Oil reservoir shaft scaling prediction method, device and equipment and storage medium

Abstract

The invention provides an oil reservoir shaft scaling prediction method and device, equipment and a storage medium, and belongs to the technical field of oil field production flow guarantee. The method comprises the steps that characteristic parameters, related to shaft scaling, of all shaft sections are obtained from production field monitoring equipment; the obtained characteristic parameters are input into a constructed wellbore scaling prediction model, a prediction result is generated, the prediction result comprises whether each wellbore section is scaled or not and / or the scaling amount of each wellbore section, and the wellbore scaling prediction model is obtained after hyper-parameter accelerated optimization is conducted on an initially constructed machine learning model through an agent model. According to the method, three means are combined: characteristic parameters are acquired by using production field monitoring equipment, a shaft scaling prediction model based on a machine learning method and a hyper-parameter optimization method based on an agent model are adopted, so that oil reservoir shaft scaling prediction which is acceptable in cost and relatively high in prediction precision and prediction efficiency is realized; the method is especially suitable for the development process of hypersalinity oil reservoirs.

Description

Technical Field

[0001] This application belongs to the field of oilfield production flow support technology, specifically relating to a method for predicting scale buildup in oil reservoir wellbore, a device for predicting scale buildup in oil reservoir wellbore, a computer device, and a machine-readable storage medium. Background Technology

[0002] Wellbore scaling is one of the most common problems in the petroleum industry. It can occur in both production and injection wells, and its formation can severely impact the productivity of production wells or the injection capacity of injection wells. Evaporation, mixing with immiscible water, and changes in thermodynamic conditions can lead to the formation of mineral scale, such as carbonates, sulfates, and hydroxides. In many oilfields, scale formation due to various reasons has been widely reported. Scale deposits in the wellbore and reduce the efficiency of valves, production tubing, pumps, perforations, and downhole completion equipment.

[0003] In the development of high-salinity oil reservoirs, scaling is a common problem in both production and injection wells, severely impacting reservoir development. Wellbore scaling prediction helps to anticipate and guide the implementation of scale inhibition and prevention measures. Numerous factors influence wellbore scaling, including temperature, pressure, and water composition, and these factors interact with each other. Currently, commonly used scaling prediction methods include the saturation index equation method and mathematical modeling. Methods for predicting single-calcium carbonate scaling include the Langelinie saturation index method, the Davis-Stiff saturation index method, the Ryznar stability index method, and mathematical modeling methods, while methods for predicting mixed scaling include the Oddo-Tomson saturation index method. However, both the saturation index equation method and the mathematical modeling method have limited considerations, inconsistent calculation accuracy, and relatively cumbersome calculation processes, making them unsuitable for the needs of production sites. Summary of the Invention

[0004] The purpose of this application is to provide a method for predicting scale buildup in oil reservoir wellbore, a device for predicting scale buildup in oil reservoir wellbore, a computer device, and a machine-readable storage medium, in order to overcome at least one of the defects in existing methods for predicting scale buildup in oil reservoir wellbore based on the saturation index equation method and the thermodynamic solubility method.

[0005] To achieve the above objectives, the first aspect of this application provides a method for predicting scale formation in oil reservoir wellbore, comprising: Characteristic parameters related to wellbore scaling in each wellbore section were obtained from on-site monitoring equipment. The characteristic parameters are input into the constructed wellbore scaling prediction model to generate prediction results, which include whether each wellbore section is scaling and / or the amount of scaling in each wellbore section. The characteristic parameters include multiple of the following: well temperature, well pressure, wellhead production parameters, and water quality analysis data. The wellbore scaling prediction model is a model obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination. The optimal hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed surrogate model is met. The surrogate model takes the hyperparameter combination of the machine learning model as input and the prediction error of the machine learning model as output. In the optimization process of generating candidate sample points, the hyperparameter combination is used as the decision variable and the sampling strategy evaluation function is used as the fitness function.

[0006] In a specific embodiment of this application, the wellbore scaling prediction model includes: A single-layer classification model is used to predict whether each wellbore section will be scaled, using the aforementioned feature parameters as input. A two-layer prediction model is used to predict the amount of scaling in a specific wellbore section using specific feature parameters as input. The specific wellbore section is the wellbore section whose prediction result is scaling after being output by a one-layer classification model. The specific feature parameters are the feature parameters of the specific wellbore section.

[0007] In a specific embodiment of this application, there are multiple wellbore scaling prediction models, and the method further includes: For any well section, if the prediction result shows that the number of the layer classification models for the well section with scale reaches a first preset value, then the well section is determined to have scale. If the prediction result shows that the number of the layer classification models for the well section without scale reaches a second preset value, then the well section is determined to have no scale. For any wellbore section, the scale amounts predicted by each of the two-layer prediction models are weighted and summed, and the resulting weighted sum is taken as the final scale amount of the wellbore section. The weighting coefficients of each scale amount are the weighting coefficients that minimize the prediction error of the combined prediction model. The combined prediction model takes the feature parameters as input and the weighted sum of the scale amounts predicted by each of the two-layer prediction models within itself as output.

[0008] In a specific embodiment of this application, the wellbore scaling prediction model is obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination, including: Construct the first training set using historical feature parameters; After training the initial classification network using the first training set, the initial layer classification model in the machine learning model is established. The hyperparameter combination in the initial single-layer classification model is updated to the first optimal hyperparameter combination to obtain the optimized single-layer classification model. The first optimal hyperparameter combination is the best candidate sample point when the termination condition of iteratively updating the first surrogate model is met. The first surrogate model takes the hyperparameter combination of the single-layer classification model as input and the prediction error of the single-layer classification model as output. In the process of generating candidate sample points, the hyperparameter combination of the single-layer classification model is used as the decision variable and the evaluation function of the first sampling strategy is used as the fitness function. The historical feature parameters that predicted no scaling were removed from the first training set using the optimized one-layer classification model; the second training set was obtained. After training the initial prediction network using the second training set, the initial two-layer prediction model in the machine learning model is established. The hyperparameter combination in the initial two-layer prediction model is updated to the second optimal hyperparameter combination to obtain the optimized two-layer prediction model. The second hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed second surrogate model is met. The second surrogate model takes the hyperparameter combination of the two-layer prediction model as input and the prediction error of the two-layer prediction model as output. In the process of generating candidate sample points, the hyperparameter combination of the two-layer prediction model is used as the decision variable and the evaluation function of the second sampling strategy is used as the fitness function. The wellbore scaling prediction model consists of an optimized single-layer classification model and an optimized two-layer prediction model.

[0009] In specific embodiments of this application, the amount of scale in the well section includes the amount of scale of various scale types, including multiple types of scale such as SrCO3, NaCl, CaCO3, CaSO4·2H2O, CaSO4, SrSO4, and Mg(OH)2.

[0010] In a specific embodiment of this application, when the wellbore is a production well, the wellhead production parameters include oil production, water production, gas production, and CO2 content in the produced gas phase. The water quality analysis data includes the pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity of the produced water. When the wellbore is an injection well, the wellhead production parameters include at least the injection volume, and the water quality analysis data includes the pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity of the injected water.

[0011] In a specific embodiment of this application, the initially constructed proxy model is iteratively updated using the following preset optimization method to obtain the optimal hyperparameter combination: Generate candidate sample points; Determine whether the termination condition for iterative update has been met; if so, use the current best candidate sample point as the optimal hyperparameter combination; otherwise, use the candidate sample point as the input parameter and the prediction error of the machine learning model with the candidate sample point as the hyperparameter combination as the label value to construct a new sample point, and use the new sample point to update the surrogate model, and then jump to execute the previous step. The generation of candidate sample points includes: Population initialization is accomplished through Kent mapping; Calculate the fitness value of each individual in the population; Generate a new population and calculate the fitness value of each individual in the new population; Determine whether the termination condition for candidate sample point optimization has been met. If so, select the individual with the best current fitness as the candidate sample point; otherwise, proceed to the previous step. Among these, the generation of new populations includes: Update the positions of individuals in the current population; After updating the position, perform Gaussian mutation on the individuals, and form a new population from the individuals after Gaussian mutation.

[0012] In specific embodiments of this application, the types of machine learning models include random forest, extreme random tree, AdaBoost, XGBoost, and Bagging meta-estimator.

[0013] In a specific embodiment of this application, the first proxy model is an Extreme Learning Machine.

[0014] In a specific embodiment of this application, the second proxy model is an interpolation model based on the Cauchy RBF function.

[0015] In a specific embodiment of this application, Gaussian mutation is performed on the individuals after position updates using the following formula: ; in, This represents the decision variable elements in the individuals after Gaussian mutation. This represents the decision variable elements in the individual before Gaussian mutation. This represents a random number that conforms to a standard normal distribution.

[0016] A second aspect of this application provides a reservoir wellbore scaling prediction device, comprising: The acquisition module is used to acquire characteristic parameters related to wellbore scaling in each wellbore section from the production site monitoring equipment. The prediction module is used to input the feature parameters into the constructed wellbore scaling prediction model and generate prediction results, which include whether each wellbore section is scaling and / or the amount of scaling in each wellbore section. The characteristic parameters include multiple of the following: well temperature, well pressure, wellhead production parameters, and water quality analysis data. The wellbore scaling prediction model is a model obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination. The optimal hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed surrogate model is met. The surrogate model takes the hyperparameter combination of the machine learning model as input and the prediction error of the machine learning model as output. In the optimization process of generating candidate sample points, the hyperparameter combination is used as the decision variable and the sampling strategy evaluation function is used as the fitness function.

[0017] In a specific embodiment of this application, the initially constructed proxy model is iteratively updated using the following preset optimization method to obtain the optimal hyperparameter combination: Generate candidate sample points; Determine whether the termination condition for iterative update has been met; if so, use the current best candidate sample point as the optimal hyperparameter combination; otherwise, use the candidate sample point as the input parameter and the prediction error of the machine learning model with the candidate sample point as the hyperparameter combination as the label value to construct a new sample point, and use the new sample point to update the surrogate model, and then jump to execute the previous step. The generation of candidate sample points includes: Population initialization is accomplished through Kent mapping; Calculate the fitness value of each individual in the population; Generate a new population and calculate the fitness value of each individual in the new population; Determine whether the termination condition for candidate sample point optimization has been met. If so, select the individual with the best current fitness as the candidate sample point; otherwise, proceed to the previous step. Among these, the generation of new populations includes: Update the positions of individuals in the current population; After updating the position, perform Gaussian mutation on the individuals, and form a new population from the individuals after Gaussian mutation.

[0018] A third aspect of this application provides a computer device, comprising: The memory is configured to store instructions; and The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the reservoir wellbore scaling prediction method according to the first aspect of this application.

[0019] A fourth aspect of this application provides a machine-readable storage medium storing instructions for causing a machine to perform the reservoir wellbore scaling prediction method according to the first aspect of this application.

[0020] To comprehensively consider prediction cost, speed, and accuracy, the above technical solution combines three methods: collecting characteristic parameters using on-site monitoring equipment, a wellbore scaling prediction model based on machine learning, and a hyperparameter optimization method based on a surrogate model. This achieves wellbore scaling prediction in oil reservoirs with acceptable cost, high prediction accuracy, and high prediction efficiency, and is particularly suitable for the development of high-salinity oil reservoirs. Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 A schematic diagram illustrating the process of predicting scale buildup in oil reservoir wellbore according to an embodiment of this application is shown. Figure 2 The illustration shows a flowchart of the process for constructing a wellbore scaling prediction model according to an embodiment of this application; Figure 3 The schematic diagram illustrates a flowchart of a preset optimization method according to an embodiment of this application; Figure 4 This illustration schematically shows another process diagram of the reservoir wellbore scaling prediction method according to an embodiment of this application; Figure 5 This diagram illustrates the components of a high-mineralization wellbore scaling monitoring system. Figure 6 The schematic diagram illustrates the process of scaling prediction in a specific application; Figure 7 This diagram illustrates the prediction effect of wellbore scaling on the training set. Figure 8 This diagram illustrates the prediction effect of wellbore scaling on the test set. Figure 9 This illustration shows the scores of a single-layer classification model on the training and test sets. Figure 10 This illustration shows the scores of the two-layer prediction model on the training and test sets. Figure 11a The diagram illustrates the prediction performance of a two-layer prediction model based on random forest on the test set. Figure 11b This diagram illustrates the scaling prediction performance of a two-layer prediction model based on extreme random trees on the test set. Figure 11c This diagram illustrates the scaling prediction performance of a two-layer prediction model based on Bagging meta-estimation on the test set. Figure 11d The diagram illustrates the scaling prediction performance of the AdaBoost-based two-layer prediction model on the test set. Figure 11e The diagram illustrates the scaling prediction performance of the XGBoost-based two-layer prediction model on the test set. Figure 12a The diagram illustrates the scaling prediction performance of a two-layer prediction model based on random forest on the training set. Figure 12b The diagram illustrates the scaling prediction performance of a two-layer prediction model based on extreme random trees on the training set. Figure 12c The diagram illustrates the scaling prediction performance of a two-layer prediction model based on Bagging meta-estimation on the training set. Figure 12d The diagram illustrates the scaling prediction performance of a two-layer prediction model based on AdaBoost on the training set. Figure 12e The diagram illustrates the scaling prediction performance of a two-layer prediction model based on XGBoost on the training set. Figure 13 The diagram illustrates the prediction effect of the combined prediction model on scale formation. Figure 14 This schematic diagram illustrates the composition of the reservoir wellbore scaling prediction device provided in an embodiment of this application; Figure 15 A schematic block diagram of a computer device provided in an embodiment of this application is shown.

[0022] Explanation of reference numerals in the attached figures In the diagram, 1. Tubing; 2. Interlayer packer; 3. ICV intelligent completion valve; 4. PDG downhole pressure and temperature monitor; 5. Transmission cable; 6. Reservoir downhole production system; 7. Subsea pipeline riser storage and transportation assembly; 8. FPSO production platform; 9. Oil, gas and water multiphase flow meter; 10. Computer equipment; 11. Water quality testing equipment. Detailed Implementation

[0023] The specific implementation methods of the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation methods described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application.

[0024] If the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0025] In existing technologies, the saturation exponent equation method is often used to predict the scaling trend of oilfield water. For example, the Davis-Stiff saturation exponent equation is shown below: (Formula 1); (Formula 2); (Formula 3); In Equations 1 to 3, Indicates the saturation index; Indicates the pH value of the water sample; This represents a correction factor, derived from the ionic strength at different temperatures. With correction factor The relationship diagram can be found; express The negative logarithm of concentration; The negative logarithm of total alkalinity; express Concentration, in moles per liter; express Concentration, in moles per liter; Indicates ionic strength; Indicates the first i Seed ion concentration; Indicates the first i The valence of the ions.

[0026] After calculating the saturation index Then, the scaling trend is determined as follows: If If so, it indicates a tendency to scale buildup; if If, then it is determined to be a critical state; if If so, it is determined that there is no tendency for scaling.

[0027] In existing technologies, to predict the scaling trend of oilfield water, an inorganic scaling trend prediction mathematical model is often used. This model combines classical solution theory, ion adsorption theory, and the basic principles of non-isothermal flow and fluid mass transfer in porous media. However, this model consists of a system of partial differential equations, making it difficult to obtain an exact solution. A better approach is to discretize the formation oil-water two-phase non-isothermal flow model and the scale-forming ion transport and deposition model in porous media using finite difference methods, based on appropriate initial and boundary conditions. This transforms the problem into a method that implicitly solves the discrete solutions at each grid node using numerical methods. Specifically, it involves calculating the dynamic changes of formation pressure, temperature, and scale-forming ion concentration with time and distance. The calculation results are then substituted into an inorganic scaling saturation index prediction model to obtain the dynamic changes of the saturation index with time and distance. The scaling trend of oilfield water is then determined based on the critical saturation index. As mentioned above, when using the saturation exponential equation method or mathematical model method to predict scaling trends, there are problems such as considering only one factor or having a complicated calculation process. The accuracy of the calculation results is also difficult to guarantee, which cannot meet the current oilfield production site's demand for real-time and accurate scaling prediction.

[0028] To overcome the above-mentioned defects, embodiments of this application provide a method for predicting scale formation in oil reservoir wellbore, including: Obtain characteristic parameters that are correlated with wellbore scaling; The acquired feature parameters are input into the constructed wellbore scaling prediction model to generate prediction results, which include whether the wellbore is scaling and / or the amount of scaling in the wellbore.

[0029] Specifically, in this application, the wellbore scaling prediction model is obtained by training and optimizing the parameters of an initially constructed machine learning model.

[0030] It should be understood that the characteristic parameters related to wellbore scaling can be obtained based on expert experience, or by performing correlation analysis on the multidimensional characteristic parameters related to wellbore scaling, and selecting multiple highly correlated characteristic parameters from the correlation analysis results as input parameters for the wellbore scaling prediction model in this application.

[0031] The above embodiments cannot determine the specific location of scale buildup in the wellbore. Therefore, as an improvement to this application, scale buildup in the reservoir wellbore is predicted using the following method: Obtain characteristic parameters related to wellbore scaling in each wellbore section; Input the feature parameters into the constructed wellbore scaling prediction model to generate prediction results, which include whether each wellbore section is scaling and / or the amount of scaling in each wellbore section.

[0032] As an improvement to the above embodiments, Figure 1This schematically illustrates a flowchart of a method for predicting scale buildup in oil reservoir wellbore according to an embodiment of this application, as shown below. Figure 1 As shown, the method for predicting scale buildup in oil reservoir wellbore provided in this application includes the following steps: Step 102: Obtain characteristic parameters related to wellbore scaling from the production site monitoring equipment for each wellbore section.

[0033] Specifically, in this application, the characteristic parameters include multiple of the following: well temperature, well pressure, wellhead production parameters, and water quality analysis data.

[0034] Step 104: Input the obtained feature parameters into the constructed wellbore scaling prediction model to generate prediction results. The prediction results include whether each wellbore section is scaling and / or the amount of scaling in each wellbore section.

[0035] Specifically, in this application, the wellbore scaling prediction model is a model obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination. The optimal hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed surrogate model is met. The surrogate model takes the hyperparameter combination of the machine learning model as input and the prediction error of the machine learning model as output. In the process of generating candidate sample points, the hyperparameter combination of the machine learning model is used as the decision variable and the sampling strategy evaluation function is used as the fitness function.

[0036] It's important to understand that establishing a surrogate model typically involves three processes: selecting the initial sampling method, selecting the surrogate model, and selecting the sampling strategy when updating candidate sample points. The sampling strategy can be based on minimizing the surrogate response, maximizing the search space, or achieving the highest combined value. The sampling strategy evaluation function mentioned in step 104 refers to an evaluation function that assesses whether the surrogate response is minimized, whether the search space is maximized, or a combined evaluation function that assesses the combined value of the surrogate response and the search space. The termination condition for iteratively updating the surrogate model can be, for example, reaching a first threshold number of iterations.

[0037] As an example, the initial experimental design method employs symmetric Latin hypercube sampling. Symmetric Latin hypercube sampling is an improved Latin hypercube sampling method designed to enhance sample homogeneity and representativeness. Compared to traditional Latin hypercube sampling, symmetric Latin hypercube sampling optimizes sample distribution by introducing the principle of symmetry, typically exhibiting better performance when dealing with high-dimensional spaces. Its main idea is to arrange the samples symmetrically in multidimensional space to ensure a more uniform distribution of sample points in each dimension. Its main steps and key elements are as follows: 1) Determine the parameter range and sample size: First, for each variable Xi ( i =1, 2, ..., d Determine the range of values. a i , b i and sample size n .

[0038] 2) Constructing the Latin hypercube: Dividing the range of each variable into... n a uniform interval I ij : (Formula 4); In Formula 4, j Indicates the numbering of the uniform interval. j =1, 2, ..., n .

[0039] 3) Symmetrical selection of sample points: In each uniform interval I ij Select sample points x ij The selection method involves symmetrically arranging these points. Typically, sample points are randomly selected within the interval to ensure a uniform distribution of samples across each dimension. For each variable... X i The selection of generated sample points can be achieved through symmetry, ensuring that the sample points are evenly distributed relative to the median.

[0040] 4) Randomly arrange the samples: Randomly arrange the generated samples to break any potential structural biases, so that the final samples exhibit better uniformity in high-dimensional space.

[0041] 5) Generate the final sample: Combine the sample points of all variables to obtain the final sample point set. .

[0042] In one comparative embodiment, only a temperature acquisition device installed at the wellhead is used to collect the wellhead temperature. Then, numerical simulation and other methods are used to calculate the temperature gradient distribution within the wellbore, thereby obtaining the wellbore temperature for each wellbore section. Furthermore, a pressure monitoring device installed at the wellhead is used to monitor the wellhead pressure. Then, numerical simulation and other methods are used to calculate the pressure gradient distribution within the wellbore, thereby obtaining the wellbore pressure for each wellbore section. Because numerical simulation and other methods are combined, the accuracy and real-time performance of the wellbore temperature and pressure obtained in the above comparative embodiment are poor. In this application, as in the above embodiment, the characteristic parameters are obtained from production site monitoring equipment, which ensures the accuracy and real-time performance of the data source used for wellbore scaling prediction.

[0043] The above embodiments have at least the following beneficial effects: 1) It is known that the scaling conditions vary at different locations in the wellbore. By predicting whether and / or the amount of scaling in different wellbore sections, the location of scaling in the wellbore can be determined. For example, if the prediction result shows that the bottom wellbore section is scaled while other wellbore sections are not scaled, then the scaling location in the wellbore at the current moment can be determined to be in the bottom wellbore section.

[0044] 2) The accuracy of the wellbore scaling prediction model depends on the input feature parameters. By acquiring feature parameters from the production site monitoring equipment, compared with other numerical simulations and theoretical formula calculations, the accuracy of the acquired feature parameters depends only on the acquisition accuracy of each production site monitoring equipment. This avoids the complex numerical simulation and theoretical calculation process, ensures the accuracy and real-time performance of the feature parameters, and improves the accuracy of wellbore scaling prediction.

[0045] 3) By optimizing the hyperparameters of the initially constructed machine learning model using a machine learning-based surrogate optimization method, the construction speed of the wellbore scaling prediction model is accelerated. That is, a better combination of hyperparameters is obtained at a faster speed, which improves the construction speed and accuracy of the wellbore scaling prediction model. This improves the accuracy of the wellbore scaling prediction results, accelerates the online maintenance and update efficiency of the wellbore scaling prediction model, and thus improves the efficiency of wellbore scaling prediction.

[0046] 4) To obtain the well temperature and pressure of each well section through monitoring, temperature and pressure monitoring equipment needs to be installed in each well section. This method increases the cost of well scale prediction. However, wellhead production parameters and water quality analysis data can be obtained from actual production equipment at the oilfield production site. In summary, the above embodiments are based on the technical concept of combining production site monitoring equipment with machine learning methods. Compared with the saturation exponential equation method and thermodynamic solubility method, the overall cost of well scale prediction is not significantly increased while improving prediction accuracy and efficiency.

[0047] In summary, the wellbore scaling prediction method implemented in the above embodiments of this application comprehensively considers prediction cost, prediction speed and prediction accuracy, and achieves wellbore scaling prediction in oil reservoirs with acceptable cost, high prediction accuracy and prediction efficiency, and is especially suitable for the development process of high salinity oil reservoirs.

[0048] As an example, the wellbore scaling prediction model is used to predict whether scaling will occur in each wellbore section and the amount of scaling in each wellbore section. Accordingly, the wellbore scaling prediction model includes a one-layer classification model and a two-layer prediction model, wherein: A single-layer classification model is used to predict whether scale will form in each section of the wellbore, using feature parameters as input. A two-layer prediction model is used to predict the amount of scaling in a specific wellbore section using specific feature parameters as input. The specific wellbore section is the wellbore section whose prediction result is scaling after being output by a one-layer classification model. The specific feature parameters are the feature parameters of the specific wellbore section.

[0049] As can be seen from the above, the wellbore scaling prediction model, which consists of a single-layer classification model and a two-layer prediction model, can predict whether scaling will occur in each section of the wellbore, and predict the amount of scaling through characteristic parameters when scaling occurs in some or all sections of the wellbore.

[0050] As a preferred embodiment of this application, the amount of scale in the well section includes scale of various types, including SrCO3, NaCl, CaCO3, CaSO4·2H2O, CaSO4, SrSO4, and Mg(OH)2.

[0051] As an example, the scale amount output by the wellbore scale prediction model for each wellbore section includes the scale amount of seven scale types: SrCO3 scale amount, NaCl scale amount, CaCO3 scale amount, CaSO4·2H2O scale amount, CaSO4 scale amount, SrSO4 scale amount, and Mg(OH)2 scale amount.

[0052] As an example, when the wellbore is a production well, the wellhead production parameters include: oil production, water production, gas production, and CO2 content in the produced gas phase. Water quality analysis data include: pH value of produced water, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity.

[0053] As an example, when the wellbore is an injection well, the wellhead production parameters include at least the injection volume, and the water quality analysis data include: injection water pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity.

[0054] As an example, machine learning models can include random forests, extreme random trees, AdaBoost, XGBoost, and Bagging meta-estimators, among others.

[0055] As an improvement to the above embodiments, Figure 2 This illustration schematically shows a flowchart of the wellbore scaling prediction model construction process according to an embodiment of this application, such as... Figure 2 As shown, the wellbore scaling prediction model is obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination, specifically including the following steps: Step 201: Construct the first training set using historical feature parameters; Step 202: After training the initial classification network using the first training set, establish the initial layer classification model in the machine learning model; Step 203: Update the hyperparameter combination in the initial one-layer classification model to the first optimal hyperparameter combination to obtain the optimized one-layer classification model. The first optimal hyperparameter combination is the best candidate sample point when the termination condition of iteratively updating the initially constructed first surrogate model is met. The first surrogate model takes the hyperparameter combination of the one-layer classification model as input and the prediction error of the one-layer classification model as output. In the process of generating candidate sample points, the hyperparameter combination of the one-layer classification model is used as the decision variable and the evaluation function of the first sampling strategy is used as the fitness function. Step 204: Delete the historical feature parameters of the prediction results generated by the optimized one-layer classification model in the first training set that are not scaled, and obtain the second training set; Step 205: After training the initial prediction network using the second training set, establish the initial two-layer prediction model in the machine learning model. Step 206: Update the hyperparameter combination in the initial two-layer prediction model to the second optimal hyperparameter combination to obtain the optimized two-layer prediction model. The second hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed second surrogate model is met. The second surrogate model takes the hyperparameter combination of the two-layer prediction model as input and the prediction error of the two-layer prediction model as output. In the process of generating candidate sample points, the hyperparameter combination of the two-layer prediction model is used as the decision variable and the second sampling strategy evaluation function is used as the fitness function. Step 207: The wellbore scaling prediction model is composed of the optimized single-layer classification model and the optimized two-layer prediction model.

[0056] Specifically, the first sampling strategy evaluation function refers to an evaluation function used to evaluate whether the response of the first surrogate model is minimized, or whether the search space of the first surrogate model is maximized, or whether the comprehensive value of the surrogate response and search space of the first surrogate model is maximized, etc. Similarly, the second sampling strategy evaluation function refers to an evaluation function used to evaluate whether the response of the second surrogate model is minimized, or whether the search space of the second surrogate model is maximized, or whether the comprehensive value of the surrogate response and search space of the second surrogate model is maximized, etc.

[0057] In a comparative embodiment, the hyperparameter optimization processes of the single-layer classification model and the two-layer prediction model are performed independently, and both are trained based on the first training set as the data source. By comparison, it can be seen that in the above embodiment of this application, by first optimizing the single-layer classification model, then reconstructing the first training set using the optimized single-layer classification model, and finally using the reconstructed second training set as the data source to train the initial prediction network to establish the initial two-layer prediction model, and then performing the hyperparameter optimization process of the two-layer prediction model, it is evident that by deleting historical feature parameters with prediction results indicating no fouling from the training set, the influence of these useless data on the optimization process of the two-layer prediction model is avoided, thus improving the accuracy and efficiency of the two-layer prediction model optimization.

[0058] As examples, the first surrogate model is the Extreme Learning Machine, and the second surrogate model is an interpolation model based on the Cauchy RBF function. Here, the Cauchy RBF function refers to an RBF function using a Cauchy kernel.

[0059] The principle of the interpolation model based on the Cauchy RBF function is as follows: (Formula 5) (Formula 6); (Formula 7); in, Indicates input data; This represents the output value of the proxy model; This indicates the number of points that have been evaluated by the function; Represents the RBF function coefficients at the points mentioned above; Represents the Cauchy RBF function; This represents the linear tail of the interpolation model based on the Cauchy RBF function; This represents the points that have already been evaluated using the function; This represents the coefficient values ​​based on the Cauchy RBF function; All represent coefficients based on the linear tail of the Cauchy RBF function; T This indicates transpose.

[0060] As another improvement to the above embodiments. Figure 3 The schematic diagram illustrates a flowchart of a preset optimization method according to an embodiment of this application, such as... Figure 3 As shown, the initially constructed proxy model is iteratively updated using the following preset optimization method to obtain the optimal combination of hyperparameters: Step 301: Generate candidate sample points by improving the SSA algorithm; Step 302: Determine whether the termination condition for iterative update has been reached; if yes, use the current best candidate sample point as the optimal hyperparameter combination; otherwise, use the candidate sample point as the input parameter and the prediction error of the machine learning model with the candidate sample point as the hyperparameter combination as the label value to construct new sample points, and use the new sample points to update the surrogate model, and then jump to execute step 301.

[0061] Specifically, in this application, step 301, generating candidate sample points, includes the following steps: Step 401: Population initialization is completed through Kent mapping; Step 402: Calculate the fitness value of each individual in the population; Step 403: Update the positions of individuals in the current population; Step 404: Perform Gaussian mutation on the individuals after the position update, and form a new population from the individuals after Gaussian mutation; Step 405: Calculate the fitness value of each individual in the new population; Step 406: Determine whether the termination condition for candidate sample point optimization has been met. If so, select the individual with the best current fitness as the candidate sample point; otherwise, proceed to step 403.

[0062] In this application, specifically, the termination condition for candidate sample point optimization may be that the cumulative number of fitness values ​​calculated reaches a second threshold, etc. During Gaussian mutation, each decision variable element in an individual undergoes the Gaussian mutation process.

[0063] As described in the above embodiment, an improved SSA algorithm is proposed for generating candidate sample points. Specifically, the improved SSA algorithm (Sparrow Search algorithm) combines the Kent mapping and Gaussian mutation. Compared to the Logistic mapping, the Kent mapping exhibits better uniformity of traversal. Furthermore, Gaussian mutation applies random perturbation after updating each individual's position, thereby increasing population diversity. Based on this, the improved SSA algorithm proposed in this application enhances global optimization capabilities, thus improving the accuracy of wellbore scaling prediction.

[0064] As an example, the expression for the Kent mapping is as follows: (Form 8); In Formula 8, Represents a random sequence; This represents a control parameter, with a value range between (0,1); express The initial population obtained after mapping.

[0065] As an example, the expression for Gaussian mutation is as follows: (Form 9); In Formula Nine, This represents the decision variable elements in the individuals after Gaussian mutation. This represents the decision variable elements in the individual before Gaussian mutation. This represents a random number that conforms to a standard normal distribution.

[0066] As another improvement to the above embodiments, there are multiple wellbore scaling prediction models, such as... Figure 4 As shown, the method for predicting scale buildup in oil reservoir wellbore provided in this application embodiment further includes the following steps: Step 106: For any well section, if the prediction result shows that the number of layer classification models for scaling in the well section reaches the first preset value, then the well section is determined to be scaling; if the prediction result shows that the number of layer classification models for non-scaling in the well section reaches the second preset value, then the well section is determined to be non-scaling. Step 108: For any well section, the scale amount predicted by each two-layer prediction model is weighted and summed, and the weighted sum is taken as the final scale amount of the well section. The weighting coefficient of each scale amount is the weighting coefficient that minimizes the prediction error of the combined prediction model. The combined prediction model includes each two-layer prediction model. The combined prediction model takes feature parameters as input and the weighted sum of scale amounts predicted by each of the two-layer prediction models within itself as output.

[0067] As in the above embodiment, the prediction results of multiple single-level classification models are used for comprehensive judgment, which avoids the error in wellbore scaling caused by the prediction error of a single single-level classification model. At the same time, the prediction of scaling amount is performed by combining prediction models, which avoids the inaccurate prediction of wellbore scaling amount caused by the prediction error of a single two-level prediction model. Thus, high-precision prediction of wellbore scaling in oil reservoirs is achieved.

[0068] As a preferred approach, different machine learning methods are used in each single-layer classification model, and different machine learning methods are used in each two-layer prediction model.

[0069] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0070] In a specific application, to predict scaling in production and injection wellbores during the development of high-salinity oil reservoirs, characteristic parameters are collected using on-site monitoring equipment in oilfield production. Figure 5A schematic diagram illustrating the components of a high-mineralization wellbore scaling monitoring system is shown. Figure 5 In this process, PDG downhole pressure and temperature monitoring gauges 4 are installed inside the tubing 1 in each section of wellbore in Well A to collect temperature and pressure data. The wellbore is divided into upper, middle, and lower sections by three interlayer packers 2. The PDG downhole pressure and temperature monitoring gauges 4 are located between the interlayer packers 2 and the ICV intelligent completion valve 3. The monitoring data from the PDG downhole pressure and temperature monitoring gauges 4 are transmitted to the FPSO (Floating Production Storage and Offloading) via transmission cable 5. The computer equipment 10 within the offloading production platform consists of tubing 1, interlayer packers 2, ICV intelligent completion valves 3, PDG downhole pressure and temperature monitors 4, and transmission cables 5, forming a reservoir downhole production system 6. An oil-gas-water multiphase flow meter 9 is installed on the subsea pipeline riser storage and transportation assembly 7. This multiphase flow meter 9 is used to achieve online measurement of the three-phase production of oil, gas, and water, thereby obtaining parameters such as production well output, water cut, and gas-oil ratio. For injection wells, it records the daily water injection volume. The monitoring data from the oil-gas-water multiphase flow meter 9 is transmitted to the computer equipment 10. A water quality testing device 11 is also installed near the subsea pipeline riser storage and transportation assembly 7. The analytical data from the water quality testing device 11 is transmitted to the computer equipment 10. The oil-gas-water multiphase flow meter 9, computer equipment 10, and water quality testing device 11 are all located on the FPSO production platform 8.

[0071] After receiving the characteristic parameters, computer device 10 predicts whether scale has formed in each section of well A and the amount of scale of seven scale types in each section using the well scale prediction method in any of the above embodiments. Specifically, it predicts whether scale has formed in each section of well A and the amount of scale of seven scale types in each section using the well scale prediction method in any of the above embodiments. Figure 6 The process shown is carried out on Figure 5 Scaling prediction for the well shown: Step S1: Construct a dataset for training and testing the wellbore scaling prediction model. The specific process is as follows: 1) The PDG downhole pressure and temperature monitor 4 records and uploads one set of pressure and temperature data every few seconds when monitoring the pressure and temperature values ​​at each measuring point. This method results in a very large amount of pressure and temperature data per day, leading to problems such as storage space saturation, high computing power requirements, and difficulty in selecting data points. Therefore, the daily uploaded pressure and temperature data is first arithmetically averaged, and anomaly detection is performed. If the pressure and temperature deviate too far from the average, the anomaly is removed. Finally, the data points after removing anomalies are arithmetically averaged again, and this average is recorded as the average pressure and temperature data for the measuring point on that day. The previously stored original pressure and temperature data values ​​are then deleted.

[0072] 2) Real-time collection of water quality analysis data and produced gas component analysis data. Produced water quality analysis data includes: pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity. Produced gas component analysis data includes CO2 content in the produced gas phase. Injection water quality analysis data includes: injection water pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity. Table 1 shows the specific characteristic parameters used for predicting scaling in production wells.

[0073] Table 1

[0074] 3) Collect the wellbore scaling information under the characteristic parameters obtained in steps 1) and 2). First, determine whether the wellbore is scaling under each characteristic parameter. Second, if scaling occurs, identify the type of scaling and calculate the amount of scaling.

[0075] 4) Construct training and testing sets from the data obtained in steps 1) to 3) above. Before building the machine learning model, the feature parameters need to be normalized to improve the prediction accuracy of the machine learning model. The normalization preprocessing expression is as follows: (Form 10); In formula ten, This represents the value of a certain dimension in the feature parameters. This represents the minimum value of that dimension. This represents the maximum value of that dimension. This is the normalized value, which is between [-1, 1].

[0076] Step S2: Construct five initial one-layer classification models, namely, a classification model based on random forest, a classification model based on extreme random tree, a classification model based on AdaBoost, a classification model based on XGBoost, and a classification model based on Bagging meta-estimation.

[0077] Step S3: Use the Extreme Learning Machine as a surrogate model to accelerate and optimize the hyperparameters of each single-layer classification model to obtain the optimized single-layer classification model. Then, use the optimized single-layer classification model to reconstruct the dataset. Step S4: Construct five initial two-layer prediction models using the reconstructed dataset. The five two-layer prediction models are a prediction model based on random forest, a prediction model based on extreme random tree, a prediction model based on AdaBoost, a prediction model based on XGBoost, and a prediction model based on Bagging meta-estimation.

[0078] Step S5: The hyperparameters of each two-layer prediction model are accelerated and optimized using an interpolation model based on the Cauchy RBF function, resulting in optimized two-layer prediction models. These optimized models are then combined to form a composite prediction model. Weights are assigned to each model based on their performance, and the prediction results are combined using a weighted method, as follows: (Form 11); In Formula 11, This represents the final scale output of the combined prediction model. This represents the weighting coefficients (weights) assigned to each two-layer prediction model, and satisfies... , Indicates the number is i The amount of scaling predicted by the two-layer prediction model.

[0079] The above weights can be combined to form a weight vector. Therefore, the sum of squared errors of the combined prediction model can be obtained as follows: ,in, Given the error information matrix composed of the prediction errors of each two-layer prediction model, find the value that makes the sum of squared errors equal to... Minimum weight vector This will give us the final combined prediction model.

[0080] Specifically, in this application example, the weight vector composed of the weighting coefficients of each two-layer prediction model in the combined prediction model is [0.1549, 0.8451, 0, 0, 0]. T .

[0081] Step S6 involves testing the optimized single-layer classification models and combined prediction models. For example... Figure 7 As shown, there are 27 groups of unscaled wellbore examples and 69 groups of scaled wellbore examples. All optimized single-layer classification models are completely accurate in their predictions, meaning 100% of the single-layer classification models are accurate. For example... Figure 8 As shown, there are 2 groups of unscaled wellbore examples and 22 groups of scaled wellbore examples. All optimized single-layer classification models are completely accurate in their predictions, meaning 100% of the single-layer classification models are accurate. For example... Figure 9 As shown, the single-layer classification model achieves a score of 1 on both the training and test sets, reaching 100% accuracy. Figure 10 As shown, the two-layer prediction model achieves high scores on both the training and test sets. Figures 11a to 11e It can be seen that on the test set, the predicted scaling points are evenly distributed around the 45° line, indicating good prediction performance. From... Figures 12a to 12eIt can also be seen that on the training set, the predicted scaling points are evenly distributed around the 45° line, indicating good prediction performance. From... Figure 13 It can be seen that the combined prediction model has improved the prediction accuracy of scale amount compared with the single two-layer prediction model.

[0082] Step S7: Obtain real-time characteristic parameters of each wellbore section of Well A from the production site monitoring equipment. Statistically analyze the prediction results of each optimized single-layer classification model. For any wellbore section, if 60% or more of the single-layer classification models determine scaling, then scaling is considered to have occurred in that wellbore section; conversely, if 60% or more of the single-layer classification models determine no scaling, then no scaling is considered to have occurred in that wellbore section. If scaling has occurred in that wellbore section, then input the real-time characteristic parameters into the combined prediction model to generate the scaling amount corresponding to the seven scaling types.

[0083] Corresponding to the reservoir wellbore scaling prediction method in the above embodiments, Figure 14 The diagram illustrates the components of the reservoir wellbore scaling prediction device 500 provided in the embodiments of this application. For ease of explanation, only the parts relevant to the embodiments of this application are shown.

[0084] like Figure 14 As shown, the reservoir wellbore scaling prediction device 500 includes: The acquisition module 501 is used to acquire characteristic parameters related to wellbore scaling in each wellbore section from the production site monitoring equipment. The prediction module 502 is used to input the feature parameters into the constructed wellbore scaling prediction model and generate prediction results, including whether each wellbore section is scaled and / or the amount of scale in each wellbore section. The characteristic parameters include multiple of the following: well temperature, well pressure, wellhead production parameters, and water quality analysis data. The wellbore scaling prediction model is a model obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination. The optimal hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed surrogate model is met. The surrogate model takes the hyperparameter combination of the machine learning model as input and the prediction error of the machine learning model as output. In the optimization process of generating candidate sample points, the hyperparameter combination is used as the decision variable and the sampling strategy evaluation function is used as the fitness function.

[0085] As an embodiment of this application, the reservoir wellbore scaling prediction device 500 can achieve the following: Figure 1 The embodiments shown are as well as other related method embodiments in this application.

[0086] The process by which each module in the reservoir wellbore scaling prediction device 500 provided in this application implements its respective function can be found in the foregoing. Figure 1 The descriptions of the embodiments shown and other related method embodiments are not repeated here.

[0087] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiments section, and will not be repeated here. Furthermore, all of the above modules can be applied to computing devices that include memory and a processor.

[0088] Figure 15 A schematic block diagram of a computer device according to an embodiment of the present application is shown. In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown below. Figure 15 As shown in the figure, the computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05, and a memory (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A06. The network interface A02 is used for communication with external terminals via a network connection. When the computer program is executed by the processor A01, it implements a method for predicting scale formation in oil reservoir wellbore. The display screen A04 can be an LCD screen or an e-ink display screen. The input device A05 can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0089] Those skilled in the art will understand that Figure 15 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0090] In one embodiment, the reservoir wellbore scaling prediction device 500 provided in this application can be implemented as a computer program, which can be implemented in various ways, such as... Figure 15 The computer device shown operates on the device. The memory of the computer device can store various program modules that make up the reservoir wellbore scaling prediction device 400. The computer program, composed of the various program modules, causes the processor to execute the steps in the reservoir wellbore scaling prediction methods of the various embodiments of this application described in this specification.

[0091] In one embodiment, this application also provides a machine-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the reservoir wellbore scaling prediction method in the above embodiments.

[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0093] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting scale formation in oil reservoir wellbore, characterized in that, include: Characteristic parameters related to wellbore scaling in each wellbore section were obtained from on-site monitoring equipment. The characteristic parameters are input into the constructed wellbore scaling prediction model to generate prediction results, which include whether each wellbore section is scaling and / or the amount of scaling in each wellbore section. The characteristic parameters include multiple of the following: well temperature, well pressure, wellhead production parameters, and water quality analysis data. The wellbore scaling prediction model is a model obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination. The optimal hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed surrogate model is met. The surrogate model takes the hyperparameter combination of the machine learning model as input and the prediction error of the machine learning model as output. In the optimization process of generating candidate sample points, the hyperparameter combination is used as the decision variable and the sampling strategy evaluation function is used as the fitness function.

2. The method for predicting scale formation in oil reservoir wellbore according to claim 1, characterized in that, The wellbore scaling prediction model includes: A single-layer classification model is used to predict whether each wellbore section will be scaled, using the aforementioned feature parameters as input. A two-layer prediction model is used to predict the amount of scaling in a specific wellbore section using specific feature parameters as input. The specific wellbore section is the wellbore section whose prediction result is scaling after being output by a one-layer classification model. The specific feature parameters are the feature parameters of the specific wellbore section.

3. The method for predicting scale formation in oil reservoir wellbore according to claim 2, characterized in that, The wellbore scaling prediction model includes multiple models, and the method further includes: For any well section, if the prediction result shows that the number of the layer classification models for the well section with scale reaches a first preset value, then the well section is determined to have scale. If the prediction result shows that the number of the layer classification models for the well section without scale reaches a second preset value, then the well section is determined to have no scale. For any wellbore section, the scale amounts predicted by each of the two-layer prediction models are weighted and summed, and the resulting weighted sum is taken as the final scale amount of the wellbore section. The weighting coefficients of each scale amount are the weighting coefficients that minimize the prediction error of the combined prediction model. The combined prediction model takes the feature parameters as input and the weighted sum of the scale amounts predicted by each of the two-layer prediction models within itself as output.

4. The method for predicting scale formation in oil reservoir wellbore according to claim 2, characterized in that, The wellbore scaling prediction model is obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination, including: Construct the first training set using historical feature parameters; After training the initial classification network using the first training set, the initial layer classification model in the machine learning model is established. The hyperparameter combination in the initial single-layer classification model is updated to the first optimal hyperparameter combination to obtain the optimized single-layer classification model. The first optimal hyperparameter combination is the best candidate sample point when the termination condition of iteratively updating the first surrogate model is met. The first surrogate model takes the hyperparameter combination of the single-layer classification model as input and the prediction error of the single-layer classification model as output. In the process of generating candidate sample points, the hyperparameter combination of the single-layer classification model is used as the decision variable and the evaluation function of the first sampling strategy is used as the fitness function. The historical feature parameters that predicted no scaling were removed from the first training set using the optimized one-layer classification model; the second training set was obtained. After training the initial prediction network using the second training set, the initial two-layer prediction model in the machine learning model is established. The hyperparameter combination in the initial two-layer prediction model is updated to the second optimal hyperparameter combination to obtain the optimized two-layer prediction model. The second hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed second surrogate model is met. The second surrogate model takes the hyperparameter combination of the two-layer prediction model as input and the prediction error of the two-layer prediction model as output. In the process of generating candidate sample points, the hyperparameter combination of the two-layer prediction model is used as the decision variable and the evaluation function of the second sampling strategy is used as the fitness function. The wellbore scaling prediction model consists of an optimized single-layer classification model and an optimized two-layer prediction model.

5. The method for predicting scale formation in oil reservoir wellbore according to claim 1, characterized in that, The scale buildup in the wellbore section includes scale buildup of various types, including multiple types of scale such as SrCO3, NaCl, CaCO3, CaSO4·2H2O, CaSO4, SrSO4, and Mg(OH)2.

6. The method for predicting scale formation in oil reservoir wellbore according to claim 1, characterized in that, When the wellbore is a production well, the wellhead production parameters include oil production, water production, gas production, and CO2 content in the produced gas phase. The water quality analysis data includes the pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity of the produced water. When the wellbore is an injection well, the wellhead production parameters include at least the injection volume. The water quality analysis data includes the pH value, calcium ion concentration, magnesium ion concentration, strontium ion concentration, sodium ion concentration, chloride ion concentration, acetate ion concentration, bicarbonate ion concentration, sulfate ion concentration, and total alkalinity of the injected water.

7. The method for predicting scale formation in oil reservoir wellbore according to claim 1, characterized in that, The initially constructed proxy model is iteratively updated using the following preset optimization method to obtain the optimal combination of hyperparameters: Generate candidate sample points; Determine whether the termination condition for iterative update has been met; if so, use the current best candidate sample point as the optimal hyperparameter combination; otherwise, use the candidate sample point as the input parameter and the prediction error of the machine learning model with the candidate sample point as the hyperparameter combination as the label value to construct a new sample point, and use the new sample point to update the surrogate model, and then jump to execute the previous step. The generation of candidate sample points includes: Population initialization is accomplished through Kent mapping; Calculate the fitness value of each individual in the population; Generate a new population and calculate the fitness value of each individual in the new population; Determine whether the termination condition for candidate sample point optimization has been met. If so, select the individual with the best current fitness as the candidate sample point; otherwise, proceed to the previous step. Among these, the generation of new populations includes: Update the positions of individuals in the current population; After updating the position, perform Gaussian mutation on the individuals, and form a new population from the individuals after Gaussian mutation.

8. The method for predicting scale formation in oil reservoir wellbore according to claim 1, characterized in that, The types of machine learning models include random forest, extreme random tree, AdaBoost, XGBoost, and Bagging meta-estimator.

9. The method for predicting scale formation in oil reservoir wellbore according to claim 4, characterized in that, The first proxy model is the Extreme Learning Machine.

10. The method for predicting scale formation in oil reservoir wellbore according to claim 4, characterized in that, The second proxy model is an interpolation model based on the Cauchy RBF function.

11. The method for predicting scale formation in oil reservoir wellbore according to claim 7, characterized in that, Gaussian mutation is performed on the individuals after location updates using the following formula: ； in, This represents the decision variable elements in the individuals after Gaussian mutation. This represents the decision variable elements in the individual before Gaussian mutation. This represents a random number that conforms to a standard normal distribution.

12. A device for predicting scale buildup in oil reservoir wellbores, characterized in that, include: The acquisition module is used to acquire characteristic parameters related to wellbore scaling in each wellbore section from the production site monitoring equipment. The prediction module is used to input the feature parameters into the constructed wellbore scaling prediction model and generate prediction results, which include whether each wellbore section is scaling and / or the amount of scaling in each wellbore section. The characteristic parameters include multiple of the following: well temperature, well pressure, wellhead production parameters, and water quality analysis data. The wellbore scaling prediction model is a model obtained by updating the initially constructed machine learning model using the optimal hyperparameter combination. The optimal hyperparameter combination is the best candidate sample point when the termination condition for iteratively updating the initially constructed surrogate model is met. The surrogate model takes the hyperparameter combination of the machine learning model as input and the prediction error of the machine learning model as output. In the optimization process of generating candidate sample points, the hyperparameter combination is used as the decision variable and the sampling strategy evaluation function is used as the fitness function.

13. The reservoir wellbore scaling prediction device according to claim 12, characterized in that, The initially constructed proxy model is iteratively updated using the following preset optimization method to obtain the optimal combination of hyperparameters: Generate candidate sample points; Determine whether the termination condition for iterative update has been met; if so, use the current best candidate sample point as the optimal hyperparameter combination; otherwise, use the candidate sample point as the input parameter and the prediction error of the machine learning model with the candidate sample point as the hyperparameter combination as the label value to construct a new sample point, and use the new sample point to update the surrogate model, and then jump to execute the previous step. The generation of candidate sample points includes: Population initialization is accomplished through Kent mapping; Calculate the fitness value of each individual in the population; Generate a new population and calculate the fitness value of each individual in the new population; Determine whether the termination condition for candidate sample point optimization has been met. If so, select the individual with the best current fitness as the candidate sample point; otherwise, proceed to the previous step. Among these, the generation of new populations includes: Update the positions of individuals in the current population; After updating the position, perform Gaussian mutation on the individuals, and form a new population from the individuals after Gaussian mutation.

14. A computer device, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the reservoir wellbore scaling prediction method according to any one of claims 1 to 11.

15. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to perform the reservoir wellbore scaling prediction method according to any one of claims 1 to 11.

Images