Method and device for predicting sand production degree of ultra-deep gas well

Abstract

The invention discloses a method and device for predicting the sand production degree of an ultra-deep gas well, and the method comprises the steps: dividing a gas well in a research region according to the sand production degree, and obtaining the type of the sand production degree; key characteristic parameters in each gas well are collected and combined with sand production degree categories, and a sample data set is formed; the characteristic factors with the correlation coefficient values larger than a preset value are screened out to serve as main control factors, and a screened data set is formed. And carrying out repeated random sampling with replacement on the screened data set to obtain a plurality of training data sets. On each training data set, feature factors are randomly selected for sample division, and optimal division feature factors are selected by comparing critical Gini indexes. A plurality of decision trees are constructed according to the optimal division feature factors, a classification model is formed, and the model predicts the sand production degree of the gas well based on the input main control factors. The problems of overcoming the limitation of an existing sand production degree prediction method and improving the prediction accuracy and practicability are solved.

Description

Technical Field

[0001] The present application relates to the technical field of oil and gas exploration and development, and in particular to a method and device for predicting the degree of sand production in ultra-deep gas wells. Background Art

[0002] As global energy demand continues to grow, the demand for natural gas, as a clean and efficient energy source, is increasing. Against this background, ultra-deep, high-pressure, fractured, low-porosity sandstone gas reservoirs have become an important area for unconventional natural gas exploration and development due to their huge reserves and high production capacity. Such gas reservoirs are usually characterized by dense reservoirs, developed fractures, and active water bodies, which bring unique challenges and opportunities to the exploitation of natural gas. The Keshen and Bozi Dabei gas fields in the Kelasu tectonic belt of the Kuqa Depression in the Tarim Basin, as typical ultra-deep, high-pressure, fractured, low-porosity sandstone gas reservoirs, have extremely high development value. However, during the development process, the problem of sand production in gas wells has become increasingly prominent, and the proportion of sand production wells has exceeded 50%, which has seriously restricted the efficient development of gas fields. Sand burial and sand plugging occur frequently, resulting in reduced or even shutdown of gas wells, which has brought tremendous pressure to the production and operation of gas fields. The degree of sand production in different gas wells varies significantly, which directly determines the choice of sand production management strategy. A small amount of sand production can sometimes improve the seepage channel, reduce the skin effect, and have a positive impact on production; however, severe sand production will seriously interfere with normal production, and active sand control measures must be taken to limit the entry of sand into the wellbore. Therefore, accurately predicting the degree of sand production in ultra-deep and high-pressure gas wells is of great significance for formulating scientific and reasonable sand production management strategies and ensuring stable and efficient production of gas fields.

[0003] Physical simulation and numerical calculation methods face many challenges in practical applications. The uncertainty of the original geological model brings large errors to the prediction results, the cumbersome complexity of data input increases the difficulty and cost of operation, and the problem of error accumulation over time also limits the long-term prediction ability of this method.

[0004] Therefore, how to overcome the limitations of existing sand production prediction methods and improve prediction accuracy and practicality has become a problem restricting the efficient development of ultra-deep, high-pressure, fractured, low-porosity sandstone gas reservoirs. Summary of the invention

[0005] In the embodiments of the present application, a method for predicting the sand production degree of ultra-deep gas wells is provided to solve the problem of how to overcome the limitations of existing sand production degree prediction methods and improve the prediction accuracy and practicality.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting the degree of sand production in ultra-deep gas wells, including: comprehensively considering wellbore sand samples, equipment wear and production parameter fluctuations, dividing the gas wells in the study area according to the degree of sand production, and obtaining sand production degree categories; wherein the sand production degree categories include no sand production, mild sand production, moderate sand production and severe sand production; collecting key characteristic parameters in each gas well, combining the key characteristic parameters with the sand production degree categories to form a sample data set; wherein the key characteristic parameters include characteristic factors in geology, completion and development; measuring the correlation coefficient value between each key characteristic parameter in the sample data set and the sand production degree, selecting the characteristic factor with a correlation coefficient value greater than a preset value as the main control factor, and forming a screening data set. ; Repeated random sampling with replacement is performed on the screening data set to obtain multiple training data sets containing different samples; for each training data set, feature factors are randomly selected, and the samples contained in the current node are divided into two subsets according to a preset partitioning threshold; by comparing the critical Gini index after partitioning with different feature factors, the feature factor that minimizes the critical Gini index after partitioning is selected as the optimal partitioning feature factor; multiple decision trees are constructed for each training data set according to the optimal partitioning feature factor to form a classification model; wherein the classification model predicts the degree of sand production of the gas well based on the input main controlling factors; the performance of the classification model is evaluated, and the parameters of the classification model are adjusted according to the evaluation results to optimize the performance of the classification model.

[0007] In a possible implementation, the comprehensive wellbore sand samples, equipment wear and production parameter fluctuations are used to classify the gas wells in the study area according to the degree of sand production. Before obtaining the sand production degree category, it also includes: determining the reservoir location and rock type in the study area.

[0008] In a possible implementation, the correlation coefficient value between each key characteristic parameter and the sand production degree in the measurement sample data set includes: according to the formula The correlation coefficient value between each key characteristic parameter and the degree of sand production in the measured sample data set; Key characteristic factors The degree of sand production The correlation coefficient value between is the covariance between the key characteristic factors and the sand production degree, is the observed value of the key characteristic factor in the i-th sample, is the observed value of the sand production degree in the i-th sample, is the average value of the key feature factors in the sample data set, is the average value of sand production degree in the sample data set, n is the number of gas wells in the sample data set, is the standard deviation of the key characteristic factor, is the standard deviation of sand production.

[0009] In a possible implementation, the samples contained in the current node are divided into two subsets according to a preset partitioning threshold, including: dividing the samples contained in the current node into two subsets according to whether the value of the characteristic factor is the preset partitioning threshold; wherein the first subset includes all samples whose value on the characteristic factor is the preset partitioning threshold, and the second subset includes the remaining samples.

[0010] In a possible implementation, the method of comparing the critical Gini indexes after division of different feature factors and selecting the feature factor that minimizes the critical Gini index after division as the optimal division feature factor includes: calculating the Gini indexes of the first subset and the second subset, and comparing the critical Gini indexes after division of different feature factors, and selecting the feature factor that minimizes the critical Gini index after division as the optimal division feature factor.

[0011] In one possible implementation, the critical Gini index is calculated as: ;in, is the critical Gini index, indicating the characteristic factor Use preset segmentation thresholds After subset division, an indicator to measure the classification uncertainty of the filtered data set is is the total sample size, is the size of the first subset, is the size of the second subset, is the Gini index of the first subset, is the Gini index of the second subset.

[0012] In a possible implementation, multiple decision trees are constructed for each training data set according to the optimal partitioning feature factors to form a classification model, including: in the process of building the decision tree, each node will be divided according to the current optimal partitioning feature factor as the partitioning standard, and the samples in the node will be divided into different sub-nodes according to different values ​​of the optimal partitioning feature factor until the stopping condition is met; the stopping condition is that the number of samples in the node reaches a preset threshold; on each training data set, the random forest classification algorithm is used to independently train the decision tree, and for each training data set, a random forest containing multiple decision trees is generated; for each test sample, it is input into all the generated random forests for prediction, and each decision tree in each random forest will give a prediction result; the number of votes obtained for each sand production degree category in the prediction result is counted, and the decision trees with more votes are retained, and they are combined to form a final random forest composed of multiple decision trees, and the final random forest is used as the classification model.

[0013] In a second aspect, an embodiment of the present application provides a device for predicting the degree of sand production in ultra-deep gas wells, including: a module for obtaining a sand production degree category, which is used to comprehensively consider wellbore sand samples, equipment wear and production parameter fluctuations, and classify the gas wells in the study area according to the degree of sand production to obtain a sand production degree category; wherein the sand production degree categories include no sand production, slight sand production, moderate sand production and severe sand production; a module for forming a sample data set, which is used to collect key characteristic parameters in each gas well, and combine the key characteristic parameters with the sand production degree category to form a sample data set; wherein the key characteristic parameters include characteristic factors in geology, completion and development; a module for forming a screening data set, which is used to measure the correlation coefficient value between each key characteristic parameter in the sample data set and the sand production degree, and select the characteristic factor with a correlation coefficient value greater than a preset value as the main control factor to form a screening data set; A training data set acquisition module is used to perform repeated random sampling with replacement on the screening data set to obtain multiple training data sets containing different samples; a partitioning module is used to randomly select characteristic factors for each training data set, and divide the samples contained in the current node into two subsets according to a preset partitioning threshold; a comparison module is used to compare the critical Gini index after partitioning with different characteristic factors, and select the characteristic factor that minimizes the critical Gini index after partitioning as the optimal partitioning characteristic factor; a construction module is used to construct multiple decision trees for each training data set according to the optimal partitioning characteristic factor to form a classification model; wherein the classification model predicts the sand production degree of the gas well based on the input main controlling factors; an evaluation module is used to evaluate the performance of the classification model, and adjust the parameters of the classification model according to the evaluation results to optimize the performance of the classification model.

[0014] In a third aspect, an embodiment of the present application provides an ultra-deep gas well sand production prediction server, comprising a memory and a processor; the memory is used to store computer-executable instructions; the processor is used to execute the computer-executable instructions to implement the method described in the first aspect or any possible implementation method of the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores executable instructions, and when a computer executes the executable instructions, the method described in the first aspect or any possible implementation method of the first aspect can be implemented.

[0016] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects: The embodiment of the present application provides a method for predicting the sanding degree of ultra-deep gas wells. By comprehensively classifying the sanding degree according to the sand samples of the wellbore, the wear of the equipment and the fluctuation of the production parameters, the sanding degree of the gas wells in the study area can be scientifically and accurately classified, which provides a solid foundation for the subsequent prediction work. By measuring the correlation coefficient value between each key characteristic parameter and the sanding degree in the sample data set, and selecting the characteristic factor with a correlation coefficient value greater than the preset value as the main control factor, the characteristic factor with a significant impact on the sanding degree can be effectively screened out. The repeated random sampling method with replacement is used to obtain multiple training data sets containing different samples, which can make full use of existing data resources and increase the diversity of the classification model. At the same time, the random selection of characteristic factors for sample division further enhances the generalization ability of the classification model, so that it can maintain stable prediction performance when facing new data. By comparing the critical Gini index after the division of different characteristic factors, the characteristic factor that minimizes the critical Gini index after division is selected as the optimal division characteristic factor, a decision tree with a more reasonable structure and better prediction performance can be constructed. The combination of multiple decision trees further improves the stability and accuracy of the classification model. Based on the input of the main control factors, the classification model constructed in this application can efficiently and accurately predict the sand production degree of gas wells. By evaluating the performance of the model and adjusting the parameters according to the evaluation results, the prediction accuracy and efficiency of the model can be continuously optimized, providing strong technical support for the exploitation and management of gas wells. It solves the problem of how to overcome the limitations of existing sand production prediction methods and improve prediction accuracy and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments of the present application or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flow chart of a method for predicting the degree of sand production in ultra-deep gas wells provided in an embodiment of the present application; Figure 2 A specific flow chart of constructing multiple decision trees for each training data set according to the optimal division feature factors to form a classification model provided in an embodiment of the present application; Figure 3 A schematic diagram of ranking the importance of characteristic factors according to the calculated correlation coefficient values ​​provided in an embodiment of the present application; Figure 4 A schematic diagram of the final random forest provided in the embodiment of the present application; Figure 5A schematic diagram of a device for predicting the degree of sand production in ultra-deep gas wells provided in an embodiment of the present application; Figure 6 A schematic diagram of an ultra-deep gas well sand production prediction server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0020] The following describes some of the techniques involved in the embodiments of the present application to facilitate understanding, and they should be considered as merely exemplary. Therefore, it should be appreciated by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, some descriptions of well-known functions and structures are omitted in the following description.

[0021] The present application embodiment provides a method for predicting the degree of sand production in ultra-deep gas wells. Figure 1 As shown, the method includes steps S101 to S108. Among them, Figure 1 This is only an execution order shown in the embodiment of the present application, and does not represent the only execution order of a method for predicting the degree of sand production in ultra-deep gas wells. Figure 1 The steps shown may be performed in parallel or reversed.

[0022] S101: Based on the wellbore sand samples, equipment wear and production parameter fluctuations, the gas wells in the study area are divided according to the degree of sand production to obtain the sand production degree categories, which include no sand production, slight sand production, moderate sand production and severe sand production.

[0023] Specifically, the research area in the embodiment of the present application is located in Baicheng County, Aksu Prefecture, Xinjiang Uygur Autonomous Region, covering Keshen and Bozi Dabei gas fields. The reservoirs of these two gas fields are deeply buried in the Kelasu tectonic belt of the Kuche depression in the northern part of the Tarim Basin, specifically located in the Keshen fault and Baicheng fault area. The main gas-producing stratum is the Cretaceous Bashijiqik Formation, with a vertical depth of 7000 to 8000 meters, and the formation pressure is in the high pressure range of 116 to 136MPa. During the sedimentary period, the Cretaceous Bashijiqik Formation formed a fan or braided delta sedimentary system, which is characterized by multi-stage river channel sand bodies superimposed vertically, and multiple fan bodies are connected to each other in the plane. The rock types thus formed are mainly medium and fine-grained lithic feldspar sandstone and feldspar lithic sandstone. The sedimentary facies distribution of these rock types in each lithological segment is relatively stable, and together constitute a large-scale and uniformly connected high-quality reservoir. However, the reservoir has a low porosity of only 4% to 8%, and the permeability is also in the extremely low range of 0.01 to 0.1 mD. The geological conditions of this type of ultra-deep, high-pressure, fractured, low-porosity sandstone gas reservoirs in the Tarim Basin are significantly different from those in other basins around the world. Affected by multiple periods of complex tectonic movements, the microcracks and faults in the Kelasu tectonic belt are complex, and the reservoir space types are also diverse. This complex geological and engineering condition has led to an extremely large number of geological, completion and development factors that affect the sand production of gas wells. The traditional single factor analysis method seems to be powerless in the face of this situation, and it is difficult to accurately identify the main controlling factors and their combined impact on the different sand production degrees of gas wells. Therefore, this application has made a detailed classification of the 93 gas wells in the study area according to the degree of sand production by comprehensively considering the composition of the wellbore sand sample, the degree of nozzle wear, and the fluctuation amplitude of oil pressure / production, and clarified the four sand production degree categories of no sand production, mild sand production, moderate sand production, and severe sand production.

[0024] Based on the wellbore sand samples, equipment wear and production parameter fluctuations, the gas wells in the study area are divided according to the degree of sand production. Before obtaining the sand production degree category, it also includes: determining the reservoir location and rock type in the study area.

[0025] Specifically, when determining the location of the reservoir in the study area, it is necessary to conduct a detailed investigation and analysis of the geological background of the study area. Taking the Kelasu tectonic belt in the Tarim Basin as an example, it is located in Baicheng County, Aksu Prefecture, Xinjiang Uygur Autonomous Region. The reservoir is located in the Keshen Fault and Baicheng Fault Zone of the Kelasu structure in the Kuche Depression in the northern Tarim Basin. Through geological exploration and logging data, the vertical depth range of the reservoir is determined. For example, the vertical depth of the Cretaceous Bashijiqik Formation can reach 7,000 to 8,000 meters. At the same time, the main gas-producing strata are clarified to provide a geological basis for the subsequent classification and prediction of the sand production degree. When determining the rock type, the rock type of the reservoir is determined by core observation, thin section identification and X-ray diffraction analysis of the coring well. In the Kelasu tectonic belt of the Tarim Basin, the reservoir rock types are mainly medium- and fine-grained lithic feldspar sandstone and feldspar lithic sandstone. Further analysis of the physical properties of the rock, such as porosity, permeability, and gas saturation, as well as mechanical properties of the rock, such as Poisson's ratio and elastic modulus, are crucial to understanding the sand production mechanism of the reservoir. The depth, stratigraphic position, and tectonic background of the reservoir directly affect its stress state and rock integrity. For example, in an ultra-deep and high-pressure environment, the reservoir rock may be subjected to strong compaction and tectonic stress, resulting in reduced stability of the rock skeleton and easy sand production. The physical and mechanical properties of different rock types vary greatly, and their effects on the degree of sand production are also different. For example, rocks with a low cement content are more likely to break and produce sand when subjected to external forces. After determining the reservoir location and rock type in the study area, the sand production mechanism of the reservoir can be analyzed more accurately, providing a reliable geological basis for the subsequent classification and prediction of the degree of sand production.

[0026] S102: Collect key characteristic parameters of each gas well, and combine the key characteristic parameters with the sand production degree category to form a sample data set. The key characteristic parameters include characteristic factors in geology, completion and development.

[0027] Table 1 is a sample data set of an embodiment of the present application. As shown in Table 1, the characteristic factors in geology include porosity, permeability, gas saturation, original formation pressure, natural fracture density, natural fracture length, vertical principal stress, vertical effective stress, rock Poisson's ratio, rock elastic modulus and cement content, a total of 11 characteristic factors. These factors provide important information about the physical properties of the reservoir and the mechanical properties of the rock. The characteristic factors in completion include completion methods (such as open hole completion or casing perforation completion), reservoir transformation methods (such as acid fracturing or SRV volume fracturing), perforation density and reservoir opening degree, a total of 4 characteristic factors. These factors reflect the connection between the wellbore and the reservoir, and the impact of reservoir transformation on wellbore stability. The characteristic factors in development include daily gas production, production pressure difference, daily water production, unimpeded flow and production time, a total of 5 characteristic factors. These factors provide important clues about the production dynamics of gas wells and fluid flow characteristics. The 20 characteristic factors of each gas well are combined with the sand production degree category to form a sample data set containing 93 samples.

[0028] Table 1

[0029] S103: measuring the correlation coefficient between each key characteristic parameter in the sample data set and the sand production degree, selecting characteristic factors with correlation coefficient values ​​greater than a preset value as main control factors, and forming a screening data set.

[0030] Measuring the correlation coefficient between each key characteristic parameter and the degree of sand production in the sample data set, including: According to the formula The correlation coefficient between each key characteristic parameter and the degree of sand production in the measured sample data set. Key characteristic factors The degree of sand production The correlation coefficient value between is the covariance between the key characteristic factors and the sand production degree, is the observed value of the key characteristic factor in the i-th sample, is the observed value of the sand production degree in the i-th sample, is the average value of the key feature factors in the sample data set, is the average value of sand production degree in the sample data set, n is the number of gas wells in the sample data set, is the standard deviation of the key characteristic factor, is the standard deviation of sand production.

[0031] Specifically, in order to quantify the degree of association between each key characteristic parameter and the degree of sand production in the sample data set, we used the statistical indicator of the correlation coefficient value. The correlation coefficient value is a value between -1 and 1, which is used to measure the degree of linear correlation between two variables. When the correlation coefficient is 1, it means that there is a complete positive linear correlation between the two variables; when the correlation coefficient is -1, it means that there is a complete negative linear correlation between the two variables; and when the correlation coefficient is 0, it means that there is no linear correlation between the two variables. Through calculation, the correlation coefficient value between each key characteristic parameter and the degree of sand production can be obtained. In order to screen out the main control factors that have a significant impact on the degree of sand production, a preset value is set, and the correlation coefficient value between each key characteristic parameter and the degree of sand production is compared with this preset value. If the correlation coefficient value is greater than the preset value, it is considered that this characteristic factor has a significant impact on the degree of sand production and is used as the main control factor. All the screened main control factors are combined to form a screening data set. This screening data set contains all the characteristic factors that have a significant impact on the degree of sand production.

[0032] Table 2 shows the correlation coefficient between each key characteristic parameter and the degree of sand production in the sample data set calculated in the embodiment of the present application. The geological factors in Table 2 are characteristic factors in geology, the completion factors are characteristic factors in completion, and the development factors are characteristic factors in development.

[0033] Table 2

[0034] Figure 3A schematic diagram of sorting the importance of characteristic factors according to the calculated correlation coefficient values ​​provided for the embodiment of the present application. The sorting results show that the main controlling factors of the sand production degree of ultra-deep high-pressure gas wells mainly include 8 characteristic factors in the three categories of geology, completion and development, which are: cement content, production pressure difference, reservoir opening degree, production time, rock elastic modulus, reservoir transformation method, perforation density and natural fracture density. The absolute values ​​of the correlation coefficients between these characteristic uniform velocities and the degree of sand production are all greater than 0.2. Since the absolute values ​​of the correlation coefficients are used to calculate the importance of characteristic variables, it shows that they have a significant impact on the degree of sand production. Therefore, the preset value in this application can be 0.2, and of course, it can also be other values. This application is not limited to the above values. The characteristic factors in geology are the internal reasons for the difference in the degree of sand production. Among them, the influence of cement content is the highest. The reservoir cement types of ultra-deep high-pressure fractured low-porosity sandstone gas reservoirs in the Tarim Basin are mainly calcite and dolomite. If the natural micro-fractures are well developed, and the cementation degree and diagenetic strength of the particles are low, the rock skeleton particles will be subjected to greater compressive stress, thereby increasing the risk of near-wellbore formation collapse and sand production. Rock elastic modulus and natural fracture density are also important geological characteristic factors, and their influence on the degree of sand production cannot be ignored. The characteristic factors in completion, namely the degree of completion transformation, have a certain influence on the difference in sand production. Among them, the degree of reservoir opening has a greater impact. The micro-fractures in the reservoir in the study area are well developed, and the fracture network system formed by a large degree of fracturing transformation will further destroy the mechanical equilibrium environment of the reservoir. If the reservoir opening degree and perforation density are low, it will cause a high gas flow rate, thereby causing shear damage to the rock skeleton, which is more likely to cause sand production in the gas well. Reservoir transformation method and perforation density are also important completion characteristic factors. The characteristic factors in development are external boosters that cause differences in the degree of sand production. Among them, the production pressure difference has the highest degree of influence. Excessive production pressure difference of gas wells can easily cause unfavorable factors such as proppant breakage in the fractures near the wellbore area, migration of formation particles, and premature water breakthrough, all of which will increase the trend of sand production in gas wells. As production time goes by, these unfavorable factors will further aggravate the sand production problem. Therefore, production time and production pressure difference are important development characteristic factors.

[0035] S104: Perform repeated random sampling with replacement on the screening data set to obtain multiple training data sets containing different samples.

[0036] Specifically, this process is achieved by adopting the bootstrap aggregating algorithm (English: Bootstrap aggregating, referred to as: Bagging). Specifically, it involves generating multiple training data sets from the screening data set through repeated random sampling with replacement. Each training data set contains some samples in the original data set, but because it is a sampling with replacement, the same sample may appear multiple times in the same training data set, or it may not appear in a training data set at all. In this way, each training data set has a certain difference and diversity, so that multiple decision trees with different decision boundaries and prediction capabilities can be trained. The advantage of this method is that it can make full use of the information of the original data set and improve the stability and accuracy of the classification model by constructing multiple decision trees. At the same time, since each training data set is generated independently, model training can be performed in parallel, thereby improving computational efficiency.

[0037] S105: For each training data set, a feature factor is randomly selected, and the samples included in the current node are divided into two subsets according to a preset division threshold.

[0038] Dividing the samples included in the current node into two subsets according to a preset partitioning threshold includes: dividing the samples included in the current node into two subsets according to whether the value of the characteristic factor is the preset partitioning threshold. The first subset includes all samples whose values ​​on the characteristic factor are the preset partitioning threshold, and the second subset includes the remaining samples.

[0039] Specifically, for the selected characteristic factors, a series of candidate partitioning thresholds need to be determined. These candidate thresholds can be any values ​​within the range of the characteristic factor values, but in order to improve efficiency and accuracy, some strategies are usually adopted to reduce the number of candidate thresholds. For example, analyze the distribution of the characteristic factor in the training data set, including its minimum value, maximum value, average value, and possible distribution forms, such as normal distribution, skewed distribution, etc. According to the results of the data distribution analysis, set a reasonable preset partitioning threshold. The preset partitioning threshold can be determined based on statistics such as the mean, median, quartiles, etc. of the data, or it can be set based on experience or professional knowledge.

[0040] S106: By comparing the critical Gini indexes after different feature factors are divided, the feature factor that minimizes the critical Gini index after division is selected as the optimal division feature factor.

[0041] By comparing the critical Gini indexes after different feature factors are divided, the feature factor that minimizes the critical Gini index after division is selected as the optimal division feature factor, including: by calculating the Gini index of the first subset and the second subset, and comparing the critical Gini indexes after different feature factors are divided, the feature factor that minimizes the critical Gini index after division is selected as the optimal division feature factor.

[0042] The critical Gini index is calculated as: .in, is the critical Gini index, indicating the characteristic factor Use preset segmentation thresholds After subset division, an indicator to measure the classification uncertainty of the filtered data set is is the total sample size, is the size of the first subset, is the size of the second subset, is the Gini index of the first subset, is the Gini index of the second subset.

[0043] Specifically, when constructing a decision tree, it is necessary to compare the critical Gini index of different feature factors after division. The feature factor that minimizes the critical Gini index after division is selected as the optimal division feature factor, because this indicates that the feature factor can minimize the uncertainty of sample classification.

[0044] The Gini index of the first subset is calculated as: .in, is the Gini index of the first subset, is the total number of categories in the first subset, is the number of samples of the kth class in the first subset. Similarly, the Gini index of the second subset is calculated as: .in, is the Gini index of the second subset, is the total number of categories in the second subset, is the number of samples of the kth class in the second subset.

[0045] Specifically, by calculating the critical Gini index after division of different feature factors and selecting the feature factor that minimizes the critical Gini index after division as the optimal division feature factor, the uncertainty of sample classification can be minimized, thereby constructing a more accurate and efficient decision tree model.

[0046] S107: construct multiple decision trees for each training data set according to the optimal division feature factors to form a classification model, wherein the classification model predicts the sand production degree of the gas well based on the input main control factors.

[0047] Figure 2 The specific flow chart of the classification model is as follows: Figure 2 As shown, it includes steps S201 to S204.

[0048] S201: During the construction of the decision tree, each node is divided according to the current optimal division feature factor as the division standard, and the samples in the node are divided into different sub-nodes according to different values ​​of the optimal division feature factor until the stop condition is met. The stop condition is that the number of samples in the node reaches a preset threshold.

[0049] Specifically, at each node of the decision tree, the samples will be divided according to the optimal partitioning feature factor of the current node, and the samples in the node will be assigned to different child nodes according to the different values ​​of the optimal partitioning feature factor. This process will be recursively performed until the stop condition is met. The construction process of the decision tree will not go on indefinitely, but will stop when the stop condition is met. In the present application, the set stop condition is that the number of samples in the node reaches a preset threshold. When the number of samples in a node reaches a preset threshold, the node will no longer continue to be divided, but will become a leaf node. The leaf node is the terminal node of the decision tree and is no longer divided. At the leaf node, the predicted category of the node will be determined based on the sand production degree of the samples in the node. Usually, the predicted category of the leaf node is the sand production degree category with the largest number of samples in the node. In order to improve the stability and accuracy of the classification model, multiple decision trees will be repeatedly constructed based on different training data sets and optimal partitioning feature factors. These decision trees will together form a classification model.

[0050] S202: On each training data set, a random forest classification algorithm is used to independently train a decision tree, and for each training data set, a random forest containing multiple decision trees is generated.

[0051] Specifically, in the above steps, multiple training data sets based on the optimal partitioning feature factors have been obtained. Each training data set contains the main controlling factors affecting the sand production degree of gas wells and the corresponding sand production degree categories. The core idea of ​​the random forest algorithm is to randomly select features from the original feature space and build a decision tree on each training subset, and finally generate a random forest containing multiple decision trees.

[0052] S203: For each test sample, it is input into all generated random forests for prediction, and each decision tree in each random forest will give a prediction result.

[0053] S204: Count the number of votes obtained for each sand production degree category in the prediction results, retain the decision trees with more votes, and combine them to form a final random forest composed of multiple decision trees, and use the final random forest as a classification model.

[0054] Specifically, for the prediction result of each test sample, the number of votes obtained for each sanding degree category is counted. The number of votes refers to how many decision trees classify it as the sanding degree category in the prediction of the test sample. For each sanding degree category, find the decision trees that get the most votes on that category. A threshold can be set to only retain decision trees with votes exceeding the threshold, or to retain the top K decision trees with the highest votes in each category, where K is a preset value. All the retained decision trees are combined to form a final random forest consisting of multiple decision trees. This final random forest is screened based on the voting results, so it is more likely to contain decision trees that are helpful for the classification task, and the final random forest is used as the classification model. When predicting, the new sample can be input into each decision tree in the classification model, and then the votes for each sanding degree category are counted, and the category with the most votes is selected as the final prediction result.

[0055] Figure 4 This is a schematic diagram of the final random forest provided in the embodiment of the present application. Figure 4As shown in the figure, the root node of the decision tree is the production pressure difference. The decision tree comprehensively considers 8 main control factors affecting sand production, including cement content, production pressure difference, reservoir opening degree, production time, rock elastic modulus, reservoir transformation method, perforation density and natural fracture density. It is worth noting that some characteristic factors are selected multiple times in the decision process. The "samples" in the decision tree indicates the number of samples participating in the current classification step, "value" shows the sample proportion of the four sand production categories (no sand production, slight sand production, moderate sand production, and severe sand production), while "class" gives the classification results, and "gini" is the critical Gini index. The overall depth of this decision tree is 9, including 16 non-leaf nodes (each node represents a characteristic factor) and 17 leaf nodes (each leaf node represents the discriminant value of a target variable). These 17 leaf nodes correspond to 17 decision branches, among which the number of branches for no sand production, slight sand production, moderate sand production and severe sand production are 3, 3, 5 and 6 respectively. The final random forest constructed in this application can not only accurately classify and predict the degree of sand production in newly commissioned wells, but also quickly assist in determining the degree of sand production in suspected sand-producing gas wells. If conditions permit, the prediction results can be further corrected in combination with online monitoring of sand production, nozzle wear, or analytical test results of sand samples, so as to obtain a more accurate sample data set of the degree of sand production in gas wells. However, when actually classifying and predicting the degree of sand production in ultra-deep and high-pressure gas wells, it is necessary to pay attention to the reliability of different parameter sources and fully consider the influence of various characteristic factors. To avoid oversimplification (underfitting) or complication (overfitting) of decision trees, the mine development experience, indoor laboratory tests, and theoretical calculation results should be comprehensively considered. On the basis of clarifying the main controlling factors of the degree of sand production in gas wells, a method for predicting the degree of sand production in ultra-deep and high-pressure gas wells that is convenient to operate, efficient, and has practical value should be formed with the help of random forest machine learning methods.

[0056] S108: Evaluate the performance of the classification model, and adjust the parameters of the classification model according to the evaluation results to optimize the performance of the classification model.

[0057] In order to verify the advantages of the classification model in the classification prediction of the sand production degree of ultra-deep high-pressure gas wells, this application conducted a reliability verification experiment. This experiment evaluated the effectiveness of random forest in dealing with such complex classification problems by comparing the performance of two classification prediction models, decision tree and random forest. The random forest model is the classification model mentioned above in this application.

[0058] The screening data set is artificially divided into a training set and a test set to evaluate the predictive ability of the model. The test set ratio is set to 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8 and 0.9 nine schemes to comprehensively examine the performance of the model under different data set division conditions. Both classification prediction models select the Gini index as the basis for division to ensure the consistency and fairness of the experiment. The decision tree model is used as a benchmark model for comparison with the random forest model, i.e., the classification model mentioned above in this application. The decision tree and random forest are respectively trained by machine learning using the training data set to obtain the classification prediction model and related parameters of the sand production degree of the two classification algorithms.

[0059] Table 3 is a comparison of parameters and prediction accuracy of two classification prediction models under different data set partitioning conditions provided by the embodiment of the present application. It should be noted that the random forest model in Table 3 is the classification model mentioned above in this application. Under different test set proportions, the prediction accuracy of the random forest model, i.e., the classification model, is generally higher than that of the decision tree model, especially when the test set proportion is high (such as 0.6, 0.7, 0.8, 0.9), the prediction accuracy of the random forest model can still maintain a relatively high level, while the prediction accuracy of the decision tree model is significantly reduced. The random forest model shows advantages in the number of decision trees and the reuse rate of training samples. With the increase in the number of decision trees, the random forest model can make more full use of the training samples to form a more robust classification prediction model, thereby improving the accuracy of the classification results. The experimental results show that the classification model used in this application has obvious advantages for the classification problem of sand production degree in ultra-deep high-pressure gas wells. Regardless of whether the test set proportion is low or high, the classification model can maintain a relatively stable classification result, showing high applicability and reliability. Therefore, the classification model has broad application prospects in dealing with such complex classification problems.

[0060] Table 3

[0061] By collecting geological data, production history data and related experimental data of multiple gas wells in Keshen, Bozi and Dabei areas, and through in-depth analysis and processing of these data, the main controlling factors affecting the sand production degree of gas wells were extracted. The sand production time and degree of three gas wells in Keshen, 10 gas wells in Bozi and two gas wells in Dabei under the maximum production pressure difference were accurately predicted. This application evaluates the sand production risk of these gas wells through a classification model and gives the corresponding prediction results. Table 4 shows the eight main controlling factors of sand production degree corresponding to the 15 gas wells provided in the embodiment of this application and the prediction results of their classification model, covering the sand production of multiple gas wells. As shown in Table 4, a total of 8 gas wells are predicted to be able to maintain a stable production state under the maximum production pressure difference, and no obvious sand production will occur. Two gas wells are predicted to have slight sand production under certain conditions, but the overall impact is small. The classification model also predicts that 3 gas wells may have more obvious sand production under the maximum production pressure difference, and corresponding measures need to be taken to reduce the risk of sand production. For example, the risk of sand production can be reduced by adjusting the production pressure difference, optimizing production parameters, etc. Two gas wells are predicted to have serious sand production under the maximum production pressure difference, which will have a serious impact on production. More stringent measures need to be taken to manage and control the risk of sand production. For example, it is possible to consider using sand control technology, installing sand control equipment, etc. to reduce the impact of sand production on production. Through practical application, the classification model of this application has shown high accuracy and reliability. This model not only provides strong support for the prediction of sand production risks in gas wells, but also helps decision makers to better formulate production plans and management strategies. At the same time, the model also provides useful reference and guidance for the sand control work of gas wells, which helps to ensure the safe production of gas wells. The classification model of this application has played an important role in the prediction and management of sand production in gas wells in the Kelasu fractured low-porosity sandstone gas reservoir, and has provided strong technical support for the development and production of oil and gas fields.

[0062] Table 4

[0063] The present application also provides a device 500 for predicting the degree of sand production in ultra-deep gas wells. Figure 5 As shown, the device includes: a sand production degree classification module 501, a sample data set formation module 502, a screening data set formation module 503, a training data set acquisition module 504, a division module 505, a comparison module 506, a construction module 507 and an evaluation module 508.

[0064] The sand production degree classification module 501 is used to classify the gas wells in the study area according to the sand production degree based on the wellbore sand sample, equipment wear and production parameter fluctuations, and obtain the sand production degree classification, wherein the sand production degree classification includes no sand production, slight sand production, moderate sand production and severe sand production.

[0065] The sample data set generation module 502 is used to collect key characteristic parameters of each gas well, and combine the key characteristic parameters with the sand production degree category to form a sample data set. The key characteristic parameters include characteristic factors in geology, completion and development.

[0066] The screening data set forming module 503 is used to measure the correlation coefficient value between each key characteristic parameter in the sample data set and the sand production degree, select the characteristic factor with a correlation coefficient value greater than a preset value as the main control factor, and form the screening data set.

[0067] The module 504 for obtaining training data sets is used to perform repeated random sampling with replacement on the screening data set to obtain a plurality of training data sets containing different samples.

[0068] The partitioning module 505 is used to randomly select characteristic factors for each training data set, and to partition the samples included in the current node into two subsets according to a preset partitioning threshold.

[0069] The comparison module 506 is used to compare the critical Gini indexes after the division of different feature factors, and select the feature factor that minimizes the critical Gini index after the division as the optimal division feature factor.

[0070] The construction module 507 is used to construct multiple decision trees for each training data set according to the optimal division characteristic factors to form a classification model. The classification model predicts the sand production degree of the gas well based on the input main control factors.

[0071] The evaluation module 508 is used to evaluate the performance of the classification model and adjust the parameters of the classification model according to the evaluation results to optimize the performance of the classification model.

[0072] Some modules in the apparatus described in the present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0073] The devices or modules described in the above application embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described in various modules according to their functions. When implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware. Of course, the module that implements a certain function can also be implemented by combining multiple sub-modules or sub-units.

[0074] The methods, devices or modules described in this application can be implemented in the form of computer-readable program codes. The controller can be implemented in any appropriate manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program codes (such as software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (English: Application Specific Integrated Circuit, referred to as: ASIC), programmable logic controllers and embedded microcontrollers. Examples of controllers include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program codes, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included in it for implementing various functions can also be regarded as structures within the hardware component. Or even, the means for realizing various functions may be regarded as both a software module for realizing the method and a structure within a hardware component.

[0075] like Figure 6 As shown, an embodiment of the present application also provides an ultra-deep gas well sand production degree prediction server, including a memory 601 and a processor 602; the memory 601 is used to store computer executable instructions; the processor 602 is used to execute computer executable instructions to implement an ultra-deep gas well sand production degree prediction method described above in the embodiment of the present application.

[0076] The embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores executable instructions. When a computer executes the executable instructions, the method for predicting the degree of sand production in an ultra-deep gas well described in the embodiment of the present application can be implemented.

[0077] It can be seen from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, or it can be reflected in the implementation process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the method described in the embodiment of the present application.

[0078] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. All or part of this application can be used in many general or special computer system environments or configurations.

[0079] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present application.

Claims

1. A method for predicting the degree of sand production in ultra-deep gas wells, characterized in that: include: Based on the sand samples of wellbore, equipment wear and fluctuation of production parameters, the gas wells in the study area are divided according to the degree of sand production to obtain the sand production degree categories; among them, the sand production degree categories include no sand production, slight sand production, moderate sand production and severe sand production; Collect key characteristic parameters of each gas well, and combine the key characteristic parameters with the sand production degree category to form a sample data set; wherein the key characteristic parameters include characteristic factors in geology, completion and development; Measure the correlation coefficient value between each key characteristic parameter and the sand production degree in the sample data set, select the characteristic factors with correlation coefficient values ​​greater than the preset values ​​as the main control factors, and form a screening data set; Repeated random sampling with replacement is performed on the screening data set to obtain multiple training data sets containing different samples; For each training data set, feature factors are randomly selected, and the samples contained in the current node are divided into two subsets according to the preset partition threshold; By comparing the critical Gini indexes after different feature factors are divided, the feature factor that minimizes the critical Gini index after division is selected as the optimal division feature factor; According to the optimal division feature factors, multiple decision trees are constructed for each training data set to form a classification model; wherein the classification model predicts the sand production degree of the gas well based on the input main control factors; The performance of the classification model is evaluated, and the parameters of the classification model are adjusted according to the evaluation results to optimize the performance of the classification model.

2. The method for predicting the degree of sand production in ultra-deep gas wells according to claim 1, characterized in that: The comprehensive wellbore sand samples, equipment wear and production parameter fluctuations are used to classify the gas wells in the study area according to the sand production degree. Before obtaining the sand production degree category, it also includes: determining the reservoir location and rock type in the study area.

3. The method for predicting the degree of sand production in ultra-deep gas wells according to claim 1, characterized in that: The correlation coefficient value between each key characteristic parameter and the sand production degree in the measurement sample data set includes: According to the formula The correlation coefficient value between each key characteristic parameter and the degree of sand production in the measured sample data set; Key characteristic factors The degree of sand production The correlation coefficient value between is the covariance between the key characteristic factors and the sand production degree, is the observed value of the key characteristic factor in the i-th sample, is the observed value of the sand production degree in the i-th sample, is the average value of the key feature factors in the sample data set, is the average value of sand production degree in the sample data set, n is the number of gas wells in the sample data set, is the standard deviation of the key characteristic factor, is the standard deviation of sand production.

4. The method for predicting the degree of sand production in ultra-deep gas wells according to claim 1, characterized in that: The dividing the samples contained in the current node into two subsets according to the preset division threshold comprises: According to whether the value of the characteristic factor is a preset division threshold, the samples contained in the current node are divided into two subsets; wherein the first subset includes all samples whose value on the characteristic factor is the preset division threshold, and the second subset includes the remaining samples.

5. The method for predicting the degree of sand production in ultra-deep gas wells according to claim 4, characterized in that: The method of comparing the critical Gini indexes after division of different characteristic factors and selecting the characteristic factor with the smallest critical Gini index after division as the optimal division characteristic factor includes: By calculating the Gini index of the first subset and the second subset and comparing the critical Gini index sizes after division by different feature factors, the feature factor that minimizes the critical Gini index after division is selected as the optimal division feature factor.

6. The method for predicting the degree of sand production in ultra-deep gas wells according to claim 5, characterized in that: The critical Gini index is calculated as: ;in, is the critical Gini index, indicating the characteristic factor Use preset segmentation thresholds After subset division, an indicator to measure the classification uncertainty of the filtered data set is is the total sample size, is the size of the first subset, is the size of the second subset, is the Gini index of the first subset, is the Gini index of the second subset.

7. The method for predicting the degree of sand production in ultra-deep gas wells according to claim 1, characterized in that: The method of constructing multiple decision trees for each training data set according to the optimal partitioning feature factors to form a classification model includes: In the process of building a decision tree, each node will be divided according to the current optimal division feature factor as the division standard, and the samples in the node will be divided into different sub-nodes according to the different values ​​of the optimal division feature factor until the stopping condition is met; the stopping condition is that the number of samples in the node reaches a preset threshold; On each training data set, the decision tree is trained independently using the random forest classification algorithm. For each training data set, a random forest containing multiple decision trees is generated; For each test sample, it is input into all generated random forests for prediction, and each decision tree in each random forest will give a prediction result; The number of votes for each sand production degree category in the prediction results is counted, and the decision trees with the most votes are retained. They are combined to form a final random forest composed of multiple decision trees, and the final random forest is used as the classification model.

8. An ultra-deep gas well sand production prediction device, characterized in that: include: The module for obtaining sand production degree classification is used to classify the gas wells in the study area according to the sand production degree based on the wellbore sand samples, equipment wear and production parameter fluctuations, and obtain the sand production degree categories; among them, the sand production degree categories include no sand production, slight sand production, moderate sand production and severe sand production; A sample data set forming module is used to collect key characteristic parameters in each gas well, and combine the key characteristic parameters with the sand production degree category to form a sample data set; wherein the key characteristic parameters include characteristic factors in geology, completion and development; A screening data set module is formed, which is used to measure the correlation coefficient value between each key characteristic parameter in the sample data set and the sand production degree, and select the characteristic factor with a correlation coefficient value greater than a preset value as the main control factor to form a screening data set; A training data set acquisition module is used to perform repeated random sampling with replacement on the screening data set to obtain multiple training data sets containing different samples; A partitioning module is used to randomly select feature factors for each training data set and divide the samples contained in the current node into two subsets according to a preset partitioning threshold; A comparison module, for comparing the critical Gini indexes after division of different characteristic factors, and selecting the characteristic factor with the smallest critical Gini index after division as the optimal division characteristic factor; A construction module is used to construct multiple decision trees for each training data set according to the optimal division feature factors to form a classification model; wherein the classification model predicts the sand production degree of the gas well based on the input main control factors; The evaluation module is used to evaluate the performance of the classification model and adjust the parameters of the classification model according to the evaluation results to optimize the performance of the classification model.

9. An ultra-deep gas well sand production prediction server, characterized in that: including memory and processor; The memory is used to store computer executable instructions; The processor is configured to execute the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores executable instructions, and when a computer executes the executable instructions, the method according to any one of claims 1 to 7 can be implemented.

Images