Dynamic prediction and inducement identification method for wellbore integrity failure risk

Abstract

The dynamic prediction and inducement identification method for the shaft integrity failure risk is characterized in that different influence sequences of shaft integrity failure risk factors under different working conditions are fully considered, and key risk factors are determined by adopting a principal component analysis method; constructing a shaft integrity failure fault tree comprising a top event, a middle event and a bottom event by utilizing a failure mode and an effect analysis method; determining a monitoring index according to the service condition of the predicted well; according to real-time changes of field data, dynamic updating of a prior probability, a conditional probability and a transition probability is realized in combination with a sliding window and a long-short-term memory network; combining the shaft integrity failure fault tree and the dynamic Bayesian network to establish a shaft integrity failure dynamic risk prediction model; and based on the dynamic changes of the prior probability, the conditional probability and the transition probability, utilizing the shaft integrity failure dynamic risk prediction model to realize shaft integrity failure dynamic risk prediction and inducement identification.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and natural gas drilling and production engineering, and in particular to a method for dynamically predicting wellbore integrity failure risks and identifying inducements. Background Art

[0002] Oil and gas resources, one of the most important energy sources in today's human society, are a cornerstone supporting modern industry, transportation, and power supply. As shallow and mid-layer oil and gas exploration and development technologies mature, proven oil and gas reserves continue to expand, production growth slows, and discoveries become increasingly challenging. Global oil and gas exploration and development is shifting toward deep and ultra-deep oil and gas resources, leading to increasingly complex downhole environments. High temperatures, high pressures, and corrosive fluids pose a serious threat to wellbore integrity. A failure in wellbore integrity can lead to serious safety incidents such as blowouts, equipment damage, and hazardous substance leaks. Examples include the P-31A well accident in the Norwegian North Sea and the Macondo accident in the Gulf of Mexico, which resulted in casualties, economic losses, and irreversible damage to the ecological environment.

[0003] In actual production, because wellbore integrity covers its entire lifecycle, from design, drilling, completion, production, workover, production adjustment, to final abandonment, and faces various service conditions such as high temperature, high pressure, high corrosion, and erosion wear, the risk factors for wellbore integrity failure are complex and variable, with significant temporal dependencies. The correlation between risk status and risk factors changes dynamically over time. Therefore, constructing a dynamic risk prediction method to capture the time-varying patterns of wellbore integrity failure risk is key to predicting wellbore integrity failure risks and identifying their causes. By keeping abreast of the latest risk status and causes of wellbore integrity failure, preventive measures and emergency response methods can be implemented in a timely manner, significantly reducing accident risks, effectively preventing secondary damage, and ensuring that the wellbore remains safe throughout its lifecycle to guarantee production.

[0004] In recent years, with the development of artificial intelligence (AI) technology, machine learning methods have been applied to wellbore integrity assessment. For example, Patent Application No. 202411692600.9, "A Method for Real-Time Monitoring and Risk Prediction of Wellbore Integrity," proposes using a CNN-LSTM-Attention model for wellbore integrity risk prediction. Patent Application No. 202311589541.8, "An Intelligent Detection Method for Oil and Gas Wellbore Safety," proposes using a recurrent neural network model for wellbore integrity prediction. While these methods can predict wellbore integrity risks, their static learning based on historical data results in static, intermittent, discontinuous, and holistic assessment results. They are unable to dynamically update model parameters based on the real-time wellbore integrity status, making it difficult to locate and identify failure events. However, wellbore integrity research spans multiple stages, from design to abandonment. Risk assessment involves numerous steps, a long time span, numerous influencing factors, and complex interactions between these factors that change dynamically over time. Previous static methods cannot effectively reflect the interactions between risk factors and their dynamic changes over time, making it difficult to dynamically predict wellbore integrity failure risks. In view of the temporal dynamics and uncertainty of wellbore integrity failure risk, the present invention proposes a dynamic prediction and cause identification method for wellbore integrity failure risk, which can be used to solve the problems of dynamic changes in wellbore integrity failure risk over time and the difficulty in tracing key causes. Summary of the Invention

[0005] The purpose of this invention is to propose a method for dynamic prediction and cause identification of wellbore integrity failure risk, which can provide dynamic prediction of wellbore integrity failure risk conditions at oil and gas development sites and locate accident causes, helping on-site managers to grasp the risk conditions in a timely manner, take effective preventive measures, and block secondary accidents from occurring.

[0006] To achieve the above objectives, the present invention provides a method for dynamic prediction and cause identification of wellbore integrity failure, characterized in that: the method fully considers the different influence orders of wellbore integrity failure risk factors under different working conditions, and uses principal component analysis to determine key risk factors; utilizes failure mode and effect analysis to construct a wellbore integrity failure fault tree including top events, intermediate events, and bottom events; determines monitoring indicators based on the service conditions of the predicted well; dynamically updates prior probabilities, conditional probabilities, and transition probabilities based on real-time changes in field data by combining a sliding window and a long-short-term memory network; establishes a dynamic risk prediction model for wellbore integrity failure by combining the wellbore integrity failure fault tree and a dynamic Bayesian network; and utilizes the dynamic risk prediction model for wellbore integrity failure based on the dynamic changes in prior probabilities, conditional probabilities, and transition probabilities to achieve dynamic risk prediction and cause identification for wellbore integrity failure. The specific technical solutions adopted are as follows:

[0007] Step 1: Collect and organize cases of wellbore integrity failure events under different service conditions (including but not limited to high-temperature and high-pressure gas wells, heavy oil thermal recovery wells, deep shale gas wells, deepwater wells, unconventional oil and gas wells, and coalbed methane wells) and establish a wellbore integrity failure case library.

[0008] Step 2: Use principal component analysis to reduce the dimensionality of failure events in the wellbore integrity failure case database and determine the impact order and key risk factors of each failure event, as follows:

[0009] (1) Establish a risk matrix where X1, X2, X3, …, X N are risk indicators that represent the characteristics of each failure event, and M is the number of failure events.

[0010] (2) Use the Z-score method to standardize the M×N matrix to obtain the standard matrix Z. The calculation formula of the Z-score method is as follows:

[0011] Where Z ij is the jth standardized data of the i-th indicator, X ij is the jth original data of the i-th indicator, is the average value of the i-th indicator data, S i is the standard deviation of the i-th indicator data.

[0012] (3) Calculate the covariance matrix C, Where Z is the standard matrix, Z T is the transpose of the standard matrix, M is the number of samples. Perform eigendecomposition on the covariance matrix C and obtain the eigenvalues ​​λ1≥λ2≥…≥λ N and eigenvectors u1,u2,…,u N .

[0013] (4) According to the cumulative variance contribution rate of each indicator, determine the impact order of each failure event. When the cumulative variance contribution rate is greater than 75% to 95%, the corresponding indicator is a key risk factor. The calculation formula is as follows:

[0014] Where η i is the variance contribution rate, η Σ (p) is the cumulative variance contribution rate.

[0015] Step 3: Based on the impact order of each failure event and key risk factors determined in step 2, the potential failure model and effect analysis method are used to construct a wellbore integrity failure fault tree, including top events, intermediate events, and bottom events.

[0016] Step 4: Based on the current service conditions of the predicted well, determine the high-frequency failure events from the bottom events of the wellbore integrity failure fault tree and the historical data of wellbore integrity failure under similar conditions as the corresponding monitoring indicators.

[0017] Step 5: Based on the dynamic Bayesian network and the wellbore integrity failure fault tree established in step 3, an initial network model of wellbore integrity failure is established.

[0018] Step 6: Based on the correlation between each event and time in the wellbore integrity failure fault tree, each risk event is defined as a static node and a dynamic node. The static node is a node whose failure will not affect any node at the next moment, and the dynamic node is a node whose failure will affect a node at the next moment. Combine the dynamic Bayesian network and the wellbore integrity failure initial network model to establish a wellbore integrity failure transfer network.

[0019] Step 7: Use maximum likelihood estimation to determine the initial prior probability, initial condition probability, and initial transition probability of each bottom event in the wellbore integrity failure fault tree.

[0020] Step 8: Determine the initial time t0, window width W, and sliding step size S of the sliding window based on the data transmission frequency of the field sensor.

[0021] Step 9: Using the middle event of the wellbore integrity failure fault tree as the latent variable and the bottom event as the observed variable, the standard matrix Z obtained after the standardization process in step 2 is organized into a one-dimensional vector containing the latent variables and the observed variables, and a sliding window width of W continuous data points is used as a set of input sequences X t =[x t-49 ,x t-48 ,...,x t ] to train, and use this group as input to predict the next data value after this group.

[0022] Step 10: Take the failure probability and hidden variable state as the output layer, combined with the input sequence X t Build a multi-layer long short-term memory network and add a Dropout layer after each layer.

[0023] Step 11: Select the root mean square error (MSE) as the loss function. The formula is as follows:

[0024] Where n is the current number of samples, is the model prediction value, is the actual monitoring value.

[0025] Step 12: Use the Adam optimizer to update the model weights and biases in real time.

[0026] Step 13: When the new data point x t+1 When it arrives, update the sliding window to form a new input sequence X t+1 =[x t-48 ,x t-47 ,...,x t+1 ], the long short-term memory network model is used to predict the conditional probability and latent variable state prediction value at the current moment.

[0027] Step 14: Use the Sigmoid function method to transform the latent variable state prediction value to obtain the transition probability, and repeat steps 13 and 14 to achieve periodic updates of conditional probability and transition probability.

[0028] Step 15: Extract the input sequence X t+1 =[x t-48 ,x t-47 ,...,x t+1 ] corresponding to the monitoring indicators in the dataset D i , i is the number of monitoring indicators.

[0029] Step 16: Combine the piecewise linear mapping and the Sigmoid function to establish a prior probability mapping table.

[0030] Step 17: Based on the prior probability mapping table, the data set D i Map it to the prior probability of the corresponding bottom event. Repeat steps 15 and 17 to achieve periodic update of the prior probability.

[0031] Step 18: Combine the wellbore integrity failure transfer network model, initial prior probability, initial condition probability and initial transfer probability to establish a wellbore integrity failure dynamic risk prediction model.

[0032] Step 19: Use the box plot method to establish a calculation method for the comprehensive score S of wellbore integrity failure risk. The formula is as follows:

[0033] Where S is the comprehensive failure score; x is the failure probability on the nth day; n is equal to the window width W; and k is a positive integer.

[0034] Step 20: Divide the comprehensive score S of the wellbore integrity failure risk into different risk levels and establish a wellbore integrity failure risk level table.

[0035] Step 21: Using Bayesian network analysis software, with the sliding window width W as the risk assessment unit, combined with the wellbore integrity failure dynamic risk prediction model and the method of periodically updating prior probability, conditional probability, and transition probability, the risk prediction and cause identification of wellbore integrity failure are performed after each periodic update:

[0036] (1) Risk prediction: The dynamic risk prediction model of wellbore integrity failure and the probability of each node are input into the Bayesian network analysis software to obtain the wellbore integrity failure risk probability distribution. The interquartile range of the wellbore integrity failure risk probability distribution is then calculated using Formula 4 and used as the comprehensive score of the wellbore integrity failure probability for this risk assessment unit. The risk level of this risk assessment unit is obtained according to Table 1 to help on-site construction personnel understand the risk situation.

[0037] (2) Cause identification: When the risk level is at a higher risk level, risk accidents are most likely to occur. Bayesian network analysis software is used to perform accident inversion to obtain the probability that each bottom event may cause wellbore integrity failure. The bottom event with a higher probability is the possible cause of the accident. At this time, targeted preventive measures should be taken to prevent secondary accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is the technical roadmap of the present invention.

[0039] Figure 2 Schematic diagram of the fault tree for wellbore integrity failure.

[0040] Figure 3 Schematic diagram of the wellbore integrity failure transfer network model

[0041] Figure 4 Schematic diagram of the predicted distribution of wellbore integrity failure risk probability. DETAILED DESCRIPTION

[0042] The objects, technical features and advantages of the present invention will now be described in detail with reference to the accompanying drawings. The implementation methods shown and described in the accompanying drawings are merely exemplary and do not limit the applicable scope of the present invention.

[0043] Referring to the accompanying drawings, the present invention proposes a method for dynamically predicting wellbore integrity failure risk and identifying its causes, the method mainly comprising the following steps:

[0044] Step 1: Collect and organize cases of wellbore integrity failure events under different service conditions (high-temperature and high-pressure gas wells, heavy oil thermal recovery wells, deep shale gas wells, deepwater wells, unconventional oil and gas wells, and coalbed methane wells) and establish a wellbore integrity failure case library.

[0045] Step 2: Use principal component analysis to reduce the dimensionality of failure events in the wellbore integrity failure case database and determine the impact order and key risk factors of each failure event, as follows:

[0046] (1) Establish a risk matrix

[0047] (2) Using formula 1 to calculate the risk matrix X8×37 Perform normalization to obtain the standard matrix Z.

[0048] (3) Calculate the covariance matrix C.

[0049] (4) According to formula 2, determine the impact order of each failure event. When the cumulative variance contribution rate is greater than 75% to 95%, the corresponding indicator is a key risk factor.

[0050] Step 3: Based on the impact order and key risk factors of each failure event determined in step 2, use the potential failure model and effect analysis method to construct a wellbore integrity failure fault tree, including top events, intermediate events, and bottom events, as shown in the attached figure. Figure 2 shown.

[0051] Step 4: The predicted well is a high-temperature, high-pressure gas well. Based on the bottom events of the wellbore integrity failure fault tree and the historical data of wellbore integrity failures under similar working conditions, the remaining wall thickness of the tubing and the annular pressure are determined as monitoring indicators. The corresponding bottom events are corrosion damage and abnormal pressure, respectively.

[0052] Step 5: Based on the dynamic Bayesian network and the wellbore integrity failure fault tree established in step 3, an initial network model of wellbore integrity failure is established.

[0053] Step 6: Based on the correlation between each event and time in the wellbore integrity failure fault tree, each risk event is defined as a static node and a dynamic node. The static node is a node whose failure will not affect any node at the next moment, and the dynamic node is a node whose failure will affect a node at the next moment. Combining the dynamic Bayesian network and the wellbore integrity failure initial network model, a wellbore integrity failure transfer network is established, as shown in the attached figure. Figure 3 shown.

[0054] Table 1 is attached Figure 3 Each node in the corresponding event name and its status.

[0055] Table 1

[0056] Step 7: Use maximum likelihood estimation to determine the initial prior probability, initial condition probability, and initial transition probability of each bottom event in the wellbore integrity failure fault tree.

[0057] Step 8: According to the data transmission frequency of the field sensor, determine the initial time t0 of the sliding window to be 0, the window width W to be 15, and the sliding step S to be 15.

[0058] Step 9: Take the intermediate events of the wellbore integrity failure fault tree (cement sheath integrity failure, tubing integrity failure, annulus pressure, wellhead displacement, wellhead device failure, packer failure, downhole safety valve failure, and improper human operation) as latent variables and the bottom events (interface detachment, corrosion damage, corrosion damage, casing strength failure, tubing buckling deformation, abnormal pressure, wellhead leakage, wellhead lift, wellhead deviation, corrosion wear, sealing packing failure, tubing axial deformation, rubber sleeve damage, locking mechanism failure, sealing leakage, mechanical failure, corrosion failure, insufficient operating experience, system management omissions, and insufficient regular training) as observed variables. The standard matrix Z obtained after the standardization process in step 2 is organized into a one-dimensional vector containing latent variables and observed variables. The continuous data points with a sliding window width of W are used as a set of input sequences X. t =[x t-49 ,x t-48 ,...,x t ] to train, and use this group as input to predict the next data value after this group.

[0059] Step 10: Take the failure probability and hidden variable state as the output layer, combined with the input sequence X t A multi-layer long short-term memory network is built, and a dropout layer is added after each layer (the dropout rate is set to 0.2 to prevent overfitting).

[0060] Step 11: Select the root mean square error (MSE) as the loss function.

[0061] Step 12: Use the Adam optimizer to update the model weights and biases in real time.

[0062] Step 13: When the new data point x t+1 When it arrives, update the sliding window to form a new input sequence X t+1 =[x t-48 ,x t-47 ,...,x t+1 ], the failure probability and latent variable state prediction value at the current moment are obtained through the long short-term memory network model.

[0063] Step 14: Use the Sigmoid function method to transform the latent variable state prediction value to obtain the transition probability, and repeat steps 13 and 14 to achieve periodic updates of conditional probability and transition probability.

[0064] Step 15: Extract the input sequence X t+1 =[x t-48 ,x t-47 ,...,x t+1 ] corresponding to the monitoring indicators in the dataset D i , i is the number of monitoring indicators.

[0065] Step 16: Combine the piecewise linear mapping and the Sigmoid function to establish a prior probability mapping table.

[0066] Step 17: Based on the prior probability mapping table, the data set D i Map it to the prior probability of the corresponding bottom event. Repeat steps 15 and 17 to achieve periodic update of the prior probability.

[0067] Step 18: Combine the wellbore integrity failure transfer network model, initial prior probability, initial condition probability and initial transfer probability to establish a wellbore integrity failure dynamic risk prediction model.

[0068] Step 19: Use the box plot method to establish a calculation method for the comprehensive score S of wellbore integrity failure risk. The formula is as follows:

[0069] Where S is the comprehensive failure score; x is the failure probability on the nth day; n is equal to the window width W; and k is a positive integer.

[0070] Step 20: The comprehensive score S of the wellbore integrity failure risk is divided into five risk levels, and a wellbore integrity failure risk level table is established.

[0071] Table 2 is the wellbore integrity failure risk level table.

[0072] Table 2

[0073] Step 21: Using Bayesian network analysis software, with sliding window width W as the risk assessment unit, combined with the wellbore integrity failure dynamic risk prediction model and the method of periodically updating prior probability, conditional probability, and transition probability, the risk prediction and cause identification of wellbore integrity failure are performed using two update cycles (days 15 and 30) as examples:

[0074] (1) Risk prediction: The dynamic risk prediction model of wellbore integrity failure and the probability of each node are input into the Bayesian network analysis software to obtain the probability distribution of wellbore integrity failure, as shown in the attached figure. Figure 4 In the first 15 days, the S of window 1 is 0.545, and the risk level is medium according to Table 1; in the last 15 days, the S of window 2 is 0.561, and the risk level is medium according to Table 1.

[0075] (2) Cause identification: When the risk level reaches medium or above, risk accidents are likely to occur. Bayesian network analysis software is used to perform accident inversion to obtain the probability that each bottom event may cause wellbore integrity failure. Analysis of the probability that each bottom event may cause wellbore integrity failure shows that the probability P of wellhead uplift after the eighth day (D8) is significantly higher than that of other bottom events. The results are shown in Table 3. Therefore, wellhead uplift is considered to be the cause of the accident. At this time, on-site construction management personnel should focus on engineering maintenance and accident prevention measures related to the wellhead uplift problem.

[0076] Table 3 shows the probability of wellbore integrity failure caused by various bottom events on day 8.

[0077] Table 3

[0078] In Table 3, D1-D20 represent the bottom events in the wellbore integrity failure fault tree, corresponding to: interface detachment, corrosion damage, corrosion damage, casing strength failure, tubing buckling, abnormal pressure, wellhead leakage, wellhead uplift, wellhead deviation, corrosion wear, sealing packing failure, tubing axial deformation, rubber sleeve damage, locking mechanism failure, sealing leakage, mechanical failure, corrosion failure, insufficient operating experience, system management omissions, and insufficient regular training.

[0079] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic prediction of wellbore integrity failure risk and identification of causes, characterized in that: include: Step 1: Collect and organize wellbore integrity failure event cases under different service conditions (including but not limited to high-temperature and high-pressure gas wells, heavy oil thermal recovery wells, deep shale gas wells, deepwater wells, unconventional oil and gas wells, and coalbed methane wells) and establish a wellbore integrity failure case library; Step 2: Use principal component analysis to reduce the dimensionality of failure events in the wellbore integrity failure case database and determine the impact order and key risk factors of each failure event, as follows: (1) Establish a risk matrix where X1, X2, X3, …, X N are risk indicators representing the characteristics of each failure event, and M is the number of failure events; (2) Use the Z-score method to standardize the M×N matrix to obtain the standard matrix Z. The calculation formula of the Z-score method is as follows: Where Z ij is the jth standardized data of the i-th indicator, X ij is the jth original data of the i-th indicator, is the average value of the i-th indicator data, S i is the standard deviation of the i-th indicator data; (3) Calculate the covariance matrix C, Where Z is the standard matrix, Z T is the transpose of the standard matrix, M is the number of samples. Perform eigendecomposition on the covariance matrix C and obtain the eigenvalues ​​λ1≥λ2≥…≥λ N and eigenvectors u1,u2,…,u N ; (4) According to the cumulative variance contribution rate of each indicator, determine the impact order of each failure event. When the cumulative variance contribution rate is greater than 75% to 95%, the corresponding indicator is a key risk factor. The calculation formula is as follows: Where η i is the variance contribution rate, η ∑ (p) is the cumulative variance contribution rate; Step 3: Based on the impact order of each failure event and key risk factors determined in step 2, a potential failure model and effect analysis method are used to construct a wellbore integrity failure fault tree, including top events, intermediate events, and bottom events; Step 4: Based on the current service condition of the predicted well, determine the high-frequency failure events from the bottom events of the wellbore integrity failure fault tree and the historical data of wellbore integrity failure under similar conditions as the corresponding monitoring indicators; Step 5: Based on the dynamic Bayesian network and the wellbore integrity failure fault tree established in step 3, an initial network model of wellbore integrity failure is established; Step 6: Based on the correlation between each event and time in the wellbore integrity failure fault tree, each risk event is defined as a static node and a dynamic node. A static node is a node whose failure will not affect any node at the next moment, while a dynamic node is a node whose failure will affect a node at the next moment. A wellbore integrity failure transfer network is established by combining the dynamic Bayesian network and the initial wellbore integrity failure network model. Step 7: Use maximum likelihood estimation to determine the initial prior probability, initial condition probability, and initial transition probability of each bottom event in the wellbore integrity failure fault tree; Step 8: Determine the initial time t0, window width W, and sliding step size S of the sliding window based on the data transmission frequency of the field sensor; Step 9: Using the middle event of the wellbore integrity failure fault tree as the latent variable and the bottom event as the observed variable, the standard matrix Z obtained after the standardization process in step 2 is organized into a one-dimensional vector containing the latent variables and the observed variables, and a sliding window width of W continuous data points is used as a set of input sequences X t =[x t-49 ,x t-48 ,...,x t ] to train, and use this group as input to predict the next data value after this group; Step 10: Take the failure probability and hidden variable state as the output layer, combined with the input sequence X t Build a multi-layer long short-term memory network and add a dropout layer after each layer; Step 11: Select the root mean square error (MSE) as the loss function. The formula is as follows: Where n is the current number of samples, is the model prediction value, is the actual monitoring value; Step 12: Use the Adam optimizer to update the model weights and biases in real time; Step 13: When the new data point x t+1 When it arrives, update the sliding window to form a new input sequence X t+1 =[x t-48 ,x t-47 ,...,x t+1 ], using the long short-term memory network model to predict the conditional probability and latent variable state prediction value at the current moment; Step 14: Use the Sigmoid function method to convert the latent variable state prediction value to obtain the transition probability, and repeat steps 13 and 14 to achieve periodic updates of conditional probability and transition probability; Step 15: Extract the input sequence X t+1 =[x t-48 ,x t-47 ,...,x t+1 ] corresponding to the monitoring indicators in the dataset D i , i is the number of monitoring indicators; Step 16: Combine the piecewise linear mapping and the Sigmoid function to establish a prior probability mapping table; Step 17: Based on the prior probability mapping table, the data set D i Map it to the prior probability of the corresponding bottom event. Repeat steps 15 and 17 to achieve periodic update of the prior probability; Step 18: Combine the wellbore integrity failure transfer network model, initial prior probability, initial condition probability and initial transfer probability to establish a wellbore integrity failure dynamic risk prediction model; Step 19: Use the box plot method to establish a calculation method for the comprehensive score S of wellbore integrity failure risk. The formula is as follows: Where S is the comprehensive failure score; x is the failure probability on the nth day; n is equal to the window width W; and k is a positive integer. Step 20: Divide the wellbore integrity failure risk comprehensive score S into different risk levels and establish a wellbore integrity failure risk level table; Step 21: Using Bayesian network analysis software, with the sliding window width W as the risk assessment unit, combined with the wellbore integrity failure dynamic risk prediction model and the method of periodically updating prior probability, conditional probability, and transition probability, the risk prediction and cause identification of wellbore integrity failure are performed after each periodic update: (1) Risk prediction: The dynamic risk prediction model of wellbore integrity failure and the probability of each node are input into the Bayesian network analysis software to obtain the wellbore integrity failure risk probability distribution. The interquartile range of the wellbore integrity failure risk probability distribution is then calculated using Formula 4 and used as the comprehensive score of the wellbore integrity failure probability for this risk assessment unit. The risk level of this risk assessment unit is obtained according to Table 1 to help on-site construction personnel understand the risk status. (2) Cause identification: When the risk level is at a higher risk level, risk accidents are more likely to occur. Bayesian network analysis software is used to perform accident inversion to obtain the probability that each bottom event may cause wellbore integrity failure. The bottom event with a higher probability is the possible cause of the accident. At this time, targeted preventive measures should be taken to prevent secondary accidents.

Images