefficients belonging to the well group, and the absolute values of the obtained multiple groups of Pearson correlation coefficients are sorted, and the construction fracturing parameters corresponding to the absolute values of the top Pearson correlation coefficients from large to small are used as variables for subsequent training of the model. Including more construction fracturing parameters may increase the complexity of the model and lead to overfitting, while including fewer construction fracturing parameters may lose some important information. Therefore, the sorting range of the Pearson correlation coefficient size can be determined based on the correlation between the construction fracturing parameters and the gas production, the performance and complexity of the model, and the comprehensive consideration of the calculation efficiency.
[0034] Step S4: Use the first few groups of construction fracturing parameters of each well group and the production parameters and gas production data of the gas wells in the well group as variables to input the Transformer model, and train to obtain the production prediction model corresponding to each well group.
[0035] The Transformer architecture is a sequence-to-sequence model. The decoder part of the Transformer outputs a high-dimensional feature vector. After multiple layers of processing including the self-attention mechanism, as well as optimization processing of the ReLU activation function and batch normalization layer, the prediction result can be output through the fully connected layer.
[0036] The model training process is as follows: a certain number of gas wells in any well group obtained by division are selected, and the construction fracturing parameters of these gas wells within the range of the absolute value of the Pearson correlation coefficient of the well group, as well as the production parameters and daily gas production of these gas wells are used as training sets and test sets, and input into the Transformer model to train and obtain the production prediction model corresponding to the well group. After that, this step is repeated until the production prediction model corresponding to all well groups is obtained. The structure of the production prediction model is as follows: Figure 1 shown.
[0037] Step S5: Determine the well group to which the gas well to be predicted belongs, and select the production prediction model corresponding to the well group to perform prediction on the gas well to be predicted.
[0038] It can be seen from step S4 that if the production prediction model needs to be used, it needs to be used for the corresponding well group. Therefore, it is necessary to judge the well group described in the gas well to be predicted. The judgment method is: collect the construction fracturing parameters of the gas well to be predicted, and bring them into steps S1 and S2 for dimensionality reduction and clustering, and obtain the well group category to which the gas well to be predicted belongs, and select the production prediction model corresponding to the well group category to which it belongs to predict it.
[0039] Example:
[0040] Taking a shale gas field in southern Sichuan as an example, a total of 66 shale gas wells with a long production cycle from November 2021 to October 2023 were selected to establish a prediction model. First, the construction fracturing parameters were collected, mainly including: wellbore trajectory parameters: oblique depth of point A of horizontal well, oblique depth of point B of horizontal well, vertical depth of point A of horizontal well, vertical depth of point B of horizontal well; fracturing design and implementation parameters: horizontal section length, designed fracturing section length, actual fracturing section length, average segmented section length, designed number of sections, actual number of sections; fracturing construction detail parameters: total number of clusters, number of perforation clusters in the main section, average cluster spacing; fluid dynamics parameters: displacement range, pump pressure extreme value; fracturing fluid composition parameters: total sand-carrying liquid volume, liquid strength, total acid liquid volume, total slick water volume, total linear glue volume, total cross-linking liquid volume; proppant parameters: total sand volume, sand adding intensity, total amount of proppant of different particle sizes, maximum sand concentration, and proportion of quartz sand; fracturing effect evaluation parameters: average pump stop pressure.
[0041] Since the above construction fracturing parameters have a high correlation, principal component analysis is used in this embodiment to reduce their dimension into two principal components, and K-Means clustering algorithm is used to divide them into three well groups, namely well group 1, well group 2, and well group 3. Figure 2 As shown, Figure 2 In the figure, the horizontal axis represents the first principal component, the vertical axis represents the second principal component, and points of different shapes represent different well groups.
[0042] After clustering the well groups, the Pearson correlation coefficients of each well group were calculated using the construction fracturing parameters and gas production of the 66 wells, and the construction fracturing parameters corresponding to the absolute values of the first 8 Pearson correlation coefficients of each well group were selected from large to small as the variables required for training. Figures 3 to 5 shown.
[0043] Then, for each group of wells, two gas wells were selected as the test set, and the remaining gas wells were selected as the training set. The construction fracturing parameters, production parameters, and daily gas production of the test set and training set were input into the Transformer architecture, and the gas production prediction model of each well group was trained separately. The input construction fracturing parameters are shown in Table 1:
[0044] Table 1 Input fracturing parameters
[0045]
[0046] The daily gas production prediction model obtained from each well group is used to predict the daily gas production of the six gas wells in well groups 1, 2, and 3. The evaluation indicators of the prediction results are shown in Table 2:
[0047] Table 2 Test gas well evaluation index table
[0048]
[0049]
[0050] From the data in Table 2, we can see that R 2 The generally high values indicate that the model has a significant fit and reliability in predicting gas well production. At the same time, the lower MAPE (mean absolute percentage error) and SMAPE (symmetric mean absolute percentage error) values indicate that the prediction error is small and the accuracy is high. These results comprehensively reflect that the gas production prediction model in the present invention has excellent performance in predicting the production of different well groups.
[0051] Then, the gas wells H42-4, H50-5, and H3-11 were selected to use different models to predict gas production to compare the results. In addition to the prediction model of the Transformer framework used in the present invention, the conventional LSTM prediction model and GRU prediction model in the field were also used to predict the gas production of the above gas wells. The specific comparison results are as follows: Figure 6 , 7 shown.
[0052] from Figure 6 It can be seen that when the actual gas production is at the highest or lowest point, only the production prediction curve of the Transformer model is closer to the actual gas production curve, which shows that the Transformer model in the present invention can more accurately reflect the actual state of the well. Figure 7 It can be seen that the evaluation indicators of the Transformer model are better than those of the LSTM and GRU models, whether in the H3-11 well, the H42-4 well, or the H50-5 well.
[0053] In the description of the present invention, it should be pointed out that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" etc. indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and should not be understood as a limitation on the present invention.
[0054] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the embodiments of the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A shale gas production prediction method based on fracturing construction parameter clustering, characterized in that: The following steps are involved: Step S1: collect construction fracturing parameters, production parameters and daily gas production of gas wells in the predicted area, and pre-process the data, wherein the construction fracturing parameters include wellbore trajectory parameters: horizontal well A point oblique depth, horizontal well B point oblique depth, horizontal well A point vertical depth, horizontal well B point vertical depth; fracturing design and implementation parameters: horizontal section length, designed fracturing section length, actual fracturing section length, average segmented section length, designed number of sections, actual number of sections; fracturing construction detail parameters: total number of clusters, number of perforation clusters in the main section, average cluster spacing; fluid dynamics parameters: displacement range, pump pressure extreme value; fracturing fluid composition parameters: total sand-carrying liquid volume, liquid strength, total acid liquid volume, total slippery water volume, total linear glue volume, total cross-linking liquid volume; proppant parameters: total sand volume, sand addition strength, total amount of proppant with different particle sizes, maximum sand concentration, quartz sand ratio; fracturing effect evaluation parameters: average pump stop pressure; The production parameters include: casing pressure, production time, and daily water production; Step S2: performing dimensionality reduction processing on the construction fracturing parameters, and clustering the data after dimensionality reduction, and dividing the gas wells in the to-be-predicted area into multiple groups of wells, wherein the dimensionality reduction processing method is principal component analysis, and the dimensionality reduction processing process is: compressing the multi-dimensional information into two principal components by principal component analysis, wherein the first principal component is the eigenvector corresponding to the maximum eigenvalue of the data covariance matrix, and the second principal component is the eigenvector corresponding to the second largest eigenvalue of the data covariance matrix; The clustering method is K-Means algorithm; Step S3: Calculate the Pearson correlation coefficient between the daily gas production of each well group and the construction fracturing parameters, and obtain the first few groups of construction fracturing parameters of each well group after sorting the absolute values of the Pearson correlation coefficients from large to small; Step S4: using the first few groups of construction fracturing parameters of each well group and the production parameters and gas production data of the gas wells in the well group as variables to input into the Transformer model, and training to obtain the production prediction model corresponding to each well group; Step S5: Determine the well group to which the gas well to be predicted belongs, and select the production prediction model corresponding to the well group to perform prediction on the gas well to be predicted.
2. The shale gas production prediction method based on fracturing construction parameter clustering according to claim 1 is characterized in that: The construction fracturing parameters obtained in step S3 and sorted from the largest to the smallest absolute value of the Pearson correlation coefficient of each well group are the first 8 groups of construction fracturing parameters.
