Drilling parameter optimization method based on variational autoencoder artificial intelligence model

Abstract

The present application relates to a drilling parameter optimization method based on a variational autoencoder artificial intelligence model, comprising the following steps: first, collecting drilling engineering parameter data, logging data and drilling fluid performance data; performing data preprocessing on the collected well history data, which includes data screening, outlier processing and data standardization processing, and finally obtaining the preprocessed drilling engineering parameter data, logging data and drilling fluid performance data. The preprocessed data is used as input to construct and train an unsupervised variational autoencoder artificial intelligence model, and the hidden variable space feature variable is output. The hidden variable space feature variable is associated with the mechanical drilling speed to realize the recommendation of the optimal drilling parameter combination from the target mechanical drilling speed to the hidden variable space feature variable. The present application can more accurately and conveniently recommend the optimal drilling parameters at different well sections, thereby ensuring the smooth development of drilling speed and efficiency improvement.

Description

Technical Field

[0001] This invention relates to the field of drilling parameter optimization in drilling engineering, and more specifically, to a drilling parameter optimization method based on a variational autoencoder artificial intelligence model. Background Technology

[0002] Since the development of drilling engineering technology, increasing speed and efficiency has always been one of the goals pursued by drillers. The key to increasing speed and efficiency lies in two points: first, equipment and tools, as the foundation for speed improvement, play a crucial role in this process; second, given the current tools and equipment conditions, improving the matching between equipment, tools, and drilling parameters to fully utilize the potential of equipment and tools is also a key approach to increasing speed and efficiency. Currently, the problem of low matching between drilling parameters and drilling equipment persists. This problem severely restricts drilling speed and can lead to damage and failure of downhole tools and other complex downhole situations, resulting in significant economic losses. For this reason, improving the matching between equipment, tools, and drilling parameters has always been a concern for petroleum workers both domestically and internationally, and it has also been one of the effective ways to increase speed and efficiency.

[0003] As a key factor in improving drilling speed, current research on drilling parameter optimization mainly includes mechanical energy specificity and drilling rate equations. Mechanical energy specificity calculates the rock-breaking energy of the drill bit, assesses the bottom-hole operating conditions based on this energy, and then optimizes drilling parameters such as pressure on bit (Pb), rotational speed (SRT), and displacement. Drilling rate equations, on the other hand, use multiple regression to fit the regression relationship between drilling parameters such as Pb, Db, and displacement, as well as logging data, and mechanical drilling rate for roller cone or PDC bits. The regression equation determines the combination of drilling parameters for different target drilling rates. However, both of these methods still suffer from excessive human intervention and strong subjectivity in practical field applications. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a drilling parameter optimization method based on a variational autoencoder artificial intelligence model, which can more accurately and conveniently propose optimal drilling parameters for different well sections, thereby ensuring the smooth progress of drilling speed-up and efficiency improvement.

[0005] The technical solution adopted by this invention to solve its technical problem is: constructing a drilling parameter optimization method based on a variational autoencoder artificial intelligence model, including the following steps:

[0006] S1. By organizing the collected well history data, 14 parameters required for predicting the plugging formula are obtained: bulk modulus, Young's modulus, Poisson's ratio, uniaxial compressive strength of rock, maximum horizontal principal stress, minimum horizontal principal stress, shear modulus, formation type, drilling pressure, rotation speed, displacement, torque, drilling fluid density, and drilling fluid viscosity.

[0007] S2. The collected well history data is preprocessed, including the removal of linearly correlated data, outlier handling, and data normalization, to finally obtain the preprocessed engineering parameter data.

[0008] S3. Using the preprocessed data as input, and taking the stabilization and minimization of reconstruction error and KL loss as the standard, establish and optimize an unsupervised variational autoencoder artificial intelligence drilling parameter optimization model.

[0009] S4. After the variational autoencoder training is completed, the target mechanical drilling speed range is selected by corresponding latent variable parameters output by the variational autoencoder, and the variational autoencoder is used again to give the corresponding drilling parameter combination based on the latent variable parameters after the selection range.

[0010] S5. Based on the geological parameter range and the preset range of engineering parameters for a specific stratum, the recommended combination of drilling parameters under the target drilling rate is finally obtained.

[0011] According to the above scheme, the specific steps for deleting linearly correlated data in step S2 are as follows:

[0012] The Spearman correlation coefficient was used to calculate the correlation between parameters.

[0013] Based on the correlation between each feature parameter, evaluate the linear correlation between each parameter and remove data with high linear correlation.

[0014] The specific steps for handling outliers in data preprocessing are as follows: outlier removal is performed on the features, and data that deviates from the mean by three times the standard deviation are deleted.

[0015] The specific steps of the normalization process in the data processing are as follows:

[0016] The data is preprocessed using the maximum-minimum normalization method;

[0017] The min-max normalization method involves subtracting the minimum value from the data and then dividing by the difference between the maximum and minimum values, as shown in the following formula:

[0018]

[0019] In the formula, X i For the original data, Xnorm For new data, X max X represents the largest data value in the feature column. min This is the minimum data value in the feature column.

[0020] According to the above scheme, in step S3, the method for establishing and optimizing the unsupervised variational autoencoder artificial intelligence drilling parameter optimization model is as follows:

[0021] The number of hidden layers in the fully connected neural network model in the encoder and decoder is determined. Using the preprocessed data as the encoder input data, the activation functions of the fully connected neural networks in the decoder and encoder are determined to achieve nonlinear fitting of multidimensional data to low-dimensional data in the latent variable space. At the same time, the decoder output layer data is compared with the encoder input layer data to determine the fitting accuracy. The gradient optimization algorithm is determined to improve the model training accuracy and accelerate the minimization of KL loss and reconstruction loss.

[0022] According to the above scheme, the number of hidden layers in the fully connected neural network model in the encoder and decoder is determined by empirical methods.

[0023] According to the above scheme, the Selu activation function is used for the fully connected neural network, as shown in the following formula:

[0024]

[0025] In the formula, λ and α are the model preset parameters, x is the output value of the hidden layer node, selu(x) is the value after processing by the hidden layer activation function, and the output layer uses a linear function.

[0026] According to the above scheme, nonlinear fitting of multidimensional data to low-dimensional data in the latent variable space is achieved by sampling the encoder output value according to a normal distribution to obtain the low-dimensional data of the latent variables.

[0027] Based on the above scheme, the objective function for training in determining the gradient optimization algorithm to improve model training accuracy and accelerate the minimization of KL loss and reconstruction loss is as follows:

[0028]

[0029] in, It is used to characterize the degree of similarity between the latent variable distribution and the standard normal distribution. The reconstruction loss = -∫q(z|x)log p(x|z)dz is used to characterize the degree of similarity between the dataset output by the upsizing of the latent variable space and the original dataset.

[0030] According to the above scheme, in step S4, the specific steps for correspondingly selecting the target mechanical drilling rate range and the latent variables, and then using a variational autoencoder to provide the corresponding drilling parameter combination based on the latent variable parameters after selecting the range are as follows:

[0031] By selecting the target mechanical drilling speed range, the corresponding range of latent variable parameters can be determined in the latent variable space; by randomly selecting latent variable points within the corresponding latent variable space, the required number of latent variable points can be set here; through the decoder of the variational autoencoder, the latent variable points can be upgraded one by one to the actual drilling parameter combination dimension, realizing the initial recommendation of drilling parameter combination.

[0032] According to the above scheme, in step S5, the specific steps for screening based on the geological parameter range and the preset range of engineering parameters for a specific stratum are as follows:

[0033] Geological screening involves removing recommended drilling parameter combinations from those that exceed the range of original logging data for the specified formation.

[0034] Engineering screening is a secondary screening of drilling parameter combinations after geological screening. First, the recommended drilling combinations are input into the theoretical model of traditional drilling rate calculation and calculation is performed. Recommended combinations that differ from the target drilling rate by more than the user's expectation are deleted. For the retained drilling parameter combinations, a second screening is performed based on the applicable scope of the engineering machinery.

[0035] The drilling parameter optimization method based on the variational autoencoder artificial intelligence model of the present invention has the following beneficial effects:

[0036] This invention uses artificial intelligence methods to optimize drilling parameters, which can more accurately and conveniently propose optimal drilling parameters for different well sections, thereby ensuring the smooth progress of drilling speed and efficiency improvement, and solving the problems of poor matching of drilling parameters and strong subjectivity in drilling parameter optimization during the drilling process. Attached Figure Description

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0038] Figure 1 This is a data processing flowchart of the drilling parameter optimization method based on the variational autoencoder artificial intelligence model of the present invention;

[0039] Figure 2 This is a schematic diagram of the variational autoencoder model structure;

[0040] Figure 3 This is a flowchart illustrating the geological and engineering screening process after the variational autoencoder recommends drilling parameter combinations. Detailed Implementation

[0041] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0042] Figure 1 This is a data preprocessing flowchart in the technical solution of this invention. Data preprocessing includes linear correlation data removal, outlier handling, and normalization. Linear correlation data removal involves checking the correlation between data points. Since both the encoder and decoder in the model are fully connected neural networks, highly collinear features cannot provide new information and consume computational resources. By evaluating the magnitude of linear correlation, highly collinear data is removed. Outlier handling refers to smoothing the data to the maximum extent possible while preserving the original data as much as possible, and removing outliers. Data normalization involves using Min-Max standardization to map all features of the data to the same scale.

[0043] Figure 2 This is a schematic diagram of the neural network structure in the technical solution of the present invention. In order to fully consider the influence of formation conditions, drilling parameters, and drilling fluid properties on mechanical drilling speed, this example uses preprocessed logging data, drilling engineering parameters, and drilling fluid performance parameters as input to train a variational autoencoder model, which satisfies the nonlinear projection of high-dimensional drilling parameter combinations to low-dimensional latent variable space.

[0044] The drilling parameter optimization method based on the variational autoencoder artificial intelligence model of this invention includes the following steps:

[0045] 1) Collect historical data on plugging and sealing formula wells based on data mining.

[0046] By processing the collected drilling parameters, logging data, and drilling fluid performance data, 14 parameters required for drilling parameter optimization were obtained: bulk modulus, Young's modulus, Poisson's ratio, uniaxial compressive strength of rock, maximum horizontal principal stress, minimum horizontal principal stress, shear modulus, formation, drilling pressure, rotational speed, displacement, torque, drilling fluid density, and drilling fluid viscosity. The collected data were used as standard samples for training the model.

[0047] 2) The collected data will be preprocessed, including removing linearly correlated data, handling outliers, and normalizing the data to obtain the preprocessed data.

[0048] A further feature of the above data preprocessing is that:

[0049] 2-1) The data preprocessing described above is completed by analyzing the original dataset and constructing a data cleaning model and algorithm to ensure the training accuracy of the artificial intelligence model and the stability of its application.

[0050] The data preprocessing described above mainly includes removing linearly correlated data, handling outliers, and normalization. It is an indispensable step before building an artificial intelligence model. The main processing steps are as follows:

[0051] 2-2-1) Since the data feature columns may not meet the requirements of normal distribution, Spearman correlation coefficient is used to calculate the correlation between feature columns, and highly linearly correlated feature columns are deleted;

[0052] 2-2-2) For data with deleted linear correlation columns, in order to improve the signal-to-noise ratio, ensure the model's high robustness, and enhance the model's generalization ability, outlier removal processing is performed on the features, removing data that deviate from the mean by three times the standard deviation.

[0053] 2-2-3) In order to improve the model accuracy and accelerate the convergence speed of model training, the max-min normalization method is used to preprocess the data;

[0054] The min-max normalization method involves subtracting the minimum value from the data and then dividing by the difference between the maximum and minimum values, as shown in the following formula:

[0055]

[0056] In the above formula, X i For the original data, X norm For new data, X max X represents the largest data value in the feature column. min This is the minimum data value in the feature column.

[0057] 3) Using the preprocessed feature data as input, construct an unsupervised artificial intelligence variational autoencoder model;

[0058] A further feature of the above technical solution is that the variational autoencoder model construction scheme is as follows;

[0059] 3-1) The variational autoencoder consists of two parts: an encoder and a decoder, both of which are fully connected neural networks. Furthermore, the encoder is a 4-layer fully connected neural network, and the decoder is a 3-layer fully connected neural network. A gradient optimization algorithm is chosen to improve model training accuracy and accelerate the minimization of KL loss and reconstruction loss. The objective function for training is as follows:

[0060]

[0061] in It is used to characterize the degree of similarity between the latent variable distribution and the standard normal distribution. The reconstruction loss = -∫q(z|x)log p(x|z)dz is used to characterize the degree of similarity between the dataset output by the upsizing of the latent variable space and the original dataset.

[0062] 3-2) Determine the activation function. The Selu activation function is used for the hidden layer in the encoder and decoder, and a linear function is used for the output layer.

[0063]

[0064] Where λ and α are preset parameters of the model, x is the output value of the hidden layer node, and selu(x) is the value after processing by the hidden layer activation function. The output layer uses a linear function.

[0065] 3-3) The encoder determines the statistical mean and variance characteristics of the input combination distribution and samples it within the standard normal distribution to obtain the latent variable parameters in low dimension.

[0066] 3-4) Based on the obtained latent variable parameters, the latent variable space of the known mechanical drilling speed and latent variable parameters is obtained.

[0067] Figure 3 The drilling parameter combination screening implementation process of the technical solution of this invention is as follows:

[0068] 4) Match the latent variable parameters given by the encoder with the known mechanical drilling speed, and then give the corresponding latent variable range under the target mechanical drilling speed range.

[0069] 5) Within the range of latent variables, randomly select latent variable points and use them as input to the decoder in the variational autoencoder. The output is an unfiltered combination of drilling parameters.

[0070] 6) Within the above drilling parameter combinations, perform geological screening. Delete drilling parameter combinations that exceed the range of known formation characteristics (bulk modulus, Young's modulus, Poisson's ratio, uniaxial compressive strength of rock, maximum horizontal principal stress, minimum horizontal principal stress, shear modulus, formation type). This ensures that the recommended parameter combinations are suitable for the target formation.

[0071] 7) Input the drilling parameter combinations selected in step 6) into the traditional rate of penetration (RLP) calculation model to verify the recommended mechanical rate of penetration (MRP) values ​​for these combinations. Delete any drilling parameter combinations that exceed the user's tolerance.

[0072] 8) The drilling parameter combinations selected in 7) are further screened based on the objective applicable conditions of the drilling machinery and equipment, and parameter combinations within the scope of engineering construction capabilities are retained.

[0073] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A drilling parameter optimization method based on a variational autoencoder artificial intelligence model, characterized in that, Includes the following steps: S1. By organizing the collected well history data, 14 parameters required for predicting the plugging formula are obtained: bulk modulus, Young's modulus, Poisson's ratio, uniaxial compressive strength of rock, maximum horizontal principal stress, minimum horizontal principal stress, shear modulus, formation type, drilling pressure, rotation speed, displacement, torque, drilling fluid density, and drilling fluid viscosity. S2. The collected well history data is preprocessed, including the removal of linearly correlated data, outlier handling, and data normalization, to finally obtain the preprocessed engineering parameter data. S3. Using the preprocessed data as input, and taking the stabilization and minimization of reconstruction error and KL loss as the standard, establish and optimize an unsupervised variational autoencoder artificial intelligence drilling parameter optimization model. S4. After the variational autoencoder training is completed, the target mechanical drilling speed range is selected by corresponding latent variable parameters output by the variational autoencoder, and the variational autoencoder is used again to give the corresponding drilling parameter combination based on the latent variable parameters after the selection range. S5. Based on the geological parameter range and the preset range of engineering parameters for a specific stratum, the recommended combination of drilling parameters under the target drilling rate is finally obtained.

2. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 1, characterized in that, In step S2, the specific steps for deleting linearly correlated data are as follows: The Spearman correlation coefficient was used to calculate the correlation between parameters. Based on the correlation between each feature parameter, evaluate the linear correlation between each parameter and remove data with high linear correlation. The specific steps for handling outliers in data preprocessing are as follows: outlier removal is performed on the features, and data that deviates from the mean by three times the standard deviation are deleted. The specific steps of the normalization process in the data processing are as follows: The data is preprocessed using the maximum-minimum normalization method; The min-max normalization method involves subtracting the minimum value from the data and then dividing by the difference between the maximum and minimum values, as shown in the following formula: In the formula, X i For the original data, X norm For new data, X max X represents the largest data value in the feature column. min This is the minimum data value in the feature column.

3. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 1, characterized in that, In step S3, the method for establishing and optimizing the unsupervised variational autoencoder artificial intelligence drilling parameter optimization model is as follows: The number of hidden layers in the fully connected neural network model in the encoder and decoder is determined. Using the preprocessed data as the encoder input data, the activation functions of the fully connected neural networks in the decoder and encoder are determined to achieve nonlinear fitting of multidimensional data to low-dimensional data in the latent variable space. At the same time, the decoder output layer data is compared with the encoder input layer data to determine the fitting accuracy. The gradient optimization algorithm is determined to improve the model training accuracy and accelerate the minimization of KL loss and reconstruction loss.

4. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 3, characterized in that, The number of hidden layers in the fully connected neural network model in the encoder and decoder is determined empirically.

5. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 3, characterized in that, The fully connected neural network uses the Selu activation function, as shown below: In the formula, λ and α are the model preset parameters, x is the output value of the hidden layer node, selu(x) is the value after processing by the hidden layer activation function, and the output layer uses a linear function.

6. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 3, characterized in that, Nonlinear fitting of multidimensional data to low-dimensional data in the latent variable space is achieved by sampling the encoder output values ​​according to a normal distribution to obtain low-dimensional latent variable data.

7. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 3, characterized in that, Among the gradient optimization algorithms used to improve model training accuracy and accelerate the minimization of KL loss and reconstruction loss, the training objective function is as follows: in, It is used to characterize the degree of similarity between the latent variable distribution and the standard normal distribution. The reconstruction loss = -∫q(z|x)logp(x|z)dz is used to characterize the degree of similarity between the dataset output by the upsizing of the latent variable space and the original dataset.

8. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 1, characterized in that, In step S4, the specific steps for correspondingly selecting the target mechanical drilling rate range and the latent variables, and then using a variational autoencoder to provide the corresponding drilling parameter combination based on the latent variable parameters after selecting the range are as follows: By selecting the target mechanical drilling speed range, the corresponding range of latent variable parameters can be determined in the latent variable space; by randomly selecting latent variable points within the corresponding latent variable space, the required number of latent variable points can be set here; through the decoder of the variational autoencoder, the latent variable points can be upgraded one by one to the actual drilling parameter combination dimension, so as to achieve a preliminary recommendation of the drilling parameter combination.

9. The drilling parameter optimization method based on a variational autoencoder artificial intelligence model according to claim 1, characterized in that, In step S5, the specific steps for screening based on the geological parameter range and the preset engineering parameter range of a specific stratum are as follows: Geological screening involves removing recommended drilling parameter combinations from those that exceed the range of original logging data for the specified formation. Engineering screening is a secondary screening of drilling parameter combinations after geological screening. First, the recommended drilling combinations are input into the theoretical model of traditional drilling rate calculation and calculation is performed. Recommended combinations that differ from the target drilling rate by more than the user's expectation are deleted. For the retained drilling parameter combinations, a second screening is performed based on the applicable scope of the engineering machinery.

Images