A Deep Learning-Based Corrosion Risk Assessment Method, Device, Electronic Equipment, and Storage Medium for CO2-Driven Surface Gathering and Transportation Pipelines
By combining deep learning with the entropy weight method and the Delphi method, the weights of corrosion risk factors in CO2-driven surface gathering and transportation pipelines are determined. A risk probability analysis model is constructed using Bayesian networks and variational autoencoders, overcoming the shortcomings of existing evaluation methods and achieving accurate technical results. This solves the problem of single-risk assessment that existing technologies have failed to effectively address, providing a more scientific and precise assessment of corrosion risk levels.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 中国石油大学(北京)克拉玛依校区
- Filing Date
- 2026-03-25
- Publication Date
- 2026-06-19
AI Technical Summary
Existing methods for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines often rely on a single risk assessment method, which cannot accurately assess corrosion risk within the pipeline. These methods suffer from issues such as strong subjectivity, stringent data requirements, or insufficient quantification.
A deep learning-based approach was adopted, combining the entropy weight method and the Delphi method to determine the objective and subjective weights of corrosion risk factors. A corrosion risk probability analysis model was constructed using Bayesian networks and variational autoencoders, and the model was used to evaluate the corrosion risk level by comprehensively determining the weights.
It enables more accurate and reliable corrosion risk assessment of CO2-driven surface gathering and transportation pipelines, enhances uncertainty modeling capabilities and model interpretability, reduces reliance on manual feature selection, and improves the scientific rigor and accuracy of the assessment.
Smart Images
Figure CN122241599A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas field pipeline technology, specifically a method, apparatus, electronic device, and storage medium for assessing corrosion risk within CO2-driven surface gathering and transportation pipelines based on deep learning. Background Technology
[0002] CO2 corrosion has long plagued the oil and gas industry. During CO2-enhanced oil recovery, a series of corrosion-related chemical reactions occur, affecting the lifespan and safety of surface gathering and transportation pipelines. Specifically, CO2 dissolves in water to form carbonic acid, increasing the acidity of the pipeline fluid and thus exacerbating corrosion of pipelines and equipment. As the lifeline of the surface gathering and transportation system, the internal corrosion risk of gathering and transportation pipelines directly impacts the safe production and development efficiency of oil and gas fields. Therefore, accurately assessing the internal corrosion risk of CO2-enhanced surface gathering and transportation pipelines is of great practical significance.
[0003] Existing methods for assessing the internal corrosion risk of CO2-driven surface gathering and transportation pipelines mainly include qualitative assessment, semi-quantitative assessment, and quantitative assessment methods, specifically: Qualitative evaluation methods: Based on domestic and international standards (such as NACE SP0775, ISO 21457, SY / T 0087.2), the corrosion level is determined by referring to tables or rules based on factors such as pipe material, medium composition (CO2 partial pressure, H2S content, pH value, chloride ion concentration, etc.), temperature, and flow rate. Alternatively, experts in corrosion, materials, and processes can be organized to jointly assess the risk through meetings or the Delphi method, based on design documents, operational history, and on-site observations. The disadvantages of this method are its reliance on expert experience, high subjectivity, low accuracy, emphasis on qualitative descriptions of the risk nature and potential impact, and lack of quantitative analysis.
[0004] Quantitative evaluation method: Relying on mathematical models and statistical data to accurately quantify the probability and impact of risks, but it has strict requirements for data completeness.
[0005] For example: Existing patent document 1, publication number CN118070934A, discloses a method and apparatus for determining the internal corrosion risk of a gas field gathering and transmission pipeline. The method includes: constructing an internal corrosion risk assessment index system based on the internal corrosion test results of the gas field gathering and transmission pipeline, wherein the index system includes each index and its interrelationships; establishing a hierarchical analysis structure model using the analytic hierarchy process (AHP) based on the constructed internal corrosion risk assessment index system and the actual data of the gas field gathering and transmission pipeline to be inspected, and determining the weight values of each index in the internal corrosion index system of the gas field gathering and transmission pipeline to be inspected; and determining the risk value of internal corrosion of the gas field gathering and transmission pipeline based on the actual data of the gas field gathering and transmission pipeline to be inspected, the determined weight values of each index, and the predetermined internal corrosion risk standard.
[0006] Existing patent document 2, publication number CN110298540B, discloses a method for assessing corrosion risk in surface pipelines of oil and gas fields, including the following steps: dividing the oil and gas field into blocks; collecting corrosion influencing factors in pipelines of each block, including factors of fluid media and operating conditions, as well as the corresponding field-detected corrosion rate v; predicting the corrosion rate based on the collected corrosion influencing factors of pipelines in each block, and establishing a pipeline corrosion rate prediction model for different blocks; classifying leakage consequences; and assessing corrosion risk.
[0007] Semi-quantitative evaluation method: This method combines the advantages of qualitative and quantitative approaches, assigning risk levels based on expert judgment or empirical data to complete the corrosion risk assessment. The drawbacks of this method are that the quantitative results are not strictly statistical, and the weighting and scoring still contain subjective elements; the scores do not represent the absolute probability of failure or corrosion rate. Summary of the Invention
[0008] This invention provides a method, apparatus, electronic device, and storage medium for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines based on deep learning. It overcomes the shortcomings of the prior art and can effectively solve the problem that existing CO2-driven surface gathering and transportation pipeline corrosion risk assessment methods mostly use a single risk assessment method and cannot accurately assess the corrosion risk in the pipeline.
[0009] One of the technical solutions of this invention is achieved through the following measures: a deep learning-based method for assessing the corrosion risk inside CO2-driven surface gathering and transportation pipelines, comprising: Identify the main corrosion risk factors in each CO2-driven surface gathering and transportation pipeline section within the target block; The objective and subjective weights of various major corrosion risk factors are determined, and the objective and subjective weights are input into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning. By utilizing the comprehensive weights of various major corrosion risk factors, the corrosion risk level of the CO2-driven surface gathering and transportation pipeline in the target area is determined.
[0010] The following are further optimizations and / or improvements to the above-mentioned technical solution: The above-mentioned method utilizes various major corrosion risk factors and their corresponding comprehensive weights to determine the corrosion risk level within CO2-driven surface gathering and transportation pipelines in the target area, including: Input the comprehensive weights of various major corrosion risk factors in the target block into the corrosion risk probability analysis model to obtain the corrosion risk probability distribution results of the target block; By substituting the corrosion risk probability distribution of the target block into the corrosion risk level classification criteria, the corresponding corrosion risk level of the CO2-driven surface gathering and transportation pipeline is obtained. The corrosion risk level classification criteria are as follows: .
[0011] The process of constructing the above corrosion risk probability analysis model includes: Multiple samples were obtained and divided into training and testing sets according to the proportion. Each sample included the corrosion risk probability distribution result of the historical block and the comprehensive weight of various major corrosion risk factors in the historical block. The pre-set network model is trained using the training set, and training ends when the training stopping condition is met, thus obtaining the corrosion risk probability analysis model. The trained corrosion risk probability analysis model was tested using a test set, the model parameters were optimized, and a corrosion risk probability analysis model that meets the test evaluation requirements was output.
[0012] The aforementioned preset network model is a combination of a Bayesian network and a variational autoencoder, and the corresponding total loss function is as follows: in, This is the total loss function; This is the variational lower bound of the VAE; This represents the inference loss of a Bayesian network.
[0013] The above-mentioned determination of the objective and subjective weights of various major corrosion risk factors includes: The objective weights of various major corrosion risk factors were determined using the entropy weight method. The subjective weights of various major corrosion risk factors were determined using the Delphi method.
[0014] The construction process of the above-mentioned comprehensive weight determination model includes: A number of samples were obtained and divided into training and testing sets according to the proportion. Each sample included the comprehensive weight label of various major corrosion risk factors in the historical block and the objective and subjective weights of various major corrosion risk factors in the historical block. The training set is used to train the preset initial network. The training ends when the training stopping condition is met, and the comprehensive weight determination model is obtained. The preset initial network can be a feedforward neural network. The trained integrated weight determination model is tested using a test set, the model parameters of the integrated weight determination model are optimized, and the integrated weight determination model that meets the test evaluation requirements is output.
[0015] The main corrosion risk factors identified above for each CO2-driven surface gathering and transportation pipeline section in the target block include: The corrosion influencing factors of each CO2-driven surface gathering and transportation pipeline section in the historical block were obtained. The corrosion influencing factors include routine operation data and data detected by corrosion probes, field clip experiments and indoor weightlessness experiments. Corrosion correlation analysis was performed on corrosion influencing factors, and corrosion influencing factors whose analysis results exceeded the set threshold were taken as the main corrosion risk factors. Based on the main corrosion risk factor types identified through screening, the main corrosion risk factors for each CO2-driven surface gathering and transportation pipeline section in the target block were obtained.
[0016] The second technical solution of the present invention is achieved through the following measures: a deep learning-based CO2-driven surface gathering and transportation pipeline corrosion risk assessment device, comprising: The basic data acquisition unit identifies the main corrosion risk factors in each CO2-driven surface gathering and transportation pipeline section within the target block. The comprehensive weight analysis unit determines the objective and subjective weights of various major corrosion risk factors, and inputs the objective and subjective weights into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning. The corrosion risk level assessment unit uses the comprehensive weights of various major corrosion risk factors to determine the corrosion risk level inside the CO2-driven surface gathering and transportation pipeline in the target area.
[0017] The following are further optimizations and / or improvements to the above-mentioned technical solution: The aforementioned corrosion risk level assessment unit includes: The probability distribution acquisition module takes the comprehensive weights of various major corrosion risk factors in the target block as input to the corrosion risk probability analysis model and obtains the corrosion risk probability distribution results of the target block. The corrosion risk classification module inputs the corrosion risk probability distribution results of the target block into the corrosion risk level classification conditions to obtain the corresponding corrosion risk level of the CO2-driven surface gathering and transportation pipeline. The corrosion risk level classification conditions are as follows: .
[0018] The aforementioned comprehensive weight analysis unit includes: The objective weight determination module uses the entropy weight method to determine the objective weights of various major corrosion risk factors; The subjective weight determination module uses the Delphi method to determine the subjective weights of various major corrosion risk factors; The comprehensive weight determination module takes objective and subjective weights as input to the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning.
[0019] The third technical solution of the present invention is achieved through the following measures: an electronic device, including a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps in the deep learning-based method for assessing the corrosion risk in CO2-driven surface gathering and transportation pipelines.
[0020] The fourth technical solution of the present invention is achieved through the following measures: a storage medium storing a computer program that can be read by a computer, the computer program being configured to execute the steps in the deep learning-based method for assessing the corrosion risk in CO2-driven surface gathering and transportation pipelines.
[0021] This invention combines multiple risk assessment methods with real-time corrosion monitoring data from the field, and improves and optimizes the model through deep learning to obtain accurate corrosion risk assessment results for CO2-driven surface gathering and transportation pipelines. Specifically, it includes: This invention utilizes a comprehensive weight determination model built through deep learning, which organically combines subjective weights determined by expert scoring and objective weights determined by entropy weighting. This fully considers multiple factors, making the comprehensive weight allocation of various major corrosion risk factors in the target area more reasonable and scientific, and providing an effective data foundation for subsequent accurate corrosion risk level classification.
[0022] In this invention, the corrosion risk probability analysis model uses the comprehensive weights of various corrosion risk factors as a low-dimensional latent representation of the input data to the Bayesian network. This combines the fitting ability and complex pattern recognition capabilities of neural networks with the uncertainty handling and reasoning capabilities of Bayesian networks, resulting in more accurate and reliable corrosion risk probability analysis results, enhancing uncertainty modeling capabilities and model interpretability. Furthermore, a variational autoencoder (VAE) is introduced to optimize the Bayesian network's inference process, reducing its reliance on manual feature selection and increasing its robustness and inference accuracy. Attached Figure Description
[0023] Appendix Figure 1 This is a schematic diagram of the corrosion risk assessment method provided in an embodiment of the present invention.
[0024] Appendix Figure 2 This is a schematic flowchart of a method for determining the main corrosion risk factors of a target block in an embodiment of the present invention.
[0025] Appendix Figure 3 This is a schematic diagram of the method for obtaining the comprehensive weight of various major corrosion risk factors in the target block provided in the embodiments of the present invention.
[0026] Appendix Figure 4A schematic flowchart illustrating the method for determining the corrosion risk level within a CO2-driven surface gathering and transportation pipeline in a target block, as provided in an embodiment of the present invention.
[0027] Appendix Figure 5 This is a schematic diagram of the corrosion risk assessment device provided in an embodiment of the present invention. Detailed Implementation
[0028] The present invention is not limited to the following embodiments, and specific implementation methods can be determined according to the technical solutions and actual conditions of the present invention.
[0029] Those skilled in the art will understand that, unless otherwise stated, in the embodiments of this application, "module" or "unit" refers to a computer program or part of a computer program with a predetermined function, which works together with other related parts to achieve a predetermined goal, and can be implemented, wholly or partially, using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.
[0030] In addition, in the embodiments of this application, "multiple" refers to two or more, and "first" and "second" are used to distinguish descriptions and should not be construed as implying relative importance.
[0031] This application provides a method, apparatus, and electronic device for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines based on deep learning. This deep learning-based corrosion risk assessment apparatus for CO2-driven surface gathering and transportation pipelines can be integrated into a computer device, which can be a server, a terminal, or other similar device; it can also be executed jointly by a terminal and a server. The above examples should not be construed as limiting this application.
[0032] The aforementioned terminals may include mobile phones, wearable smart devices, tablets, laptops, personal computers (PCs), and in-vehicle computers, etc., and this application does not limit them. This application also does not limit the number of terminal devices.
[0033] The aforementioned server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. This application does not impose any restrictions on this.
[0034] For example, computer equipment identifies the main corrosion risk factors in each CO2-driven surface gathering and transportation pipeline section in the target block, determines the objective and subjective weights of each main corrosion risk factor, and inputs the objective and subjective weights into the comprehensive weight determination model to obtain the comprehensive weight of each main corrosion risk factor. The comprehensive weight determination model is obtained through deep learning. Using the comprehensive weight of each main corrosion risk factor, the corrosion risk level in the CO2-driven surface gathering and transportation pipeline of the target block is determined.
[0035] Based on this, the technical solution of this application will be described and explained below with reference to several examples.
[0036] Example 1: As shown in the attached document Figure 1 As shown in the figure, this invention discloses a deep learning-based method for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines, including: Step S110: Determine the main corrosion risk factors for each CO2-driven surface gathering and transportation pipeline section in the target block; Step S120: Determine the objective and subjective weights of various major corrosion risk factors, and input the objective and subjective weights into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning. Step S130: Using the comprehensive weight of various major corrosion risk factors, determine the corrosion risk level inside the CO2-driven surface gathering and transportation pipeline in the target block.
[0037] This invention discloses a deep learning-based method for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines. It uses big data analytics to determine the objective and subjective weights of various major corrosion risk factors, and then utilizes deep learning technology to obtain the comprehensive weights of these factors based on their objective and subjective weights. This provides a solid data foundation for accurately determining the corrosion risk level of CO2-driven surface gathering and transportation pipelines in the target area.
[0038] Example 2: As shown in the attached document Figure 2 As shown, the embodiments of the present invention are further optimizations of the above embodiments, wherein the specific steps for determining the main corrosion risk factors of the target block include: Step S210: Obtain the corrosion influencing factors for each CO2-driven surface gathering and transportation pipeline section in the historical block, wherein the corrosion influencing factors include: (1) Routine operating data: historical block temperature, pressure, flow rate, etc.; (2) The corrosion rate of historical blocks obtained by using corrosion probes, etc.; (3) The average corrosion rate of historical blocks obtained by on-site hanging plate experiments, etc.; (4) The corrosion rate of historical blocks obtained by indoor weightlessness test and the quantitative relationship between corrosion rate and CO2 partial pressure.
[0039] Step S220: Perform a corrosion impact correlation analysis on the corrosion influencing factors, and take the corrosion influencing factors whose analysis results are greater than the set threshold as the main corrosion risk factors. Step S230: Based on the main corrosion risk factor types obtained from the screening, identify the main corrosion risk factors for each CO2-driven surface gathering and transportation pipeline section in the target block.
[0040] This embodiment uses corrosion impact correlation analysis to screen out the main corrosion risk factors among the corrosion influencing factors, providing an effective data basis for the accurate assessment of corrosion risk in CO2-driven surface gathering and transportation pipelines.
[0041] Example 3: As shown in the attached document Figure 3 As shown, this embodiment of the invention is a further optimization of the above embodiment, wherein the objective weights and subjective weights of various major corrosion risk factors are determined, and the objective weights and subjective weights are input into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors in the target block, including: Step S310: Determine the objective weights of various major corrosion risk factors using the entropy weight method, including: (1) Based on the main corrosion risk factors of each CO2-driven surface gathering and transportation pipeline section in the target block, a matrix of main corrosion risk factors was established, and it was preprocessed and standardized. Specifically: First, to eliminate the impact of extreme outliers, quantile pruning was introduced to preprocess the data of the main corrosion risk factor matrix before standardization to remove outliers. Next, the Min-Max standardization method is used to standardize the matrix of main corrosion risk factors, resulting in a standardized matrix that eliminates dimensional differences. The formula for the Min-Max standardization method is shown below: in, The data is from the matrix of major corrosion risk factors. For the first j The minimum value of the main corrosion risk factors. For the first j The maximum value of the main corrosion risk factors, The data is standardized, representing the standardized result scaled to the range [0,1].
[0042] (2) Calculate the information entropy of each major corrosion risk factor in the standardized matrix to measure the uncertainty of each major corrosion risk factor. That is, the greater the information entropy, the smaller the information content and the smaller the weight of the major corrosion risk factor. in, The first in the standardized matrix j Information entropy of major corrosion risk factors For the first j The main corrosion risk factors are in the first category. i The percentage of data in the row k It is a constant. n This refers to the number of CO2-driven surface gathering and transportation pipeline sections.
[0043] (3) Calculate the objective weight of each major corrosion risk influencing factor using information entropy; in, For the first j The objective weights of major corrosion risk factors. For the first j Information entropy of major corrosion risk factors k It is a constant. m The total number of types of major corrosion risk factors.
[0044] Step S320: Determine the subjective weights of various major corrosion risk factors using the Delphi method, including: (1) Set with n One expert was used to obtain the expert score for each of the major corrosion risk factors using the Delphi method, and the average expert score for each major corrosion risk factor was calculated. in, For the first j Average expert scores for major corrosion risk factors For experts i For the j Expert scores for major corrosion risk factors n The total number of experts; The above settings have n Each expert was assigned an expert score for various major corrosion risk factors using the Delphi method, including: Identify problems and objectives: Based on the identified main corrosion risk factors, construct a relevant assessment indicator system; Select an expert panel: Select a group of experts in the field, with a number of 5 to 10 people, to ensure sufficient diversity and representativeness; Design a questionnaire and conduct the first survey: Design a questionnaire to ask each expert to evaluate the importance of various major corrosion risk factors. Experts score various major corrosion risk factors based on their existing knowledge and experience, and obtain corresponding expert scores. Use a proportional scale to compare the importance of each factor. The answers from experts are anonymous. Statistics and Feedback: Statistical analysis of the results of the first survey, including calculating the average expert score, standard deviation, variance and other statistical measures for each major corrosion risk factor; feeding back the statistical results of the expert scores for each major corrosion risk factor to the expert panel, and requiring experts to readjust their expert scores based on the feedback results in the second round of survey; Repeat the above steps to conduct 2-4 rounds of questionnaire surveys, and provide feedback on the results of each round of questionnaire surveys to the expert panel. The experts adjust their opinions based on the feedback results until they reach a relatively consistent opinion on the expert scores of various major corrosion risk factors.
[0045] (2) The average expert scores of each major corrosion risk factor are standardized to obtain the subjective weights of each major corrosion risk factor, and the sum of the subjective weights of each major corrosion risk factor is 1. in, For the first j Subjective weights of major corrosion risk factors m The total number of types of major corrosion risk factors. For the first j Average expert scores for major corrosion risk factors.
[0046] Step S330: Input the objective and subjective weights of various major corrosion risk factors into the comprehensive weight determination model to obtain the comprehensive weight of each major corrosion risk factor; This step, which involves determining the model's construction process by integrating weights, includes: (1) Obtain several samples and divide them into training set and test set according to the proportion. Each sample includes the comprehensive weight label of various major corrosion risk factors in the historical block and the objective weight and subjective weight of various major corrosion risk factors in the historical block. (2) Train the preset initial network using the training set, and end the training when the training stopping condition is met to obtain the comprehensive weight determination model; The aforementioned preset initial network can be any type of neural network, such as a feedforward neural network.
[0047] A feedforward neural network consists of an input layer, hidden layers, and an output layer, as detailed below: The input layer receives the objective and subjective weights of the main corrosion risk factors, and the input vector can be represented by the following formula: Wherein, input vector x The dimension can be 2m; for the first j The output of a neuron in a hidden layer is the weighted sum of the input vector and the neuron's weights, calculated using an activation function, as shown in the following formula: in, This is a subjective weight matrix. Here, the objective weight matrix is used, and ReLU is the ReLU activation function. For the first j The weight vector of each neuron. For bias terms; Propagation from hidden layer to output layer: There are k hidden layers, each outputting a vector. The neurons in the output layer map the outputs of the hidden layers to the final combined weights using a weight matrix and a bias term, as shown in the following formula: in, The final output is the overall weight. For the output layer i The weight vector of each neuron. For the output layer i Bias terms for each neuron.
[0048] The training stopping conditions mentioned above include the loss function stabilizing and the number of iterations reaching its maximum value. The loss function can be set as needed. In this embodiment, mean squared error can be selected as the loss function, and its formula is shown below: in, The final output is the overall weight. The true target weight is the overall weight of the sample.
[0049] (3) Test the trained integrated weight determination model using the test set, optimize the model parameters of the integrated weight determination model, and output the integrated weight determination model that meets the test evaluation requirements.
[0050] This embodiment utilizes a comprehensive weight determination model built through deep learning, organically combining the subjective weights determined by expert scoring and the objective weights determined by entropy weighting. This fully considers multiple factors, making the comprehensive weight allocation of various major corrosion risk factors in the target block more reasonable and scientific, and providing an effective data foundation for subsequent accurate corrosion risk level classification.
[0051] Example 4: As shown in the appendix Figure 4 As shown, this embodiment of the invention is a further optimization of the above embodiment, wherein the corrosion risk level inside the CO2-driven surface gathering and transportation pipeline in the target block is determined by utilizing various major corrosion risk factors and their corresponding comprehensive weights, including: Step S410: Input the comprehensive weights of various major corrosion risk factors in the target block into the corrosion risk probability analysis model to obtain the corrosion risk probability distribution results of the target block; The process of constructing the above corrosion risk probability analysis model includes: (1) Obtain multiple samples and divide them into training set and test set according to the proportion. Each sample includes the corrosion risk probability distribution result of historical blocks and the comprehensive weight of various major corrosion risk factors in historical blocks. (2) The pre-set network model is trained using the training set. The training ends when the training stopping condition is met, and the corrosion risk probability analysis model is obtained. It should be noted that the preset network model can be selected as needed, and can be, but is not limited to, a network model combining Bayesian networks (BN) and variational autoencoders (VAEs), specifically including: First, the comprehensive weight of various major corrosion risk factors is determined. w The input is fed into a variational autoencoder (VAE) for feature extraction and dimensionality reduction.
[0052] Variational Autoencoders (VAEs) are generative probabilistic models that combine deep learning and Bayesian inference, primarily consisting of an encoder and a decoder. Unlike traditional autoencoders that directly map the input to a fixed hidden vector, VAE encoders map the input data to a latent continuous probability distribution. Latent variables are then sampled from this distribution, and the decoder reconstructs the data. This embodiment introduces a VAE model to handle the complex and nonlinearly coupled weights of various major corrosion risk factors in the target block. Through VAE's powerful nonlinear dimensionality reduction and feature representation capabilities, noise and redundancy in the original weight data can be effectively eliminated, transforming it into a low-dimensional, dense, latent continuous feature space. This latent representation not only preserves the core distribution patterns of the original data but also provides a smoother and more robust prior input for the subsequent Bayesian network, thus overcoming the dependence on manual feature selection in traditional methods.
[0053] Based on the above principles, the specific structure of the variational autoencoder is as follows: The encoder maps the combined weights w of various major corrosion risk factors to latent variables. z（The probability distribution (representing the low-dimensional, dense latent features of the original input data) is used to estimate latent variables through a neural network. z mean μ (w) and variance σ (w) 2 The formula is expressed as follows: Decoder, the decoder part is derived from latent variables z The reconstructed probability of ground pipeline corrosion risk is generated, representing a latent variable. z Given the weights of the corrosion risk, the goal of the decoder is to enable the latent variables to reconstruct w from the latent variable z, as expressed in the following formula: Here, Decoder represents the decoding operation.
[0054] Subsequently, the latent variables output by the variational autoencoder are... z As input priors for Bayesian networks (BNs), these latent variables are used by Bayesian networks for decision-making and inference under uncertainty.
[0055] Bayesian networks (BNs) are probabilistic graphical models that combine probability theory and graph theory. They intuitively represent the conditional dependencies between multiple variables using a directed acyclic graph (DAG). In a Bayesian network, network nodes represent different random variables (such as latent feature attributes and corrosion risk levels in this embodiment), and directed edges between nodes represent causal or probabilistic dependencies between variables. Each node is also associated with a conditional probability distribution (CPD) or conditional probability table (CPT) to quantitatively describe the influence of the parent node on the state of its child node. This embodiment introduces Bayesian networks to leverage their powerful advantages in handling uncertainty, complex dependencies, and the fusion of prior knowledge, transforming the high-dimensional abstract features extracted by VAEs into highly interpretable corrosion risk probability results.
[0056] Based on the above principles, the specific inference process of the Bayesian network in this embodiment includes: Define a node: The latent feature variables extracted and output by VAE are used here as prior input nodes for Bayesian network inference. Y For corrosion risk level or probability of occurrence (target variable); Conditional Probability Distribution (CPD): The conditional probabilities of each node in a Bayesian network mainly include: P(Z|w) is the conditional probability of the latent variable Z given the comprehensive weight of the input (approximated by the encoder of VAE); P(Y|Z) is the conditional probability of the corrosion risk level occurring given the latent variable Z, based on comprehensive reasoning of the latent variable.
[0057] The continuous inference results of the Bayesian network are mapped to a corrosion risk probability distribution in the interval [0,1] using the Sigmoid activation function, as shown in the following formula: in, This represents the final corrosion risk probability P(Corrosion); e is the base of the natural logarithm; and x is the linear inference result after fusing latent variables in the Bayesian network output layer.
[0058] The training stopping conditions described above include the loss function converging and stabilizing, and the number of iterations reaching its maximum value. The total loss function... L total It needs to be minimized during training, and its formula is as follows: in, This is the total loss function; The reconstruction and regularization loss of VAE (i.e., the negative variational lower bound). For the inference loss of the Bayesian network; The specific formula is expressed as follows: in, The reconstruction loss is represented by the loss generated through the latent variable z. w Reconstruction error; Let KL divergence represent the difference between the latent variable distribution generated by the encoder and the prior distribution. p The difference in (z) is used to select the standard normal distribution as the prior.
[0059] Negative log-likelihood (NLL) is used as the inference loss for Bayesian networks. L BN The formula is expressed as follows: Where N is the total number of samples; Let be the latent variables extracted from the i-th sample using VAE; For the corresponding actual corrosion risk target variable; These are the model parameters in the Bayesian network.
[0060] (3) Test the trained corrosion risk probability analysis model using the test set, optimize the model parameters of the corrosion risk probability analysis model, and output a corrosion risk probability analysis model that meets the test evaluation requirements.
[0061] Step S420: Substitute the corrosion risk probability distribution results of the target block into the corrosion risk level classification conditions to obtain the corresponding corrosion risk level of the CO2-driven surface gathering and transportation pipeline. The corrosion risk level classification conditions are as follows: .
[0062] In this embodiment, the corrosion risk probability analysis model uses the comprehensive weights of various corrosion risk factors as a low-dimensional latent representation of the input data of the Bayesian network. This combines the fitting ability and complex pattern recognition capability of the neural network with the uncertainty handling and reasoning capabilities of the Bayesian network, resulting in more accurate and reliable corrosion risk probability analysis results, enhancing uncertainty modeling capabilities and model interpretability. Furthermore, a variational autoencoder (VAE) is introduced to optimize the inference process of the Bayesian network, reducing its reliance on manual feature selection and increasing its robustness and inference accuracy.
[0063] It should also be noted that all models built based on deep learning in the embodiments of the present invention can be incrementally trained by periodically consulting with experts, thereby improving the accuracy of long-term predictions.
[0064] Example 5: As shown in the attached document Figure 5 As shown in the figure, this invention discloses a deep learning-based corrosion risk assessment device for CO2-driven surface gathering and transportation pipelines, comprising: The basic data acquisition unit identifies the main corrosion risk factors in each CO2-driven surface gathering and transportation pipeline section within the target block. The comprehensive weight analysis unit determines the objective and subjective weights of various major corrosion risk factors, and inputs the objective and subjective weights into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning. The corrosion risk level assessment unit uses the comprehensive weights of various major corrosion risk factors to determine the corrosion risk level inside the CO2-driven surface gathering and transportation pipeline in the target area.
[0065] The corrosion risk level assessment unit includes: The probability distribution acquisition module takes the comprehensive weights of various major corrosion risk factors in the target block as input to the corrosion risk probability analysis model and obtains the corrosion risk probability distribution results of the target block. The corrosion risk classification module inputs the corrosion risk probability distribution results of the target block into the corrosion risk level classification conditions to obtain the corresponding corrosion risk level of the CO2-driven surface gathering and transportation pipeline. The corrosion risk level classification conditions are as follows: .
[0066] The comprehensive weight analysis unit includes: The objective weight determination module uses the entropy weight method to determine the objective weights of various major corrosion risk factors; The subjective weight determination module uses the Delphi method to determine the subjective weights of various major corrosion risk factors; The comprehensive weight determination module takes objective and subjective weights as input to the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning.
[0067] The specific steps of each unit / module in this embodiment are the same as those in embodiments 1 to 4, and will not be repeated here.
[0068] Example 6: This embodiment of the invention discloses a storage medium storing a computer program that can be read by a computer. The computer program is configured to execute the steps in the deep learning-based method for assessing the corrosion risk inside CO2-driven surface gathering and transportation pipelines when it runs.
[0069] The aforementioned storage media may include, but are not limited to, USB flash drives, read-only memory, portable hard drives, magnetic disks, optical disks, and other media capable of storing computer programs.
[0070] Example 7: This embodiment of the invention discloses an electronic device, including a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement the steps in the deep learning-based method for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines.
[0071] The processor described above can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. It can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The memory can include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, portable hard drives, magnetic disks, or optical disks.
[0072] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0073] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0075] The above content is only a specific embodiment of this application, which has strong adaptability and implementation effect. However, the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. Therefore, equivalent changes made in accordance with the claims of this application are still within the scope of this application.
Claims
1. A deep learning-based CO2 flooding ground gathering pipeline internal corrosion risk evaluation method, characterized in that, include: Identify the main corrosion risk factors in each CO2-driven surface gathering and transportation pipeline section within the target block; The objective and subjective weights of various major corrosion risk factors are determined, and the objective and subjective weights are input into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning. By utilizing the comprehensive weights of various major corrosion risk factors, the corrosion risk level of the CO2-driven surface gathering and transportation pipeline in the target area is determined.
2. The deep learning-based CO2 flooding ground gathering pipeline internal corrosion risk assessment method according to claim 1, characterized in that, The method of determining the corrosion risk level of CO2-driven surface gathering and transportation pipelines in the target area by utilizing various major corrosion risk factors and their corresponding comprehensive weights includes: Input the comprehensive weights of various major corrosion risk factors in the target block into the corrosion risk probability analysis model to obtain the corrosion risk probability distribution results of the target block; By substituting the corrosion risk probability distribution of the target block into the corrosion risk level classification criteria, the corresponding corrosion risk level of the CO2-driven surface gathering and transportation pipeline is obtained. The corrosion risk level classification criteria are as follows: 。 3. The deep learning-based CO2 flooding ground gathering pipeline internal corrosion risk assessment method according to claim 2, characterized in that, The process of constructing the corrosion risk probability analysis model includes: Multiple samples were obtained and divided into training and testing sets according to the proportion. Each sample included the corrosion risk probability distribution result of the historical block and the comprehensive weight of various major corrosion risk factors in the historical block. The pre-set network model is trained using the training set, and training ends when the training stopping condition is met, thus obtaining the corrosion risk probability analysis model. The trained corrosion risk probability analysis model was tested using a test set, the model parameters were optimized, and a corrosion risk probability analysis model that meets the test evaluation requirements was output. The preset network model is a combination of a Bayesian network and a variational autoencoder, and the corresponding total loss function is as follows: where, is the total loss function; is the variational lower bound of the VAE; is the inference loss of the Bayesian network.
4. The deep learning-based CO2 flooding surface gathering pipeline internal corrosion risk assessment method according to any one of claims 1 to 3, characterized in that, The determination of the objective and subjective weights of various major corrosion risk factors includes: The objective weights of various major corrosion risk factors were determined using the entropy weight method. The subjective weights of various major corrosion risk factors were determined using the Delphi method.
5. The method for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines based on deep learning according to any one of claims 1 to 4, characterized in that, The construction process of the comprehensive weight determination model includes: A number of samples were obtained and divided into training and testing sets according to the proportion. Each sample included the comprehensive weight label of various major corrosion risk factors in the historical block and the objective and subjective weights of various major corrosion risk factors in the historical block. The training set is used to train the preset initial network. The training ends when the training stopping condition is met, and the comprehensive weight determination model is obtained. The preset initial network can be a feedforward neural network. The trained integrated weight determination model is tested using a test set, the model parameters of the integrated weight determination model are optimized, and the integrated weight determination model that meets the test evaluation requirements is output.
6. The method for assessing corrosion risk in CO2-driven surface gathering and transportation pipelines based on deep learning according to any one of claims 1 to 5, characterized in that, The main corrosion risk factors identified in each CO2-driven surface gathering and transportation pipeline section within the target block include: The corrosion influencing factors of each CO2-driven surface gathering and transportation pipeline section in the historical block were obtained. The corrosion influencing factors include routine operation data and data detected by corrosion probes, field clip experiments and indoor weightlessness experiments. Correlation analysis of corrosion influencing factors was conducted, and corrosion influencing factors whose analysis results exceeded the set threshold were taken as the main corrosion risk factors. Based on the main corrosion risk factor types identified through screening, the main corrosion risk factors for each CO2-driven surface gathering and transportation pipeline section in the target block were obtained.
7. A deep learning-based corrosion risk assessment device for CO2-driven surface gathering and transportation pipelines using the method described in any one of claims 1 to 6, characterized in that, include: The basic data acquisition unit identifies the main corrosion risk factors in each CO2-driven surface gathering and transportation pipeline section within the target block. The comprehensive weight analysis unit determines the objective and subjective weights of various major corrosion risk factors, and inputs the objective and subjective weights into the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning. The corrosion risk level assessment unit uses the comprehensive weights of various major corrosion risk factors to determine the corrosion risk level inside the CO2-driven surface gathering and transportation pipeline in the target area.
8. The deep learning-based corrosion risk assessment device for CO2-driven surface gathering and transportation pipelines according to claim 7, characterized in that, The corrosion risk level assessment unit includes: The probability distribution acquisition module takes the comprehensive weights of various major corrosion risk factors in the target block as input to the corrosion risk probability analysis model and obtains the corrosion risk probability distribution results of the target block. The corrosion risk classification module inputs the corrosion risk probability distribution results of the target block into the corrosion risk level classification conditions to obtain the corresponding corrosion risk level of the CO2-driven surface gathering and transportation pipeline. The corrosion risk level classification conditions are as follows: ； or / and, The comprehensive weight analysis unit includes: The objective weight determination module uses the entropy weight method to determine the objective weights of various major corrosion risk factors; The subjective weight determination module uses the Delphi method to determine the subjective weights of various major corrosion risk factors; The comprehensive weight determination module takes objective and subjective weights as input to the comprehensive weight determination model to obtain the comprehensive weights of various major corrosion risk factors. The comprehensive weight determination model is obtained through deep learning.
9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps of the method as claimed in any one of claims 1 to 6.
10. A storage medium, characterized in that, The storage medium stores a computer program that can be read by a computer, the computer program being configured to execute the steps of the method as described in any one of claims 1 to 6 when it is run.