Wellbore ECD intelligent regulation and control method and system based on machine learning and closed-loop control

Abstract

The invention relates to a shaft ECD intelligent regulation and control method and system based on machine learning and closed-loop control, and the method comprises the steps: obtaining original data of drilling parameters, and constructing an ECD prediction model based on the original data; acquiring real-time input data; inputting the real-time input data into the ECD prediction model for prediction to obtain an ECD prediction value; calculating according to the ECD predicted value and the target ECD value to obtain an ECD error value; and selecting a model intelligent regulation and control mode or a PID intelligent regulation and control mode according to the ECD error value to carry out ECD regulation and control. According to the method, the shaft ECD is predicted through the machine learning model, and the drilling parameters are adjusted in real time through the PID controller, so that the ECD is kept within the preset range. The double-layer control system combines the real-time feedback advantage of PID and the predictive capacity of machine learning, it is ensured that variables are accurately regulated and controlled in the drilling process, the ECD is maintained within the safety range, the risks such as stratum fracture and blowout are reduced, and the drilling efficiency and safety are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of petroleum drilling and production engineering, and more specifically, to a wellbore ECD intelligent control method and system based on machine learning and closed-loop control. Background Art

[0002] As oil and gas exploration extends into deepwater and deep formations, the narrow safe density window in high-temperature, high-pressure formations has become increasingly prominent, significantly increasing the difficulty of wellbore pressure control and the complexity of drilling operations. The narrow density window problem essentially concerns the safe operating range of the drilling fluid's safe density window (i.e., equivalent circulating density, ECD), which is determined by formation pore pressure, loss pressure, and fracture pressure. If the ECD exceeds the safe range during drilling, serious accidents such as lost circulation, kicks, and even blowouts can occur, leading to extended operation cycles and surging costs.

[0003] Currently, ECD prediction relies primarily on downhole measurement tools or mathematical models, but both have significant limitations. For one thing, extensive drilling data indicates that the operating limit of high-temperature measurement-while-drilling tools is 175°C. When temperatures exceed this limit, downhole measurement instruments and screw drill bits may not function properly or even be damaged, leading to increased drilling time and construction costs. Furthermore, the use of mathematical models to calculate equivalent circulating density (ECD) involves numerous assumptions and limitations regarding downhole pressure, temperature, and mud type. Engineers must select or construct appropriate models based on the characteristics of individual wellbores and the physical properties of the formation. Consequently, the calculated results are significantly influenced by empirical factors and have limited accuracy, which in turn impacts safe and efficient drilling operations. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a wellbore ECD intelligent control method and system based on machine learning and closed-loop control to address the problems existing in the prior art.

[0005] The technical solution adopted by the present invention to solve the technical problem is to construct a wellbore ECD intelligent control method based on machine learning and closed-loop control, including the following steps:

[0006] Acquiring raw data of drilling parameters and building an ECD prediction model based on the raw data;

[0007] Get real-time input data;

[0008] Inputting the real-time input data into the ECD prediction model for prediction to obtain an ECD prediction value;

[0009] Calculate the ECD error value based on the ECD prediction value and the target ECD value;

[0010] According to the ECD error value, the model intelligent control mode or the PID intelligent control mode is selected to perform ECD control.

[0011] In the wellbore ECD intelligent control method based on machine learning and closed-loop control described in the present invention, the steps of obtaining raw data of drilling parameters and constructing an ECD prediction model based on the raw data include:

[0012] Acquiring raw data of the drilling parameters;

[0013] Analyzing and processing the raw data to obtain key influencing parameters;

[0014] A model is constructed based on the data of the key influencing parameters to obtain the ECD prediction model.

[0015] In the wellbore ECD intelligent control method based on machine learning and closed-loop control described in the present invention, the analysis and processing of the raw data to obtain key influencing parameters includes:

[0016] Performing data cleaning on the raw data to obtain preprocessed data;

[0017] performing standardization processing on the preprocessed data to obtain standardized data;

[0018] Correlation analysis is performed on the standardized data to obtain the key influencing parameters.

[0019] In the wellbore ECD intelligent control method based on machine learning and closed-loop control described in the present invention, performing correlation analysis on the standardized data to obtain the key influencing parameters includes:

[0020] Analyzing the correlation between any two drilling parameters by using a mutual information method to obtain a mutual information value between any two drilling parameters;

[0021] Parameter screening is performed according to the mutual information value between any two drilling parameters to obtain the key influencing parameters.

[0022] In the wellbore ECD intelligent control method based on machine learning and closed-loop control of the present invention, the model construction based on the data of the key influencing parameters to obtain the ECD prediction model includes:

[0023] Dividing the data of the key influencing parameters into a training set, a validation set, and a test set;

[0024] Build an initial prediction model based on machine learning methods;

[0025] Training the initial prediction model using the training set to obtain a training model;

[0026] Using the validation set to validate the training model to obtain a validated prediction model;

[0027] The verified prediction model is tested and evaluated using the test set to obtain the ECD prediction model.

[0028] In the wellbore ECD intelligent control method based on machine learning and closed-loop control of the present invention, the training model is verified using the verification set to obtain a verified prediction model, which includes:

[0029] Validating the training model using the validation set after each training cycle;

[0030] Record the loss value of the training set and the loss value of the validation set, draw a loss function curve graph, and save the verified prediction model.

[0031] In the wellbore ECD intelligent control method based on machine learning and closed-loop control of the present invention, the calculation based on the ECD predicted value and the target ECD value to obtain the ECD error value includes:

[0032] The target ECD value is subtracted from the ECD predicted value to obtain the ECD error value.

[0033] In the wellbore ECD intelligent control method based on machine learning and closed-loop control described in the present invention, the selection of a model intelligent control method or a PID intelligent control method for ECD control according to the ECD error value includes:

[0034] comparing the ECD error value with an error threshold;

[0035] If the ECD error value is greater than the error threshold, the model intelligent control mode is selected to perform ECD control;

[0036] If the ECD error value is less than the error threshold, the PID intelligent control mode is selected to perform ECD control.

[0037] The present invention also provides a wellbore ECD intelligent control system based on machine learning and closed-loop control, comprising:

[0038] A model building unit, configured to obtain raw data of drilling parameters and build an ECD prediction model based on the raw data;

[0039] A data acquisition unit, used for acquiring real-time input data;

[0040] An ECD prediction model is used to make predictions based on the real-time input data to obtain an ECD prediction value;

[0041] an error calculation unit, configured to calculate an ECD error value based on the ECD prediction value and the target ECD value;

[0042] The control mode selection and determination unit is used to select a model intelligent control mode or a PID intelligent control mode to perform ECD control according to the ECD error value.

[0043] The wellbore ECD intelligent control system based on machine learning and closed-loop control of the present invention also includes:

[0044] A PID controller is used to determine the target ECD value and perform ECD control according to the PID intelligent control method.

[0045] The implementation of the wellbore ECD intelligent control method and system based on machine learning and closed-loop control of the present invention has the following beneficial effects: including: obtaining the original data of drilling parameters and building an ECD prediction model based on the original data; obtaining real-time input data; inputting the real-time input data into the ECD prediction model for prediction to obtain an ECD prediction value; calculating according to the ECD prediction value and the target ECD value to obtain an ECD error value; selecting a model intelligent control method or a PID intelligent control method for ECD control according to the ECD error value. The present invention predicts the wellbore ECD through a machine learning model, and then uses a PID controller to adjust the drilling parameters in real time to keep the ECD within a preset range. The two-layer control system combines the real-time feedback advantages of PID and the predictive ability of machine learning to ensure precise control of variables during the drilling process, maintain ECD within a safe range, reduce the risks of formation rupture, blowout, etc., and improve drilling efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0047] Figure 1 This is a flow chart of an embodiment of a wellbore ECD intelligent control method based on machine learning and closed-loop control provided by the present invention;

[0048] Figure 2 It is a schematic diagram of the LSTM model provided by the present invention;

[0049] Figure 3 This is a principle block diagram of the wellbore ECD intelligent control system based on machine learning and closed-loop control of the present invention;

[0050] Figure 4It is a schematic diagram of the automatic control of the "PID+ΔECD" double-layer collaborative feedback provided by the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] In response to the problems existing in the existing wellbore ECD control, the present invention provides an intelligent wellbore ECD control method based on machine learning and closed-loop control, which realizes precise pressure management during the drilling process through the coupling of data-driven prediction and dynamic compensation control mechanisms. By constructing a machine learning model based on drilling data, the limitations of the downhole high-temperature environment on the measuring tools are broken through, and high-precision prediction of wellbore ECD is achieved; the PID control algorithm and the prediction deviation compensation mechanism are further integrated to establish a two-layer collaborative control system with error self-correction capabilities, which controls the dynamic fluctuation range of ECD within the allowable error range. Accurately predicting ECD not only helps to improve the accuracy and efficiency of oil and gas exploration and development, but also provides a means to reasonably control the wellbore pressure during drilling, ensure construction safety, and maximize reservoir protection.

[0053] refer to Figure 1 In a preferred embodiment, the wellbore ECD intelligent control method based on machine learning and closed-loop control includes the following steps:

[0054] Step S100: obtaining raw data of drilling parameters and constructing an ECD prediction model based on the raw data.

[0055] Optionally, in some embodiments, obtaining the raw data of drilling parameters and constructing an ECD prediction model based on the raw data includes: obtaining the raw data of drilling parameters; analyzing and processing the raw data to obtain key influencing parameters; constructing a model based on the data of key influencing parameters to obtain an ECD prediction model. The drilling parameters include surface parameters and downhole parameters. Surface parameters include, but are not limited to, pump speed, displacement, riser pressure, hook load, mud density, plastic viscosity, inlet temperature, outlet temperature, etc. Downhole parameters include: drilling speed, drill bit speed, drilling pressure, torque, annular pressure, cuttings concentration, well depth, wellbore diameter, formation pressure, temperature, etc. The raw data of drilling parameters can be obtained in an existing conventional manner, which is not specifically limited by the present invention.

[0056] Optionally, in some embodiments, analyzing and processing the raw data to obtain key influencing parameters includes: performing data cleaning on the raw data to obtain preprocessed data; performing standardization on the preprocessed data to obtain standardized data; and performing correlation analysis on the standardized data to obtain key influencing parameters. Performing correlation analysis on the standardized data to obtain key influencing parameters includes: analyzing the correlation between any two drilling parameters using a mutual information method to obtain a mutual information value between the any two drilling parameters; and screening parameters based on the mutual information value between the any two drilling parameters to obtain key influencing parameters.

[0057] Specifically, the data of surface parameters and downhole parameters are first extracted, and then the median filling method is used to process the missing values ​​and outliers in the data of surface parameters and downhole parameters to obtain preprocessed data, and then the preprocessed data is normalized to obtain standardized data. By standardizing or normalizing the data, features of different scales can be within the same range, preventing certain features from having an unbalanced impact on the model. Standardizing a data set means moving the distribution of each variable to a unit scale, that is, the mean is 0 and the standard deviation is 1. This is a necessary step in the PCA (Principal Component Analysis) algorithm. The values ​​of the input parameters are standardized using the following formula:

[0058]

[0059] Where: Y is the normalized input parameter, X is the input parameter to be normalized, and σ is the standard deviation of the input parameter.

[0060] Then, based on the drilling logs and the analysis of factors affecting drilling parameters, variables with high correlation with the target variable (i.e., wellbore equivalent circulating density (ECD)) were selected as input feature parameters of the prediction model, i.e., key influencing parameters. Specifically, the correlation between variables was analyzed using the mutual information method. Mutual information is a measure of the interdependence between two random variables that can reflect both linear and nonlinear relationships between the variables. Suppose there are two random variables X and Y. Their mutual information I(X,Y) is defined as:

[0061]

[0062] Where: p(x,y) is the joint probability distribution of X and Y, p(x) and p(y) are the marginal probability distributions of X and Y respectively. To facilitate calculation and understanding, the form of mutual information joint entropy is converted to the entropy subtraction form:

[0063] MI(X;Y)=H(X)+H(Y)-H(X,Y)(1.3);

[0064] Where MI(X;Y) is the subtraction representation of the mutual information I(X;Y), H(X,Y) is the joint entropy of X and Y, and H(X) and H(Y) are the unconditional entropies of X and Y, respectively.

[0065] Normalize the mutual information values ​​between variables to facilitate the evaluation of the correlation between variables. The normalization can refer to the following formula:

[0066]

[0067] Where: NMI(X;Y) is the normalized mutual information value.

[0068] After calculating the normalized mutual information value between any two variables using formula (1.4), the correlation between the two can be determined based on the mutual information value, thereby screening out key influencing parameters. Specifically, when the mutual information value is between [0.5, 0.8], the two parameters are considered to be moderately correlated, and when the mutual information value is greater than 0.8, it indicates that there is a strong correlation between the two parameters. That is, when the mutual information value between two parameters is between [0.5, 0.8] or above 0.8, they can be determined to be key influencing parameters. Among them, the variables referred to in the present invention are drilling parameters.

[0069] Mutual information analysis was used to quantify the linear relationship between features and the target variable. Thirteen key influencing parameters with high correlations with the target variable were selected from surface and downhole parameters. These 13 key influencing parameters include outlet temperature, inlet temperature, drilling rate, rotational speed, torque, weight on bit, pump rate, standpipe pressure, displacement, hook load, cuttings concentration, mud density, and plastic viscosity. These key influencing parameters encompass key operational parameters during drilling and have direct or indirect impacts on the measurement of wellbore equivalent circulating density (ECD). Temperature, density, and viscosity affect the physical properties of the mud, thereby altering annular pressure. Mechanical parameters such as drilling rate, rotational speed, torque, and weight on bit influence the efficiency of cuttings generation and removal, directly impacting annular pressure. Fluid parameters such as pump rate, displacement, and standpipe pressure determine the efficiency and pressure balance of the mud circulation system. Hook load is related to the operational stability and load capacity of the drilling equipment. These key influencing parameters collectively influence the fluid dynamics and pressure distribution within the wellbore, thus affecting the accurate measurement of ECD.

[0070] Furthermore, in an embodiment of the present invention, after obtaining the key influencing parameters, a wavelet filtering algorithm can be used to perform noise processing on the data of the key influencing parameters screened out in the correlation analysis. Wavelet filtering denoising is a method of removing signal noise through wavelet transform, which combines the localization characteristics of Fourier transform and can perform multi-scale analysis of signals in the time domain and frequency domain. This makes wavelet transform more effective than traditional methods in denoising, especially when processing unstable or highly variable signals. The wavelet coefficients can be expressed as follows:

[0071]

[0072] Where: W(a,b) is the wavelet coefficient, x(t) is the original signal, α(t) is the wavelet basis coefficient, a is the scale coefficient, and b is the translation coefficient.

[0073] Optionally, in some embodiments, a model is constructed based on the data of key influencing parameters to obtain an ECD prediction model, including: dividing the data of key influencing parameters to obtain a training set, a validation set, and a test set; constructing an initial prediction model based on a machine learning method; training the initial prediction model with the training set to obtain a training model; validating the training model with the validation set to obtain a verified prediction model; testing and evaluating the verified prediction model with the test set to obtain an ECD prediction model. Verifying the training model with the validation set to obtain the verified prediction model includes: validating the training model with the validation set after each training cycle; recording the loss value of the training set and the loss value of the validation set, drawing a loss function curve, and saving the verified prediction model.

[0074] Specifically, after completing the denoising process of the key influencing parameters, the data is divided to obtain a training set, a validation set, and a test set. The split ratio of the training set, validation set, and test set is 70:15:15. The training set is used to create training data for the network, and the desired goals are used to change the bias and weights in the network to achieve active optimal performance. The validation set is used to prevent the model from overfitting and ensure the generalization of new data. The test set is used to test the model on new data. After completing the data division, the data of the training set, validation set, and test set are standardized respectively. In some embodiments, the Z-score normalization method can be used to perform a linear transformation on the original data to eliminate the influence of different magnitudes between the various features. The formula for the normalization process is as follows:

[0075]

[0076] Where: z is the standardized value, x is the original data, μ is the mean of the original data, σ is the standard deviation of the original data. After standardization, the mean of the data is converted to 0 and the standard deviation is converted to 1.

[0077] Optionally, in the embodiment of the present invention, the LSTM model can be selected as the ECD prediction model. The principle of the LSTM model is as follows Figure 2 shown.

[0078] The tanh function is used as the activation function for the neural network. The output range of the tanh function is [-1, 1]. When passing through the activation function, the mean of the data is closer to zero, which helps speed up the training of the neural network and can capture complex patterns and relationships in the data. To optimize the weights of the neural network, the Adam optimizer with an adaptive learning rate is used. This optimizer trains the neural network by minimizing the loss function (MSE, Mean Squared Error). In each iteration, the Adam optimizer uses one sample or a small batch of samples to update the model parameters. The model training process selects the training function with the smallest MSE to ensure the best model performance.

[0079]

[0080] yi is the actual value, is the predicted value and N is the number of samples.

[0081] After each training cycle, the training model is evaluated using the validation set. The loss values ​​of the training set and validation set are recorded and the loss function curve is plotted. The best performing model (i.e., the verified prediction model) is saved. The verified prediction model is then tested and evaluated using the test set to obtain the ECD prediction model. Specifically, the performance of the model is evaluated using the test set on the best performing model. The mean absolute error (MAE) and the coefficient of determination (R) are used. 2 , mean absolute percentage error (MAPE) and root mean square error (RMSE) are used to evaluate the model performance. MAE is the average of the absolute errors between the predicted value and the actual value. The smaller the MAE value, the higher the prediction accuracy of the model. 2 The value of is usually between 0 and 1. The closer to 1, the better the model fit. MAPE is the average of the absolute percentage differences between the predicted value and the actual value, and the error is expressed in percentage form. RMSE is the square root of the average of the squares of the differences between the predicted value and the actual value. The smaller the RMSE value, the smaller the prediction error of the model and the better the prediction performance. 2 The highest training function trains the model.

[0082] The formula is as follows:

[0083]

[0084] Among them, y i is the actual value, is the predicted value, N is the number of samples, is the average of the actual values.

[0085] Step S200: Acquire real-time input data.

[0086] Specifically, the real-time input data is real-time data of key influencing parameters.

[0087] Step S300: inputting real-time input data into the ECD prediction model for prediction to obtain an ECD prediction value.

[0088] After constructing the ECD prediction model, real-time input data is fed into the model to generate a real-time ECD, or predicted ECD value. Specifically, real-time data on key influencing parameters such as drilling fluid density, well depth, drilling rate, drill bit speed, annular pressure, wellbore diameter, formation pressure, and temperature are fed into the ECD prediction model to predict the current wellbore ECD.

[0089] Step S400: performing calculation based on the ECD prediction value and the target ECD value to obtain an ECD error value.

[0090] Optionally, in an embodiment of the present invention, the target ECD value may be set according to formation pressure and wellbore stability requirements, so that the set target ECD value is used as a reference value of the PID controller.

[0091] In some embodiments, calculating the ECD error value based on the predicted ECD value and the target ECD value includes: subtracting the target ECD value from the predicted ECD value to obtain the ECD error value. The error calculation formula is as follows:

[0092] e(t)=r(t)-y(t)(1.12);

[0093] Where: r(t) is the target ECD, g / cm 3 ; y(t) is the actual ECD, g / cm 3 .

[0094] Step S500: selecting a model intelligent control mode or a PID intelligent control mode to perform ECD control according to the ECD error value.

[0095] Optionally, in an embodiment of the present invention, selecting a model intelligent control method or a PID intelligent control method for ECD control according to the ECD error value includes: comparing the ECD error value with an error threshold; if the ECD error value is greater than the error threshold, selecting the model intelligent control method for ECD control; if the ECD error value is less than the error threshold, selecting the PID intelligent control method for ECD control.

[0096] refer to Figure 3The present invention also provides a wellbore ECD intelligent control system based on machine learning and closed-loop control.

[0097] Specifically, such as Figure 3 As shown, the wellbore ECD intelligent control system based on machine learning and closed-loop control includes:

[0098] The model building unit is used to obtain the original data of drilling parameters and build an ECD prediction model based on the original data.

[0099] The data acquisition unit is used to acquire real-time input data.

[0100] The ECD prediction model is used to make predictions based on real-time input data to obtain ECD prediction values.

[0101] The error calculation unit is used to calculate the ECD error value based on the ECD prediction value and the target ECD value.

[0102] The control mode selection and determination unit is used to select the model intelligent control mode or the PID intelligent control mode for ECD control according to the ECD error value.

[0103] Furthermore, the wellbore ECD intelligent control system based on machine learning and closed-loop control also includes: a PID controller, which is used to determine the target ECD value and perform ECD control according to the PID intelligent control method.

[0104] The PID (Proportional-Integral-Derivative) controller is a commonly used feedback control system used to control variables in complex systems to approach their target values. A PID controller calculates the error between the current and target values ​​in real time and adjusts the system input based on this error, gradually bringing the output to the desired value. In drilling operations, a PID controller can be used to adjust drilling parameters in real time based on the error between the predicted wellbore ECD and the target wellbore ECD to maintain wellbore stability and drilling safety.

[0105] PID controller calculation: The PID controller consists of three main components: proportional control (P), integral control (I), and differential control (D). The formula for the PID controller is:

[0106]

[0107] Where e(t) is the current error, u(t) is the controller output, and K p , K i , K d are the proportional, integral and differential coefficients respectively.

[0108] From formula (1.12), we can see that due to the existence of the integral term, the PID control algorithm will produce the accumulation of control error, resulting in an increase in error. Therefore, the incremental PID control algorithm is used to eliminate the integral term error:

[0109] Δu(t)=K p e(t)+K i e(t)+K d [Δe(t)-Δe(t-1)](1.14);

[0110] Where: △e(t) is the error change value at the current moment, △e(t-1) is the error change value at the previous moment, and △u(t) is the control amount increment.

[0111] The outputs of the proportional, integral, and differential components are summed to produce the total output of the PID controller. Based on this total output, drilling parameters are adjusted in real time to bring the real-time wellbore ECD closer to the target wellbore ECD. The adjusted wellbore ECD is monitored in real time, and the error is recalculated. Based on the newly calculated error, the PID controller further adjusts drilling parameters, forming a closed-loop control loop to continuously optimize the wellbore ECD.

[0112] like Figure 4 As shown in the figure, in this embodiment of the present invention, the ECD prediction model (machine learning model) serves as the upper-level controller, responsible for predicting the wellbore ECD and providing the predicted ECD value to the lower-level control. The PID controller, serving as the lower-level controller, adjusts the drilling operating parameters (i.e., the aforementioned key influencing parameters) to achieve the target ECD value based on the wellbore ECD prediction value provided by the upper-level controller and the real-time error (i.e., the calculated ECD error value). The upper-level and lower-level controllers work together to continuously adjust the drilling operating parameters through real-time feedback to achieve precise control of the wellbore ECD.

[0113] Specifically, first, the target ECD value is determined based on formation pressure, wellbore stability and other relevant factors. Then, the difference between the ECD prediction value output by the ECD prediction model and the target ECD value is used to obtain ΔECD (ECD error value), and ΔECD is compared with the error threshold. If ΔECD exceeds the error threshold, the PID control is skipped and the ECD prediction model is directly called to predict the required adjustment amount of the drilling operation parameters to quickly reduce the error. According to the predicted adjustment amount, the control parameters (that is, the aforementioned input characteristic parameters) are adjusted in real time to make the actual wellbore ECD quickly approach the target wellbore ECD. When the error is reduced to within the error threshold, the system automatically switches to PID control mode. At this time, the PID controller fine-tunes the drilling operation parameters based on the real-time error to ensure that the actual wellbore ECD remains stable within the target range.

[0114] In addition, the system also adjusts PID parameters (such as gain, integral, and differential coefficients) in real time based on historical control effects and changes in the operating environment to improve control accuracy and response speed. Downhole sensor data is monitored in real time. If the actual ECD deviates from the predicted value, timely feedback is provided to the machine learning model for online updating or correction of model parameters. During long-term operation, the parameters of the PID controller are adaptively adjusted according to the actual drilling situation to adapt to different downhole conditions. This two-layer control system is deployed in actual drilling operations to control the wellbore ECD in real time. Based on field feedback data, the parameters of the machine learning model and PID controller are continuously optimized to improve the stability and accuracy of the system. Through two-layer collaborative feedback, both the speed of wellbore ECD adjustment and the accuracy and stability of wellbore ECD prediction are guaranteed.

[0115] The present invention uses surface parameters and downhole parameters as input features, adopts a machine learning algorithm to train a prediction model, and establishes a nonlinear mapping relationship between surface parameters and wellbore ECD. During the drilling process, the model outputs the predicted wellbore ECD value in real time. The PID controller dynamically adjusts the pump speed and throttle valve opening by monitoring parameters such as wellhead pressure in real time to ensure that the wellbore ECD is stable within the target range. Based on the deviation (ΔECD) between the ECD value predicted by the machine learning model and the target ECD value, the control parameters are adjusted in advance. When the predicted ECD value deviates from the target range, the system automatically calculates ΔECD and dynamically adjusts the drilling parameters through the PID controller to ensure that the wellbore ECD is always within the safety window. The innovative fusion of data-driven prediction (machine learning) and traditional feedback control (PID) forms a "prediction-compensation-control" closed-loop system, without relying on downhole measurement tools, breaking through the technical limitations of ultra-high temperature environments. Combining the forward-looking prediction of machine learning with the real-time feedback of PID control, adaptive and precise control of wellbore pressure is achieved.

[0116] This method uses a machine learning model to predict the wellbore ECD, then uses a PID controller to adjust drilling parameters in real time to maintain the ECD within a preset range. This two-tier control system combines the real-time feedback advantages of PID with the predictive power of machine learning to ensure precise control of variables during the drilling process, maintaining the ECD within a safe range, reducing risks such as formation fractures and blowouts, and improving drilling efficiency and safety. Furthermore, based on the ECD values ​​predicted by machine learning, combined with reliability theory and pressure balance theory, early warnings can be issued for downhole engineering risks such as lost circulation, well kicks, and wellbore collapse.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0118] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0119] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0120] The above embodiments are intended only to illustrate the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the present invention and implement it accordingly. They are not intended to limit the scope of protection of the present invention. All equivalent variations and modifications within the scope of the claims of the present invention are intended to be covered by the claims of the present invention.

Claims

1. A wellbore ECD intelligent control method based on machine learning and closed-loop control, characterized in that: The following steps are involved: Acquiring raw data of drilling parameters and building an ECD prediction model based on the raw data; Get real-time input data; Inputting the real-time input data into the ECD prediction model for prediction to obtain an ECD prediction value; Calculate the ECD error value based on the ECD prediction value and the target ECD value; According to the ECD error value, the model intelligent control mode or the PID intelligent control mode is selected to perform ECD control.

2. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 1, characterized in that: The obtaining of raw data of drilling parameters and constructing an ECD prediction model based on the raw data includes: Acquiring raw data of the drilling parameters; Analyzing and processing the raw data to obtain key influencing parameters; A model is constructed based on the data of the key influencing parameters to obtain the ECD prediction model.

3. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 2, characterized in that: The analyzing and processing of the raw data to obtain key influencing parameters includes: Performing data cleaning on the raw data to obtain preprocessed data; performing standardization processing on the preprocessed data to obtain standardized data; Correlation analysis is performed on the standardized data to obtain the key influencing parameters.

4. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 3, characterized in that: The performing correlation analysis on the standardized data to obtain the key influencing parameters includes: Analyzing the correlation between any two drilling parameters by using a mutual information method to obtain a mutual information value between any two drilling parameters; Parameter screening is performed according to the mutual information value between any two drilling parameters to obtain the key influencing parameters.

5. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 2, characterized in that: The model is constructed based on the data of the key influencing parameters to obtain the ECD prediction model, which includes: Dividing the data of the key influencing parameters into a training set, a validation set, and a test set; Build an initial prediction model based on machine learning methods; Training the initial prediction model using the training set to obtain a training model; Using the validation set to validate the training model to obtain a validated prediction model; The verified prediction model is tested and evaluated using the test set to obtain the ECD prediction model.

6. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 5, characterized in that: The validation set is used to validate the training model to obtain a validated prediction model, which includes: Validating the training model using the validation set after each training cycle; Record the loss value of the training set and the loss value of the validation set, draw a loss function curve graph, and save the verified prediction model.

7. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 1, characterized in that: The calculating according to the ECD predicted value and the target ECD value to obtain the ECD error value includes: The target ECD value is subtracted from the ECD predicted value to obtain the ECD error value.

8. The method for intelligent control of wellbore ECD based on machine learning and closed-loop control according to claim 1, characterized in that: The selecting of a model intelligent control mode or a PID intelligent control mode for ECD control according to the ECD error value includes: comparing the ECD error value with an error threshold; If the ECD error value is greater than the error threshold, the model intelligent control mode is selected to perform ECD control; If the ECD error value is less than the error threshold, the PID intelligent control mode is selected to perform ECD control.

9. A wellbore ECD intelligent control system based on machine learning and closed-loop control, characterized in that: include: A model building unit, configured to obtain raw data of drilling parameters and build an ECD prediction model based on the raw data; A data acquisition unit, used for acquiring real-time input data; An ECD prediction model is used to make predictions based on the real-time input data to obtain an ECD prediction value; an error calculation unit, configured to calculate an ECD error value based on the ECD prediction value and the target ECD value; The control mode selection and determination unit is used to select a model intelligent control mode or a PID intelligent control mode to perform ECD control according to the ECD error value.

10. The wellbore ECD intelligent control system based on machine learning and closed-loop control according to claim 9, characterized in that: Also includes: A PID controller is used to determine the target ECD value and perform ECD control according to the PID intelligent control method.

Images