Marine oil and gas exploration well control manifold erosion rate prediction method
By constructing a support vector machine model and optimizing its parameters, combined with genetic algorithms and simulation data, the problems of accuracy and efficiency in predicting the erosion rate of well-controlled manifolds were solved, achieving efficient and accurate prediction of the erosion rate of well-controlled manifolds.
Patent Information
- Application Number
- CN202511641868.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-02-17
AI Technical Summary
Existing technologies struggle to accurately predict the erosion rate of well control manifolds, making it difficult to quantify the risk of well control manifold failure and resulting in low prediction efficiency.
A support vector machine model combined with a genetic algorithm was used to optimize the penalty function and kernel function parameters. The well control manifold erosion rate prediction model was constructed using manifold diameter, bending angle, fluid velocity, fluid density, and fluid viscosity as input parameters, and trained by combining simulation data with real data.
It improves the accuracy and efficiency of well control manifold erosion rate prediction, reduces prediction errors, and enables more accurate prediction of the maximum erosion rate.
Smart Images
Figure BDA0005680894640000021 
Figure BDA0005680894640000031 
Figure BDA0005680894640000041
Abstract
Description
Technical Field
[0001] This invention belongs to the field of marine oil and gas exploration technology, and specifically relates to a method for predicting the erosion rate of the control pipe manifold in marine oil and gas exploration wells. Background Technology
[0002] In offshore oil and gas exploration operations, well control manifolds, as an important component of well control equipment, can effectively control well kicks and overflows, reducing the occurrence of blowout accidents. When an overflow occurs during drilling, the medium flow velocity is very high, and it carries a large amount of rock cuttings, sand, and other solid particles, which can easily cause erosion and wear on the inner wall of the well control manifold. In severe cases, the well control manifold may fail in a short period of time due to perforation or leakage.
[0003] Predicting the erosion rate of well control manifolds can help anticipate the risk of well control manifold failure. However, due to the uncertainty of parameters such as flow velocity, pressure, and particle diameter during a blowout, numerical simulations are limited in scope and time-consuming, making it difficult to quantify the damage caused by erosion to the well control manifold. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for predicting the erosion rate of well control manifolds in marine oil and gas exploration. This method can accurately predict the maximum erosion rate of well control manifolds and improve prediction efficiency.
[0005] The technical solution provided by this invention is as follows:
[0006] A method for predicting the erosion rate of the control manifold in marine oil and gas exploration wells includes the following steps:
[0007] Step 1: Construct a support vector machine model;
[0008] The input parameters of the support vector machine model are: manifold diameter, manifold bend angle, fluid velocity, fluid density, and fluid viscosity, and the output parameter of the support vector machine is: the maximum erosion rate of the well control manifold.
[0009] Step 2: Use a genetic algorithm to optimize the penalty function coefficients and kernel function parameters of the support vector machine model to obtain an optimized support vector machine model;
[0010] Step 3: Train the optimized support vector machine model to obtain the well control manifold erosion rate prediction model;
[0011] Step 4: Obtain the manifold diameter, manifold bend angle, fluid velocity, fluid density, and fluid viscosity of the manifold to be predicted, and input them into the manifold erosion rate prediction model to obtain the maximum erosion rate of the manifold to be predicted.
[0012] Preferably, in step two, the penalty function coefficients and kernel function parameters of the support vector machine model are optimized, including the following steps:
[0013] Step 1: Randomly initialize the penalty function coefficients and kernel function parameters of the support vector machine model to obtain multiple arrays containing the penalty function coefficients and kernel function parameters, which serve as the initial population;
[0014] Step 2: Calculate the fitness of each array in the population;
[0015] Step 3: Select the array with high fitness according to the set ratio and move it directly to the next generation; perform crossover and mutation on the remaining arrays in the population and then move them to the next generation.
[0016] Step 4: Repeat steps 2-3 until the maximum number of iterations is reached; select the array with the highest fitness as the penalty function coefficients and kernel function parameters of the support vector machine model.
[0017] Preferably, in step 2, the fitness calculation formula is:
[0018]
[0019] in, y represents the predicted maximum erosion rate of the well control manifold. i The value represents the actual maximum erosion rate of the well control manifold; N represents the number of training samples, and w1 and w2 are both weighting coefficients.
[0020] Preferably, the values of w1 and w2 satisfy: w1∈(0.3,1), w2∈(0.3,1), and w1+w2=1.
[0021] Preferably, in step 3, 30% of the arrays with high fitness are selected to directly enter the next generation, and the proportions of crossover and mutation are 60% and 10%, respectively.
[0022] Preferably, before step three, a sample set is constructed, including the following steps:
[0023] Step a: Determine the value range of each input parameter based on the actual situation;
[0024] Step b: Take values at even intervals within the range of each input parameter;
[0025] In this case, each input parameter can take the same number of values;
[0026] Step c: Obtain multiple samples using an orthogonal experiment to form the sample set.
[0027] Preferably, step four further includes:
[0028] When the fluid viscosity is greater than 1.5 kg × 10 -3 When the erosion rate is / (m·s), the maximum erosion rate of the well-controlled manifold to be predicted, output by the well-controlled manifold erosion rate prediction model, is corrected to:
[0029]
[0030] Where μ represents fluid viscosity, μ0 represents reference fluid viscosity, and E0 is the maximum erosion rate of the well control manifold to be predicted, output by the well control manifold erosion rate prediction model.
[0031] The beneficial effects of this invention are:
[0032] The method for predicting the erosion rate of well control manifolds in marine oil and gas exploration provided by this invention can accurately predict the maximum erosion rate of well control manifolds and improve prediction efficiency. Detailed Implementation
[0033] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.
[0034] During the erosion and wear process of well control manifolds, solid particles such as rock cuttings and sand grains change with the erosion. However, numerical simulations struggle to adjust the particle properties in real time. Existing prediction methods only consider the initial properties of the particles and require the assumption that the particles are spatially uniformly distributed and undamaged. However, the actual spatial distribution of particles is not uniform but changes in real time. Therefore, this assumption leads to significant errors in the prediction results.
[0035] This invention provides a method for predicting the erosion rate of the manifold in offshore oil and gas exploration wells. Considering the variations in the erosion fluid, it uses variables that are easy to collect in real time as input parameters for the prediction model. Furthermore, by reasonably setting the model parameters, more accurate prediction results can be obtained. The specific implementation process of this invention is as follows.
[0036] I. Constructing a Support Vector Machine Model
[0037] SVM nonlinear regression analysis establishes a new linear model by introducing a kernel function and mapping it to a high-dimensional space.
[0038] The input parameters of the support vector machine model are set as: manifold diameter, manifold bending angle, fluid velocity, fluid density, and fluid viscosity. The output parameter of the support vector machine is set as: the maximum erosion rate of the well control manifold.
[0039] Second, the penalty function coefficients and kernel function parameters of the support vector machine model are optimized to obtain an optimized support vector machine model.
[0040] The key to constructing an accurate SVM regression prediction model lies in selecting appropriate penalty function coefficients C and kernel function parameters g. The penalty function coefficients C adjust the relationship between model complexity and training error, aiming to improve the model's generalization ability. In a sample set, if the penalty factor is too small, it reduces the probability of penalizing abnormal data in the training data, leading to large training errors and potential underlearning. Conversely, if the penalty factor is too large, it places higher demands on the accuracy of the training data, potentially resulting in overlearning. The kernel function parameter g affects the fitting performance of the SVM model, controlling the number of support vectors and its generalization ability. In a training model, a smaller g results in more support vectors and relatively poorer generalization ability; a larger g results in fewer support vectors, reducing prediction accuracy.
[0041] This invention uses a genetic algorithm to optimize the penalty function coefficient C and kernel function parameter g of the support vector machine model. The specific process is as follows.
[0042] 1. Randomly initialize the penalty function coefficients and kernel function parameters of the support vector machine model, resulting in multiple arrays containing the penalty function coefficients and kernel function parameters, which serve as the initial population. That is, each array containing the penalty function coefficients and kernel function parameters represents an individual in the population.
[0043] 2. Calculate the fitness of each array in the population.
[0044] The execution process of a genetic algorithm involves selecting chromosomes with higher fitness values to participate in the reproduction of the next generation. In contrast, the execution process of a support vector machine (SVM) model concerns a minimization problem. The metric for evaluating the training and learning process is the error between the actual output value and the expected output value. A smaller error indicates better training and learning, resulting in more accurate predictions. Therefore, the reciprocal of the error is typically used as the fitness function.
[0045] However, experiments have shown that the model obtained by constructing the fitness function using a single error still has a large error. As a preferred embodiment, this invention sets the fitness function as follows:
[0046]
[0047] in, y represents the predicted maximum erosion rate of the well control manifold. i The value represents the actual maximum erosion rate of the well control manifold; N represents the number of training samples, and w1 and w2 are both weighting coefficients.
[0048] As a preferred option, the values of w1 and w2 satisfy: w1∈(0.3,1), w2∈(0.3,1), and w1+w2=1.
[0049] By improving the fitness function, the rationality of the parameter selection process can be increased, thereby improving the accuracy of the model.
[0050] 3. Heredity produces the next generation
[0051] Select the array with high fitness according to the set ratio and directly enter the next generation; crossover and mutation are performed on the remaining arrays in the population and then they enter the next generation.
[0052] In this embodiment, 30% of the arrays with high fitness are selected to directly enter the next generation, while the proportions of crossover and mutation are 60% and 10%, respectively.
[0053] 4. Repeat the genetic optimization process until the maximum number of iterations is reached; select the array with the highest fitness as the penalty function coefficients and kernel function parameters of the support vector machine model.
[0054] In this embodiment, the maximum number of iterations is set to 500.
[0055] Third, the optimized support vector machine model is trained to obtain the well control manifold erosion rate prediction model.
[0056] Since model training requires a large amount of sample data, and obtaining real data takes a long time, this invention obtains samples by combining real data with simulation data.
[0057] The specific process for constructing the simulation sample set is as follows.
[0058] a. Determine the value range of each input parameter based on the actual situation.
[0059] In this embodiment, the manifold diameter ranges from [80, 160] mm, the manifold bending angle ranges from [30°, 90°], the fluid velocity ranges from [30, 120] m / s, and the fluid density ranges from [850, 1500] kg / m³. 3 The range of fluid viscosity is set to [1×10]. -3 3×10 -3 ]kg / (m·s.
[0060] b. Take values at uniform intervals within the range of each input parameter; wherein, the number of values for each input parameter is the same.
[0061] c. The selected parameter values are used to obtain multiple samples through orthogonal experiments using Fluent simulation software, forming the simulation sample set.
[0062] 4. Obtain the manifold diameter, manifold bend angle, fluid velocity, fluid density, and fluid viscosity of the well control manifold to be predicted, and input them into the well control manifold erosion rate prediction model to obtain the maximum erosion rate of the well control manifold to be predicted.
[0063] Studies have found that fluid viscosity is unstable in the actual working environment of well control manifolds, and increased fluid viscosity enhances the viscous binding effect of particles, leading to a decrease in the erosion rate. As a preferred option, step four further includes: when the fluid viscosity is greater than 1.5 kg / (m·s), correcting the maximum erosion rate of the well control manifold to be predicted output by the well control manifold erosion rate prediction model to:
[0064]
[0065] Where μ represents the fluid viscosity, μ0 represents the reference fluid viscosity, and μ0 = 1.5 × 10⁻⁶. -3 kg / (m·s); E0 is the maximum erosion rate of the well-controlled manifold to be predicted, output by the well-controlled manifold erosion rate prediction model.
[0066] By correcting the maximum erosion rate output by the prediction model based on fluid viscosity, errors in the output results caused by excessive changes in fluid viscosity can be avoided, thereby further improving the accuracy of the maximum erosion rate prediction results.
[0067] Example
[0068] In this embodiment, 20 sets of real data were used as experimental data. The manifold diameter, manifold bending angle, fluid velocity, fluid density, and fluid viscosity from these 20 sets of data were used as input parameters for the model. The model outputs the maximum erosion rate of the well-controlled manifold through a well-controlled manifold erosion rate prediction model. The specific prediction results are shown in Table 1.
[0069] Table 1 Model Prediction Results
[0070]
[0071]
[0072] The average relative error of the model prediction results, calculated using the relative errors in Table 1, is 6.12%. Among these, the relative errors for samples 1, 5, 9, and 16 exceed 10%. Observation revealed that the viscosities of samples 1, 5, 9, and 16 are 2.35 × 10⁻⁶. -3 kg / (m·s), 2.03×10 -3 kg / (m·s), 2.13×10 -3 kg / (m·s) and 1.98×10 -3 kg / (m·s) are all greater than 1.5×10 -3kg / (m·s); by formula The prediction results after correcting the maximum erosion rate output by the model are shown in Table 2.
[0073] Table 2 Correction Prediction Results
[0074]
[0075] Table 2 shows that the errors in the maximum erosion rate were significantly reduced after correction using the formula. Combining the four corrected sets of data with the 16 sets of data directly output from the model yielded the final prediction result. The calculated average relative error of the final prediction result was 4.4%.
[0076] In summary, the method for predicting the erosion rate of well control manifolds in marine oil and gas exploration provided by this invention can accurately predict the maximum erosion rate of well control manifolds and improve prediction efficiency.
[0077] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.
Claims
1. A method for predicting the erosion rate of the control manifold in marine oil and gas exploration wells, characterized in that, The method comprises the following steps: Step one, constructing a support vector machine model; Wherein, the input parameters of the support vector machine model are: manifold diameter, manifold bending angle, fluid velocity, fluid density and fluid viscosity, and the output parameter of the support vector machine is: the maximum erosion rate of the well control manifold; Step two, optimizing the penalty function coefficient and the kernel function parameter of the support vector machine model by using genetic algorithm, to obtain an optimized support vector machine model; Step three, training the optimized support vector machine model to obtain a well control manifold erosion rate prediction model; Step four, obtaining the manifold diameter, manifold bending angle, fluid velocity, fluid density and fluid viscosity of the well control manifold to be predicted, and inputting the well control manifold erosion rate prediction model to obtain the maximum erosion rate of the well control manifold to be predicted.
2. The method of predicting the rate of erosion of a marine oil and gas exploration well control manifold according to claim 1, wherein, In the step two, the penalty function coefficient and the kernel function parameter of the support vector machine model are optimized, comprising the following steps: Step 1, randomly initializing the penalty function coefficient and the kernel function parameter of the support vector machine model to obtain a plurality of arrays containing the penalty function coefficient and the kernel function parameter as the initial population; Step 2, calculating the fitness of each array in the population; Step 3, selecting arrays with high fitness directly into the next generation according to a set proportion; the remaining arrays in the population are crossed and mutated to enter the next generation; Step 4, repeating steps 2-3 until the maximum number of iterations is reached; selecting the array with the highest fitness as the penalty function coefficient and the kernel function parameter of the support vector machine model.
3. The method of predicting the rate of erosion of a marine oil and gas exploration well control manifold according to claim 2, wherein, In the step 2, the fitness calculation formula is: wherein, y represents the predicted value of the maximum erosion rate of the well control manifold, y i y represents the actual value of the maximum erosion rate of the well control manifold; N represents the number of training samples, and w1 and w2 are both weight coefficients.
4. The method of predicting the rate of erosion of a marine oil and gas exploration well control manifold according to claim 3, wherein, The values of w1 and w2 satisfy: w1∈(0.3,1), w2∈(0.3,1), and w1+w2=1.
5. The method of predicting the rate of erosion of a marine oil and gas exploration well control manifold according to claim 3 or 4, wherein, In the step 3, the arrays with high fitness are directly selected into the next generation according to a proportion of 30%, and the crossing and mutation proportions are 60% and 10%, respectively.
6. The method of predicting the rate of erosion of a marine oil and gas exploration well control manifold according to claim 5, wherein, Before the step three, it further comprises constructing a sample set, comprising the following steps: Step a, determining the value range of each input parameter according to the actual situation; Step b, uniformly interval sampling in the value range of each input parameter; Wherein, the number of each input parameter value is the same; Step c, obtaining a plurality of samples by using orthogonal experiment to form the sample set.
7. The method of predicting the rate of erosion of a marine oil and gas exploration well control manifold according to claim 6, wherein, In the step four, it further comprises: When the fluid viscosity is greater than 1.5 x 10 -3 kg / (m·s), the maximum erosion rate of the well control manifold to be predicted output by the well control manifold erosion rate prediction model is corrected as follows: Wherein, μ represents the fluid viscosity, μ0 represents the reference fluid viscosity; E0 is the maximum erosion rate of the well control manifold to be predicted output by the well control manifold erosion rate prediction model.