Construction parameter optimization method and device based on dessert prediction, medium and equipment

Abstract

The invention discloses a construction parameter optimization method and device based on dessert prediction, a medium and equipment, and the method comprises the steps: obtaining a data set of a shale gas well production process, and determining a dessert indication factor; training a network model by taking the data set as input and the dessert indication factor as output to obtain a dessert indication factor prediction model; optimizing the dessert indicator factor prediction model by using a Bayesian hyper-parameter optimization algorithm, and inputting a data set obtained in real time into the optimized dessert indicator factor prediction model to obtain a prediction result; inverting a prediction result into three-dimensional space distribution characteristics of desserts in the shale reservoir; and determining contribution values and partial dependency relationships of the plurality of geological parameters and the plurality of engineering parameters to a prediction result based on the three-dimensional space distribution characteristics, deploying construction parameters according to the contribution values and the partial dependency relationships, and synchronously updating a sweet spot indication factor prediction result to realize spatial dynamic optimization and adjustment of the construction scheme. Through the scheme, the construction parameters can be scientifically optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of unconventional oil and gas exploitation, and in particular to a construction parameter optimization method, device, medium and equipment based on sweet spot prediction. Background Art

[0002] In oil and gas extraction, the complex geological environment and severe construction conditions greatly limit the quality of oil and gas extraction. Accurately predicting sweet spots and optimizing construction parameters are the key to improving the efficiency of oil and gas extraction.

[0003] At present, many experts and scholars have done a lot of work in this regard, but there are still many shortcomings, the main shortcomings of which include: (1) In the process of sweet spot prediction, the existing technology often focuses too much on geological factors or engineering factors, resulting in an incomplete characterization of the development potential of oil and gas reservoirs. For example, the patent application document with the authorization number CN111461386B and the name of the shale gas sweet spot prediction method based on BP neural network records the use of geological characteristic parameters through the BP neural network model to establish a mapping relationship corresponding to the shale gas content, obtain the optimal weight of each geological characteristic parameter, and obtain the sweet spot distribution prediction model of the shale reservoir by quantitatively superimposing each parameter plane map. (2) When designing fracturing construction parameters, only the one-way effect from the sweet spot to the construction plan is considered, ignoring the reaction of the construction work to the sweet spot evaluation and the linkage effect of the construction work on the selection of sweet spots for multiple wells and multiple platforms.

[0004] Therefore, the existing construction parameter optimization methods in the shale oil and gas production process cannot accurately optimize the construction parameters of multiple wells and multiple platforms under complex geological conditions and multiple working conditions. Summary of the invention

[0005] Based on this, it is necessary to provide construction parameter optimization methods, devices, media and equipment based on sweet spot prediction to address the technical problem that existing technologies cannot accurately optimize construction parameters of multiple wells and multiple platforms under complex geological conditions and multiple working conditions.

[0006] The present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a construction parameter optimization method based on sweet spot prediction, the method comprising:

[0008] Acquire multiple geological parameters, multiple engineering parameters and production capacity data related to the production process of shale gas wells, obtain a data set, and determine a sweet spot indicator factor;

[0009] Taking the dataset as the input and the dessert indication factor as the output, training a network model constructed based on the LightGBM algorithm to obtain a dessert indication factor prediction model; optimizing the dessert indication factor prediction model using the Bayesian hyperparameter optimization algorithm to obtain an optimized dessert indication factor prediction model;

[0010] Inputting the dataset related to the production process of the shale gas well obtained in real time into the optimized dessert indication factor prediction model to obtain the prediction result of the dessert indication factor; inverting the prediction result into the three-dimensional spatial distribution characteristics of the sweet spots in the reservoir of the shale;

[0011] Based on the three-dimensional spatial distribution characteristics, determining the contribution values and partial dependence relationships of the multiple geological parameters and the multiple engineering parameters to the prediction result, optimizing the construction parameters of the shale gas well according to the contribution values and partial dependence relationships, and whenever a shale gas well is produced using the optimized construction parameters, optimizing the prediction result of the dessert indication factor based on the dataset after construction, and further optimizing the construction parameters based on the optimized prediction result of the dessert indication factor.

[0012] Further, the network model constructed based on the LightGBM algorithm, its construction method specifically includes:

[0013] Constructing a first objective function based on the LightGBM algorithm, performing a second-order Taylor expansion on the loss function of the first objective function, and substituting the regularization term expression into the first objective function to obtain the second objective function of the LightGBM algorithm;

[0014] Determining the extreme point of the second objective function, optimizing the second objective function according to the extreme point to obtain the third objective function of the LightGBM algorithm;

[0015] Constructing the dessert indication factor prediction model according to the third objective function;

[0016] Among them, the expression of the first objective function is:

[0017]

[0018] Among them, n refers to the total number of samples, y i represents the true value of the i-th sample, y io represents the predicted value of the i-th sample, f t represents the t-th subtree, l represents the loss function, and Ω represents the regularization term;

[0019] The expression of the second objective function is:

[0020]

[0021] The expression of the third objective function is as follows:

[0022]

[0023] where, G j represents the set of first-order derivatives, H j represents the set of second-order derivatives, λ is a constant, T represents the number of leaves, and γ represents the complexity of the generated leaf nodes.

[0024] Furthermore, according to the third objective function, the dessert indication factor prediction model is constructed, which specifically includes:

[0025] Multiple decision trees are generated according to the third objective function, and the multiple decision trees are combined according to the weights of each decision tree in the multiple decision trees to obtain the dessert indication factor prediction model;

[0026] Among them, the generation process of any one of the multiple decision trees specifically includes:

[0027] The LightGBM algorithm is used to determine whether the training set is a high-dimensional sparse matrix. In the case where the training set is a high-dimensional sparse matrix, the GOSS algorithm is used to determine the features in the training set whose gradients are greater than a preset threshold, and the features whose gradients are greater than the preset threshold are retained, and the EFB algorithm is used to bundle the mutually exclusive features in the high-dimensional sparse matrix into a single feature;

[0028] In the case where the training set is not a high-dimensional sparse matrix, the histogram algorithm is used to determine the optimal splitting node of any one of the decision trees; the histogram difference algorithm based on the histogram algorithm is used to determine the child nodes of the optimal splitting node;

[0029] The Leaf-wise growth algorithm is used to guide the growth of any one of the decision trees according to the optimal splitting node and the child nodes of the optimal splitting node.

[0030] Furthermore, whenever a shale gas well is produced using the optimized construction parameters, the prediction result of the dessert indication factor is optimized based on the dataset after construction, and the construction parameters are further optimized based on the optimized prediction result of the dessert indication factor, which specifically includes:

[0031] The i-th shale gas well is produced according to the optimized construction parameters, and the (i + 1)-th dataset related to the shale gas well production process after the production of the i-th shale gas well is obtained, where i is a positive integer, i < n, and n is the number of shale gas wells;

[0032] Input the (i + 1)-th dataset into the optimized sweet spot indicator prediction model to obtain the (i + 1)-th prediction result of the sweet spot indicator; use the Kriging interpolation method to invert the (i + 1)-th prediction result into the (i + 1)-th three-dimensional spatial distribution characteristics of the sweet spot in the reservoir of the shale.

[0033] Based on the (i + 1)-th three-dimensional spatial distribution characteristics, use the SHAP method and the PDP method to determine the contribution values and partial dependence relationships of multiple geological parameters and multiple engineering parameters in the (i + 1)-th dataset on the (i + 1)-th prediction result, and optimize the (i + 1)-th construction parameters of the shale gas well according to the contribution values and partial dependence relationships of the (i + 1)-th prediction result.

[0034] Furthermore, optimizing the sweet spot indicator prediction model by using the Bayesian hyperparameter optimization algorithm to obtain an optimized sweet spot indicator prediction model specifically includes:

[0035] Use the Bayesian hyperparameter optimization algorithm to determine the best hyperparameter combination of the sweet spot indicator prediction model;

[0036] Input the best hyperparameter combination into the sweet spot indicator prediction model to obtain the optimized sweet spot indicator prediction model.

[0037] Furthermore, determining the contribution values and partial dependence relationships of the multiple geological parameters and the multiple engineering parameters on the prediction result specifically includes:

[0038] Use the SHAP method to calculate the contribution values of each geological parameter and each engineering parameter in the dataset on the prediction result of the sweet spot indicator under the coupled action;

[0039] Use the PDP method to calculate the partial dependence relationships of each geological parameter and each engineering parameter in the dataset on the prediction result of the sweet spot indicator under the coupled action.

[0040] Furthermore, before taking the dataset as input, it further includes:

[0041] Delete the null values and abnormal data in the dataset to obtain a preprocessed dataset.

[0042] In a second aspect, the present invention provides a construction parameter optimization device based on sweet spot prediction, including:

[0043] An acquisition module, configured to acquire multiple geological parameters, multiple engineering parameters, and production capacity data related to the production process of a shale gas well to obtain a dataset, and determine a sweet spot indicator;

[0044] A training module, configured to take a data set as input and the dessert indication factor as output, train a network model constructed based on the LightGBM algorithm to obtain a dessert indication factor prediction model; and optimize the dessert indication factor prediction model by using a Bayesian hyperparameter optimization algorithm to obtain an optimized dessert indication factor prediction model;

[0045] A prediction module, configured to input a data set related to the production process of a shale gas well, which is obtained in real time, into the optimized dessert indication factor prediction model to obtain a prediction result of the dessert indication factor; and invert the prediction result into three-dimensional spatial distribution characteristics of the sweet spots in the reservoir of the shale;

[0046] An optimization module, configured to determine contribution values and partial dependence relationships of the multiple geological parameters and the multiple engineering parameters to the prediction result based on the three-dimensional spatial distribution characteristics, and whenever a shale gas well is produced by using optimized construction parameters, optimize the prediction result of the dessert indication factor based on the data set after construction, and further optimize the construction parameters based on the optimized prediction result of the dessert indication factor.

[0047] The present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method for optimizing construction parameters based on dessert prediction is implemented.

[0048] The present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the method for optimizing construction parameters based on dessert prediction is implemented.

[0049] At least one technical solution adopted by the present invention can achieve the following beneficial effects: The present invention constructs a sweet spot indicator factor prediction model based on the Light Gradient Boosting Machine (LightGBM) algorithm, which can reduce the depth of decision trees in the sweet spot indicator factor prediction model, improve the training speed of the sweet spot indicator factor prediction model, better process complex data, and thus improve the prediction accuracy of the sweet spot indicator factor prediction model. Then, the Bayesian optimization algorithm is used to optimize the hyperparameters of the sweet spot indicator factor prediction model to obtain an optimized sweet spot indicator factor prediction model with the best parameters, which can further improve the accuracy of predicting the sweet spot indicator factor. And in the process of predicting the sweet spot indicator factor, the above solution combines geological parameters and engineering parameters related to the production process of shale gas wells, realizes the combination of geological factors and engineering factors, comprehensively predicts the sweet spot indicator factor, makes the prediction of the sweet spot indicator factor more scientific and reasonable, and realizes the precise optimization of the construction parameters of shale gas wells under complex geological conditions. Finally, the prediction results of the sweet spot indicator factor are analyzed to obtain the analysis results. According to the analysis results, the construction parameters of the shale gas wells are optimized. And whenever the production of a shale gas well is completed, the dataset changed due to the production of the shale gas well is obtained, the prediction results of the sweet spot indicator factor are optimized based on the obtained latest dataset, and the construction parameters are further optimized based on the optimized prediction results of the sweet spot indicator factor, realizing the accurate optimization of the construction parameters of multiple wells and multiple platforms under multiple working conditions, significantly improving the scientificity of optimizing the construction parameters, providing theoretical support and technical guarantee for the exploration and development of unconventional oil and gas resources, and having wide academic and practical application values. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0051] Figure 1 is a flowchart of a construction parameter optimization method based on sweet spot prediction provided by the present invention;

[0052] Figure 2 is another flowchart of a construction parameter optimization method based on sweet spot prediction provided by the present invention;

[0053] Figure 3 is a three-dimensional spatial distribution feature map of the sweet spot of the reservoir obtained by Kriging interpolation inversion provided by the present invention;

[0054] Figure 4 is a feature importance ranking map obtained by using the SHAP method provided by the present invention;

[0055] Figure 5 SHAP dependence graph of gas content characteristics provided by the present invention;

[0056] Figure 6 Decision margin graph of gas content characteristics provided by the present invention;

[0057] Figure 7 Schematic structural diagram of a construction parameter optimization device based on sweet spot prediction provided by the present invention;

[0058] Figure 8 Schematic structural diagram of a computer device provided by the present invention. Detailed implementation manners

[0059] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0060] At present, the server mentioned in the present invention can be a server set up on a business platform, or a device such as a desktop computer or a laptop computer that can execute the solution of the present invention. For the convenience of description, only the server is used as the execution subject for description below. The technical solutions provided by each embodiment of the present invention will be described in detail below with reference to the drawings.

[0061] Refer to Figure 1 , the construction parameter optimization method based on sweet spot prediction in the present invention specifically includes the following steps:

[0062] S10: Obtain a plurality of geological parameters, a plurality of engineering parameters, and production capacity data related to the production process of a shale gas well to obtain a data set, and determine a sweet spot indicator factor.

[0063] In this embodiment, the engineering parameters include but are not limited to: horizontal section length of the fracturing stage, number of fracturing stages, number of clusters, sand addition intensity, liquid usage intensity, construction pressure, average cluster spacing, total sand volume, total liquid volume. The geological parameters include but are not limited to: gas content, porosity, total organic carbon (TOC), brittle mineral content, thickness of high-quality reservoir, pressure coefficient. The production capacity data is data about oil and gas production. The sweet spot indicator factor refers to a characteristic parameter that can reflect the position distribution of sweet spots in the underground three-dimensional space. The selection methods of the sweet spot indicator factor include but are not limited to: selecting multiple parameters from the data set as the sweet spot indicator factor according to historical experience, or selecting multiple parameters from the data set according to the collected data.

[0064] S20: Use the dataset as input and the dessert indication factor as output to train a network model constructed based on the LightGBM algorithm to obtain a dessert indication factor prediction model; use the Bayesian hyperparameter optimization algorithm to optimize the dessert indication factor prediction model to obtain an optimized dessert indication factor prediction model.

[0065] In this embodiment, the dataset is divided into a training set and a test set according to a preset ratio. In this embodiment, the dataset is divided into a training set and a test set at a ratio of 7:3.

[0066] Preferably, before using the dataset as input, it further includes:

[0067] Delete the null values and abnormal data in the dataset to obtain a preprocessed dataset.

[0068] Specifically, in one or more embodiments of the present invention, for the network model constructed based on the LightGBM algorithm, its construction method specifically includes:

[0069] S201: Construct a first objective function based on the LightGBM algorithm, perform a second-order Taylor expansion on the loss function of the first objective function, and substitute the regularization term expression into the first objective function to obtain a second objective function of the LightGBM algorithm.

[0070] In this embodiment, the first objective function refers to the initial objective function constructed based on the LightGBM algorithm, and its specific expression is:

[0071] Among them, the expression of the first objective function is:

[0072]

[0073] Among them, n refers to the total number of samples, y i represents the true value of the i-th sample, y io represents the predicted value of the -th sample, f t represents the t-th subtree, l represents the loss function, and Ω represents the regularization term.

[0074] Since the t-th tree generated by LightGBM is adding a new prediction function on the basis of the (t - 1)-th tree, represent the t-th tree with the (t - 1)-th tree, perform a second-order Taylor expansion on the loss function expression in the first objective function, and substitute the regularization term expression to obtain the second objective function, and its specific expression is:

[0075]

[0076] Among them, G j represents the set of first-order derivatives, and H j represents the set of second-order derivatives.

[0077] S202: Determine the extreme point of the second objective function, optimize the second objective function according to the extreme point, and obtain the third objective function of the LightGBM algorithm.

[0078] In this embodiment, the third objective function refers to the optimal form of the second objective function. To find the optimal form of the second objective function, that is, it is necessary to first find the minimum value of the second objective function. According to formula (2), it is found that the second objective function is a quadratic function about W j . Therefore, finding the minimum value is transformed into finding the minimum extreme value. The expression of the third objective function obtained according to the extreme point is:

[0079]

[0080] where, G j represents the set of first-order derivatives, H j represents the set of second-order derivatives, λ is a constant, T represents the number of leaves, and γ represents the complexity of the generated leaf nodes.

[0081] S203: Construct a dessert indication factor prediction model according to the third objective function.

[0082] In this embodiment, constructing a dessert indication factor prediction model according to the third objective function specifically includes:

[0083] Generate multiple decision trees according to the third objective function, and combine the decision trees according to the weights of the decision trees in the multiple decision trees to obtain a dessert indication factor prediction model.

[0084] Among them, the generation process of any one of the multiple decision trees specifically includes:

[0085] Use the LightGBM algorithm to determine whether the training set is a high-dimensional sparse matrix. In the case where the training set is a high-dimensional sparse matrix, use the GOSS algorithm to determine the features in the training set whose gradients are greater than the preset threshold, retain the features whose gradients are greater than the preset threshold, and use the EFB algorithm to bundle the mutually exclusive features in the high-dimensional sparse matrix into a single feature.

[0086] In the case where the training set is not a high-dimensional sparse matrix, use the histogram algorithm to determine the optimal splitting node of any one decision tree; use the histogram difference algorithm based on the histogram algorithm to determine the child nodes of the optimal splitting node.

[0087] Use the Leaf-wise growth algorithm to guide the growth of any one decision tree according to the optimal splitting node and the child nodes of the optimal splitting node.

[0088] Specifically, in this embodiment, the LightGBM algorithm scans the shale gas training set to determine whether it is a high-dimensional sparse matrix. If it is a high-dimensional coefficient matrix, GOSS and EFB operations are performed: non-zero features often carry key information, and the gradients of the corresponding samples may be large. GOSS retains the samples with larger gradients and randomly samples the samples with smaller gradients. There are a large number of mutually exclusive features in the high-dimensional sparse matrix, and EFB bundles the mutually exclusive features into a single feature, significantly reducing the dimension.

[0089] If it is not a high-dimensional sparse matrix, the histogram algorithm discretizes the data features into k discrete values, constructs a histogram, and statistically calculates the cumulative statistics of each discrete value in the histogram. When performing feature selection, the optimal tree node splitting point is found according to the statistics.

[0090] Secondly, when splitting a node and generating the child nodes of the node, the histogram subtraction acceleration algorithm based on the histogram algorithm obtains the histogram of the other child node by subtracting the histogram of one of the child nodes from the histogram of the node.

[0091] Finally, the Leaf-wise growth algorithm is used to guide the growth of decision trees in the LightGBM model, and the decision trees are combined according to the weights of each decision tree to form a prediction model framework of the sweet spot indicator factor based on the LightGBM machine learning algorithm.

[0092] S30: Input the data set related to the production process of the shale gas well obtained in real time into the optimized sweet spot indicator factor prediction model to obtain the prediction result of the sweet spot indicator factor; invert the prediction result into the three-dimensional spatial distribution characteristics of the sweet spot in the reservoir of the shale.

[0093] In this embodiment, the Kriging method is used to invert the three-dimensional spatial distribution characteristics of the reservoir sweet spot from the sweet spot indicator factor data. By observing the values z i , y i ) of the sweet spot indicator factor at several discrete well coordinates (x i =(x i , y i ) on the known shale gas block, the sweet spot indicator factor at any coordinate (x, y) on the known shale gas block is estimated. Then the estimated value at (x0, y0) The calculation formula is as follows:

[0094]

[0095] Among them, λ i is the weight coefficient that can satisfy the set of optimal coefficients with the smallest difference between the estimated value at the point (x0, y0) and the true value Z0, that is Meet the conditions of unbiased estimation simultaneously

[0096] S40: Based on the three-dimensional spatial distribution characteristics, determine the contribution values and partial dependence relationships of multiple geological parameters and multiple engineering parameters to the prediction result, optimize the construction parameters of the shale gas well according to the contribution values and partial dependence relationships, and whenever a shale gas well is produced using the optimized construction parameters, optimize the prediction result of the sweet spot indicator factor based on the dataset after construction, and further optimize the construction parameters based on the optimized prediction result of the sweet spot indicator factor.

[0097] In this embodiment, the Shapley Additive exPlanations (SHAP) method and the Probability Distribution Plot (PDP) method are used to determine the contribution values and partial dependence relationships of multiple geological parameters and multiple engineering parameters to the prediction result. Among them, the SHAP method is a method for explaining and debugging machine learning models. It is based on the Shapley value in game theory and is used to quantify the contribution of each feature to the model prediction. SHAP can help understand how the model makes predictions, identify which features have the greatest impact on the prediction result, and how these features interact with each other. The PDP is a visualization tool used to show the dependence relationship between the target variable and a set of input features. It observes the change of the model output by changing the value of the feature of interest while keeping the values of other features unchanged. The PDP can help understand the average impact of features on the model prediction and the specific impact of features under different samples.

[0098] In one or more embodiments of the present invention, determining the contribution values and partial dependence relationships of multiple geological parameters and multiple engineering parameters to the prediction result specifically includes:

[0099] Use the SHAP method to calculate the contribution values of each geological parameter and each engineering parameter in the dataset to the prediction result of the sweet spot indicator factor under the coupled action.

[0100] Use the PDP method to calculate the partial dependence relationship of each geological parameter and each engineering parameter in the dataset to the prediction result of the sweet spot indicator factor under the coupled action.

[0101] In this embodiment, according to the visualization interpretation results of the SHAP method and the PDP method, quantitatively evaluate the influence of multiple geological engineering parameters on sweet spot prediction under the coupled action. Finally, according to the coupled relationship of geological engineering parameters and the quantitative evaluation results, implement the regulatory control parameter values during the construction process to achieve the purpose of optimizing the construction parameters.

[0102] Among them, the SHAP method is an interpretation method based on game theory, used to explain the output of machine learning models, and quantifies the importance of features by calculating the contribution value of each feature to the model prediction. The calculation formula for the contribution value is as follows:

[0103]

[0104] Among them, φ i represents the contribution value of feature i, N represents the complete set of features, S is any subset of N but does not include feature i, M represents the total number of features, and f() represents the model prediction function.

[0105] The PDP method shows the marginal effect of each feature on the prediction result of the machine learning model, and is especially suitable for understanding the relationship between features. In the dessert indicator factor prediction model, the partial dependence plot (Probability Distribution Plot, PDP) reveals the linear or non-linear relationship between the estimated ultimate recovery (EUR) of oil and gas wells and each feature. Then, the calculation formula for the partial dependence function of the PDP method is as follows:

[0106]

[0107] Among them, x s represents the selected feature, X c represents the set of other features, f() represents the model prediction function, and for the selected feature x s , by taking the expectation or integral of all possible values of X c , the partial dependence relationship of the selected feature on the model prediction result is obtained.

[0108] Based on Figure 1The construction parameter optimization method based on sweet spot prediction shown herein. The present invention constructs a sweet spot indicator factor prediction model based on the LightGBM algorithm, which can reduce the depth of decision trees in the sweet spot indicator factor prediction model and improve the training speed of the sweet spot indicator factor prediction model, and can better process complex data, thereby improving the prediction accuracy of the sweet spot indicator factor prediction model. Then, the Bayesian optimization algorithm is used to optimize the hyperparameters of the sweet spot indicator factor prediction model to obtain an optimized sweet spot indicator factor prediction model with the best parameters, which can further improve the accuracy of predicting the sweet spot indicator factor. And in the process of predicting the sweet spot indicator factor, the above solution combines geological parameters and engineering parameters related to the production process of shale gas wells, realizes the combination of geological factors and engineering factors, comprehensively predicts the sweet spot indicator factor, makes the prediction of the sweet spot indicator factor more scientific and reasonable, and realizes the precise optimization of the construction parameters of shale gas wells under complex geological conditions. Finally, the prediction results of the sweet spot indicator factor are analyzed to obtain the analysis results, and the construction parameters of the shale gas well are optimized according to the analysis results. And whenever the production of a shale gas well is completed, the dataset changed due to the production of the shale gas well is obtained, the prediction results of the sweet spot indicator factor are optimized based on the obtained latest dataset, and the construction parameters are further optimized based on the optimized prediction results of the sweet spot indicator factor, realizing the accurate optimization of the construction parameters of multiple wells and multiple platforms under multiple working conditions, significantly improving the scientificity of optimizing the construction parameters, providing theoretical support and technical guarantee for the exploration and development of unconventional oil and gas resources, and having broad academic and practical application values.

[0109] When applying the construction parameter optimization method based on sweet spot prediction provided by the present invention, it is not necessary to execute according to Figure 1 the order of the steps shown. The specific execution order of each step can be determined as needed, and the present invention does not limit this.

[0110] In addition, in one or more embodiments of the present invention, the Bayesian hyperparameter optimization algorithm is used to optimize the sweet spot indicator factor prediction model to obtain an optimized sweet spot indicator factor prediction model, which specifically includes:

[0111] Use the Bayesian hyperparameter optimization algorithm to determine the best hyperparameter combination of the sweet spot indicator factor prediction model.

[0112] In this embodiment, the Bayesian hyperparameter optimization method is used to identify the sweet spot indicator factor prediction model built based on the LightGBM algorithm as an unknown function, continuously take points and measure within the defined model hyperparameter interval, and then update the Gaussian process distribution according to the measured historical data, repeating multiple times, so as to find the optimal function and its extreme point, and the extreme point is the hyperparameter combination corresponding to the optimal function.

[0113] Input the optimal hyperparameter combination into the dessert indicator factor prediction model to obtain an optimized dessert indicator factor prediction model.

[0114] In this embodiment, input the hyperparameter combination into the corresponding dessert indicator factor prediction model framework, and use the Mean Absolute Percentage Error (MAPE) to calculate the accuracy of the dessert indicator factor prediction model. The formula is:

[0115]

[0116] where n represents the number of samples, y i represents the true value, and y io represents the predicted value.

[0117] In addition, in one or more embodiments of the present invention, whenever a shale gas well is produced using the optimized construction parameters, the prediction result of the dessert indicator factor is optimized based on the dataset after construction, and the construction parameters are further optimized based on the optimized prediction result of the dessert indicator factor. Specifically, it includes:

[0118] Produce the i-th shale gas well according to the optimized construction parameters, and obtain the (i + 1)-th dataset related to the shale gas well production process after the production of the i-th shale gas well. i is a positive integer, i < n, and n is the number of shale gas wells.

[0119] Input the (i + 1)-th dataset into the optimized dessert indicator factor prediction model to obtain the (i + 1)-th prediction result of the dessert indicator factor; use the Kriging interpolation method to invert the (i + 1)-th prediction result into the (i + 1)-th three-dimensional spatial distribution characteristics of the sweet spots in the shale reservoir.

[0120] Based on the (i + 1)-th three-dimensional spatial distribution characteristics, use the SHAP method and the PDP method to determine the contribution values and partial dependence relationships of multiple geological parameters and multiple engineering parameters in the (i + 1)-th dataset on the (i + 1)-th prediction result, and optimize the (i + 1)-th construction parameters of the shale gas well according to the contribution values and partial dependence relationships of the (i + 1)-th prediction result.

[0121] In this embodiment, after each shale gas well is produced, the geological parameters, engineering parameters, and three-dimensional spatial distribution of the sweet spots will all change. Therefore, the prediction results of the sweet spot indicator factor will deviate, resulting in a reduction in the optimization effect of the construction parameters for the subsequent production of shale gas wells. Through this embodiment, the latest geological parameters and engineering parameters can be obtained after the production of each shale gas well. Considering the reaction of the shale gas well construction operation on the sweet spot prediction, the prediction accuracy of the sweet spot indicator factor prediction model can be optimized. The Kriging interpolation method is used to obtain the latest three-dimensional spatial distribution of the sweet spots, and then the construction parameters for the subsequent production of shale gas wells are optimized in real time, realizing the linkage of the optimization of the construction parameters in the production process of multiple shale gas wells, and further improving the optimization effect of the construction parameters for the production of shale gas wells.

[0122] It should be noted that this solution can optimize the prediction results of the sweet spot indicator factor based on the dataset after construction after producing multiple shale gas wells, and further optimize the construction parameters based on the prediction results of the optimized sweet spot indicator factor, not limited to one.

[0123] Specifically, taking the field application in the Luzhou block in the southern part of the Sichuan Basin as an example, combined with the appendix Figure 2 This further illustrates the implementation of the present invention:

[0124] Refer to Figure 2 , a construction parameter optimization method based on spatial sweet spot prediction specifically includes:

[0125] Step 1, collect original on-site data: Collect the geological engineering parameters and production capacity data during the production of shale gas wells in the Luzhou block to form a dataset. Among them, the engineering parameters include the horizontal section length of the fracturing section, the number of fracturing sections, the number of clusters, the sand addition intensity, the liquid usage intensity, the construction pressure, the average cluster spacing, the total sand volume, and the total liquid volume. The geological parameters include gas content, porosity, TOC, brittle mineral content, high-quality reservoir thickness, and pressure coefficient. Finally, data from 53 shale gas wells are collected.

[0126] Step 2, select the sweet spot indicator factor: According to the dataset formed in Step 1, conduct dataset parameter analysis and select the geological parameters and engineering parameters to evaluate the sweet spot indicator factor. The method of deleting null values and abnormal data is used to preprocess the dataset to ensure the accuracy of the dataset. Calibrate the sweet spot indicator factor data as the target value. The dataset is divided into a training set for training the model and a test set for validating the model according to a ratio of 7:3.

[0127] Step 3, delete null values and outliers: Use the method of deleting null values and outliers to process the original dataset to obtain the overall dataset. The overall dataset is divided into a training set and a test set according to a ratio of 7:3.

[0128] Step 4, construct a dessert indicator factor prediction model: The specific steps are as follows: Combine the training set and test set in Step 3. In the Python language environment, based on the LightGBM machine learning algorithm and the Bayesian hyperparameter optimization algorithm, construct an optimal dessert indicator factor prediction model. The steps are as follows:

[0129] Calculate the optimal form of the LightGBM objective function:

[0130]

[0131] where, G j represents the set of first-order derivatives, H j represents the set of second-order derivatives, γ is the complexity of the new leaf node, T is the number of leaves, and λ is a constant.

[0132] On the basis of finding the optimal objective function, LightGBM scans and analyzes the training set and determines that the dataset does not belong to a high-dimensional sparse matrix. So it skips the two operations of GOSS and EFB. LightGBM enables the histogram algorithm to discretize various features in the dataset into k specific discrete values and constructs a corresponding histogram structure on this basis, carefully counting statistics such as the cumulative frequency of each discrete value in the histogram. In the process of feature selection to determine the best tree node splitting point, make full use of the information contained in these statistics. Through rigorous calculation and comparison, accurately find the splitting point position that can make the model performance reach the optimal. In the stage of generating the child nodes of the node, the histogram subtraction acceleration method derived from the histogram algorithm plays a key role. By performing a difference operation on the histogram of the current node and the histogram of one of the child nodes, the histogram information of the other child node can be quickly and accurately deduced, greatly accelerating the speed of node generation and model construction. Finally, use the Leaf-wise growth algorithm to guide the growth direction of the decision tree in the LightGBM model, generate a decision tree, and based on this, generate multiple decision trees to form the framework of the dessert indicator factor prediction model.

[0133] Use the Bayesian hyperparameter optimization method to regard the dessert indicator factor prediction model framework built based on the LightGBM machine learning algorithm as an unknown function, continuously take points and measure within the defined model hyperparameter interval, and then update the Gaussian process distribution according to the measured historical data, repeatedly for many times, so as to find the optimal function and its extreme point. The extreme point is the hyperparameter combination corresponding to the optimal function. The Bayesian hyperparameter optimization results are shown in Table 1:

[0134] Table 1 LightGBM Hyperparameter Combination Table

[0135]

[0136] The accuracy of the prediction model of the dessert indication factor is calculated using MAPE, and the result shows that the error rate of the dessert indication factor prediction model is 13.7%.

[0137] Step 5: According to the optimal prediction model selected in Step 4, input the geographical coordinates and the predicted values of the dessert index factors into the Kriging method, then the estimated value at (x0, y0) The calculation formula is as follows:

[0138]

[0139] where λ i is the weight coefficient that can satisfy a set of optimal coefficients with the smallest difference between the estimated value at the point (x0, y0) and the true value Z0, that is simultaneously satisfying the condition of unbiased estimation

[0140] Set the number of neighboring points to 10, the proportion of test points to 80%, and the grid spacing to 2. Finally, the three-dimensional spatial distribution characteristics of the reservoir sweet spots are inverted from the dessert indication factor data, as Figure 3 shown.

[0141] Step 6: According to the optimal prediction model selected in Step 4, and based on the contribution degree of the geological engineering parameters to the prediction model, exclude the parameters with insignificant contributions. Finally, 10 parameters are selected and their contribution distributions are shown, as Figure 4 shown. Then, the SHAP method and the PDP method are used to display the decision margins, which show the influence of individual features in the entire dataset and consider the interaction effects existing in the features. Here, the gas content is taken as an example to show, as Figure 5 and Figure 6 shown. The gas content and the sand addition intensity have a positive contribution to the model prediction output. When the gas content is less than 5.8 m 3 / t and the sand addition intensity is increased, the contributions of the gas content and the sand addition intensity to the model output are not obvious. When the gas content exceeds 5.8 m 3 / t and the sand addition intensity is greater than 2.0 t / m, the contributions of the gas content and the sand addition intensity to the model are more prominent. Therefore, during the engineering construction process, the control of the sand addition intensity parameter can be regulated in real time according to the value of the gas content, so as to optimize the construction parameters.

[0142] The above is the construction parameter optimization method based on sweet spot prediction provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding construction parameter optimization device based on sweet spot prediction, as Figure 7 shown, including:

[0143] An acquisition module, configured to acquire a plurality of geological parameters, a plurality of engineering parameters, and production capacity data related to the production process of a shale gas well, obtain a data set, and determine a sweet spot indication factor.

[0144] A training module, configured to use the data set as an input and the sweet spot indication factor as an output to train a network model constructed based on the LightGBM algorithm to obtain a sweet spot indication factor prediction model; use the Bayesian hyperparameter optimization algorithm to optimize the sweet spot indication factor prediction model to obtain an optimized sweet spot indication factor prediction model.

[0145] A prediction module, configured to input the data set related to the production process of the shale gas well obtained in real time into the optimized sweet spot indication factor prediction model to obtain a prediction result of the sweet spot indication factor; invert the prediction result into three-dimensional spatial distribution characteristics of the sweet spot in the reservoir of the shale.

[0146] An optimization module, configured to determine the contribution values and partial dependence relationships of a plurality of geological parameters and a plurality of engineering parameters to the prediction result based on the three-dimensional spatial distribution characteristics; optimize the construction parameters of the shale gas well according to the contribution values and partial dependence relationships, and whenever a shale gas well is produced using the optimized construction parameters, optimize the prediction result of the sweet spot indication factor based on the data set after construction, and further optimize the construction parameters based on the optimized prediction result of the sweet spot indication factor.

[0147] Specific limitations on the device for optimizing construction parameters based on sweet spot prediction can refer to the limitations on the method for optimizing construction parameters based on sweet spot prediction in the above text, which will not be elaborated here. Each module in the device for optimizing construction parameters based on sweet spot prediction can be implemented in whole or in part through software, hardware, and their combination. The modules can be embedded in the processor in the computer device in the form of hardware or be independent of it, or be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0148] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program can be used to execute the Figure 1 Method for optimizing construction parameters based on sweet spot prediction provided above.

[0149] The present invention also provides Figure 8 The structural schematic diagram of the computer device shown in Figure 8 As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, other hardware required for other services may also be included. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the Figure 1 Method for optimizing construction parameters based on sweet spot prediction provided above.

[0150] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the described embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the various methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0151] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope recorded by the present invention.

Claims

1. A construction parameter optimization method based on dessert prediction, characterized in that Including: Obtain multiple geological parameters, multiple engineering parameters, and production capacity data related to the production process of a shale gas well, obtain a data set, and determine a sweet spot indication factor; Use the data set as the input and the sweet spot indication factor as the output to train a network model constructed based on the LightGBM algorithm to obtain a sweet spot indication factor prediction model; Optimize the sweet spot indication factor prediction model using the Bayesian hyperparameter optimization algorithm to obtain an optimized sweet spot indication factor prediction model; Input the data set related to the production process of the shale gas well obtained in real time into the optimized sweet spot indication factor prediction model to obtain the prediction result of the sweet spot indication factor; invert the prediction result into the three-dimensional spatial distribution characteristics of the sweet spot in the reservoir of the shale; Based on the three-dimensional spatial distribution characteristics, determine the contribution values and partial dependence relationships of the multiple geological parameters and the multiple engineering parameters to the prediction result, optimize the construction parameters of the shale gas well according to the contribution values and partial dependence relationships, and whenever a shale gas well is produced using the optimized construction parameters, optimize the prediction result of the sweet spot indication factor based on the data set after construction, and further optimize the construction parameters based on the optimized prediction result of the sweet spot indication factor.

2. The construction parameter optimization method based on dessert prediction according to claim 1, wherein The construction method of the network model constructed based on the LightGBM algorithm specifically includes: Construct a first objective function based on the LightGBM algorithm, perform a second-order Taylor expansion on the loss function of the first objective function, and substitute the regularization term expression into the first objective function to obtain a second objective function of the LightGBM algorithm; Determine the extreme point of the second objective function, optimize the second objective function according to the extreme point to obtain a third objective function of the LightGBM algorithm; Construct the sweet spot indication factor prediction model according to the third objective function; Among them, the expression of the first objective function is: Among them, n refers to the total number of samples, y i represents the true value of the i-th sample, y io represents the predicted value of the sample, f t represents the t-th subtree, l represents the loss function, and Ω represents the regularization term; The expression of the second objective function is: The expression of the third objective function is: Among them, G j represents the set of first-order derivatives, H j represents the set of second-order derivatives, λ is a constant, T represents the number of leaves, and γ represents the complexity of the generated leaf nodes.

3. The construction parameter optimization method based on dessert prediction according to claim 2, wherein Construct the sweet spot indication factor prediction model according to the third objective function, specifically including: Generate multiple decision trees according to the third objective function, and combine the decision trees according to the weights of the decision trees in the multiple decision trees to obtain the sweet spot indication factor prediction model; Among them, the generation process of any one of the multiple decision trees specifically includes: Use the LightGBM algorithm to determine whether the training set is a high-dimensional sparse matrix. When the training set is a high-dimensional sparse matrix, use the GOSS algorithm to determine the features in the training set with gradients greater than a preset threshold, retain the features with gradients greater than the preset threshold, and use the EFB algorithm to bundle mutually exclusive features in the high-dimensional sparse matrix into a single feature; When the training set is not a high-dimensional sparse matrix, use the histogram algorithm to determine the optimal splitting node of any one of the decision trees; use the histogram difference algorithm based on the histogram algorithm to determine the child nodes of the optimal splitting node; Using the leaf-wise growth algorithm, guide the growth of any decision tree according to the optimal splitting node and the child nodes of the optimal splitting node.

4. The construction parameter optimization method based on dessert prediction according to claim 1, characterized in that Whenever a shale gas well is produced using the optimized construction parameters, optimize the prediction result of the sweet spot indication factor based on the dataset after construction, and further optimize the construction parameters based on the optimized prediction result of the sweet spot indication factor. Specifically, it includes: Produce the i-th shale gas well according to the optimized construction parameters, and obtain the (i + 1)-th dataset related to the production process of the i-th shale gas well after production. i is a positive integer, i < n, and n is the number of shale gas wells; Input the (i + 1)-th dataset into the optimized sweet spot indication factor prediction model to obtain the (i + 1)-th prediction result of the sweet spot indication factor; use the Kriging interpolation method to invert the (i + 1)-th prediction result into the (i + 1)-th three-dimensional spatial distribution characteristics of the sweet spot in the reservoir of the shale. Based on the (i + 1)-th three-dimensional spatial distribution characteristics, use the SHAP method and the PDP method to determine the contribution values and partial dependence relationships of multiple geological parameters and multiple engineering parameters in the (i + 1)-th dataset on the (i + 1)-th prediction result, and optimize the (i + 1)-th construction parameters of the shale gas well according to the contribution values and partial dependence relationships of the (i + 1)-th prediction result.

5. The construction parameter optimization method based on dessert prediction according to claim 1, characterized in that Optimize the sweet spot indication factor prediction model using the Bayesian hyperparameter optimization algorithm to obtain an optimized sweet spot indication factor prediction model. Specifically, it includes: Use the Bayesian hyperparameter optimization algorithm to determine the best hyperparameter combination of the sweet spot indication factor prediction model; Input the best hyperparameter combination into the sweet spot indication factor prediction model to obtain the optimized sweet spot indication factor prediction model.

6. The construction parameter optimization method based on dessert prediction according to claim 1, characterized in that Determine the contribution values and partial dependence relationships of the multiple geological parameters and the multiple engineering parameters on the prediction result. Specifically, it includes: Use the SHAP method to calculate the contribution values of each geological parameter and each engineering parameter in the dataset to the prediction result of the sweet spot indication factor under the coupled action; Use the PDP method to calculate the partial dependence relationships of each geological parameter and each engineering parameter in the dataset to the prediction result of the sweet spot indication factor under the coupled action.

7. The construction parameter optimization method based on dessert prediction according to claim 1, characterized in that, Before using the dataset as input, it also includes: Delete the null values and abnormal data in the dataset to obtain a preprocessed dataset.

8. The construction parameter optimization device based on dessert prediction, characterized in that It includes: An acquisition module for acquiring multiple geological parameters, multiple engineering parameters, and production capacity data related to the production process of a shale gas well to obtain a dataset, and determining a sweet spot indication factor; A training module for using the dataset as input and the sweet spot indication factor as output to train a network model constructed based on the LightGBM algorithm to obtain a sweet spot indication factor prediction model; Optimize the sweet spot indication factor prediction model using the Bayesian hyperparameter optimization algorithm to obtain an optimized sweet spot indication factor prediction model; A prediction module, configured to input a data set related to the production process of a shale gas well obtained in real time into the optimized sweet spot indicator factor prediction model to obtain a prediction result of the sweet spot indicator factor; and invert the prediction result into three-dimensional spatial distribution characteristics of sweet spots in the reservoir of the shale. An optimization module, configured to determine contribution values and partial dependence relationships of the multiple geological parameters and the multiple engineering parameters on the prediction result based on the three-dimensional spatial distribution characteristics, optimize the construction parameters of the shale gas well according to the contribution values and partial dependence relationships, and whenever a shale gas well is produced using the optimized construction parameters, optimize the prediction result of the sweet spot indicator factor based on the data set after construction, and further optimize the construction parameters based on the optimized prediction result of the sweet spot indicator factor.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and when the computer program is executed by a processor, it implements the construction parameter optimization method based on sweet spot prediction according to any one of claims 1 to 7.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the construction parameter optimization method based on sweet spot prediction according to any one of claims 1 to 7.

Images