CN-121976794-A - Deep water drilling gas invasion monitoring method

CN121976794ACN 121976794 ACN121976794 ACN 121976794ACN-121976794-A

Abstract

The invention belongs to the technical field of deepwater oil-gas field drilling gas invasion monitoring, and particularly relates to a deepwater drilling gas invasion monitoring method. The monitoring method extracts nonlinear dynamic characteristics of ultrasonic signals through multi-scale fuzzy divergence entropy, and obtains a gas content value based on inversion of a support vector machine model trained by samples, so that quantitative monitoring and early warning of small changes of the gas content are realized. A deep water drilling gas invasion monitoring method comprises the following steps of collecting ultrasonic echo signals of gas-liquid two-phase flow, extracting multi-scale fuzzy divergence entropy values in the ultrasonic echo signals, and inputting feature vectors in the multi-scale fuzzy divergence entropy values into a trained support vector machine classification model to obtain a gas content recognition result.

Inventors

LI WEI
LI XIAO
ZHANG SHUHANG
ZHANG ZIHENG
LIU JIANLEI
Zhou Langfei
LIU SHENGBO
YIN XIAOKANG
YUAN XINAN

Assignees

中国石油大学(华东)

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (6)

1. The deepwater drilling gas invasion monitoring method is characterized by comprising the following steps of: s1, collecting ultrasonic echo signals of gas-liquid two-phase flow; s2, extracting a multi-scale fuzzy divergence entropy value in an ultrasonic echo signal; S3, inputting the feature vectors in the multi-scale fuzzy divergence entropy values into a trained support vector machine classification model to obtain a gas content recognition result; the step S3 specifically comprises the following steps: s31, constructing a training subset and a testing subset required by a unified recognition model; Step S32, carrying out MFDE operation on the training subset and the testing subset to respectively obtain a feature vector matrix of the training subset and a feature vector matrix of the testing subset; Step S33, splicing feature vector matrixes of the training subsets according to columns to obtain a full-working-condition training set, endowing category labels to feature vectors of all the training subsets in the full-working-condition training set, and merging to obtain training sample label vectors; The feature vector matrix of the test subset is spliced according to columns to obtain a full-working-condition test set, category labels are given to the feature vectors of all the test subset in the full-working-condition test set, and the feature vectors are combined to obtain test sample label vectors; step S34, a full-working-condition test set and a training sample label vector are used as inputs, and a MFDE-SVM algorithm with gas intrusion feature recognition capability is obtained through training; And comparing the predictive label vector with the test sample label vector to construct an confusion matrix, wherein the confusion matrix is used for evaluating the recognition performance of the trained support vector machine classification model on the deep water drilling gas content under different working conditions.
2. The deep water drilling gas invasion monitoring method according to claim 1, wherein the process of extracting the multiscale ambiguity entropy value in the ultrasonic echo signal in the step S2 specifically comprises the following steps: s21, performing multi-scale segmentation on the acquired ultrasonic echo signals of the gas-liquid two-phase flow, and converting the original time sequence of the ultrasonic echo signals into A plurality of multi-scale time sequences; Wherein, the Is the number of scale factors; the original time sequence of the ultrasonic echo signals satisfies: After multi-scale segmentation processing, the multi-scale time sequence of the ultrasonic echo signals meets the following conditions: (1) Wherein, the method comprises the steps of, ; For the data length after coarse graining, the following is satisfied: ; representing the scale factor currently being calculated, satisfies: ; step S22, on-scale Then, carrying out phase space reconstruction on a multi-scale time sequence matrix formed by the multi-scale time sequence of the ultrasonic echo signals to obtain a high-order matrix ; Wherein, multiscale time series matrix satisfies: Higher order matrix The method comprises the following steps: (2); Step S23, calculating a high-order matrix Cosine similarity between adjacent trajectories; Step S24, dividing the value domain interval and calculating to obtain a membership matrix ; Wherein the value range [ -1,1] is divided into A kind of electronic device Each interval; And, membership matrix The method comprises the following steps: (6); Wherein, the For cosine membership For the interval Is a fuzzy membership degree of (1), satisfying: (5); , ; the number of the divided intervals; for embedding dimensions, a dynamic structure for capturing signals; For similar margin, sensitivity for controlling entropy; Step S25, calculating to obtain membership probability matrix And obtaining a state probability matrix by calculating the state probability of each interval ; Wherein the membership probability matrix The method comprises the following steps: (7) Probability of From membership degree By the formula Obtaining after conversion; probability matrix of membership Summing each column of the two groups to obtain the number of cosine similarity falling in each interval ; State probability matrix The method comprises the following steps: wherein the state probability of each section The method comprises the following steps: ; step S26, based on membership probability matrix Constructing to obtain the embedded dimension m and the scale factor m Membership function order of The similar tolerance is The number of symbols is MFDE, and calculating to obtain a multi-scale fuzzy divergence entropy value in the ultrasonic echo signal; Wherein, MFDE algorithm satisfies: (8)。
3. the deep water drilling gas invasion monitoring method according to claim 2, wherein the step S23 calculates a higher-order matrix The cosine similarity between adjacent trajectories is specifically described as: For high order matrix Simplified representation is made; wherein the high order matrix Is a simplified representation of (1), satisfying: (3); Wherein, the As a result of the initial trajectory vector, Is in combination with Adjacent track vectors, and so on; Computing a higher order matrix Cosine similarity between adjacent tracks in the array to obtain cosine similarity matrix ; Wherein, the (4)。
4. The method of claim 1, wherein the process of constructing the training subset and the testing subset required for the unified recognition model in step S31 is specifically described as follows: Will be under a certain current working condition The R signal samples obtained below Is marked as Wherein, the method comprises the steps of, Number the samples and assign all The assembled set is denoted as ; Wherein, the Representing the current working condition Flow rate of liquid phase Characterized by, Representing the current working condition Lower air content Features; The method meets the following conditions: Collection of The method comprises the following steps: e is the working condition number; for a plurality of groups of current working conditions in a ratio of 6:4 The set obtained below In (a) and (b) Dividing to obtain training subsets And test subset ; Wherein the training subset The method comprises the following steps: Test subset The method comprises the following steps: 。
5. The method for monitoring deep water drilling gas invasion according to claim 1, wherein the process of performing MFDE operations on the training subset and the test subset in the step S32 is specifically described as the calculated training subset Is a feature vector matrix of (a) The method comprises the following steps: ; Calculated test subset Is a feature vector matrix of (a) The method comprises the following steps: 。
6. The method of claim 1, wherein the step S33 is specifically described as: The full working condition training set obtained by calculation meets the following conditions: ; the feature vectors of each training subset assigned to the category labels satisfy the following conditions: ; The training sample label vector obtained by calculation satisfies the following conditions: ; The calculated full-working condition test set meets the following conditions: ; the feature vectors of each test subset assigned to the category label satisfy the following conditions: ; The calculated test sample label vector satisfies the following conditions: 。

Description

Deep water drilling gas invasion monitoring method Technical Field The invention belongs to the technical field of deepwater oil-gas field drilling gas invasion monitoring, and particularly relates to a deepwater drilling gas invasion monitoring method. Background Under the large background of global energy transformation, the exploration and development of deep water oil and gas resources have become a key ring for maintaining energy safety. Nevertheless, deep water drilling operations still face serious safety challenges, with flooding and blowout being the greatest potential risks affecting safe production of the drilling platform. Once the gas invasion occurs and is not restrained in time, the imbalance of bottom hole pressure is directly caused, and the destructive blowout accident is extremely easy to occur. Therefore, developing an efficient gas invasion early identification technology is important to guaranteeing the safety of deep water drilling. However, after further research, the high nonlinearity and flow state variability of the gas-liquid two-phase flow in the water isolation pipe bring great challenges to deep water drilling gas invasion monitoring, namely, on one hand, weak gas invasion signals are often submerged by mechanical noise and environmental interference of a drilling platform, and on the other hand, the existing qualitative or semi-quantitative analysis means lack deep excavation capability on complex fluid dynamics characteristics, and small changes of gas content are difficult to accurately identify, so that urgent requirements of modern deep water drilling on high-precision and real-time quantitative monitoring cannot be met. Therefore, it is highly desirable for those skilled in the art to provide a completely new method for monitoring deep water drilling gas invasion. Disclosure of Invention The invention provides a deep water drilling gas invasion monitoring method, which extracts nonlinear dynamic characteristics of ultrasonic signals through multi-scale fuzzy divergence entropy, and obtains a gas content value based on inversion of a support vector machine model trained by a sample, so that quantitative monitoring and early warning of small change of the gas content are realized. In order to solve the technical problems, the invention adopts the following technical scheme: A deep water drilling gas invasion monitoring method comprises the following steps: s1, collecting ultrasonic echo signals of gas-liquid two-phase flow; s2, extracting a multi-scale fuzzy divergence entropy value in an ultrasonic echo signal; S3, inputting the feature vectors in the multi-scale fuzzy divergence entropy values into a trained support vector machine classification model to obtain a gas content recognition result; the step S3 specifically comprises the following steps: s31, constructing a training subset and a testing subset required by a unified recognition model; Step S32, carrying out MFDE operation on the training subset and the testing subset to respectively obtain a feature vector matrix of the training subset and a feature vector matrix of the testing subset; Step S33, splicing feature vector matrixes of the training subsets according to columns to obtain a full-working-condition training set, endowing category labels to feature vectors of all the training subsets in the full-working-condition training set, and merging to obtain training sample label vectors; The feature vector matrix of the test subset is spliced according to columns to obtain a full-working-condition test set, category labels are given to the feature vectors of all the test subset in the full-working-condition test set, and the feature vectors are combined to obtain test sample label vectors; step S34, a full-working-condition test set and a training sample label vector are used as inputs, and a MFDE-SVM algorithm with gas intrusion feature recognition capability is obtained through training; And comparing the predictive label vector with the test sample label vector to construct an confusion matrix, wherein the confusion matrix is used for evaluating the recognition performance of the trained support vector machine classification model on the deep water drilling gas content under different working conditions. Preferably, the process of extracting the multiscale ambiguity entropy value in the ultrasonic echo signal in the step S2 specifically includes the following steps: s21, performing multi-scale segmentation on the acquired ultrasonic echo signals of the gas-liquid two-phase flow, and converting the original time sequence of the ultrasonic echo signals into A plurality of multi-scale time sequences; Wherein, the Is the number of scale factors; the original time sequence of the ultrasonic echo signals satisfies: After multi-scale segmentation processing, the multi-scale time sequence of the ultrasonic echo signals meets the following conditions: (1) Wherein, the method comprises the steps of, ;For the data len