Drilling tool jamming prediction method and system based on physical and data dual-drive fusion

Abstract

The invention relates to a drill jamming prediction method and system based on physical and data dual-drive fusion, and belongs to the technical field of well drilling. The method comprises the following steps: preprocessing historical logging data; establishing a physical model of drill jamming; based on the data obtained through preprocessing, a data driving model of drill jamming is established; fusing the physical model of the stuck drill with the data driving model; performing model increment adjustment on the fused dual-drive model; and on the basis of the dual-drive fusion model after increment adjustment and the new logging data, predicting whether the drilling tool is stuck or not. The accuracy of the drilling tool jamming prediction method based on physical and data dual-drive fusion and after increment adjustment is greatly improved compared with the accuracy of the drilling tool jamming prediction method based on pure data drive.

Description

Technical Field

[0001] The present invention belongs to the technical field of drilling, and in particular relates to a method and system for predicting a stuck pipe based on dual-drive fusion of physics and data. Background Art

[0002] Pipe sticking is a dangerous phenomenon in the drilling process that may lead to serious accidents and losses. Predicting pipe sticking can help drilling engineers and operators realize the risk of pipe sticking in advance and take corresponding measures to ensure the safety of drilling operations and prevent accidents.

[0003] At present, there are several commonly used technologies that can be used to predict stuck drill: the first is a rule-based expert system, which formulates a series of rules based on expert knowledge and experience to predict stuck drill. These rules can be determined based on drilling parameters, rock formation properties, drilling tool status and other factors. Although the rule-based expert system can provide a certain degree of prediction accuracy and interpretability, it is limited by the subjectivity of expert knowledge and the rigidity of the rules. The second is the physical model, which is based on the mechanism of stuck drill and mechanical principles and predicts stuck drill by establishing mathematical equations or models. This method can provide an in-depth understanding and explanation of the mechanism of stuck drill, but its prediction results may have certain uncertainties. Then there is the data-driven model, which uses machine learning algorithms to analyze and model historical data to predict stuck drill. Commonly used algorithms include decision trees, support vector machines, random forests, neural networks, etc. This method can learn patterns and laws in the drilling process from data and provide more accurate prediction results, but it may lack interpretability and reliability. Summary of the invention

[0004] In response to the above problems, the present invention provides a method and system for predicting drill stuck based on the fusion of physics and data dual drive. It is a prediction method that combines physical models and data-driven models, and combines the advantages of the physical model's in-depth understanding and interpretation of the drill stuck mechanism, and the more accurate prediction results of the data-driven model, while avoiding the shortcomings of a single physical model that cannot make accurate predictions, and a single data-driven model's low interpretability and reliability; the prediction method in the present invention integrates the two models in a way that takes advantage of their strengths and avoids their weaknesses, which can improve the accuracy and reliability of drill stuck predictions. At the same time, the method can also provide an explanation of the drill stuck mechanism and the prediction results, and enhance the credibility of the prediction results. This method has important application value in actual drilling operations, and can help drilling engineers predict the risk of drill stuck, take corresponding measures, and improve the efficiency and safety of drilling operations.

[0005] The first object of the present invention is to provide a method for predicting stuck pipe based on physics and data dual-driven fusion, comprising:

[0006] Preprocessing of historical logging data;

[0007] Establish a physical model of drill sticking;

[0008] Based on the preprocessed data, a data-driven model for stuck drill is established;

[0009] Merge the physical model of stuck drill with the data-driven model;

[0010] Perform incremental adjustments to the fused dual-drive model;

[0011] Based on the incrementally adjusted dual-drive fusion model and new logging data, a prediction is made on whether the drill is stuck.

[0012] In a specific embodiment of the present invention, the preprocessing of historical logging data includes:

[0013] Use SMOTE algorithm to process historical logging data;

[0014] The grey correlation analysis method is used to screen the historical logging data obtained through one-time processing to obtain the preprocessed data.

[0015] In a specific embodiment of the present invention, the data-driven model for stuck drill is established based on the preprocessed data, including:

[0016] Normalize the preprocessed data;

[0017] Based on the data obtained through normalization, different algorithm models are used for drill sticking training;

[0018] Based on the drill-stuck training results of different algorithm models, the final algorithm model is selected;

[0019] Based on the final optimized algorithm model and historical data, a data-driven model for drill sticking is established.

[0020] In a specific embodiment of the present invention, the drill stuck training results based on different algorithm models are used to select the final algorithm model, including:

[0021] Compare the drill-stuck training results of different algorithm models;

[0022] The algorithm model with the best performance in the comparison results is taken as the final algorithm model.

[0023] In a specific embodiment of the present invention, the expression of the data-driven model of the stuck drill is as follows:

[0024] Y pre =f k (w [k] ×f k-1 (w [k-1] ×…×f2(w[2] ×f1(w [1] ×X×b [1] )+b [2] )+…+b [k-1] )+b [k] )

[0025] Among them, Y pre is the predicted value of the data-driven model for stuck drill; w [k] represents the weight coefficient matrix of the kth layer, which means the connection weight between each neuron in the previous layer and each neuron in the current layer; b [k] is the deviation of the kth layer, f k (x) represents the activation function of the kth layer, and X is the input parameter of the neural network model.

[0026] In a specific embodiment of the present invention, the fusion of the physical model of the stuck drill and the data-driven model includes:

[0027] Based on the data-driven model, the partial derivative of the well depth is obtained, and the relationship between the physical model and the weights and biases of the data-driven model is established;

[0028] Based on the relationship between the weights and biases of the physical model and the data-driven model, a fused dual-drive model is obtained.

[0029] In a specific embodiment of the present invention, the relational expression of the fused dual-drive model is as follows:

[0030]

[0031] Among them, MSE is the mean square error, which is used to evaluate the deviation between the predicted value and the true value; Y pre Predicted value of the data-driven model for stuck pipe, Y true is the true value; n is the number of training samples or test samples; A and represent the partial derivatives of the stuck drill data-driven model and stuck drill mechanism model with respect to well depth, respectively; st is the abbreviation of subject to, which means "under certain conditions", and the certain condition is the condition expressed by the expression after "{";

[0032] The expression of the fused dual-drive model means that in A and A theory Under the constraint that the difference is 0, adjust the model parameter weight w and bias b to minimize the MSE value between the predicted value and the true value.

[0033] In a specific embodiment of the present invention, after the incremental adjustment is completed, the method further includes:

[0034] The confusion matrix shows the difference between the prediction results and the actual results of the incrementally adjusted stuck pipe prediction model;

[0035] Visualize and output the probability of stuck pipe predicted by the incrementally adjusted stuck pipe prediction model.

[0036] The second object of the present invention is to provide a stuck pipe prediction system based on physics and data dual-driven fusion, comprising:

[0037] The preprocessing module is used to preprocess the historical logging data;

[0038] The physical model module is used to establish the physical model of the stuck drill;

[0039] The data model module is used to establish a data-driven model for stuck drill based on the preprocessed data;

[0040] The fusion module is used to fuse the physical model of the stuck drill with the data-driven model;

[0041] Perform incremental adjustments to the fused dual-drive model;

[0042] Based on the incrementally adjusted dual-drive fusion model and new logging data, a prediction is made on whether the drill is stuck.

[0043] A third object of the present invention is to provide an electronic device, comprising: a processor, wherein the processor is coupled to a memory;

[0044] The memory is used to store computer programs;

[0045] The processor is used to execute the computer program stored in the memory so that the electronic device executes the above method.

[0046] A fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a program or instruction, and when the program or instruction is executed on a computer, the computer executes the method as described above.

[0047] Beneficial effects of the present invention:

[0048] The present invention proposes a method and system for predicting stuck pipe based on the fusion of physics and data dual drive, which combines the physical model and the data driven model, fully utilizes the advantages of both, and finally obtains a more accurate stuck pipe prediction result.

[0049] The incremental learning method proposed in the present invention can use new data to perform incremental adjustments to the model when a prediction deviation of the dual-drive fusion model is detected, thereby improving the accuracy of the final stuck drill prediction model.

[0050] The accuracy of the drill stuck prediction method based on the fusion of physics and data dual drive and incremental adjustment proposed in the present invention is greatly improved compared with the accuracy of the drill stuck prediction method driven by pure data.

[0051] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 A flowchart of a method for predicting stuck pipe based on physics and data dual-driven fusion according to an embodiment of the present invention is shown;

[0054] Figure 2 A schematic diagram of generating new samples using the SMOTE algorithm according to an embodiment of the present invention is shown;

[0055] Figure 3 A heat map of a grey relational analysis method according to an embodiment of the present invention is shown;

[0056] Figure 4 A prediction process diagram of a neural network model according to an embodiment of the present invention is shown;

[0057] Figure 5 A diagram showing the accuracy of training data and verification data of a dual-drive fusion model obtained after fusion according to an embodiment of the present invention is shown;

[0058] Figure 6 A visualization output diagram of a confusion matrix in a dual-drive fusion model obtained after fusion according to an embodiment of the present invention is shown;

[0059] Figure 7 A visualization diagram of the predicted probability of occurrence of a stuck pipe in a dual-drive fusion model obtained after fusion according to an embodiment of the present invention is shown;

[0060] Figure 8 A framework diagram of a stuck pipe prediction system based on physics and data dual-driven fusion according to an embodiment of the present invention is shown;

[0061] Fig. 9 A framework diagram of an electronic device according to an embodiment of the present invention is shown;

[0062] In the figure: preprocessing module 1; physical model module 2; data model module 3; fusion model module 4; incremental module 5; prediction model 6; electronic device 300; processor 301; memory 302. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, a method for predicting stuck pipe based on physics and data dual-driven fusion according to an embodiment of the present invention includes:

[0065] S1. Preprocessing of historical logging data;

[0066] S2, establishing a physical model of drill sticking;

[0067] S3. Establishing a data-driven model for stuck drill based on the preprocessed data;

[0068] S4, integrating the physical model of drill sticking and the data-driven model;

[0069] S5, performing incremental model adjustment on the fused dual-drive model;

[0070] S6. Predict whether the drill is stuck based on the incrementally adjusted dual-drive fusion model and the new logging data.

[0071] Based on the dual-drive fusion model obtained after fusion and the new logging data, whether the drill is stuck is predicted.

[0072] Step S1, preprocessing the historical logging data, including:

[0073] i) Use SMOTE algorithm to process the historical logging data;

[0074] In this step, the SMOTE algorithm (full name: Synthetic Minority Over-sampling Technique) is used to solve the class imbalance problem. It generates synthetic samples to balance the categories, thereby improving the model's prediction ability for the minority class.

[0075] The idea of ​​the SMOTE algorithm is to synthesize new minority class samples. The synthesis strategy is to randomly select a sample b from its nearest neighbor for each minority class sample a, and then randomly select a point on the line between a and b as the newly synthesized minority class sample. The algorithm calculation formula is shown in formula (1):

[0076] X new =X ref +r×(X nn -X ref ) (1)

[0077] In formula (1), X new is a new sample; X ref is the reference sample; r is any value between 0 and 1; X nn is the nearest neighbor sample;

[0078] In the embodiment of the present invention, the ratio of the original sample without stuck drill (0) and the sample with stuck drill (1) is 700:6, and the sample data belongs to category imbalanced data;

[0079] Before prediction, it is best to perform data prediction processing and use the SMOTE model to generate 1:1 data. Figure 2 The visualization shown in the figure shows that the ratio of samples without stuck pipe (0) and samples with stuck pipe (1) is 700:100. Figure 2 In the figure, the black dots are samples where pipe sticking occurred, and the gray dots are samples where pipe sticking did not occur.

[0080] ii) Use the grey correlation analysis method to screen the historical logging data obtained by one-time processing to obtain the pre-processed data, specifically:

[0081] The grey correlation analysis method is used to conduct correlation analysis on the logging data obtained in one process, and the main control variables with greater influence are selected (the degree of closeness between each influencing factor and the occurrence of pipe sticking is judged according to the size of the grey correlation, so as to determine which are the main influencing factors and which are the secondary influencing factors), and irrelevant variables are screened out to reduce the dimension of input parameters;

[0082] Figure 3 The thermal diagram of the well logging data analyzed by grey correlation analysis in the embodiment of the present invention is given. Figure 3 It can be clearly concluded that the size of the correlation index of each parameter affecting drill sticking is large. In the embodiment of the present invention, characteristic parameters with a correlation coefficient greater than 0.5 are selected, and correlation analysis is carried out between these characteristic parameters, focusing on the parameters with a correlation greater than 0.7, and preferably the parameters with greater credibility are used as input parameters.

[0083] The closeness of each influencing factor to the occurrence of pipe sticking is judged according to the size of the grey correlation degree, so as to determine the main influencing factors, which are well depth, drilling speed, hook load, drilling pressure, rotary table speed, standpipe pressure, pump stroke and DC index.

[0084] For example, the training sample data after grey relational analysis is shown in Table 1.

[0085] Table 1

[0086]

[0087] In step S2, a physical model of drill stuck is established. Specifically, different key parameters correspond to different drill stuck types. For differential pressure drill stuck, the key parameter is the pressure difference between the bottom hole pressure and the formation pore pressure; for other drill stuck types such as collapse drill stuck, reduced diameter drill stuck, mud bag drill stuck, etc., the key parameters may be one or more of the key parameters such as the pressure difference between the bottom hole pressure and the collapse pressure, the big hook load F, the drill torque T, the riser pressure ps, etc. Therefore, the physical model of drill stuck includes drill stuck models such as differential pressure drill stuck model, collapse drill stuck model, reduced diameter drill stuck model, mud bag drill stuck model, etc.

[0088] The embodiment of the present invention exemplarily illustrates step S2 of the method provided by the invention by taking differential pressure sticking as an example. The establishment of the drill sticking models such as collapse drill sticking model, reduced diameter drill sticking model, mud bag drill sticking model, etc. is similar to the establishment of the differential pressure sticking physical model. The process involves the calculation and derivation process of physical quantities such as the pressure difference between the bottom hole pressure and the collapse pressure, the big hook load, the drilling tool torque, and the riser pressure. The calculation formulas of the physical quantities involved are well known to those skilled in the art, and the present invention does not specifically limit them.

[0089] At the same time, the drill stuck prediction method provided by the present invention can establish a physical model of drill stuck for any number of physical quantities such as the pressure difference between bottom hole pressure and formation pore pressure, the pressure difference between bottom hole pressure and collapse pressure, hook load, drilling tool torque, and riser pressure;

[0090] In differential pressure drilling, the differential pressure can be expressed as shown in formula (2):

[0091] ΔP=P b -P p (2)

[0092] In formula (2), P b is the bottom hole pressure, P p is the formation pore pressure.

[0093] Bottom hole pressure P b Calculated by formula (3):

[0094] P b =P HY +(P AF +P BACK) (3)

[0095] In formula (3), P HY is the annulus hydrostatic pressure, MPa; P AF is the annular circulation pressure loss, MPa; P BACK is the wellhead back pressure, MPa.

[0096] Formation pore pressure P p Calculated by formula (4):

[0097] P p =ρ p ·gh (4)

[0098] In formula (4), ρ p is the equivalent density of formation pore pressure, g / cm 3 ; g is the acceleration due to gravity, usually 9.81m / s 2 ; h is the formation depth, m.

[0099] Among them, the formation pore pressure equivalent density ρ p Calculated by formula (5):

[0100]

[0101] In formula (5), ρ o is the equivalent density of the overburden pressure gradient, g / cm 3 ρ n is the normal formation pressure equivalent density, i.e. the density of formation water, g / cm 3 ;d c is the actual calculated value of depth h, dimensionless; d cn is the normal trend line value of depth h, dimensionless.

[0102] Among them, at any depth h in the formation o The actual value of the equivalent density of the overburden pressure gradient can be calculated by formula (6):

[0103]

[0104] In formula (5), h is the burial depth, m; ρ(h) is the density of the rock mass at the burial depth, g / cm 3 .

[0105] d c The value is the corrected d index value, which can be obtained by formula (7):

[0106]

[0107] In formula (7), ρ n is the normal formation pressure equivalent density, g / cm 3 ρd is the actual drilling fluid density, g / cm 3 ; R is mechanical drilling speed, m / h; N is drill bit speed, r / min; W is drilling pressure, kN; d b is the drill diameter, mm.

[0108] Normal trend line equation d cn It can be expressed by formula (8):

[0109] d cn =10 ah+b (8)

[0110] In formula (8), h is the well depth, m; a is the slope of the trend line; and b is the intercept of the trend line.

[0111] In step S3, a data-driven model for drill sticking is established based on the data obtained by preprocessing. Similar to step S2, the data-driven model for drill sticking in step S3 includes a data-driven model for drill sticking due to differential pressure, a data-driven model for drill sticking due to collapse, a data-driven model for drill sticking due to reduced diameter, a data-driven model for drill sticking due to mud packing, and other data-driven models for drill sticking;

[0112] Specifically include:

[0113] i) Normalize the preprocessed data:

[0114] Data normalization is to scale the data according to a certain ratio so that it falls within a specific range. Common normalization methods include maximum-minimum scaling and standardization, which can improve the robustness and generalization ability of the prediction model and optimize the prediction effect;

[0115] The maximum-minimum scaling is to linearly scale the data to the interval between [0,1], and the calculation formula is shown in formula (9):

[0116]

[0117] In formula (9), X i is the current data, X min is the minimum value of all data, X max is the maximum value of all data;

[0118] The standardization is to transform the data into a distribution with a mean of 0 and a standard deviation of 1. The calculation formula is shown in formula (10):

[0119]

[0120] In formula (10), X i is the current data, μ is the mean of all data, and σ is the standard deviation of all data.

[0121] As shown in Table 2 (Comparison of the accuracy of the prediction model before and after normalization), it can be seen from the subsequent model accuracy evaluation that before the normalization (maximum-minimum scaling and standardization) process, only part of the data was predicted accurately, and after the normalization process, the accuracy was improved.

[0122] Table 2

[0123]

[0124] ii) Based on the normalized data, different algorithm models (BP neural network, support vector machine, logistic regression, decision tree, naive Bayes, random forest) are used to perform drill stuck training;

[0125] iii) based on the drill-stuck training results of different algorithm models, the final algorithm model is selected;

[0126] The above steps ii) and iii) use BP neural network, support vector machine, logistic regression, decision tree, naive Bayes and random forest to train the normalized data to find the best model and provide a model basis for subsequent predictions;

[0127] During the drill bit training process, the data obtained by normalization processing is used as sample data, and the sample set data is divided into 80% for training and 20% for testing. Different models are tried separately and their performance indicators on the validation set are compared, such as accuracy, precision, etc. The best performing model is selected for further data model construction. The model prediction accuracy corresponding to different algorithms is shown in Table 2. From the results in Table 2, the BP neural network model is preferably used as the data-driven model, and the trained prediction model is saved for easy application. For example, in Python, the pickle.dump function can be used to save the trained model.

[0128] iv) Based on the final algorithm model selected and the data obtained through normalization, a data-driven model for drill sticking is established:

[0129] From the data comparison in Table 2, it can be seen that the final algorithm model selected is the BP neural network model;

[0130] Next, create a neural network model. For example, in Python, you can use the Sequential function of the tensorflow.keras library to create a sequence model. Define two hidden layers and one output layer by adding a tf.keras.layers.Dense layer, such as Figure 4As shown, the two hidden layers have 64 and 32 neurons respectively, and the output layer has 1 neuron. Use the compile function to compile the model, specify the loss function as binary cross entropy (binary_crossentropy), and the optimizer as adam.

[0131] Then, train the model, such as using the fit function in Python: pass in the input features and target variables of the training set, set the number of training iterations (epochs) to 100, and the batch size (batch_size) to 32.

[0132] Then, make predictions on the test set, such as using the predict function in Python to predict the input features of the test set and get the prediction result y_pred. Then, use numpy's where function to convert probability values ​​greater than 0.5 to 1 and probability values ​​less than or equal to 0.5 to 0.

[0133] Finally, output the evaluation indicators, such as using the classification_report function in Python to calculate and output the classification report of the model on the test set. The classification report includes indicators such as precision, recall, and F1-score.

[0134] S4. Integrate the physical model of stuck drill and the data-driven model, including:

[0135] i) Based on the data-driven model, the partial derivative of the well depth is obtained, and the relationship between the physical model and the weights and biases of the data-driven model is established:

[0136] First, Figure 4 The calculation process of the neural network model shown is converted into the following mathematical expression as shown in formula (11):

[0137] Y pre =f k (w [k] ×f k-1 (w [k-1] ×…×f2(w [2] ×f1(w [1] ×X×b [1] )+b [2] )+…+b [k-1] )+b [k] )(11)

[0138] In formula (11), Y pre is the predicted value of the data-driven model for stuck drilling, representing the pressure difference between the predicted bottom hole pressure and the formation pore pressure Pp pre , predict the pressure difference between bottom hole pressure and collapse pressure Pc pre , Predicted hook load value Fpre , predict the value of drilling tool torque T pre And the predicted riser pressure value Ps pre Any one of the parameters; w [k] represents the weight coefficient matrix of the kth layer, which means the connection weight between each neuron in the previous layer and each neuron in the current layer; b [k] is the deviation of the kth layer, f k (x) represents the activation function of the kth layer, and X is the input parameter of the neural network model;

[0139] For example, the mathematical expression of the stuck pipe prediction neural network model (i.e., the stuck pipe data-driven model) can be expressed by the following formula:

[0140] P pre =f(w [3] ×f(w [2] ×f(w [1] ×X+b [1] )+b [2] )+b [3] ) (12)

[0141] In formula (12), P pre is the predicted value of the data-driven model for stuck drill. Here, the differential pressure between the bottom hole pressure and the formation pore pressure is used as an example to facilitate the subsequent demonstration of the partial derivative of the well depth obtained by the data-driven model and the derivation process of establishing the relationship between the weight and bias of the physical model and the data-driven model; w [1] Represents the weight coefficient of the first layer, w [2] Represents the weight coefficient of the second layer, w [3] is the weight coefficient of the third layer, b [1] is the bias of the first layer, b [2] is the bias of the second layer, b [3] is the deviation of the third layer, f i (x) represents the activation function, and X is the input parameter of the neural network model.

[0142] The essence of the neural network model is to obtain better weights and biases through repeated fitting of training samples, and continuously reduce the deviation between the predicted value and the target value. The construction of the physical model constraint conditions is based on the improved objective function to constrain the optimization process of weights and biases to avoid the generation of abnormal values ​​that do not meet the conditions.

[0143] Combining formula (2) with the partial derivative of the well depth, we can get formula (13):

[0144]

[0145] Combining formula (12) and taking partial derivative of well depth, we can get formula (14):

[0146]

[0147] Therefore, by calculating the partial derivative of the well depth for the neural network model and the differential pressure stuck drill model, the connection between the physical model and the neural network weight and bias can be established, as shown in formula (15):

[0148]

[0149] In formula (15), A theory The partial derivative of the well depth is driven by the model with data representing stuck pipe;

[0150] ii) Based on the relationship between the weights and biases of the physical model and the data-driven model, a fused dual-driven model is obtained;

[0151] The neural network objective function under the constraints of the physical model can be established and called the knowledge constraint condition of the neural network. The fused dual-drive model is shown in formula (16):

[0152]

[0153] Among them, MSE is the mean square error, which is used to evaluate the deviation between the predicted value and the true value; Y pre is the predicted value of the data-driven model for stuck drilling, representing the pressure difference between the predicted bottom hole pressure and the formation pore pressure Pp pre , the pressure difference between bottom hole pressure and collapse pressure Pc pre , Predicted hook load value F pre , predict the value of drilling tool torque T pre And the predicted riser pressure value Ps pre , Y true is the true value, representing the pressure difference between the true bottom hole pressure and the formation pore pressure Pp true , the pressure difference between the real bottom hole pressure and the collapse pressure Pc pre , Actual hook load value F true , Actual drilling tool torque value T true and the actual standpipe pressure value Ps true Any one of the physical quantities; n is the number of training samples or test samples; A and A theory They represent the partial derivatives of the stuck drill data-driven model and the stuck drill mechanism model for well depth respectively; st is the abbreviation of subject to, which means "under certain conditions". The certain condition is the condition expressed by the expression after "{". "st{" specifically means that when solving, the value of the variable cannot deviate from the value. This condition constrains the value of so that it must be strictly equal to;

[0154] Formula (16) means that between A and A theoryUnder the constraint that the difference is 0, adjust the model parameter weight w and bias b to minimize the MSE (mean square error) between the predicted value and the true value.

[0155] Step S5: perform model incremental adjustment on the fused dual-drive model:

[0156] Establish an online learning framework. When a prediction deviation of the dual-drive model after the fusion of the physical model and the data model is detected, the dual-drive model is incrementally adjusted using new logging data, so that the dual-drive model can dynamically adapt to changes in data distribution.

[0157] First, the process of dividing the training data and the test data is extracted into a function so that the training set and the test set can be re-divided every time new data arrives.

[0158] Secondly, add a loop to simulate the process of online learning, and make incremental adjustments to the model each time new data is obtained. For example, in Python, use the model.fit method to continue training the model by setting the initial_epoch parameter to the index of the last epoch of the previous training plus 1. Then re-evaluate the performance of the model.

[0159] In order for incremental learning to work, it is necessary to keep the previously trained model and training history, so that before each acquisition of new data and incremental adjustment of the model, the current model and training history are saved. For example, in Python, the model.save method and np.save method are used to keep the previously trained model and training history respectively.

[0160] The training accuracy of the drill stuck prediction model based on the dual-drive fusion of physics and data and after incremental adjustment (i.e., the dual-drive fusion model obtained by incremental adjustment) established in the embodiment of the present invention is as follows: Figure 5 As shown, the accuracy has reached a high value around the 15th training, and stabilized around the 30th training. After that, the average accuracy remained unchanged at 98%, and the final model accuracy was 98.4%.

[0161] In order to facilitate the application of the dual-drive fusion model obtained by the subsequent incremental adjustment in the evaluation, in the embodiment of the present invention, after the incremental adjustment step, the following is further included:

[0162] i) The difference between the prediction results and the actual results of the incrementally adjusted stuck drill prediction model is shown through the confusion matrix:

[0163] The confusion matrix is ​​used to intuitively and effectively evaluate the model performance. Figure 6As shown, the upper left corner is the true positive example: it indicates the number of samples correctly predicted as positive examples by the model. That is, the true label is a positive example, and the model prediction result is also a positive example. The upper right corner is the false positive example: it indicates the number of samples incorrectly predicted as positive examples by the model. That is, the true label is a negative example, but the model prediction result is a positive example. The lower left corner is the false negative example: it indicates the number of samples incorrectly predicted as negative examples by the model. That is, the true label is a positive example, but the model prediction result is a negative example. The lower right corner is the true negative example: it indicates the number of samples correctly predicted as negative examples by the model. That is, the true label is a negative example, and the model prediction result is also a negative example. Among them, the positive example is a drill stuck (1); the negative example is a drill stuck not occurring (0). As shown Figure 6 As shown, the number of drill bits that did not get stuck (0) in the real data was 278, and the number of drill bits that got stuck (1) was 42; the number of drill bits that did not get stuck (0) predicted by the drill bit prediction model based on the dual-driven fusion of physics and data was 273, and the number of drill bits that got stuck (1) was 47.

[0164] ii) Visualizing the probability of occurrence of stuck pipe predicted by the incrementally adjusted stuck pipe prediction model, that is, visualizing the probability of occurrence of stuck pipe, such as Figure 7 The closer the value is to 1, the more likely the drill will get stuck.

[0165] In step S6, based on the dual-drive fusion model obtained after incremental adjustment and the new logging data, whether the drill is stuck is predicted, that is, the new logging data is used as sample data and input into the dual-drive fusion model obtained after fusion to predict whether the drill is stuck, thus completing the prediction process.

[0166] like Figure 8 As shown, a stuck pipe prediction system based on physical and data dual-driven fusion according to an embodiment of the present invention includes:

[0167] The preprocessing module 1 is used for preprocessing the historical logging data;

[0168] The physical model module 2 is used to establish a physical model of a stuck drill;

[0169] The data model module 3 is used to establish a data-driven model for stuck drill based on the preprocessed data;

[0170] The fusion module 4 is used to fuse the physical model of the stuck drill and the data-driven model;

[0171] The incremental module 5 is used to perform incremental adjustment on the fused dual-drive model;

[0172] The prediction module 6 is used to predict whether the drill is stuck based on the incrementally adjusted dual-drive fusion model and the new logging data.

[0173] like Fig. 9As shown, some embodiments of the present invention provide an electronic device, the electronic device 300, comprising: a processor 301, the processor 301 is coupled to a memory 302;

[0174] The memory 302 is used to store computer programs;

[0175] The processor 301 is configured to execute the computer program stored in the memory 302 so that the electronic device executes the method described in the above embodiment.

[0176] In certain embodiments of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a program or instruction. When the program or instruction is executed on a computer, the computer executes the method described in the above embodiments.

[0177] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an electronic device or a device.

[0178] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting stuck pipe based on the fusion of physical and data dual drivers, characterized in that: include: Preprocessing of historical logging data; Establish a physical model of drill sticking; Based on the preprocessed data, a data-driven model for stuck drill is established; Merge the physical model of stuck drill with the data-driven model; Perform incremental adjustments to the fused dual-drive model; Based on the incrementally adjusted dual-drive fusion model and new logging data, whether the drill is stuck is predicted.

2. The method for predicting stuck pipe based on physics and data dual-drive fusion according to claim 1, characterized in that: The preprocessing of historical logging data includes: Use SMOTE algorithm to process historical logging data; The grey correlation analysis method is used to screen the historical logging data obtained through one-time processing to obtain the preprocessed data.

3. The method for predicting stuck pipe based on physics and data dual-driven fusion according to claim 1, characterized in that: The data driven model for drill sticking is established based on the preprocessed data, including: Normalize the preprocessed data; Based on the data obtained through normalization, different algorithm models are used for drill sticking training; Based on the drill-stuck training results of different algorithm models, the final algorithm model is selected; Based on the final optimized algorithm model and historical data, a data-driven model for drill sticking is established.

4. The method for predicting stuck pipe based on physics and data dual-driven fusion according to claim 3 is characterized in that: The final algorithm model is selected based on the drill stuck training results of different algorithm models, including: Compare the drill-stuck training results of different algorithm models; The algorithm model with the best performance in the comparison results is taken as the final algorithm model.

5. A method for predicting stuck pipe based on physics and data dual-drive fusion according to any one of claims 3-4, characterized in that: The expression of the data-driven model of the stuck drill is as follows: Y pre =f k (w [k] ×f k-1 (w [k-1] ×…×f2(w [2] ×f1(w [1] ×X×b [1] )+b [2] )+…+b [k-1] )+b [k] ) Among them, Y pre is the predicted value of the data-driven model for stuck drill; w [k] represents the weight coefficient matrix of the kth layer, which means the connection weight between each neuron in the previous layer and each neuron in the current layer; b [k] is the deviation of the kth layer, f k (x) represents the activation function of the kth layer, and X is the input parameter of the neural network model.

6. The method for predicting stuck pipe based on physics and data dual-driven fusion according to claim 5, characterized in that: The fusion of the physical model of the stuck drill and the data-driven model includes: Based on the data-driven model, the partial derivative of the well depth is obtained, and the relationship between the physical model and the weights and biases of the data-driven model is established; Based on the relationship between the weights and biases of the physical model and the data-driven model, a fused dual-drive model is obtained.

7. The method for predicting stuck pipe based on physics and data dual-drive fusion according to claim 6, characterized in that: The expression of the fused dual-drive model is as follows: Among them, MSE is the mean square error, which is used to evaluate the deviation between the predicted value and the true value; Y pre is the predicted value of the data-driven model for stuck drill, Y true is the true value; n is the number of training samples or test samples; A and A theory They represent the partial derivatives of the stuck drill data-driven model and the stuck drill mechanism model on the well depth respectively; st is the abbreviation of subject to, which means "under certain conditions", and the certain condition is the condition expressed by the expression after "{"; The expression of the fused dual-drive model means that in A and A theory Under the constraint that the difference is 0, adjust the model parameter weight w and bias b to minimize the MSE value between the predicted value and the true value.

8. The method for predicting stuck pipe based on physics and data dual-drive fusion according to claim 7, characterized in that: After completing the incremental adjustment, it also includes: The confusion matrix shows the difference between the prediction results and the actual results of the incrementally adjusted stuck pipe prediction model; Visualize and output the probability of stuck pipe predicted by the incrementally adjusted stuck pipe prediction model.

9. A stuck drill prediction system based on physics and data dual-driven fusion, characterized in that: include: The preprocessing module is used to preprocess the historical logging data; The physical model module is used to establish the physical model of the stuck drill; The data model module is used to establish a data-driven model for stuck drill based on the preprocessed data; The fusion module is used to fuse the physical model of the stuck drill with the data-driven model; The incremental adjustment module is used to perform incremental adjustment on the fused dual-drive model; The prediction module is used to predict whether the drill is stuck based on the incrementally adjusted dual-drive fusion model and new logging data.

10. An electronic device, characterized in that: include: a processor coupled to the memory; The memory is used to store computer programs; The processor is configured to execute the computer program stored in the memory, so that the electronic device executes the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or an instruction, and when the program or the instruction is executed on a computer, the computer is caused to perform the method according to any one of claims 1 to 8.

Images