Multi-factor shale gas flowback rate calculation method considering geological engineering flowback

Abstract

The invention discloses a multi-factor shale gas flow-back rate calculation method considering geological engineering flow-back, which comprises the following steps: S1, on the basis of on-site data acquisition, taking an acquired on-site fractured well as a data set for calculating a flow-back rate value after shale gas pressure; s2, data preprocessing is performed on the collected data samples, and geological engineering flowback variables required for establishing a flowback rate mathematical model are screened out; s3, on the basis of the geological engineering flowback variable data processing result in the step S2, establishing a variable comprehensive calculation model, and sequencing variable importance; variable reconstruction is carried out through independence judgment, principal component analysis is carried out on reconstructed variables, and processed independent principal components are adopted as input variables of a prediction model; s4, based on gene expression programming, establishing a shale gas flowback rate prediction model under the influence of multiple factors of geological engineering flowback; and S5, fitting a shale gas flowback rate calculation formula based on shale gas well geological engineering flowback data by applying the established shale gas flowback rate prediction model.

Description

Technical Field

[0001] The present invention belongs to the technical field of unconventional oil and gas production enhancement and transformation, and in particular relates to a multi-factor shale gas flowback rate calculation method considering geological engineering flowback. Background Art

[0002] Multi-stage, multi-cluster fracturing in horizontal wells increases shale gas production by injecting large amounts of fracturing fluid to create a complex fracture network in the reservoir. However, the large amount of fracturing fluid injected into the reservoir results in an extremely low flowback rate for shale fracturing. This inefficient flowback results in a large amount of fracturing fluid remaining in the reservoir, hindering gas flow and damaging the reservoir, thus limiting the effectiveness of shale reservoir production.

[0003] In recent years, numerous researchers have conducted extensive research on the issue of post-fracture fluid flowback in shale formations, using both traditional and machine learning methods, and the factors influencing it. The prevailing belief is that unreturned fracturing fluid is trapped in complex branching fractures, natural fractures, and the matrix, resulting in a significant portion of the fluid remaining in the formation and unable to be discharged. This inability is directly due to the forces of gravity, capillary forces, and surface tension within the reservoir, which influence the flowback rate. In addition to mechanical influences, other key factors influencing the flowback rate include reservoir permeability and rock wettability, well shut-in time, natural fracture strength, matrix-related physical and chemical parameters, and other reservoir properties, as well as geotechnical parameters such as fracturing stage length, flow rate, TOC content, and porosity.

[0004] As can be seen from the above research, there are many factors that influence the flowback rate of shale gas wells. To address this multi-factor influencing issue, machine learning has been introduced into petroleum industry research to process and analyze data and establish relationships between complex nonlinear data sets. Artificial neural networks (ANN), support vector machines (SVM), and relevance vector machines (RVM) are among the most widely used. The rational integration of multiple methods can outperform a single algorithm, and these methods can predict results through trained models. However, currently, no model comprehensively considers all factors influencing flowback. Developing a comprehensive fracturing fluid flowback calculation model that incorporates the geological and engineering factors affecting fracturing fluid flowback remains a challenge.

[0005] In view of this, it is necessary to propose a multi-factor shale gas flowback rate calculation method that comprehensively considers geological engineering flowback. Summary of the Invention

[0006] The purpose of the present invention is to overcome the problems of the prior art and disclose a multi-factor shale gas flowback rate calculation method taking into account geological engineering flowback. By considering the parameters that best fit the flowback rate results, a new equation is established to calculate the mathematical model of the flowback rate and the main controlling factors affecting the shale gas flowback rate are studied, thereby achieving accurate calculation and prediction of the flowback rate of shale gas wells after pressure reduction.

[0007] The object of the present invention is achieved through the following technical solutions:

[0008] A multi-factor shale gas flowback rate calculation method considering geological engineering flowback, the multi-factor shale gas flowback rate calculation method comprising the following steps:

[0009] S1: Based on the data obtained on site, the obtained on-site fracturing wells are used as a data set to calculate the shale gas backflow rate value after pressure adjustment. Each well is regarded as a data sample. The input of each sample includes 31 parameter variables, and the output variable is the backflow rate value.

[0010] S2: Preprocess the collected data samples to select the geological engineering flowback variables required to establish the flowback rate mathematical model;

[0011] S3: Based on the geological engineering flowback variable data processing results in step S2, a variable comprehensive calculation model is established and the importance of the variables is ranked;

[0012] Variables are reconstructed through independence judgment, principal component analysis is performed on the reconstructed variables, and the independent principal components after processing are used as input variables of the prediction model;

[0013] S4: Based on gene expression programming, a shale gas flowback rate prediction model was established under the influence of multiple factors of geological engineering flowback;

[0014] S5: Apply the established shale gas flowback rate prediction model and fit the shale gas flowback rate calculation formula based on the geological engineering flowback data of shale gas wells.

[0015] According to a preferred embodiment, the input parameter variables of each sample in step S1 include: liquid, proppant, engineering parameters and geological parameters.

[0016] According to a preferred embodiment, the data preprocessing of the collected data samples in step S2 includes: data normalization, divergence judgment, correlation judgment, and variable importance judgment;

[0017] The data normalization process includes: centering the data according to the minimum value, and then scaling it according to the range. The data is moved by the minimum value units and will converge to [0, 1].

[0018] The divergence judgment includes: using the dispersion coefficient to judge the degree of dispersion, where the dispersion coefficient is defined as the ratio of the standard deviation of the characteristic value to the mean; for a designed variable, if there is no divergence in this variable, then the corresponding variable has no effect on the training of the model;

[0019] The correlation determination includes: for the correlation between the variable and the flowback rate, using the Pearson correlation coefficient PCC to complete the correlation determination between the variable and the flowback rate.

[0020] The variable importance judgment includes: selecting a CatBoost model to perform importance judgment, and after completing the training of the CatBoost model on the data set, directly outputting the model's relative importance score for each variable.

[0021] According to a preferred embodiment, in step S3,

[0022] The establishment of a variable comprehensive index calculation model includes:

[0023] The correlation between variables and return rate is defined as a weighted evaluation of the three evaluation indicators, with the sum of the weights being 1. By integrating the results of different evaluation methods, a comprehensive return rate-related score Score is given to the feature. V The score is assigned a weight w3 of 0.1, and the remaining weight is evenly distributed to PCC and CatBoost, that is:

[0024] Score=w1·S PCC +w2·S CatBoost +w3·S CV .

[0025] According to a preferred embodiment, in step S3, the variable reconstruction process includes:

[0026] The Pearson correlation coefficient (PCC) was used to determine the correlation between the 31 input variables. Based on the comprehensive score of the features, the important factors affecting flowback were first selected, and then classified according to independence, and the reconstructed features were screened for GEP model training and prediction.

[0027] According to a preferred embodiment, step S3 also includes: performing principal component analysis on the reconstructed variables, eliminating the weak independence between the reconstructed variables, recombining the input reconstructed features into a new set of mutually unrelated variables, and retaining the information carried by the original variables according to actual needs, and using the processed independent principal components as input variables of the prediction model.

[0028] According to a preferred embodiment, step S4 includes:

[0029] ①Choice of basic operation symbols:

[0030] Select basic operators including +, -, *, / , exp, Inv, Min, and Max to implement the gene expression programming model;

[0031] ②Fitness function construction:

[0032] If the error selected is the absolute error, then the fitness f of a single program i i As shown in the following formula,

[0033]

[0034] If the error selected is a relative error, then the fitness f of a single program i i As shown in the following formula,

[0035]

[0036] Among them, M is the selection range, C (i,j) It is C t The individual chromosome i outside the fitness situation is the value returned by the fitness situation j, T j is the target value for fitness case j; for full fitness, C (i,j) =T j , f i =f max =C t M;

[0037] ③Result data evaluation:

[0038] In order to evaluate the performance of the trained GEP calculation model, two statistical benchmark indices are introduced, namely the mean square error RMSE and the coefficient of determination R 2 , used to describe the difference between the model calculated value and the actual value on site,

[0039]

[0040]

[0041] Among them, Y i and y i represent the actual and predicted flowback rate values, respectively; and represent the average values ​​of actual flowback rate and predicted flowback rate respectively; n is the number of data samples.

[0042] According to a preferred embodiment, step S5 includes:

[0043] ① Initialize the population, randomly generate a certain number of individuals as the initial population, and encode the individuals as linear symbol strings;

[0044] ② Calculate the fitness value of each individual in the population. If a satisfactory fitness value is obtained or other termination conditions are met, save the result and exit; otherwise, go to the next step;

[0045] ③ Save the optimal individual;

[0046] ④Selection is based on the size of the fitness value. The probability of being selected is high for the one with a large fitness value, and the probability of being selected is low for the one with a small fitness value.

[0047] ⑤ For the selected individuals, directly copy them to the next generation;

[0048] ⑥ Perform genetic operations on the selected individuals according to the preset probability, including mutation, recombination, and transposition, to form new individuals;

[0049] ⑦ A new population is formed from individuals that have undergone selection, replication, and other genetic operations, and the process returns to step ②. Finally, the calculation formula for the shale gas well flowback rate is obtained through iteration.

[0050] The aforementioned main solution of the present invention and its various further alternatives may be freely combined to form multiple solutions, all of which are solutions that can be adopted and protected by the present invention. After understanding the solutions of the present invention, those skilled in the art will understand, based on existing technology and common knowledge, that there are many possible combinations, all of which are technical solutions to be protected by the present invention, and these are not exhaustive here.

[0051] Beneficial effects of the present invention:

[0052] The method of the present invention is used to achieve accurate calculation and precise prediction of the post-fracturing flowback rate of shale gas wells. The method collects on-site geological engineering flowback data of fracturing wells, pre-processes the data samples, establishes a variable comprehensive index calculation model, reconstructs variables through independence judgment, and uses the processed independent principal components as input variables of the prediction model; based on gene expression programming, a shale gas flowback rate prediction model under the influence of multiple factors of geological engineering flowback is established, and the established shale gas flowback rate prediction model is applied to fit the shale gas flowback rate calculation formula based on the geological engineering flowback data of shale gas wells. The calculation results of the mathematical model are consistent with the actual data collected on-site. The present invention is conducive to calculating the flowback rate of shale gas fracturing fluid, has guiding significance for guiding the adjustment of shale gas well production systems, and plays an important role in improving the post-fracturing effect of shale gas wells. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1(a) is a schematic diagram of the independence among the 31 variables;

[0054] Figure 1(b) is a schematic diagram of the independence among the 18 reconstructed variables;

[0055] Figure 2 It is the comprehensive score map of 18 reconstructed features;

[0056] Figure 3 This is the independence relationship diagram between the 18 variables after PCA processing;

[0057] Figure 4(a) is the frequency distribution of the return rate in the training set;

[0058] Figure 4(b) is the frequency distribution of the flowback rate in the test set;

[0059] Figure 5(a) is a sequence diagram of the calculation results of the GEP model;

[0060] Figure 5(b) is a sequence diagram of the field data;

[0061] Figure 6(a) is a comparison chart of the target and calculated values ​​of the return flow rate in the training set;

[0062] Figure 6(b) is a comparison chart of the target and calculated values ​​of the flowback rate in the test set. DETAILED DESCRIPTION

[0063] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0064] The present invention discloses a multi-factor shale gas flowback rate calculation method considering geological engineering flowback, and the multi-factor shale gas flowback rate calculation method comprises the following steps.

[0065] Step S1: Based on the data obtained on site, the obtained on-site fracturing wells are used as a data set to calculate the shale gas backflow rate value after pressure treatment. Each well is used as a data sample. The input of each sample includes 31 parameter variables (liquid, proppant, engineering parameters and geological parameters), and the output variable is the backflow rate value. Therefore, there are a total of 31 input variables and one output variable.

[0066] Step S2: pre-process the collected data samples to screen out the geological engineering flowback variables required for establishing a flowback rate mathematical model.

[0067] The data preprocessing of the collected data samples in step S2 includes: data normalization, divergence judgment, correlation judgment, and variable importance judgment.

[0068] ① Data normalization

[0069] When the data is centered at the minimum value and then scaled by the range (maximum value - minimum value), the data moves by the minimum value units and will converge to [0,1].

[0070]

[0071] In the formula: min(x) is the minimum value of data x; max(x) is the maximum value of data x; x* is the normalized value of data x.

[0072] ②Divergent judgment

[0073] For a designed variable, if there is no divergence in this variable, for example, the variance of the variable is close to 0, it can usually be considered that the sample data does not show obvious differences in this variable, and this variable has no effect on the training of the model. Based on this, the coefficient of variation (CV) is used to judge the degree of dispersion. The coefficient of variation is defined as the ratio of the standard deviation of the eigenvalue to the mean, and the expression is:

[0074]

[0075] ③Relevance judgment

[0076] The Pearson correlation coefficient (PCC) is used to determine the correlation between variables and flowback rate. PCC is a classic statistic used to reflect the degree of linear correlation between two variables. The correlation coefficient is expressed as ρ. X,Y It represents the covariance cov(X,Y) of two variables X and Y divided by their standard deviation σ X and σ Y , the result is between -1 and 1, and the larger the absolute value of ρ, the stronger the correlation.

[0077]

[0078] ④ Variable importance judgment

[0079] Select the CatBoost model for importance judgment. After completing the training of the CatBoost model on the data set, directly output the model's relative importance score for each variable.

[0080] Step S3: Based on the data processing results of the geological engineering flowback variables in step S2, a comprehensive variable calculation model is established and the variables are ranked by importance. Variables are reconstructed through independence judgment and principal component analysis is performed on the reconstructed variables. The processed, independent principal components are used as input variables for the prediction model.

[0081] ① Variable comprehensive index calculation model

[0082] The correlation between the variable and the return rate is defined as a weighted evaluation of the three evaluation indicators (the sum of the weights is 1). By integrating the results of different evaluation methods, a comprehensive return rate correlation score is given to the feature. V The score is assigned a weight of 0.1 (w3), and the remaining weight is evenly distributed between PCC and CatBoost.

[0083] Score=w1·S PCC +w2·S CatBoost +w3·S CV (4)

[0084] ②Independence judgment

[0085] Not only are there correlations between variables and flowback rates, but there are also correlations within variables, that is, between input variables. The correlations between the 31 input variables were determined using the Pearson correlation coefficient (PCC), grouping those with poor independence into one category. This selection process was done to identify key factors influencing the flowback rate while ensuring independence between features, thereby improving model accuracy. Based on the comprehensive feature scores, the key factors influencing flowback were first selected, then categorized according to their independence, and reconstructed features were selected for GEP model training and prediction.

[0086] Principal component analysis is performed on the reconstructed variables to eliminate any weak independence between them. The input reconstructed features are recombined into a new set of independent variables while retaining the information carried by the original variables as needed. The processed independent principal components are used as the input variables of the prediction model.

[0087] Step S4: Based on gene expression programming, a shale gas flowback rate prediction model under the influence of multiple factors of geological engineering flowback is established.

[0088] ①Choice of basic operation symbols

[0089] Select basic operators such as +, -, *, / , exp, Inv, Min, Max, etc. to implement the gene expression programming model.

[0090] ②Fitness function

[0091] If the error chosen is the absolute error, then the fitness f of a single program i i As shown in formula (5), if the error selected is a relative error, then as shown in formula (6):

[0092]

[0093]

[0094] Among them, M is the selection range, C (i,j) The individual chromosome i is the fitness case j (C t The value returned (except for fitness) is T j is the target value for fitness case j. Note that for full fitness, C (i,j) =T j , f i =f max =Ct M. Note that with this fitness function, the system can find the optimal solution for itself.

[0095] ③Result data evaluation

[0096] In order to evaluate the performance of the trained GEP computational model, two statistical benchmark indices were introduced, namely the mean square error (RMSE) and the coefficient of determination (R 2 ), which is used to describe the difference between the model calculation value and the actual value on site, as shown in formulas (7) and (8). The smaller the RMSE, the better the model calculation performance. 2 The closer the value is to 1, the better the computational performance of the model.

[0097]

[0098]

[0099] Among them, Y i and y i represent the actual and predicted flowback rate values, respectively; and represent the average values ​​of actual flowback rate and predicted flowback rate respectively; n is the number of data samples.

[0100] Step S5: Apply the established shale gas flowback rate prediction model and fit the shale gas flowback rate calculation formula based on the shale gas well geological engineering flowback data.

[0101] Step S5 includes: ① Initializing the population by randomly generating a certain number of individuals as the initial population, with the individuals encoded as linear symbol strings; ② Calculating the fitness of each individual in the population. If a satisfactory fitness value is obtained or other termination conditions are met, the result is saved and the process exits; otherwise, the process proceeds to the next step; ③ Saving the optimal individual; ④ Selecting individuals based on their fitness value: individuals with higher fitness values ​​have a higher probability of being selected, while individuals with lower fitness values ​​have a lower probability of being selected; ⑤ Directly replicating the selected individuals to the next generation; ⑥ Performing genetic operations on the selected individuals, such as mutation, recombination, and transposition, according to a certain probability, to form new individuals; ⑦ Forming a new population from the individuals that have undergone selection, replication, and other genetic operations, and returning to step ②. Finally, through iteration, the formula for calculating the flowback rate of shale gas wells is obtained.

[0102] Example 1:

[0103] In this example, 189 field-fracturing wells were used as a dataset for calculating the flowback rate after shale gas pressure treatment, using field data. Each well served as a data sample, and the inputs for each sample included 31 variables (fluid, proppant, engineering parameters, and geological parameters), with the output variable being the flowback rate. Therefore, there were a total of 31 input variables and one output variable (flowback rate). Data preprocessing included sample normalization, divergence assessment, correlation assessment, and feature importance assessment.

[0104] (1) Data normalization

[0105] After the data is centered according to the minimum value and then scaled by the range (maximum value - minimum value), the data moves by the minimum value units and will converge to [0,1], and is calculated according to formula (1).

[0106] (2) Divergent judgment

[0107] For a designed variable, if there is no divergence on this variable, for example, if the variance of the variable is close to 0, it can usually be considered that the sample data does not show obvious differences on this variable, and this variable has no effect on model training. Based on this, the coefficient of variation (CV) is used to judge the degree of dispersion. The coefficient of variation is defined as the ratio of the standard deviation of the eigenvalue to the mean and is calculated according to formula (2).

[0108] After calculation, the results and ranking of the 31 characteristic dispersion coefficients designed in this embodiment are shown in Table 1. Generally speaking, the dispersion coefficient C V The larger the value, the more dispersed the data is, and the more effective the prediction of fracturing fluid flowback rate is. As can be seen from Table 1, the dispersion coefficient C of the five characteristics of total injected acid, total injected linear glue, total 100 mesh silt, single stage 100 mesh silt, and total number of fracturing clusters is V The data on this feature are relatively scattered and have strong divergence, while the dispersion coefficient C of the single-stage injected liquid volume and the proportion of slick water is V The data value on this feature is more concentrated than other features, and the divergence is relatively weak.

[0109] (3) Relevance judgment

[0110] The Pearson correlation coefficient (PCC) is used to determine the correlation between the 31 variables and the flowback rate. PCC is a classic statistic used to reflect the degree of linear correlation between two variables. The correlation coefficient is expressed as ρ. X,Y It represents the covariance cov(X,Y) of two variables X and Y divided by their standard deviation σ X and σ Y , the result is between -1 and 1. The larger the absolute value of ρ, the stronger the correlation. Calculation is performed according to formula (3).

[0111] The Pearson correlation coefficients between the characteristics and the calculated flowback rate are shown in Table 1. The results show that the correlation between the quartz sand ratio, flow rate, average number of clusters per stage, and vertical depth is relatively high, indicating that these four characteristics are strongly correlated with the flowback rate.

[0112] (4) Variable importance judgment

[0113] Table 1 shows the random forest calculation of the importance of variables to the calculated flowback rate. The results show that quartz sand is the most important. Other highly correlated features include displacement, pump-off pressure, brittleness index, and vertical depth. Less important features include sand intensity, number of fracturing stages, total number of fracturing clusters, and cluster length.

[0114] (5) Variable comprehensive index calculation model

[0115] The correlation between the variable and the return rate is defined as a weighted evaluation of the scores of the three evaluation indicators (the sum of the weights is 1). By integrating the results of different evaluation methods, a comprehensive return rate correlation score is given to the feature. V The score reflects the divergence of the feature, that is, the amount of information, but it is calculated within a feature data column, and different features have different coefficients of difference C. V The scores of C are quite different under the evaluation method. V The score is assigned only a weight of 0.1 (w3), and the remaining weight is evenly distributed between PCC and CatBoost. The comprehensive score is calculated according to formula (4) and is shown in Table 1.

[0116] Table 1 Characteristic parameters corresponding calculation results

[0117]

[0118]

[0119] (6) Independence Judgment

[0120] Not only is there a correlation between the variables and the return flow rate, but there is also a correlation within the variables, that is, between the input variables. The correlation between the 31 input variables is judged by the Pearson correlation coefficient PCC. As shown in Figure 1(a), some of the correlation coefficients are greater than 0.5, indicating that there is a strong correlation between the input variables. Strong correlation will cause GEP to have deviations in variable selection during training and prediction, resulting in poor correlation of the model. Therefore, it is necessary to combine feature importance and independence, re-divide the features, and classify those with poor independence into one category. The selection can reflect the important factors affecting the return flow rate and ensure the independence between features, thereby improving the model accuracy, as shown in Figure 1(b). Based on the comprehensive score of the features, the important factors affecting the return flow are first selected, and then classified according to independence. The classification results are shown in Table 2. The reconstructed features X1~X 18 , the factor importance distribution histogram is as follows Figure 2 As shown in the figure, it can be seen that the reconstruction features X1~X6 have higher scores and have a greater impact on the return rate. The other 12 reconstruction features have a relatively smaller impact on the return rate than X1~X6, indicating that different features carry different impact information. Since the amount of information carried by different features cannot be ignored, 18 reconstruction features (X1~X6) are selected. 18 ) for GEP model training and prediction.

[0121] Table 2 Reconstruction feature parameter correspondence table

[0122]

[0123]

[0124] (7) Principal component analysis

[0125] Perform principal component analysis on the reconstructed variables to eliminate the weak independence between the reconstructed variables and recombining the input reconstructed features into a new set of independent variables while retaining the information carried by the original variables according to actual needs. The independent principal components after processing are used as the input variables of the prediction model. The independence between the 18 variables after PCA processing is shown in Figure 2. Figure 3 .

[0126] (8) Equation fitting

[0127] Based on gene expression programming, a shale gas flowback rate prediction model, influenced by multiple geological and engineering flowback factors, was established. This model includes: ① selection of basic operation symbols; ② fitness function; and ③ result data evaluation. The shale gas flowback rate calculation formula was then fitted.

[0128] Currently, there are many studies on the selection of training set and test set sizes in machine learning. Generally, 70% is selected as the training set size. Therefore, for the 189 shale gas wells collected in this paper, 134 wells are selected as training set samples, and the remaining 55 wells are selected as test set samples. The frequency distribution histogram of the flowback rate in the training set and the test set is shown in Figure 2. Figure 4(a) 、 4(b) As shown, it can be seen that the standard deviations of the training set and the test set are basically the same, the return rate is mainly distributed between 0.2 and 0.7, the data distribution pattern is consistent, and it conforms to the normal distribution, which can be used for GEP fitting calculation.

[0129] The accuracy and effectiveness of the GEP model depend on many factors. Whether the model input variables are independent of each other plays an important role in the selection of important variables when fitting the equation. Therefore, in order to build a reliable and accurate model, the 18 reconstructed variables (X1, X2, X3...X 18 ) is used as the input variable of the GEP model to calculate the flowback rate of shale reservoirs after fracturing. To improve the calculation accuracy of the model, this study uses basic operators such as +, -, *, / , exp, Inv, Min, and Max to implement the GEP model. The fitted equation is shown in Equation (9), where y1 to y8 are shown in Equation (10), and the equation coefficients are shown in Table 3.

[0130] LR=y1+y2+y3+y4+y5+y6+y7+y8 (9)

[0131]

[0132] Table 3 Equation coefficient values

[0133] a＝-0.088 b＝2.819 c＝5.189 d＝-1.316 m＝4.800 f＝-1.039 g＝-9.398 h＝-15.328 i＝-1.230 j＝-3.029 k＝-4.700 l＝7.516

[0134] Based on the resulting GEP model, a training and test set of post-pressure flowback data for shale reservoirs was analyzed. To better understand the accuracy and reliability of the resulting model, a comparison of the actual and calculated flowback rates versus the data point sequence was plotted, as shown in Figure 5. This plot also demonstrates the accuracy of the developed model, indicating that the GEP model results correctly account for the trends in the field data. Comparing the target and calculated flowback rate values ​​in the training and test sets clearly demonstrates that the model's generalization ability and the calculation of the trends and behavior of the target values ​​in response to changes in the input parameters are another important consideration when building a model.

[0135] Figures 6(a) and 6(b) compare the return rate values ​​from field data and the GEP model. As can be seen, the GEP model successfully learned the relationship between the nonlinear return rate and its influencing variables, and the data points are mainly distributed near the diagonal, which means that there is appropriate coordination between the calculated data and the target data. The computational performance was evaluated using R and RMSE. In the case of the training set, the statistical parameters obtained by GEP were: R = 0.835, RMSE = 0.073. According to statistical recommendations, a good calculation can be evaluated when R>0.8. Therefore, the computational performance of GEP was very satisfactory in the training set. Similarly, the statistical parameters of the test set were: R = 0.815, RMSE = 0.068, which performed very well on the test set, indicating that the model trained by GEP can be used to calculate the return rate. This shows that the return rate calculation method proposed in the present invention, which comprehensively considers multiple factors of geological engineering return, is reasonable and more applicable.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-factor shale gas flowback rate calculation method considering geological engineering flowback, characterized in that: The multi-factor shale gas flowback rate calculation method comprises the following steps: S1: Based on the data obtained on site, the obtained on-site fracturing wells are used as a data set to calculate the shale gas backflow rate value after pressure adjustment. Each well is regarded as a data sample. The input of each sample includes 31 parameter variables, and the output variable is the backflow rate value. S2: Preprocess the collected data samples to select the geological engineering flowback variables required to establish the flowback rate mathematical model; S3: Based on the geological engineering flowback variable data processing results in step S2, a variable comprehensive calculation model is established and the importance of the variables is ranked; Variables are reconstructed through independence judgment, principal component analysis is performed on the reconstructed variables, and the independent principal components after processing are used as input variables of the prediction model; S4: Based on gene expression programming, a shale gas flowback rate prediction model was established under the influence of multiple factors of geological engineering flowback; S5: Apply the established shale gas flowback rate prediction model and fit the shale gas flowback rate calculation formula based on the geological engineering flowback data of shale gas wells.

2. The multi-factor shale gas flowback rate calculation method according to claim 1, characterized in that: The input parameter variables of each sample in step S1 include: liquid, proppant, engineering parameters and geological parameters.

3. The multi-factor shale gas flowback rate calculation method according to claim 1, characterized in that: In step S2, the collected data samples are preprocessed, including data normalization, divergence judgment, correlation judgment, and variable importance judgment; The data normalization process includes: centering the data according to the minimum value, and then scaling it according to the range. The data is moved by the minimum value units and will converge to [0, 1]. The divergence judgment includes: using the dispersion coefficient to judge the degree of dispersion, where the dispersion coefficient is defined as the ratio of the standard deviation of the characteristic value to the mean; for a designed variable, if there is no divergence in this variable, then the corresponding variable has no effect on the training of the model; The correlation determination includes: for the correlation between the variable and the flowback rate, using the Pearson correlation coefficient PCC to complete the correlation determination between the variable and the flowback rate. The variable importance judgment includes: selecting a CatBoost model to perform importance judgment, and after completing the training of the CatBoost model on the data set, directly outputting the model's relative importance score for each variable.

4. The multi-factor shale gas flowback rate calculation method according to claim 3, characterized in that: In step S3, The establishment of a variable comprehensive index calculation model includes: The correlation between variables and return rate is defined as a weighted evaluation of the three evaluation indicators, with the sum of the weights being 1. By integrating the results of different evaluation methods, a comprehensive return rate-related score Score is given to the feature. V The score is assigned a weight w3 of 0.1, and the remaining weight is evenly distributed to PCC and CatBoost, that is: Score=w1·S PCC +w2·S CatBoost +w3·S CV 。 5. The multi-factor shale gas flowback rate calculation method according to claim 4, characterized in that: In step S3, the variable reconstruction process includes: The Pearson correlation coefficient (PCC) was used to determine the correlation between the 31 input variables. Based on the comprehensive score of the features, the important factors affecting flowback were first selected, and then classified according to independence, and the reconstructed features were screened for GEP model training and prediction.

6. The multi-factor shale gas flowback rate calculation method according to claim 5, characterized in that: Step S3 also includes: performing principal component analysis on the reconstructed variables to eliminate the weak independence between the reconstructed variables, recombining the input reconstructed features into a new set of mutually unrelated variables, while retaining the information carried by the original variables according to actual needs, and using the processed independent principal components as input variables of the prediction model.

7. The multi-factor shale gas flowback rate calculation method according to claim 1, characterized in that: Step S4 includes: ①Choice of basic operation symbols: Select basic operators including +, -, *, / , exp, Inv, Min, and Max to implement the gene expression programming model; ②Fitness function construction: If the error selected is the absolute error, then the fitness f of a single program i i As shown in the following formula, If the error selected is a relative error, then the fitness f of a single program i i As shown in the following formula, Among them, M is the selection range, C (i,j) It is C t The individual chromosome i outside the fitness situation is the value returned by the fitness situation j, T j is the target value for fitness case j; for full fitness, C (i,j) =T j , f i =f max =C t M; ③Result data evaluation: In order to evaluate the performance of the trained GEP calculation model, two statistical benchmark indices are introduced, namely the mean square error RMSE and the coefficient of determination R 2 , used to describe the difference between the model calculated value and the actual value on site, Among them, Y i and y i represent the actual and predicted flowback rate values, respectively; and represent the average values ​​of actual flowback rate and predicted flowback rate respectively; n is the number of data samples.

8. The multi-factor shale gas flowback rate calculation method according to claim 1, characterized in that: Step S5 includes: ① Initialize the population, randomly generate a certain number of individuals as the initial population, and encode the individuals as linear symbol strings; ② Calculate the fitness value of each individual in the population. If a satisfactory fitness value is obtained or other termination conditions are met, save the result and exit; otherwise, go to the next step; ③ Save the optimal individual; ④Selection is based on the size of the fitness value. The probability of being selected is high for the one with a large fitness value, and the probability of being selected is low for the one with a small fitness value. ⑤ For the selected individuals, directly copy them to the next generation; ⑥ Perform genetic operations on the selected individuals according to the preset probability, including mutation, recombination, and transposition, to form new individuals; ⑦ A new population is formed from individuals that have undergone selection, replication, and other genetic operations, and the process returns to step ②. Finally, the calculation formula for the shale gas well flowback rate is obtained through iteration.

Images