CN-121995741-A - Throttle valve control method, device, equipment and medium based on increment learning

CN121995741ACN 121995741 ACN121995741 ACN 121995741ACN-121995741-A

Abstract

The invention provides a throttle valve control method, device, equipment and medium based on incremental learning, which comprises the steps of collecting real-time parameters of a throttle valve during drilling, wherein the real-time parameters at least comprise outlet flow, temperature, density, throttle valve opening and actual measurement casing pressure, applying a pre-trained XGBoost model to the drilling parameters to obtain the predicted opening of the throttle valve, and updating a XGBoost model according to the real-time parameters when the performance evaluation reduction degree of the XGBoost model and/or the distribution distance of the real-time parameters of the throttle valve exceeds a preset threshold. The invention directly updates the model by using new data without head training, effectively improves the accuracy of valve opening prediction, enables the model to adapt to valve characteristics along with time, and realizes accurate prediction of the valve opening corresponding to the target casing pressure under different temperatures, flow rates and pressures.

Inventors

TANG GUI
He Xianzeng
ZHANG DONGMEI
HU BIWEN
Zuo xing
LI SAI
DENG HU
LIU QING
TANG MING
LI ZHAO

Assignees

中国石油天然气集团有限公司
中国石油集团川庆钻探工程有限公司

Dates

Publication Date: 20260508
Application Date: 20241104

Claims (10)

1. The throttle valve control method based on the increment learning is characterized by comprising the following steps of: collecting real-time parameters of the throttle valve during drilling, wherein the real-time parameters at least comprise outlet flow, temperature, density, throttle valve opening and actual measured casing pressure; Applying a pre-trained XGBoost model to the drilling parameters to obtain a predicted opening of the throttle valve, wherein the XGBoost model is updated according to the real-time parameters when the performance evaluation degradation of the XGBoost model or the distribution distance of the real-time parameters of the throttle valve exceeds a preset threshold.
2. The incremental learning based throttle control method of claim 1 wherein the pre-training comprises: Collecting valve opening degrees corresponding to different target casing pressures and one or more groups of parameter data of throttle valves corresponding to the valve opening degrees; Preprocessing the parameter data of the throttle valve, wherein the preprocessing comprises the steps of screening out the input data of the XGBoost model according to the influence of the parameter data of the throttle valve on the valve opening; Dividing the preprocessed parameter data into a training data set and a test data set, and training and evaluating the XGBoost model according to the training data set and the test data set.
3. The incremental learning based throttle control method of claim 2 wherein the preprocessing further comprises: Constructing a random forest model, and applying the parameter data of the throttle valve to the random forest model; calculating the overall importance scores of the features of the random forest model by using the random forest model, and sequencing; And confirming the input data of the XGBoost model according to the sorting.
4. The incremental learning based throttle control method of claim 3 wherein the calculating of the overall importance score comprises: Calculating the difference of average square errors of the father node and the child node generated by each division of the characteristics of the random forest model to obtain a characteristic contribution value; And carrying out accumulated average and normalization processing on the characteristic contribution values to obtain the overall importance score.
5. The incremental learning-based throttle control method according to claim 4, wherein the characteristic contribution value is calculated by: In the formula (1), y i is the actually measured throttle opening; The predicted opening degree of the throttle valve of the XGBoost model is obtained, n is the total number of samples, MSE parent is the average square error of a parent node, MSE leftchild and MSE rightchild are the average square error of left and right child nodes after splitting respectively, and pL and pR are the sample proportions of left and right subtrees respectively.
6. The incremental learning based throttle control method of claim 1 wherein the calculation of the degree of degradation in the performance evaluation of the XGBoost model is: In the formula (2), RMSE is the average square root error of the XGBoost model, and y i is the actually measured throttle opening; a predicted opening of a throttle valve for the XGBoost model; n is the total number of samples; the calculation formula of the distribution distance of the real-time parameters of the throttle valve is as follows: d (P, Q) =inf γ ∈ Γ(P,Q) E (x,y)～γ [ x-y ] (3) In equation (3), P and Q are probability distributions of old data and new data, respectively, inf is the maximum lower bound, E (x,y)～γ [/x-y ] is the expected value of the cost of movement (i.e., the distance from x to y) from P to Q under the joint distribution γ.
7. The incremental learning-based throttle control method according to claim 1, wherein when the XGBoost model is updated according to the real-time parameters, a calculation formula of parameter update of the XGBoost model is: In the formula (4), θ old and θ new are parameters of the XGBoost model before and after updating, respectively; alpha is the learning rate; is a gradient of the loss function with respect to the new data D new .
8. The throttle valve control device based on the increment learning is characterized by comprising a real-time parameter collection module and a throttle valve opening degree prediction module which are connected, wherein, The real-time parameter collection module is configured to collect real-time parameters of the throttle valve during drilling, wherein the real-time parameters at least comprise outlet flow, temperature, density, throttle valve opening and measured casing pressure; The opening prediction module of the throttle valve is configured to apply a pre-trained XGBoost model to the drilling parameters to obtain a predicted opening of the throttle valve, wherein the XGBoost model is updated according to the real-time parameters when a performance evaluation degradation degree of the XGBoost model or a distribution distance of the real-time parameters of the throttle valve exceeds a preset threshold.
9. A computer device, the computer device comprising: At least one processor, and A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions comprising instructions for performing the delta learning-based throttle control method as claimed in any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the incremental learning based throttle control method of any one of claims 1-7.

Description

Throttle valve control method, device, equipment and medium based on increment learning Technical Field The present invention relates to the field of petroleum and gas, and in particular, to a throttle control method based on incremental learning, a throttle control apparatus based on incremental learning, and an apparatus and a computer-readable storage medium implementing the throttle control method based on incremental learning. Background With the continuous progress of petroleum drilling technology, drilling is gradually increased for stratum with complex structures, and the stratum usually has a narrow safe drilling density window, so that the requirement on the accuracy of the drilling technology is higher. Pressure-controlled drilling technology is a key technology for solving the challenges because of the capability of finely controlling the bottom hole pressure, so that the probability of occurrence of accidents can be remarkably reduced by effectively improving the controllability of the drilling process. In pressure-controlled drilling systems, surface choke manifolds play a critical role as a core well control device, where the choke is the core means to control fluid flow and maintain the desired casing pressure, which directly affects the stability of the bottom hole pressure and the safety of the drilling operation. Currently, the throttle valve regulating technology mainly adopts an opening degree regulating method or a casing pressure regulating method, however, the conventional methods face various challenges. First, conventional PID control algorithms have difficulty in quickly and accurately adjusting the valve to the desired position, especially in situations where a quick response is required, such an insufficient adjustment speed may result in operational delays that affect the safety and efficiency of the overall drilling operation. And secondly, the opening decision of the valve depends on the experience judgment of field engineering personnel to a great extent, so that the risk of human errors is increased, and the regulation and control precision and repeatability are limited. Again, the valve characteristics may change over time and over continued use, such as wear, corrosion, or deposit accumulation, further increasing the difficulty of precise control using conventional methods. Therefore, it is important to develop an automatic control method to automatically adjust the valve setting according to the real-time data. Disclosure of Invention The present invention aims to address at least one of the above-mentioned deficiencies of the prior art. For example, it is an object of the present invention to provide a throttle valve control method based on incremental learning, which can automatically adjust the opening of a throttle valve while quickly achieving a desired casing pressure. In order to achieve the above object, according to one aspect of the present invention, a throttle valve control method based on incremental learning is provided. The throttle valve control method based on incremental learning comprises the following steps: And collecting real-time parameters of the throttle valve during drilling, wherein the real-time parameters at least comprise outlet flow, temperature, density, throttle valve opening and measured casing pressure. Applying a pre-trained XGBoost model to the drilling parameters to obtain a predicted opening of the throttle valve, wherein the XGBoost model is updated according to the real-time parameters when the performance evaluation degradation of the XGBoost model or the distribution distance of the real-time parameters of the throttle valve exceeds a preset threshold. In one exemplary embodiment of the present invention, the pre-training may include: And collecting valve opening corresponding to different target casing pressures and one or more groups of parameter data of the throttle valves corresponding to the valve opening. The parameter data of the throttle valve may be preprocessed, which may include screening the input data of the XGBoost model according to the influence of the parameter data of the throttle valve on the valve opening. The preprocessed parameter data may be separated into a training dataset and a test dataset from which the XGBoost model may be trained and evaluated. In one exemplary embodiment of the present invention, the throttle control method based on incremental learning, the preprocessing may further include: A random forest model is constructed to which the parameter data of the throttle valve may be applied. The random forest model may be utilized to calculate an overall importance score for features of the random forest model and may be ranked. Input data for the XGBoost model may be validated according to the ordering. In one exemplary embodiment of the present invention, the calculation of the overall importance score may include: Calculating the difference of average square errors of the father node and the c