Underground drilling parameter prediction method and device based on deep learning
By building a multi-layer perceptron model based on deep learning, the problem that traditional drilling methods are difficult to achieve accurate prediction of downhole drilling parameters is solved, real-time and accurate prediction of drilling parameters is achieved, and drilling efficiency and safety are improved.
Patent Information
- Application Number
- CN202311832034.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-27
- Publication Date
- 2025-06-27
AI Technical Summary
Traditional drilling methods have challenges in facing complex geological environments and improving efficiency, reducing costs and ensuring safety, making it difficult to achieve accurate prediction of underground drilling parameters.
Using a deep learning method, a deep learning drilling downhole parameter prediction model based on multi-layer perception machines is constructed to obtain drilling data in real time and make predictions.
Real-time and accurate prediction of underground drilling parameters is achieved, helping to optimize drilling operations, improve drilling efficiency and safety, and reduce costs.
Smart Images
Figure CN120216899A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of drilling, and in particular to a downhole drilling parameter prediction method and device based on deep learning. Background Art
[0002] With the continuous growth of global energy demand and the in-depth exploration of oil and gas resources, traditional drilling methods are facing more and more challenges. In order to improve exploration and development efficiency, reduce costs, ensure operation safety, and adapt to complex geological environments, intelligent drilling technology has emerged. Intelligent drilling technology is the intersection of information technology, automation, sensor technology, and data analysis, bringing unprecedented revolutionary changes to the drilling field.
[0003] Downhole drilling parameter prediction is of great significance in the process of oil drilling. It can not only improve the efficiency and safety of drilling operations, but also help reduce costs, protect the geological environment, and provide reliable data support for real-time decision-making. Drilling parameter prediction has become an indispensable part of intelligent drilling technology. In actual operation, by analyzing formation property parameters, fluid parameters, drill string parameters, and bit parameters, as well as surface drilling parameters, parameters such as drilling rate, weight on bit, and torque in the corresponding downhole can be predicted, so as to guide drilling operations. Through accurate prediction of drilling parameters, potential drilling engineering problems can be identified in advance, and corresponding measures can be taken to avoid failure risks. Summary of the Invention
[0004] The present invention provides a downhole drilling parameter prediction method and device based on deep learning to achieve real-time and accurate prediction of downhole drilling parameters, thereby guiding drilling operations.
[0005] For this purpose, the present invention provides the following technical solutions:
[0006] A downhole drilling parameter prediction method based on deep learning, the method comprising:
[0007] Collecting historical data of adjacent wells to generate an original data set;
[0008] Preprocessing the original data set to generate a training sample set;
[0009] Using the training sample set to construct a deep learning drilling downhole parameter prediction model based on a multi-layer perceptron;
[0010] Obtaining drilling data in real time;
[0011] Using the drilling downhole parameter prediction model and the drilling data to predict real-time downhole drilling parameters.
[0012] Optionally, the types of data in the original data set include any one or more of the following: numerical type, text type.
[0013] Optionally, the preprocessing of the original data set includes any one or more of the following processes: data cleaning, dimensionality reduction.
[0014] Optionally, the data cleaning includes any one or more of the following: numerical type conversion, linear regression interpolation, outlier deletion, data standardization.
[0015] Optionally, the dimensionality reduction of the original data set includes:
[0016] Calculating the covariance matrix of the data in the standardized data set, where the covariance matrix is used to describe the correlation between different features in the data;
[0017] Performing eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector;
[0018] Sorting the eigenvalues in descending order, and selecting a set number of eigenvalues to generate a projection matrix of a set size;
[0019] Using the projection matrix to project the original data into a new multi-dimensional subspace to obtain a reduced-dimensional data set, and using the reduced-dimensional data set as a training sample set.
[0020] Optionally, the constructing of the deep learning drilling downhole parameter prediction model based on the multi-layer perceptron using the training sample set includes:
[0021] Determining the multi-layer perceptron model structure, where the multi-layer perceptron model structure includes: an input layer, a hidden layer, and an output layer;
[0022] Using the PReLU activation function for the neurons in the hidden layer;
[0023] Using the mean squared error as the loss function of the drilling downhole parameter prediction model;
[0024] Training the deep learning drilling downhole parameter prediction model based on the multi-layer perceptron using the training sample set, and using the Bayesian optimization algorithm to adjust the multi-layer perceptron model to obtain the global optimal solution;
[0025] Determining the drilling downhole parameter prediction model according to the global optimal solution.
[0026] A downhole drilling parameter prediction device based on deep learning, the device includes:
[0027] A data acquisition module, configured to acquire adjacent well historical data and generate an original data set;
[0028] A preprocessing module, configured to preprocess the original data set to generate a training sample set;
[0029] A model construction module for constructing a deep learning drilling downhole parameter prediction model based on a multi-layer perceptron using the training sample set;
[0030] A data acquisition module for acquiring drilling data in real time;
[0031] A prediction module for predicting real-time downhole drilling parameters using the drilling downhole parameter prediction model and the drilling data.
[0032] The preprocessing module includes:
[0033] A data cleaning unit for performing any one or more of the following operations on the data in the original data set: numerical type conversion, linear regression interpolation, outlier deletion, data standardization;
[0034] A dimensionality reduction element for performing dimensionality reduction processing on the data in the cleaned original data set to obtain a dimensionality-reduced data set.
[0035] Optionally, the dimensionality reduction element includes:
[0036] A calculation subunit for calculating the covariance matrix of the data in the standardized data set, and the covariance matrix is used to describe the correlation between different features in the data;
[0037] An eigenvalue decomposition subunit for performing eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector;
[0038] A sorting subunit for arranging the eigenvalues in descending order and selecting a set number of eigenvalues to generate a projection matrix of a set size;
[0039] A mapping subunit for projecting the original data into a new multi-dimensional subspace using the projection matrix to obtain a dimensionality-reduced data set, and using the dimensionality-reduced data set as the training sample set.
[0040] Optionally, the model construction module includes:
[0041] A model structure determination unit for determining the multi-layer perceptron model structure, and the multi-layer perceptron model structure includes: an input layer, a hidden layer, and an output layer;
[0042] An activation function determination unit for using the PReLU activation function for the neurons in the hidden layer;
[0043] A loss function determination unit for using the mean square error as the loss function of the drilling downhole parameter prediction model;
[0044] A training unit that trains a deep learning downhole drilling parameter prediction model based on the multi-layer perceptron using the training sample set, and adjusts the multi-layer perceptron model using the Bayesian optimization algorithm to obtain the global optimal solution;
[0045] A model generation unit for determining a downhole drilling parameter prediction model according to the global optimal solution.
[0046] The downhole drilling parameter prediction method and device based on deep learning provided by the present invention collect historical data of adjacent wells, construct a deep learning downhole drilling parameter prediction model based on a multi-layer perceptron, and use this model to realize real-time and accurate prediction of downhole drilling parameters. The solution of the present invention uses deep learning algorithms to learn from a large amount of drilling data and generate accurate prediction results, which can help operators optimize drilling operations. By predicting these downhole drilling parameters, the drilling process can be more precisely controlled, operation errors can be reduced, and drilling efficiency can be improved. Improving the downhole drilling process through deep learning technology, enhancing efficiency, safety, and cost-effectiveness, helps the oil and gas exploration and production industry achieve better results in drilling operations. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 is a flowchart of a downhole drilling parameter prediction method based on deep learning provided by the present invention;
[0048] Figure 2 is a flowchart of preprocessing the original data set in an embodiment of the present invention;
[0049] Figure 3 is a flowchart of constructing a deep learning downhole drilling parameter prediction model based on a multi-layer perceptron in an embodiment of the present invention;
[0050] Figure 4 is a schematic structural diagram of a multi-layer perceptron in an embodiment of the present invention;
[0051] Figure 5 is a schematic structural diagram of a downhole drilling parameter prediction device based on deep learning provided by the present invention;
[0052] Figure 6 is a schematic structural diagram of a model construction module in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0054] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments cannot be elaborated one by one here, but the embodiments of the present invention are not limited to the following embodiments.
[0055] As Figure 1 shown, it is a flowchart of a downhole drilling parameter prediction method based on deep learning provided by the present invention, including the following steps:
[0056] In step 101, historical data of adjacent wells are collected to generate an original data set.
[0057] Specifically, synchronized drilling parameters, logging parameters, and mud logging parameters can be retrieved from the historical databases of each oilfield to generate an original data set.
[0058] The collected data may include, but is not limited to, any one or more of the following: well depth, formation properties, formation pressure, formation permeability, bit size, drill string strength, mud density, mud viscosity, mud pH value, drilling speed, surface rotation speed, surface drilling pressure, drilling fluid flow rate, surface rotation speed, surface torque, etc., 15 characteristic inputs, and bottom hole drilling pressure, bottom hole rotation speed, bottom hole torque, 3 target outputs, and other data.
[0059] For example, in a specific application, the drilling, logging, and mud logging parameters obtained by collecting and organizing the historical data of an oilfield are shown in the following Table 1 for the generated original data set:
[0060] Table 1
[0061]
[0062]
[0063] The types of data in the original data set include any one or more of the following: numerical type, text type. For example, the two parameter data of formation properties and formation permeability in Table 1 above are text type data, and other parameter data are numerical type.
[0064] In step 102, the original data set is preprocessed to generate a training sample set.
[0065] In the embodiment of the present invention, preprocessing the original data set may include, but is not limited to, any one or more of the following processes: data cleaning, dimensionality reduction.
[0066] Figure 2 It shows a flowchart for preprocessing the original data set in the embodiment of the present invention, including the following steps:
[0067] Step 201, convert non - numerical classification information into numerical information, and convert formation properties into form codes.
[0068] For example, conversion can be performed according to the formation property comparison table shown in Table 2 below.
[0069] Table 2
[0070]
[0071] Step 202: Determine and process all missing values.
[0072] Specifically, missing values can be filled by linear interpolation. The linear interpolation formula is as follows:
[0073]
[0074] Where X is the abscissa to be estimated, i.e., the position where the missing value is located;
[0075] X1 and X2 are the abscissas of known data points;
[0076] Y1 and Y2 are the ordinates of known data points;
[0077] Y is the missing value estimated at position X.
[0078] Step 203: Identify and process all outliers.
[0079] For example, the Z-score method can be used to detect outliers.
[0080] First, calculate the standard deviation σ:
[0081]
[0082] Where n is the number of feature data; x i is the value of the i-th feature data; μ is the mean of the feature data.
[0083] Then, use the Z-score method to determine outliers:
[0084]
[0085] If the Z value of the feature data point x is greater than a set value (such as 3), then the data point x is considered an outlier.
[0086] For the identified outliers, the outlier can be deleted, and the corresponding replacement value can be calculated by interpolation.
[0087] Step 204: Standardize the data set so that each feature parameter data has a zero mean and unit variance.
[0088] The specific processing is as follows:
[0089]
[0090] where x′ is the standardized feature parameter data, and x i is the i-th feature data, is the feature mean, and σ is the feature standard deviation.
[0091] Step 205: Perform dimensionality reduction on the standardized dataset (i.e., the model input features).
[0092] Specifically, the dimensionality reduction of the standardized dataset can be performed in the following manner:
[0093] (1) Calculate the covariance matrix of the data in the standardized dataset. The covariance matrix is used to describe the correlation between different features in the data.
[0094] The covariance matrix describes the correlation between different features in the data, and its calculation formula is as follows:
[0095]
[0096] where C ij represents the element in the i-th row and j-th column of the covariance matrix, N is the number of samples, x i and x j are the i-th and j-th eigenvalue respectively, and are the i-th and j-th feature mean respectively, (i = 1, 2, 3... 15; j = 1, 2, 3... 15).
[0097] Through the above calculation, a covariance matrix C of a set size (such as 15×15) can be obtained. The covariance matrix C describes the linear relationship between each pair of features.
[0098] (2) Perform eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector, specifically as follows:
[0099] C = PDP -1
[0100] where P is a 15x15 matrix, and each column is the eigenvector of the covariance matrix C. D is a 15x15 diagonal matrix, and the elements on its diagonal are the eigenvalues of the covariance matrix C. P -1 is the inverse matrix of matrix P.
[0101] (3) Arrange the eigenvalues in descending order to obtain an eigenvalue sequence, and select a set number of eigenvalues from it to generate a projection matrix of a set size.
[0102] For example, retain the first set number (such as 10) of principal components in the eigenvalue sequence to obtain a 10×15 projection matrix.
[0103] (4) Project the original data into a new multi-dimensional subspace using the projection matrix to obtain a dimensionality-reduced dataset, and use the dimensionality-reduced dataset as the training sample set.
[0104] For example, use the projection matrix to project the original data into a new 10-dimensional subspace to obtain a new dataset. The new dataset only includes 10 principal component information, thus achieving the purpose of dimensionality reduction.
[0105] Merge the input feature dataset (10-dimensional) and the target output dataset (3-dimensional) to generate the training sample set.
[0106] The following example further illustrates the process of preprocessing the original dataset in step 102 above.
[0107] For example, in a specific embodiment, a linear interpolation processing method is used to process missing values and outliers to obtain the following dataset:
[0108]
[0109] Then, perform standardization processing on dataset D to obtain D′, ensuring that each feature has zero mean and unit variance:
[0110]
[0111] Then, perform dimensionality reduction on dataset D′ to obtain D″
[0112]
[0113] It is also possible to further divide 80% of the dimensionality-reduced dataset into the training sample set and 20% into the test sample set to obtain:
[0114] Training sample set D Train is a 13×3970 matrix, and the test sample set D Test is a 13×993 matrix.
[0115] Among them, D Train includes X train (10×3970) y train (3×3970);
[0116] D Test includes X Test (10×993) y Test (3×993).
[0117] Continue to refer to Figure 1, in step 103, a deep learning prediction model for downhole drilling parameters based on a multi-layer perceptron is constructed using the training sample set.
[0118] Figure 3 The flowchart of constructing a deep learning prediction model for downhole drilling parameters based on a multi-layer perceptron in an embodiment of the present invention is shown, including the following steps:
[0119] Step 301, set the structure of the multi-layer perceptron model.
[0120] As Figure 4 shown, the structure of the multi-layer perceptron model provided in an embodiment of the present invention includes: an input layer, a hidden layer, and an output layer.
[0121] For example, the number of neurons in the input layer can be set to 10, the number of neurons in the output layer can be set to 3 (weight on bit, rotary speed, torque), the number of hidden layers can be set to 2 layers, and the number of neurons in the hidden layer can be set to 6.
[0122] The hidden layer with an appropriate activation function can represent any decision boundary with arbitrary precision and can fit any smooth mapping with any precision. In an embodiment of the present invention, the PReLU (Parametric ReLU) non-linear activation function is used for the neurons in the hidden layer. The main function of the PReLU activation function is to improve the training stability of the neural network, increase the expression ability of the network, alleviate the vanishing gradient problem, and improve the adaptability of the network to different types of data, thereby helping to improve the performance of the deep learning model and enabling the model to fit more complex data relationships.
[0123] For any input neuron x, the output y of the PReLU activation function is as follows:
[0124]
[0125] where α is a learnable parameter, usually initialized to 0.01.
[0126] In the forward propagation process, for each neuron in the hidden layer, calculate its weighted input:
[0127]
[0128] where z is the weighted input of the neuron, w i is the weight, x i is the input feature, and b is the bias.
[0129] Apply the PReLU activation function to the weighted input z of each neuron:
[0130]
[0131] Pass the activated output y from the first hidden layer to the next hidden layer, and repeat the above process to pass it to the output layer.
[0132] Step 302, determine the loss function.
[0133] Specifically, the mean squared error (MSE) can be used as the loss function of this model so that the model can gradually approach the target value of the training data. The calculation formula of the mean squared error is as follows:
[0134]
[0135] Among them, MSE is the mean squared error, representing the average of the squared errors between the model prediction value and the true value;
[0136] n is the number of samples, representing the number of samples in the dataset;
[0137] y i is the true value of the i-th sample;
[0138] is the model prediction value of the i-th sample.
[0139] Step S303, use the Bayesian optimization algorithm to adjust the multi-layer perceptron model.
[0140] The Bayesian optimization algorithm is an iterative method for optimizing black-box functions, and its main principle is based on Bayesian statistical theory and Gaussian process models. The goal of this algorithm is to find the global optimal solution within a limited number of iterative steps.
[0141] The optimization process of the multi-layer perceptron model is as follows:
[0142] (1) Define the optimization objective
[0143] Take the mean squared error as the performance index of this optimization, and determine a set of hyperparameters including the number of neurons in the hidden layer and the learning rate as the input of the algorithm. In addition, define the search range of the hyperparameters. The number of neurons in the hidden layer is between 1 and 100, and the learning rate is between 0.001 and 0.1.
[0144] (2) Establish a surrogate model
[0145] The surrogate model is designed to approximate the performance of the MLP (Multilayer Perceptron) model under different combinations of hyperparameters. In this way, the surrogate model can be used to predict the performance of unevaluated combinations of hyperparameters without actually running the multi-layer perceptron model.
[0146] In the embodiment of the present invention, the specific process of modeling the surrogate model is as follows:
[0147] 1) Select the Gaussian function. The core of the Gaussian process is the covariance function. This model uses the Gaussian kernel function (RBF) to establish the correlation in the input space as follows:
[0148]
[0149] Among them, k(x, x′) represents the kernel function value between the input data points x and x′;
[0150] x represents an input data point, such as a feature vector containing 10 feature values;
[0151] x′ represents another input data point, such as another vector containing 10 feature values;
[0152] σ 2 is the noise variance, representing a hyperparameter of the kernel function, which controls the scale of the kernel function.
[0153] When the distance between x and x′ is farther, it means the distance between the two feature vectors is larger, and the surrogate model can approximate the performance of the multi-layer perceptron model under different hyperparameter combinations.
[0154] 2) Gaussian process modeling.
[0155] The GP model assumes that the target variable is Gaussian distributed at each input point. For each feature vector x in the input space, the target variable y satisfies the Gaussian distribution, that is:
[0156] y ∼ GP(μ(x), k(x, x′))
[0157] Among them, μ(x) is the mean function, set to 0 (i.e., no bias);
[0158] k(x, x′) is the RBF kernel function, representing the covariance between the input data points x and x′.
[0159] 3) Calculate the covariance matrix K(x, x′)
[0160] Input the training sample set into the above surrogate model, and a covariance matrix of n×n can be obtained by using the above Gaussian kernel function.
[0161] Among them, n represents the number of samples in the input training sample set;
[0162] The covariance matrix K(x, x′) contains the covariance information between all input points. The element K of K(x, x′) ij represents the covariance between the input points x i and x j That is, the correlation. This enables the Gaussian process to be used to model the joint distribution of function values at different input points.
[0163] 4) Hyperparameter Estimation
[0164] Estimate the hyperparameters of the Gaussian process model based on the selected mean function, covariance function, and observed data (covariance matrix).
[0165] In the embodiments of the present invention, maximum likelihood estimation can be used to estimate the hyperparameters, that is:
[0166] L(θ|D) = N(y|μ(x), K(x,x';θ) + σ 2 I);
[0167] where θ represents the hyperparameters, including the parameters of the kernel function and the noise variance;
[0168] D represents the observed data, including the input feature matrix x and the target vector y;
[0169] μ(x) represents the output of the mean function;
[0170] K(x,x';θ) represents the covariance matrix, which is a function of the hyperparameters θ;
[0171] σ 2 represents the noise variance;
[0172] I is the identity matrix.
[0173] Take the natural logarithm of the likelihood function to obtain the log-likelihood function, that is:
[0174] logL(θ|D) = -1 / 2 × [y^T(K(x,x';θ) + σ 2 I)^(-1)y + log|K(x,x';θ) + σ 2 I| + n × log(2π)]
[0175] This log-likelihood function contains the sample variance of the observed data and the uncertainty of the model.
[0176] By solving the equation that the partial derivative of the log-likelihood function with respect to the hyperparameters θ is equal to zero, the hyperparameter values of the maximum likelihood estimation can be obtained, that is:
[0177] θ = [θ1, θ2, σ 2
[0178] where θ1 is the length scale parameter of the kernel function; θ2 is the amplitude parameter of the kernel function; σ 2 is the noise variance.
[0179] 5) Construct a Gaussian process model.
[0180] Construct a Gaussian process model using the selected mean function, covariance function, and estimated hyperparameters, i.e.:
[0181] y ∼ GP(μ(x), k(x, x′; θ best ))
[0182] This model can be used for prediction. For a new input data point x*, the predicted value can be obtained by calculating the mean μ(x*) and covariance k(x*, x*'; θ best )
[0183] For example, the predicted value is: θ = [1, 2, 0.01].
[0184] (3) Select a sampling strategy.
[0185] For example, the Expected Improvement (EI) algorithm strategy can be used to optimize the hyperparameters. The definition of the EI function is as follows:
[0186] EI(x) = E[max(0, f(x best ) - f(x))]
[0187] where EI(x) represents the expected improvement value under the hyperparameter combination x, that is, the expectation of the improvement in model performance at this point;
[0188] x best represents the best hyperparameter combination that has been observed so far, i.e., the hyperparameter combination with the lowest MSE;
[0189] x represents other hyperparameter combinations except x best ;
[0190] f(x) represents the value of the mean squared error under the hyperparameter combination x.
[0191] (4) Initial sample collection.
[0192] That is, set the initial hyperparameters for the algorithm to start searching. For example, set the initial value of the number of hidden layers to 6 and the learning rate to 0.001.
[0193] (5) Iterative optimization
[0194] Select the point with the largest EI function value as the next sampling point because it indicates that there may be a greater opportunity for performance improvement near this point.
[0195] (6) Evaluate the mean squared error
[0196] Use the training set and the selected hyperparameter combination to train the model, and use the test set to test the performance of the model.
[0197] (7) Update the surrogate model.
[0198] When a new observation (mean squared error value), i.e., the mean squared error under the previous hyperparameter combination, is obtained, this observation is added to the existing set of observations. Then, using the existing set of observations, the surrogate model is updated to reflect the estimation of the mean squared error. At this time, the mean function and covariance function of the Gaussian process also need to be updated to adapt to the new observation.
[0199] (8) Repeat the iteration.
[0200] Repeat the above steps (5), (6), and (7): After the model is updated, continue the next round of iterative optimization. In the next iteration, according to the updated surrogate model and EI strategy, select the next hyperparameter combination to be evaluated, and repeat the entire process (i.e., evaluate the mean squared error and update the model).
[0201] (9) Output the optimal solution.
[0202] Stop the iteration when the mean squared error converges (i.e., the trend of the mean squared error changes from decreasing continuously to increasing). The Bayesian optimization algorithm terminates. At this time, the hyperparameter combination is considered the best and is used to train the final MLP model to obtain the minimum mean squared error.
[0203] Step S304, train the MLP model using the best hyperparameter combination.
[0204] Input the samples in the training sample set into the above - constructed multi - layer perceptron (MLP) model, train the model, compare the output of the model with the actual target value, and evaluate its performance.
[0205] According to the analysis of the prediction results, the hyperparameters of the Gaussian process (GP) model in the model can be adjusted, and then the above Bayesian optimization operation is repeated.
[0206] Finally, a deep - learning drilling downhole parameter prediction model based on the multi - layer perceptron is obtained.
[0207] After establishing the drilling downhole parameter prediction model, the test sample set mentioned above can be used to evaluate the model and evaluate the prediction effect of the model.
[0208] For example, accuracy, precision, mean squared error, and mean absolute error can be used to determine the prediction effect of downhole drilling parameters. The higher the accuracy and precision, and the lower the mean squared error and mean absolute error, the more accurate the model prediction. Among them:
[0209] The calculation formula for accuracy is as follows:
[0210]
[0211] Among them, TP represents the number of true positives (the number of actual positive examples correctly marked as positive by the model), TN represents the number of true negatives (the number of actual negative examples correctly marked as negative by the model), FP represents the number of false positives (the number of actual negative examples wrongly marked as positive by the model), and FN represents the number of false negatives (the number of actual positive examples wrongly marked as negative by the model).
[0212] The calculation formula for precision is as follows:
[0213]
[0214] The calculation formula for the mean squared error (MSE) is as follows:
[0215]
[0216] The calculation formula for the mean absolute error (MAE) is as follows:
[0217]
[0218] Continue to refer to Figure 1 , in step 104, drill data is obtained in real time.
[0219] The drill data to be collected can be determined according to the parameter data required to be input into the model, and the present invention does not make any limitation thereto.
[0220] In addition, for the drill data collected in real time, it can also be processed according to the preprocessing method in Figure 2 above to make the prediction results obtained from these data more accurate.
[0221] In step 105, the real-time downhole drilling parameters are predicted by using the downhole drilling parameter prediction model and the drill data.
[0222] For example, input real-time parameters:
[0223] x r =(x′1 x′2 x′3…x′ 15 )=(35 20 4 1…13.7 105 125 16.2)
[0224] Obtain the prediction result output by the model:
[0225] y=(y1 y2 y3)=(100 124 15.7)
[0226] Among them, x′1, x′2, x′3,…, x′ 15They respectively represent 15 characteristic input parameters such as well depth, formation properties, formation pressure, formation permeability, bit size, drill string strength, mud density, mud viscosity, mud pH value, drilling speed, surface rotation speed, surface drilling pressure, drilling fluid flow rate, surface rotation speed, and surface torque. y1, y2, and y3 represent 3 target outputs of bottom hole drilling pressure, bottom hole rotation speed, and bottom hole torque.
[0227] Correspondingly, the present invention also provides a downhole drilling parameter prediction device based on deep learning, as Figure 5 shown, which is a schematic structural diagram of the device.
[0228] The downhole drilling parameter prediction device 500 includes the following modules:
[0229] The data acquisition module 501 is used to collect historical data of adjacent wells and generate an original data set;
[0230] The preprocessing module 502 is used to preprocess the original data set to generate a training sample set;
[0231] The model construction module 503 is used to construct a deep learning drilling downhole parameter prediction model based on a multi-layer perceptron by using the training sample set;
[0232] The data acquisition module 504 is used to acquire drilling data in real time;
[0233] The prediction module 505 is used to predict real-time downhole drilling parameters by using the downhole drilling parameter prediction model and the drilling data.
[0234] Among them, the preprocessing module 502 may include a data cleaning unit and a dimensionality reduction unit.
[0235] Among them:
[0236] The data cleaning unit is used to perform any one or more of the following operations on the data in the original data set: numerical type conversion, linear regression interpolation, outlier deletion, and data standardization;
[0237] The dimensionality reduction unit is used to perform dimensionality reduction processing on the data in the cleaned original data set to obtain a dimensionality-reduced data set.
[0238] The dimensionality reduction unit may specifically include the following sub-units:
[0239] The calculation sub-unit is used to calculate the covariance matrix of the data in the standardized data set, and the covariance matrix is used to describe the correlation between different features in the data;
[0240] The eigen-decomposition sub-unit is used to perform eigen-value decomposition on the covariance matrix to obtain each eigen-value and the corresponding eigen-vector;
[0241] A sorting subunit, configured to sort the feature values in descending order and select a set number of feature values therefrom to generate a projection matrix of a set size;
[0242] A mapping subunit, configured to project the original data into a new multi-dimensional subspace by using the projection matrix to obtain a data set after dimensionality reduction, and use the data set after dimensionality reduction as a training sample set.
[0243] As Figure 6 shown, it is a schematic structural diagram of a model construction module in an embodiment of the present invention.
[0244] The model construction module 503 includes the following units:
[0245] A model structure determination unit 531, configured to determine the multi-layer perceptron model structure, where the multi-layer perceptron model structure includes: an input layer, a hidden layer, and an output layer;
[0246] An activation function determination unit 532, configured to use a PReLU activation function for neurons in the hidden layer;
[0247] A loss function determination unit 533, configured to use mean squared error as the loss function of the downhole drilling parameter prediction model;
[0248] A training unit 534, configured to train a deep learning downhole drilling parameter prediction model based on the multi-layer perceptron by using the training sample set, and use a Bayesian optimization algorithm to adjust the multi-layer perceptron model to obtain a global optimal solution;
[0249] A model generation unit 535, configured to determine a downhole drilling parameter prediction model according to the global optimal solution.
[0250] For the specific implementation manners of the above modules and units in the downhole drilling parameter prediction device of the present invention, reference may be made to the descriptions in the method embodiment of the present invention above, and details are not described herein again.
[0251] The downhole drilling parameter prediction method and device based on deep learning provided by the present invention collect historical data of adjacent wells, construct a deep learning downhole drilling parameter prediction model based on a multi-layer perceptron, and use this model to achieve real-time and accurate prediction of downhole drilling parameters. The solution of the present invention uses a deep learning algorithm to learn from a large amount of drilling data and generate accurate prediction results, thereby helping operators optimize drilling operations. By predicting these downhole drilling parameters, the drilling process can be more precisely controlled, operation errors can be reduced, and drilling efficiency can be improved. By improving the downhole drilling process through deep learning technology, improving efficiency, safety, and cost-effectiveness, it helps the oil and gas exploration and production industry achieve better results in drilling operations.
[0252] It should be noted that the terms "comprising" and "having" in the description, claims and above-mentioned drawings of the present invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.
[0253] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. Moreover, the system embodiments described above are only illustrative. The modules and units described as separate components may or may not be physically separated, that is, they may be located on one network unit or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative work.
[0254] The embodiments of the present invention have been introduced in detail above. Specific implementation manners are used herein to elaborate the present invention. The descriptions of the above embodiments are only used to help understand the method and system of the present invention. They are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by a person of ordinary skill in the art without creative work shall fall within the protection scope of the present invention. The content of this specification should not be construed as a limitation to the present invention. Therefore, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A downhole drilling parameter prediction method based on deep learning, characterized in that, The method includes: Collecting historical data of adjacent wells to generate an original data set; Preprocessing the original data set to generate a training sample set; Constructing a deep learning drilling downhole parameter prediction model based on a multi-layer perceptron using the training sample set; Obtaining drilling data in real time; Predicting real-time downhole drilling parameters using the drilling downhole parameter prediction model and the drilling data.
2. The downhole drilling parameter prediction method based on deep learning according to claim 1, wherein, The types of data in the original data set include any one or more of the following: numerical type, text type.
3. The downhole drilling parameter prediction method based on deep learning according to claim 1, wherein The preprocessing of the original data set includes any one or more of the following processes: data cleaning, dimensionality reduction.
4. The downhole drilling parameter prediction method based on deep learning according to claim 3, wherein The data cleaning includes any one or more of the following: numerical type conversion, linear regression interpolation, outlier deletion, data standardization.
5. The downhole drilling parameter prediction method based on deep learning according to claim 4, wherein, The dimensionality reduction of the original data set includes: Calculating the covariance matrix of the data in the standardized data set, where the covariance matrix is used to describe the correlation between different features in the data; Performing eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector; Sorting the eigenvalues in descending order, and selecting a set number of eigenvalues to generate a projection matrix of a set size; Using the projection matrix to project the original data into a new multi-dimensional subspace to obtain a dimensionality-reduced data set, and using the dimensionality-reduced data set as the training sample set.
6. The downhole drilling parameter prediction method based on deep learning according to any one of claims 1 to 5, characterized in that, The constructing of a deep learning drilling downhole parameter prediction model based on a multi-layer perceptron using the training sample set includes: Determining the multi-layer perceptron model structure, where the multi-layer perceptron model structure includes: an input layer, a hidden layer, and an output layer; Using the PReLU activation function for the neurons in the hidden layer; Using the mean squared error as the loss function of the drilling downhole parameter prediction model; Training a deep learning drilling downhole parameter prediction model based on the multi-layer perceptron using the training sample set, and using the Bayesian optimization algorithm to adjust the multi-layer perceptron model to obtain the global optimal solution; Determining the drilling downhole parameter prediction model according to the global optimal solution.
7. An underground drilling parameter prediction device based on deep learning, characterized in that, The device includes: A data acquisition module for collecting historical data of adjacent wells to generate an original data set; A preprocessing module for preprocessing the original data set to generate a training sample set; A model construction module for constructing a deep learning drilling downhole parameter prediction model based on a multi-layer perceptron using the training sample set; A data acquisition module for obtaining drilling data in real time; A prediction module for predicting real-time downhole drilling parameters using the drilling downhole parameter prediction model and the drilling data.
8. The downhole drilling parameter prediction device based on deep learning according to claim 7, wherein, The preprocessing module includes: A data cleaning unit for performing any one or more of the following operations on the data in the original data set: numerical type conversion, linear regression interpolation, outlier deletion, data standardization; A dimensionality reduction unit for performing dimensionality reduction processing on the data in the pre-cleaned original data set to obtain a dimensionality-reduced data set.
9. The downhole drilling parameter prediction device based on deep learning according to claim 8, characterized in that, The dimensionality reduction unit includes: A calculation subunit for calculating the covariance matrix of the data in the standardized data set, where the covariance matrix is used to describe the correlation between different features in the data; An eigenvalue decomposition sub-unit, configured to perform eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector; A sorting sub-unit, configured to sort the eigenvalues in descending order, and select a set number of eigenvalues therefrom to generate a projection matrix of a set size; A mapping sub-unit, configured to project the original data into a new multi-dimensional subspace by using the projection matrix to obtain a dimensionality-reduced data set, and use the dimensionality-reduced data set as a training sample set.
10. The downhole drilling parameter prediction device based on deep learning according to any one of claims 7 to 9, characterized in that, The model construction module includes: A model structure determination unit, configured to determine the multi-layer perceptron model structure, where the multi-layer perceptron model structure includes an input layer, a hidden layer, and an output layer; An activation function determination unit, configured to use the PReLU activation function for the neurons in the hidden layer; A loss function determination unit, configured to use the mean square error as the loss function of the drilling downhole parameter prediction model; A training unit, configured to train a deep learning drilling downhole parameter prediction model based on the multi-layer perceptron by using the training sample set, and adjust the multi-layer perceptron model by using the Bayesian optimization algorithm to obtain a global optimal solution; A model generation unit, configured to determine a drilling downhole parameter prediction model according to the global optimal solution.