Multi-parameter collaborative ultra-deep water drilling well overflow leakage monitoring method

Abstract

The invention relates to a multi-parameter collaborative ultra-deep water drilling well overflow leakage monitoring method. According to the technical scheme, the method comprises the following steps of 1, data collection and preprocessing, wherein key parameters such as shaft flow and vertical pressure are collected in real time through a sensor; filtering and Z-Score standardization processing are carried out on the collected data; step 2, constructing a standard mode; step 3, identifying a single-parameter overflow leakage risk; step 4, support vector machine model algorithm: the method is mainly divided into a linear support vector machine and a non-linear support vector machine; 5, performing multi-parameter collaborative analysis; and step 6, monitoring and early warning. The overflow leakage risk identification method has the advantages that the problems that an existing overflow leakage monitoring technology is poor in real-time performance and insufficient in accuracy are solved, efficient data processing is achieved by improving the symbolic aggregation approximation method, and overflow leakage risk identification is carried out by utilizing the PSO-SVM model and integrating multi-parameter characteristics.

Description

Technical Field

[0001] The invention relates to the field of petroleum drilling engineering, and in particular to a multi-parameter coordinated ultra-deep water drilling overflow and leakage monitoring method. Background Art

[0002] In the process of deepwater and ultra-deepwater drilling, there are problems such as complex formation pressure system and narrow drilling fluid safety density window, which are prone to complex accidents such as overflow and leakage. These problems will destroy the wellbore pressure balance and lead to a series of safety hazards (such as wellbore collapse, drill bit sticking, etc.), which will have a significant impact on the safety and economy of drilling operations.

[0003] Existing overflow and leakage monitoring technologies mostly rely on single parameter analysis or traditional machine learning models, which makes it difficult to achieve real-time and accurate monitoring under complex working conditions. Traditional methods have the limitations of insufficient real-time performance and high false alarm rate when dealing with high-dimensional pattern recognition and nonlinear classification problems of multi-source data. Therefore, it is particularly important to develop a monitoring method that can integrate multi-parameter collaborative analysis and improve real-time performance and accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method to address the above-mentioned defects in the prior art, thereby solving the problems of poor real-time performance and insufficient accuracy of the existing overflow and leakage monitoring technology. Efficient data processing is achieved by improving the symbolic aggregation approximation method, and a particle swarm optimized support vector machine (PSO-SVM) model is used to comprehensively identify overflow and leakage risks based on multi-parameter features.

[0005] The present invention provides a multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method, the technical solution of which is as follows: Step 1: Data collection and preprocessing: Use sensors to collect key parameters of wellbore flow and standing pressure in real time; filter and Z-Score normalize the collected data; Step 2: Standard model construction: The construction methods of standard models include: numerical simulation method, using the calculation model of drilling overflow and leakage, simulating and analyzing the dynamic changes of monitoring parameters when overflow and leakage occur, and generating standard model curves reflecting the laws of overflow and leakage changes through theoretical calculation; historical data method, filtering and denoising the historical data of field data, extracting the changing trends of key parameters during overflow and leakage, and generating standard models; expert experience method, in the absence of suitable models and historical data, combining expert experience to describe parameter changes and generate temporary standard models; Step 3: Single parameter overflow and leakage risk identification: Based on step 2, data processing is performed. First, the dimension is reduced by segment and the compression ratio is set. k, divide the Z-Score standardized time series into several sequence segments, and use the least squares method to perform linear fitting on the data contained in each sequence segment to obtain the slope sequence S =( s 1 ,s 2 ,s 3 … s v ); Step 4: Support vector machine model algorithm: This method is mainly divided into linear support vector machine and nonlinear support vector machine; Step 5: Multi-parameter collaborative analysis: Based on the similarity measurement results of multiple parameter monitoring in step 3, the similarity measurement distances and label results of multiple parameters are used as the feature vector D of the support vector machine: , in, d i For the i Similarity measure of monitoring parameters; Step 6: Monitoring and early warning: Input the feature vector into the PSO-SVM model for classification to determine whether there is overflow or leakage. The output result is +1, indicating that overflow or leakage has occurred, and the output result is -1, indicating that overflow or leakage has not occurred.

[0006] Preferably, in the above step 1: For monitoring parameters with large fluctuations, low-pass filtering is used for processing, and for monitoring parameters with small fluctuations, mean filtering is used for smoothing; standardization is based on the mean and standard deviation of the original data. The processed data conforms to the normal distribution (mean is 0, standard deviation is 1), and the calculation formula is as follows: (1), In the formula, x origin is the original data; x z-score is the standardized data; m is the mean; d is the standard deviation.

[0007] Preferably, in the above step 3: By calculating the average difference of the data contained in each sequence segment, we can get the average difference sequence M =( m 1 ,m 2 ,m 3 … mv ); After the segmented dimensionality reduction, the length is w Time Series Q Transformed to length v The slope data series S and mean difference data series M ; In addition, the domain transformation transforms the domain of the slope and mean difference data series into the domain of the classical symbolic aggregation approximation, and the transformation calculation formula is: (2), In the formula, s i is the slope sequence before the domain transformation; s' i is the slope sequence after the domain transformation; k is the compression ratio; Mean difference domain conversion calculation formula: when k When is an even number, the domain transformation formula is: (3), when k When is an odd number, the domain transformation formula is: (4), In the formula, m i is the slope sequence before the domain transformation; m' i is the slope sequence after the domain transformation.

[0008] Preferably, in the above step 3: According to the symbolic representation method in the classical symbol aggregation approximation, the size of the letter set is first selected α , the slope data sequence after domain transformation S' and mean difference data series M' Symbolic representation is performed; similarity measurement is then calculated, and each time series segment is represented by slope and average difference to form a two-dimensional vector for curve identification, that is, the time series is represented as: (5), For length w Time Series Q 1 and Q 2 ,After filtering, standardization, segmented dimension reduction and symbolic representation, the following algorithm is used to calculate their similarity; Q 1 and Q 2 is expressed as and ,Right now: and , where 1≤x≤ v ; The similarity measurement distance calculation formula between two curves is as follows: (6), In the formula, distance The () function is a distance function that measures the distance between two characters, which is the same as the character distance measurement function in the classic symbol aggregation approximation method: (7), In the formula, Representation and r and c A weight or mapping value associated with a larger value of ; Representation and r and c A weight or mapping value associated with the smaller value of .

[0009] Preferably, in the above step 4: The basic process of the linear support vector machine classification model is as follows: ① Select the error penalty factor C>0, construct and solve the constrained optimization problem, the expression is as follows: (8), Use the SMO algorithm to find the corresponding minimum value of formula (8) α The value of the vector α* Vector, that is, the optimal solution of the above formula ; ②Calculate the linear classification model parameters oh , b , the calculation expression is as follows: (9), choose α* A component of Satisfy the conditions ,calculate b* : (10), ③The final classification hyperplane is calculated as follows: (11), The final classification decision function is: (12).

[0010] Preferably, in the above step 4: The classification model of the nonlinear support vector machine is: ① Select the kernel function as Gaussian kernel function, the error penalty factor C>0, construct and solve the constrained optimization problem, the expression is as follows: (13), The SMO algorithm is also used to find the optimal solution of formula (13): ; ②Calculate the parameters of the nonlinear classification model oh , b , the calculation expression is as follows: (14), choose α* A component of Satisfy the conditions ,calculate b* : (15), ③The final classification hyperplane is calculated as follows: (16), The final classification decision function is: (17); Furthermore, the training set and test set are constructed through historical data to train the support vector machine classification model.

[0011] Preferably, in the above step 5: The single parameter identification results are summarized as input for multi-parameter analysis.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention realizes efficient data processing by improving the symbolic aggregation approximation method, and utilizes the particle swarm optimized support vector machine (PSO-SVM) model to comprehensively identify the overflow loss risk by integrating multi-parameter features; its real-time performance is high: the improved symbolic aggregation approximation method significantly improves the data processing efficiency and adapts to the real-time monitoring needs under complex working conditions; its accuracy is high: through multi-parameter collaborative analysis and optimization of classification models, the false alarm rate and missed alarm rate are significantly reduced; its adaptability is strong: the system can achieve stable monitoring performance under complex working conditions and variable parameter environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram when the present invention is applied. DETAILED DESCRIPTION

[0014] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0015] Embodiment 1, a multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method mentioned in the present invention, comprising the following steps: Step 1: Data collection and preprocessing. Use sensors to collect key parameters such as wellbore flow rate and standing pressure in real time; filter and Z-Score the collected data to remove noise signals and improve data quality. For monitoring parameters with large fluctuations, low-pass filtering is used for processing, and for monitoring parameters with small fluctuations, mean filtering is used for smoothing. Standardization is based on the mean and standard deviation of the original data. The processed data conforms to the normal distribution (mean is 0, standard deviation is 1), and the calculation formula is as follows: (1), In the formula, x origin is the original data; x z-score is the standardized data; m is the mean; d is the standard deviation; Step 2: Standard model construction. The construction methods of the standard model include: numerical simulation method, using the calculation model of drilling overflow and leakage, simulating and analyzing the dynamic changes of monitoring parameters when overflow and leakage occur, and generating standard model curves reflecting the laws of overflow and leakage changes through theoretical calculation; historical data method, filtering and denoising the historical data of field data, extracting the changing trends of key parameters during overflow and leakage, and generating standard models; expert experience method, in the absence of suitable models and historical data, combining expert experience to describe parameter changes and generate temporary standard models; Step 3: Single parameter overflow and leakage risk identification. Based on step 2, data processing is performed. First, the dimension is reduced by segment and the compression ratio is set. k , divide the Z-Score standardized time series into several sequence segments, and use the least squares method to perform linear fitting on the data contained in each sequence segment to obtain the slope sequence S =( s 1 ,s 2 ,s 3 … s v ). By calculating the average difference of the data contained in each sequence segment, we can get the average difference sequence M =( m 1 ,m 2 ,m 3 … mv ). After the segmented dimensionality reduction, the length w Time Series Q Transformed to length v The slope data series S and mean difference data series M ; Further, the domain transformation transforms the domain of the slope and mean difference data series into the domain of the classical symbol aggregation approximation, and the transformation calculation formula is: (2), In the formula, s i is the slope sequence before the domain transformation; s' i is the slope sequence after the domain transformation; k is the compression ratio.

[0016] Mean difference domain conversion calculation formula: when k When is an even number, the domain transformation formula is: (3), when k When is an odd number, the domain transformation formula is: (4), Furthermore, according to the symbolic representation method in the classical symbol aggregation approximation, the size of the letter set is first selected α , the slope data sequence after domain transformation S' and mean difference data series M' Symbolic representation is performed; further, similarity measurement calculation is performed, and the slope and average difference are used to represent each time series segment to form a two-dimensional vector for curve identification, that is, the time series can be expressed as: (5), For the length w Time Series Q 1 and Q 2 , after filtering, standardization, segmented dimension reduction and symbolic representation, the following algorithm can be used to calculate their similarity. Q 1 and Q 2 is expressed as and ,Right now: and , where 1≤x≤ v The similarity metric distance calculation formula between two curves is as follows: (6), In the formula, distance () function is the distance function between two characters, which is the same as the character distance measurement function in the classic symbol aggregation approximation method: (7); Step 4: Support vector machine model algorithm. This method is mainly divided into two types: linear support vector machine and nonlinear support vector machine; the basic process of the linear support vector machine classification model is as follows: ① Select the error penalty factor C>0, construct and solve the constrained optimization problem, the expression is as follows: (8), Use the SMO algorithm to find the corresponding minimum value of formula (8) α The value of the vector α* Vector, that is, the optimal solution of the above formula .

[0017] ②Calculate the linear classification model parameters oh , b , the calculation expression is as follows: (9), choose α* A component of Satisfy the conditions ,calculate b* : (10), ③The final classification hyperplane is calculated as follows: (11), The final classification decision function is: (12), Nonlinear support vector machine classification model: ① Select the kernel function as Gaussian kernel function, the error penalty factor C>0, construct and solve the constrained optimization problem, the expression is as follows: (13), The SMO algorithm is also used to find the optimal solution of formula (13): .

[0018] ②Calculate the parameters of the nonlinear classification model oh , b , the calculation expression is as follows: (14), choose α* A component of Satisfy the conditions ,calculate b* : (15), ③The final classification hyperplane is calculated as follows: (16), The final classification decision function is: (17), Furthermore, the training set and test set are constructed through historical data to train the support vector machine classification model; Step 5: Multi-parameter collaborative analysis. Based on the similarity measurement results of multiple parameter monitoring in step 3, the similarity measurement distances and label results of multiple parameters are used as the feature vector D of the support vector machine: , in, d i For the i Similarity measure of the monitoring parameters.

[0019] Furthermore, the single parameter identification results are summarized as input for multi-parameter analysis; Step 6: Monitoring and early warning. Input the feature vector into the PSO-SVM model for classification to determine whether there is overflow or leakage. The output result is +1, indicating that overflow or leakage has occurred, and the output result is -1, indicating that overflow or leakage has not occurred.

[0020] Example 2, a multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method mentioned in the present invention, specifically a calculation and analysis was performed on an overflow well in the South China Sea, which overflowed when it was drilled to 4383.1m.

[0021] The specific steps are as follows: ① Select four monitoring data (mud pool volume, mechanical drilling speed, outlet flow rate and vertical pressure) before overflow occurs as overflow monitoring characteristic parameters; ② After filtering, standardizing and symbolizing the four groups of overflow monitoring parameters, the similarity measurement distance between the measured data and the overflow standard mode data is calculated to characterize the similarity between the two curves. If the similarity measurement distance between the two curves is small, it means that the change trends between the two curves are similar; ③ Divide the monitoring parameter time series into 10 sequence segments and select the size of the letter set α =5, the symbol sequence of filtering and standard mode is obtained, as shown in the following table: ; ④ According to the similarity measurement distance between the overflow monitoring curve and the overflow mode curve calculated in ③, it can be seen from the results that the filtered overflow monitoring curve and the overflow standard mode curve have a high similarity, which can realize the single parameter identification of overflow risk; ⑤ The similarity metric distance of each parameter (inlet and outlet flow difference, vertical pressure, mud pool increment and mechanical drilling speed) obtained after single parameter risk identification is used as the characteristic parameter value to form the characteristic vector of the multi-parameter overflow risk identification model based on support vector machine; ⑥ Divide the data into training set and test set, use the training set to train the model, and finally use the test set to verify whether the classification is correct; ⑦ Input the training set data into the overflow risk identification model, and then use the genetic algorithm to optimize the error penalty factor: 1.00289, the kernel parameter: 0.20306; ⑧Finally, the test set data without labeled data is input into the trained multi-parameter collaborative overflow risk identification model, and the classification results obtained are consistent with the labeled data of the test set, indicating that the established multi-parameter collaborative overflow risk identification model based on down vector machine can accurately identify overflow risks.

[0022] The above are only some preferred embodiments of the present invention. Any person skilled in the art may modify the above technical solutions or modify them into equivalent technical solutions. Therefore, the corresponding simple modifications or equivalent transformations made according to the technical solutions of the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method, characterized by: The following steps are involved: Step 1: Data collection and preprocessing: The key parameters of wellbore flow and standing pressure are collected in real time through sensors; Filter and Z-Score normalize the collected data; Step 2: Standard model construction: The construction methods of standard models include: numerical simulation method, using the calculation model of drilling overflow and leakage, simulating and analyzing the dynamic changes of monitoring parameters when overflow and leakage occur, and generating standard model curves reflecting the laws of overflow and leakage changes through theoretical calculation; historical data method, filtering and denoising the historical data of field data, extracting the changing trends of key parameters during overflow and leakage, and generating standard models; expert experience method, in the absence of suitable models and historical data, combining expert experience to describe parameter changes and generate temporary standard models; Step 3: Single parameter overflow and leakage risk identification: Based on step 2, data processing is performed. First, the dimension is reduced by segment and the compression ratio is set. k , divide the Z-Score standardized time series into several sequence segments, and use the least squares method to perform linear fitting on the data contained in each sequence segment to obtain the slope sequence S =( s 1 ,s 2 ,s 3 … s v ); Step 4: Support vector machine model algorithm: This method is divided into two types: linear support vector machine and nonlinear support vector machine; Step 5: Multi-parameter collaborative analysis: Based on the similarity measurement results of multiple parameter monitoring in step 3, the similarity measurement distances and label results of multiple parameters are used as the feature vector D of the support vector machine: ， in, d i For the i Similarity measure of monitoring parameters; Step 6: Monitoring and early warning: Input the feature vector into the PSO-SVM model for classification to determine whether there is overflow or leakage. The output result is +1, indicating that overflow or leakage has occurred, and the output result is -1, indicating that overflow or leakage has not occurred.

2. The multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method according to claim 1 is characterized by: In step one: For monitoring parameters with large fluctuations, low-pass filtering is used for processing, and for monitoring parameters with small fluctuations, mean filtering is used for smoothing; standardization is based on the mean and standard deviation of the original data. The processed data conforms to the normal distribution (mean is 0, standard deviation is 1), and the calculation formula is as follows: （1）， In the formula, x origin is the original data; x z-score is the standardized data; μ is the mean; δ is the standard deviation.

3. The multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method according to claim 1 is characterized by: step Three Middle Schools: By calculating the average difference of the data contained in each sequence segment, we can get the average difference sequence M =( m 1 ,m 2 ,m 3 … m v ); After the segmented dimensionality reduction, the length is w Time Series Q Transformed to length v The slope data series S and mean difference data series M ; In addition, the domain transformation transforms the domain of the slope and mean difference data series into the domain of the classical symbolic aggregation approximation, and the transformation calculation formula is: （2）， In the formula, s i is the slope sequence before the domain transformation; s' i is the slope sequence after the domain transformation; k is the compression ratio; Mean difference domain conversion calculation formula: when k When is an even number, the domain transformation formula is: （3）， when k When is an odd number, the domain transformation formula is: （4）， In the formula, m i is the slope sequence before the domain transformation; m' i is the slope sequence after the domain transformation.

4. The multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method according to claim 3 is characterized in that: in step three: According to the symbolic representation method in the classical symbol aggregation approximation, the size of the letter set is first selected α , the slope data sequence after domain transformation S' and mean difference data series M' Symbolic representation is performed; similarity measurement is then calculated, and each time series segment is represented by slope and average difference to form a two-dimensional vector for curve identification, that is, the time series is represented as: （5）， For length w Time Series Q 1 and Q 2. After filtering, standardization, segmented dimension reduction and symbolic representation, the following algorithm is used to calculate their similarity; Q 1 and Q 2 is represented as and ,Right now: and , where 1≤x≤ v ; The similarity measurement distance calculation formula between two curves is as follows: （6）， In the formula, dist The () function is a distance function that measures the distance between two characters, which is the same as the character distance measurement function in the classic symbol aggregation approximation method: （7）， In the formula, Representation and r and c A weight or mapping value associated with a larger value of ; Representation and r and c A weight or mapping value associated with the smaller value of .

5. The multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method according to claim 1 is characterized by: step Fourth Middle School: The basic process of the linear support vector machine classification model is as follows: ① Select the error penalty factor C>0, construct and solve the constrained optimization problem, the expression is as follows: （8）， Use the SMO algorithm to find the corresponding minimum value of formula (8) α The value of the vector α* Vector, that is, the optimal solution of the above formula ; ②Calculate the linear classification model parameters ω , b , the calculation expression is as follows: （9）， choose α* A component of Satisfy the conditions ,calculate b* : （10）， ③The final classification hyperplane is calculated as follows: （11）， The final classification decision function is: （12）。 6. The multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method according to claim 5 is characterized by: step Fourth Middle School: The classification model of nonlinear support vector machine is: ① Select the kernel function as Gaussian kernel function, the error penalty factor C>0, construct and solve the constrained optimization problem, the expression is as follows: （13）， The SMO algorithm is also used to find the optimal solution of formula (13): ; ②Calculate the parameters of the nonlinear classification model ω , b , the calculation expression is as follows: （14）， choose α* A component of Satisfy the conditions ,calculate b* : （15）， ③The final classification hyperplane is calculated as follows: （16）， The final classification decision function is: （17）； Furthermore, the training set and test set are constructed through historical data to train the support vector machine classification model.

7. The multi-parameter coordinated ultra-deepwater drilling overflow and leakage monitoring method according to claim 1 is characterized by: In step five: The single parameter identification results are summarized as input for multi-parameter analysis.

Images