A method for monitoring deepwater drilling gas invasion

Abstract

The present application belongs to the technical field of gas invasion monitoring of deepwater oil and gas field drilling, and particularly relates to a deepwater drilling gas invasion monitoring method. The monitoring method extracts the nonlinear dynamic characteristics of ultrasonic signals through multi-scale fuzzy divergence entropy, and obtains the gas holdup value based on the support vector machine model trained by samples, thereby realizing quantitative monitoring and early warning of the small change of gas holdup. Compared with the existing deepwater drilling gas invasion monitoring method, the present method has significant advantages in monitoring accuracy, dynamic response characteristics and anti-noise interference performance, and can realize accurate identification and real-time quantitative monitoring of early weak gas invasion signal. A deepwater drilling gas invasion monitoring method comprises the following steps: collecting ultrasonic echo signals of gas-liquid two-phase flow; extracting multi-scale fuzzy divergence entropy values in the ultrasonic echo signals; inputting the feature vectors in the multi-scale fuzzy divergence entropy values into the trained support vector machine classification model to obtain the gas holdup recognition result.

Description

Technical Field

[0001] This invention belongs to the field of deepwater oil and gas field drilling gas invasion monitoring technology, and particularly relates to a deepwater drilling gas invasion monitoring method. Background Technology

[0002] Against the backdrop of the global energy transition, the exploration and development of deepwater oil and gas resources has become a crucial link in maintaining energy security. Nevertheless, deepwater drilling operations still face severe safety challenges; among them, spills and blowouts are the greatest potential risks affecting the safe operation of drilling platforms. Once gas intrusion occurs and is not contained in time, it will directly cause bottomhole pressure imbalance, easily leading to a catastrophic blowout accident. Therefore, developing efficient early gas intrusion identification technologies is crucial to ensuring the safety of deepwater drilling.

[0003] However, further research revealed that the highly nonlinear and variable flow regimes of the gas-liquid two-phase flow within the riser pose significant challenges to gas intrusion monitoring in deepwater drilling. On the one hand, weak gas intrusion signals are often masked by mechanical noise from the drilling platform and environmental interference. On the other hand, existing qualitative or semi-quantitative analysis methods lack the ability to delve deeper into complex fluid dynamics characteristics, making it difficult to accurately identify minute changes in gas content. Consequently, they cannot meet the urgent need for high-precision, real-time quantitative monitoring in modern deepwater drilling. Therefore, there is a pressing need for those skilled in the art to provide a novel method for monitoring gas intrusion in deepwater drilling. Summary of the Invention

[0004] This invention provides a method for monitoring gas intrusion in deep-water drilling. This method extracts the nonlinear dynamic features of ultrasonic signals through multi-scale fuzzy divergence entropy and obtains the gas content value based on a support vector machine model trained on samples, thereby realizing quantitative monitoring and early warning of minute changes in gas content. Compared with existing deep-water drilling gas intrusion monitoring methods, this method has significant advantages in monitoring accuracy, dynamic response characteristics, and anti-noise interference performance, and can achieve accurate identification and real-time quantitative monitoring of early weak gas intrusion signals.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A method for monitoring gas intrusion in deep-water drilling includes the following steps:

[0007] Step S1: Acquire ultrasonic echo signals of gas-liquid two-phase flow;

[0008] Step S2: Extract the multi-scale fuzzy divergence entropy value from the ultrasonic echo signal;

[0009] Step S3: Input the feature vectors from the multi-scale fuzzy divergence entropy values ​​into the trained support vector machine classification model to obtain the gas content identification result;

[0010] Step S3 specifically includes the following steps:

[0011] Step S31: Construct the training and testing subsets required for the unified recognition model;

[0012] Step S32: Perform MFDE operation on the training subset and the test subset to obtain the feature vector matrix of the training subset and the feature vector matrix of the test subset, respectively;

[0013] Step S33: Concatenate the feature vector matrices of the training subsets column by column to obtain the full-condition training set; and assign category labels to the feature vectors of each training subset in the full-condition training set, and merge them to obtain the training sample label vector;

[0014] The feature vector matrices of the test subsets are concatenated column by column to obtain the full-condition test set; and each feature vector of the test subset in the full-condition test set is assigned a category label, and the result is a test sample label vector after merging.

[0015] Step S34: Use the full-condition training set and training sample label vectors as input to train the MFDE-SVM algorithm with air intrusion feature recognition capability; input the full-condition test set into the trained MFDE-SVM algorithm to obtain the predicted label vector;

[0016] After comparing the predicted label vector with the test sample label vector, a confusion matrix is ​​constructed; the confusion matrix is ​​used to evaluate the recognition performance of the trained support vector machine classification model for gas content in deep water drilling under different working conditions.

[0017] Preferably, the process of extracting the multi-scale fuzzy divergence entropy value from the ultrasonic echo signal in step S2 specifically includes the following steps:

[0018] Step S21: Perform multi-scale segmentation on the acquired ultrasonic echo signal of the gas-liquid two-phase flow, and convert the original time series of the ultrasonic echo signal into... A multi-scale time series;

[0019] in, The number of scale factors;

[0020] The original time series of the ultrasonic echo signal satisfies: After multi-scale segmentation, the multi-scale time series of the ultrasonic echo signal satisfies: (1); where, ; The length of the data after coarsening satisfies: ; This represents the scale factor currently being calculated, satisfying: ;

[0021] Step S22: At scale Next, phase space reconstruction is performed on the multi-scale time series matrix composed of multi-scale time series of ultrasonic echo signals to obtain a higher-order matrix. ;

[0022] The multi-scale time series matrix satisfies: Higher-order matrices ,satisfy: (2);

[0023] Step S23: Calculate the higher-order matrix Cosine similarity between adjacent orbits;

[0024] Step S24: Divide the value range interval and calculate the membership matrix. ;

[0025] The range [-1, 1] is divided into... of A range;

[0026] And, membership matrix ,satisfy: (6);

[0027] in, cosine membership For interval The fuzzy membership degree satisfies: (5); , ; The number of intervals to be divided; The embedding dimension is used to capture the dynamic structure of the signal; This is a similarity tolerance used to control the sensitivity of the entropy value;

[0028] Step S25: Calculate the membership probability matrix ; and by calculating the state probability for each interval, the state probability matrix is ​​obtained. ;

[0029] Among them, the membership probability matrix ,satisfy: (7); probability By membership degree Through formula The result after conversion;

[0030] For the membership probability matrix Summing each column in the matrix yields the number of cosine similarities falling within each interval. ;

[0031] And, the state probability matrix ,satisfy: ; where the state probability of each interval ,satisfy: ;

[0032] Step S26: Based on the membership probability matrix The resulting embedding dimension is m and the scaling factor is m. The membership function order is Similarity tolerance is The number of signs is The MFDE algorithm was used to calculate the multi-scale fuzzy divergence entropy value in the ultrasonic echo signal;

[0033] The MFDE algorithm satisfies the following: (8).

[0034] Preferably, in step S23, the higher-order matrix is ​​calculated. The process of cosine similarity between adjacent orbits is specifically described as follows:

[0035] For higher order matrices Simplified representation;

[0036] Among them, higher-order matrices The simplified representation satisfies: (3);

[0037] in, The initial trajectory vector, To and Adjacent trajectory vectors, and so on;

[0038] Calculate higher-order matrices The cosine similarity between adjacent orbits is used to obtain the cosine similarity matrix. ;

[0039] in, (4).

[0040] Preferably, the process of constructing the training subset and test subset required for the unified recognition model in step S31 is specifically described as follows:

[0041] A certain current working condition The following R signal samples were obtained Recorded as ;in, Number the samples and put all The set formed is denoted as ;

[0042] in, Indicates the current operating condition Liquid phase flow rate feature, Indicates the current operating condition gas content feature; satisfy: ;gather ,satisfy: ;e represents the operating condition number;

[0043] Using a 6:4 ratio, several groups of current working conditions were analyzed. The set obtained below In The training subset is obtained by partitioning the dataset. With test subset ;

[0044] Among them, training subset ,satisfy: test subset ,satisfy: .

[0045] Preferably, the process of performing MFDE operation on the training subset and the test subset in step S32 is specifically described as follows: the calculated training subset eigenvector matrix ,satisfy: ;

[0046] The calculated test subset eigenvector matrix ,satisfy: .

[0047] Preferably, step S33 is specifically described as follows:

[0048] The calculated full-condition training set satisfies: ;

[0049] The feature vectors of each training subset, assigned category labels, satisfy: ;

[0050] The calculated training sample label vectors satisfy: ;

[0051] The calculated full-condition test set satisfies: ;

[0052] The feature vectors of each test subset, assigned category labels, satisfy: ;

[0053] The calculated test sample label vector satisfies: .

[0054] This invention provides a method for monitoring gas intrusion in deep-water drilling. The method includes the following steps: Step S1: Acquiring ultrasonic echo signals from a gas-liquid two-phase flow; Step S2: Extracting multi-scale fuzzy divergence entropy values ​​from the ultrasonic echo signals; Step S3: Inputting the feature vectors from the multi-scale fuzzy divergence entropy values ​​into a trained support vector machine classification model to obtain the gas content identification result.

[0055] The deep-water drilling gas intrusion monitoring method with the above characteristics has at least the following technical advantages compared with existing technologies:

[0056] (1) The deep-water drilling gas invasion monitoring method provided by the present invention adopts a multi-scale fuzzy divergence entropy algorithm. By integrating multi-scale analysis and fuzzy divergence measurement, it realizes the accurate extraction of nonlinear dynamic features related to gas content in deep-water drilling ultrasonic signals.

[0057] (2) The deep-water drilling gas invasion monitoring method provided by the present invention combines multi-scale fuzzy divergence entropy with support vector machine classification algorithm to achieve rapid identification of deep-water drilling gas invasion state and accurate quantification of gas content, and solves the technical problem of insufficient characterization of micro gas invasion state by existing time-frequency analysis methods. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the following drawings:

[0059] Figure 1 This is a flowchart illustrating a deep-water drilling gas intrusion monitoring method provided by the present invention.

[0060] Figure 2 A schematic diagram of the structure after installing an ultrasonic transducer and a ultrasonic transducer on the riser of a deep-water drilling well for gas intrusion monitoring.

[0061] Figure 3a This is a schematic diagram of the classification results obtained by the present invention based on multi-scale fuzzy divergence entropy;

[0062] Figure 3b This is a schematic diagram of the classification results obtained by existing technology based on multi-scale divergence entropy.

[0063] Figure 3c This is a schematic diagram of the classification results obtained by existing technology based on multi-scale fuzzy entropy.

[0064] Figure label:

[0065] 1. Water-proof pipe; 2. Transmitting ultrasonic transducer; 3. Receiving ultrasonic transducer; 4. Bubble. Detailed Implementation

[0066] This invention provides a method for monitoring gas intrusion in deep-water drilling. This method extracts the nonlinear dynamic features of ultrasonic signals through multi-scale fuzzy divergence entropy and obtains the gas content value based on a support vector machine model trained on samples, thereby realizing quantitative monitoring and early warning of minute changes in gas content. Compared with existing deep-water drilling gas intrusion monitoring methods, this method has significant advantages in monitoring accuracy, dynamic response characteristics, and anti-noise interference performance, and can achieve accurate identification and real-time quantitative monitoring of early weak gas intrusion signals.

[0067] like Figure 1 As shown, the present invention provides a method for monitoring gas intrusion in deep-water drilling, comprising the following steps:

[0068] Step S1: Acquire the ultrasonic echo signal of the gas-liquid two-phase flow.

[0069] To facilitate understanding by those skilled in the art, the following specific embodiments are provided to further explain and illustrate the present invention. (See references...) Figure 2 As shown, an ultrasonic transducer 2 and a ultrasonic transducer 3 are installed on the riser 1 of a deep-water well to be monitored for gas intrusion. Preferably, both the ultrasonic transducer 2 and the ultrasonic transducer 3 are at a fixed angle to the horizontal reference plane. ( The excitation frequency was 500kHz (30°) to obtain the maximum effective monitoring range and optimal sound field intensity. The ultrasonic echo signal was synchronously acquired at a sampling frequency of 5MHz, with a sampling time window set to 200ms. The final data length for each ultrasonic echo signal group was 1×10⁶ sampling points.

[0070] It is worth noting that the higher the gas content in the riser 1, the larger the contact area between the bubble 4 and the liquid phase, which will intensify nonlinear behaviors such as interface fluctuations and breakage, leading to an increase in system disorder. The relevant characteristic information is contained in the acquired ultrasonic echo signal data.

[0071] Step S2: Extract the multi-scale fuzzy divergence entropy value from the ultrasonic echo signal.

[0072] Based on completing step S1, step S2 is further implemented. As a preferred embodiment of the present invention, step S2, which extracts the multi-scale fuzzy divergence entropy value from the ultrasonic echo signal, specifically includes the following steps:

[0073] Step S21: Perform multi-scale segmentation on the acquired ultrasonic echo signal of the gas-liquid two-phase flow, and convert the original time series of the ultrasonic echo signal into... A multi-scale time series.

[0074] One point to add is that step S21 essentially involves averaging the original time series (corresponding ultrasound echo signal data) using sliding windows of different scales. After processing, the original time series will be converted... A multi-scale time series. Among them, This represents the number of scale factors.

[0075] It is worth noting that the original time series of the ultrasound echo signal satisfies: After multi-scale segmentation, the multi-scale time series of the ultrasonic echo signal satisfies: (1); where, ; The length of the data after coarsening satisfies: ; This represents the scale factor currently being calculated, satisfying: .

[0076] Step S22: At scale Next, phase space reconstruction is performed on the multi-scale time series matrix composed of multi-scale time series of ultrasonic echo signals to obtain a higher-order matrix. .

[0077] The multi-scale time series matrix satisfies: Higher-order matrices ,satisfy: (2).

[0078] Step S23: Calculate the higher-order matrix Cosine similarity between adjacent orbits.

[0079] Building upon steps S21 and S22, step S23 is further implemented. Preferably, step S23 involves calculating the higher-order matrix. The process of cosine similarity between adjacent orbits is specifically described as follows:

[0080] For higher order matrices Simplified representation.

[0081] Among them, higher-order matrices The simplified representation satisfies: (3);

[0082] in, The initial trajectory vector, To and Adjacent trajectory vectors, and so on.

[0083] Then, calculate the higher-order matrix. The cosine similarity between adjacent orbits is used to obtain the cosine similarity matrix. .in, (4).

[0084] Step S24: Divide the value range interval and calculate the membership matrix. .

[0085] The range [-1, 1] is divided into... of A range;

[0086] And, membership matrix ,satisfy: (6).

[0087] in, cosine membership For interval The fuzzy membership degree satisfies: (5); , .

[0088] The number of intervals to be divided; The embedding dimension is used to capture the dynamic structure of the signal; This is a similarity tolerance, used to control the sensitivity of the entropy value.

[0089] Step S25: Calculate the membership probability matrix ; and by calculating the state probability for each interval, the state probability matrix is ​​obtained. .

[0090] Among them, the membership probability matrix ,satisfy: (7); probability By membership degree Through formula The result after conversion.

[0091] For the membership probability matrix Summing each column in the matrix yields the number of cosine similarities falling within each interval. .

[0092] And, the state probability matrix ,satisfy: ; where the state probability of each interval ,satisfy: .

[0093] Step S26: Based on the membership probability matrix The resulting embedding dimension is m and the scaling factor is m. The membership function order is Similarity tolerance is The number of signs is The MFDE algorithm was used to calculate the multi-scale fuzzy divergence entropy value in the ultrasonic echo signal;

[0094] The MFDE algorithm satisfies the following: (8).

[0095] Step S3: Input the feature vectors from the multi-scale fuzzy divergence entropy values ​​into the trained support vector machine classification model to obtain the gas content identification result.

[0096] Based on completing step S2, step S3 is further implemented. In a preferred embodiment of the present invention, step S3 specifically includes the following steps:

[0097] Step S31: Construct the training subset and test subset required for the unified recognition model.

[0098] It is worth noting that in the process of inverting the feature vector in the multi-scale fuzzy divergence entropy value containing gas content data feature information, in order to facilitate the algorithm to complete the multi-condition classification task, it is first necessary to construct the training subset and test subset required by the unified recognition model.

[0099] The preferred step S31, which involves constructing the training and testing subsets required for the unified recognition model, is specifically described as follows:

[0100] A certain current working condition The following R signal samples were obtained Recorded as ;in, Number the samples and put all The set formed is denoted as .

[0101] in, Indicates the current operating condition Liquid phase flow rate feature, Indicates the current operating condition gas content feature; satisfy: ;gather ,satisfy: ; e represents the operating condition number.

[0102] Using a 6:4 ratio, several groups of current working conditions were analyzed. The set obtained below In The training subset is obtained by partitioning the dataset. With test subset .

[0103] Among them, training subset ,satisfy: test subset ,satisfy: .

[0104] Step S32: Perform MFDE operation on the training subset and the test subset to obtain the eigenvector matrix of the training subset and the eigenvector matrix of the test subset, respectively.

[0105] Building upon step S31, step S32 is further implemented. It is worth noting that here, after obtaining the required training and test subsets for the model, MFDE operations are further performed on these subsets. This is done because performing MFDE operations not only saves subsequent computational costs but also maximizes the computational efficiency of subsequent steps.

[0106] First, based on the comparison of parameter analysis and experimental results, select... A set that can be clearly distinguished from another set of A further preferred approach is to perform MFDE operations on the training subset and the test subset sequentially. The calculated training subset... eigenvector matrix ,satisfy: .

[0107] And, the calculated test subset eigenvector matrix ,satisfy: .

[0108] Step S33: Concatenate the feature vector matrices of the training subsets column by column to obtain the full-condition training set; assign category labels to the feature vectors of each training subset in the full-condition training set, and merge them to obtain the training sample label vector.

[0109] The feature vector matrices of the test subsets are concatenated column by column to obtain the full-condition test set; and each feature vector of the test subset in the full-condition test set is assigned a category label, and the result is a test sample label vector after merging.

[0110] Step S34: Use the full-condition training set and training sample label vectors as input to train the MFDE-SVM algorithm with air intrusion feature recognition capability; input the full-condition test set into the trained MFDE-SVM algorithm to obtain the predicted label vector;

[0111] A confusion matrix is ​​constructed by comparing the predicted label vector with the test sample label vector. This confusion matrix is ​​used to evaluate the performance of the trained support vector machine classification model in identifying gas content in deepwater drilling under different operating conditions.

[0112] Based on step S32, further steps S33 and S34 are implemented. Firstly, preferably, the full-condition training set calculated in step S33 satisfies: The feature vectors of each training subset, assigned category labels, satisfy: The calculated training sample label vector satisfies: The calculated full-condition test set satisfies: The feature vectors of each test subset, assigned category labels, satisfy: The calculated test sample label vector satisfies: .

[0113] Then, a full-condition training set was selected. and training sample label vectors The MFDE-SVM algorithm is trained using the input, resulting in an MFDE-SVM algorithm with air intrusion feature recognition capabilities. Subsequently, a full-condition test set is selected. The input is fed into the pre-trained MFDE-SVM algorithm to obtain the predicted label vector. Finally, the predicted label vectors will be... With test sample label vector After comparison, a confusion matrix can be constructed, thereby enabling the analysis of different operating conditions. The identification results of the gas content of the deep water well to be monitored.

[0114] Finally, to verify the superiority of the deep-water drilling gas invasion monitoring method provided by this invention over existing entropy methods, it was compared with existing gas invasion feature extraction methods based on multi-scale divergence entropy or multi-scale fuzzy entropy. (See references...) Figures 3a-3c As shown, where, Figure 3a This invention provides the classification results based on multi-scale fuzzy divergence entropy. Figure 3b This refers to the classification results obtained by existing technologies based on multi-scale divergence entropy; Figure 3cThe classification results are based on existing technologies using multi-scale fuzzy entropy. It can be observed that for different operating conditions with gas content of 0%, 1%, 2%, 3%, 4%, and 5%, at least 40 sets of samples are randomly selected for each condition to construct a test dataset. After feature extraction, the obtained feature vectors are input into a support vector machine (SVM) classification model for training and discriminant analysis. The classification results are quantitatively represented using a confusion matrix. The representation results show that the deep-water drilling gas intrusion monitoring method proposed in this invention achieves an accuracy rate of 98%, significantly outperforming existing inversion algorithms based on multi-scale divergence entropy (89.2%) and multi-scale fuzzy entropy (72.5%). Therefore, this invention provides a deep-water drilling gas intrusion monitoring method with significant advantages in gas intrusion feature discrimination and classification accuracy.

[0115] This invention provides a method for monitoring gas intrusion in deep-water drilling. The method includes the following steps: Step S1: Acquiring ultrasonic echo signals from a gas-liquid two-phase flow; Step S2: Extracting multi-scale fuzzy divergence entropy values ​​from the ultrasonic echo signals; Step S3: Inputting the feature vectors from the multi-scale fuzzy divergence entropy values ​​into a trained support vector machine classification model to obtain the gas content identification result.

[0116] The deep-water drilling gas intrusion monitoring method with the above characteristics has at least the following technical advantages compared with existing technologies:

[0117] (1) The deep-water drilling gas invasion monitoring method provided by the present invention adopts a multi-scale fuzzy divergence entropy algorithm. By integrating multi-scale analysis and fuzzy divergence measurement, it realizes the accurate extraction of nonlinear dynamic features related to gas content in deep-water drilling ultrasonic signals.

[0118] (2) The deep-water drilling gas invasion monitoring method provided by the present invention combines multi-scale fuzzy divergence entropy with support vector machine classification algorithm to achieve rapid identification of deep-water drilling gas invasion state and accurate quantification of gas content, and solves the technical problem of insufficient characterization of micro gas invasion state by existing time-frequency analysis methods.

[0119] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring gas intrusion in deep-water drilling, characterized in that, The steps include the following: Step S1: Acquire ultrasonic echo signals of gas-liquid two-phase flow; Step S2: Extract the multi-scale fuzzy divergence entropy value from the ultrasonic echo signal; Step S3: Input the feature vectors from the multi-scale fuzzy divergence entropy values ​​into the trained support vector machine classification model to obtain the gas content identification result; Step S3 specifically includes the following steps: Step S31: Construct the training and testing subsets required for the unified recognition model; Step S32: Perform MFDE operation on the training subset and the test subset to obtain the feature vector matrix of the training subset and the feature vector matrix of the test subset, respectively; Step S33: Concatenate the feature vector matrices of the training subsets column by column to obtain the full-condition training set; and assign category labels to the feature vectors of each training subset in the full-condition training set, and merge them to obtain the training sample label vector; The feature vector matrices of the test subsets are concatenated column by column to obtain the full-condition test set; and each feature vector of the test subset in the full-condition test set is assigned a category label, and the result is a test sample label vector after merging. Step S34: Use the full-condition training set and training sample label vectors as input to train the MFDE-SVM algorithm with air intrusion feature recognition capability; input the full-condition test set into the trained MFDE-SVM algorithm to obtain the predicted label vector; After comparing the predicted label vector with the test sample label vector, a confusion matrix is ​​constructed; the confusion matrix is ​​used to evaluate the recognition performance of the trained support vector machine classification model for gas content in deep water drilling under different working conditions.

2. The method for monitoring gas intrusion in deep-water drilling according to claim 1, characterized in that, The process of extracting the multi-scale fuzzy divergence entropy value from the ultrasonic echo signal in step S2 specifically includes the following steps: Step S21: Perform multi-scale segmentation on the acquired ultrasonic echo signal of the gas-liquid two-phase flow, and convert the original time series of the ultrasonic echo signal into... A multi-scale time series; in, The number of scale factors; The original time series of the ultrasonic echo signal satisfies: After multi-scale segmentation, the multi-scale time series of the ultrasonic echo signal satisfies: (1); where, ; The length of the data after coarsening satisfies: ; This represents the scale factor currently being calculated, satisfying: ; Step S22: At scale Next, phase space reconstruction is performed on the multi-scale time series matrix composed of multi-scale time series of ultrasonic echo signals to obtain a higher-order matrix. ; The multi-scale time series matrix satisfies: Higher-order matrices ,satisfy: (2); Step S23: Calculate the higher-order matrix Cosine similarity between adjacent orbits; Step S24: Divide the value range interval and calculate the membership matrix. ; The range [-1, 1] is divided into... of A range; And, membership matrix ,satisfy: (6); in, cosine membership For interval The fuzzy membership degree satisfies: (5); , ; The number of intervals to be divided; The embedding dimension is used to capture the dynamic structure of the signal; This is a similarity tolerance used to control the sensitivity of the entropy value; Step S25: Calculate the membership probability matrix ; and by calculating the state probability for each interval, the state probability matrix is ​​obtained. ; Among them, the membership probability matrix ,satisfy: (7); probability By membership degree Through formula The result after conversion; For the membership probability matrix Summing each column in the matrix yields the number of cosine similarities falling within each interval. ; And, the state probability matrix ,satisfy: ; where the state probability of each interval ,satisfy: ; Step S26: Based on the membership probability matrix The resulting embedding dimension is m and the scaling factor is m. The membership function order is Similarity tolerance is The number of signs is The MFDE algorithm was used to calculate the multi-scale fuzzy divergence entropy value in the ultrasonic echo signal; The MFDE algorithm satisfies the following: (8).

3. The method for monitoring gas intrusion in deep-water drilling according to claim 2, characterized in that, In step S23, the higher-order matrix is ​​calculated. The process of cosine similarity between adjacent orbits is specifically described as follows: For higher order matrices Simplified representation; Among them, higher-order matrices The simplified representation satisfies: (3); in, The initial trajectory vector, To and Adjacent trajectory vectors, and so on; Calculate higher-order matrices The cosine similarity between adjacent orbits is used to obtain the cosine similarity matrix. ; in, (4).

4. The method for monitoring gas intrusion in deep-water drilling according to claim 1, characterized in that, The process of constructing the training subset and test subset required for the unified recognition model in step S31 is specifically described as follows: A certain current working condition The following R signal samples were obtained Recorded as ;in, Number the samples and put all The set formed is denoted as ; in, Indicates the current operating condition Liquid phase flow rate feature, Indicates the current operating condition gas content feature; satisfy: ;gather ,satisfy: ;e represents the operating condition number; Using a 6:4 ratio, several groups of current working conditions were analyzed. The set obtained below In The training subset is obtained by partitioning the dataset. With test subset ; Among them, training subset ,satisfy: test subset ,satisfy: .

5. The method for monitoring gas intrusion in deep-water drilling according to claim 1, characterized in that, The process of performing MFDE operation on the training subset and the test subset in step S32 is specifically described as follows: the calculated training subset eigenvector matrix ,satisfy: ; The calculated test subset eigenvector matrix ,satisfy: .

6. The method for monitoring gas intrusion in deep-water drilling according to claim 1, characterized in that, Step S33 is specifically described as follows: The calculated full-condition training set satisfies: ; The feature vectors of each training subset, assigned category labels, satisfy: ; The calculated training sample label vectors satisfy: ; The calculated full-condition test set satisfies: ; The feature vectors of each test subset, assigned category labels, satisfy: ; The calculated test sample label vector satisfies: .

Images