CN-121980208-A - AI monitoring method for analyzing corrosion risk of drilling pipeline based on time sequence data

CN121980208ACN 121980208 ACN121980208 ACN 121980208ACN-121980208-A

Abstract

The invention provides an AI monitoring method for analyzing corrosion risk of a drilling pipeline based on time series data, which comprises the following steps of S1, multi-source time series data acquisition of the drilling pipeline of an offshore drilling platform, S2, preprocessing and standardization of original data, S3, time series data feature engineering and feature screening, S4, corrosion risk label marking and data set division, S5, AI corrosion risk prediction model training, S6, AI model performance verification and optimization, S7, real-time corrosion risk reasoning, S8, corrosion risk early warning and data feedback, and S9, historical data disc multiplexing and model iterative optimization. Aiming at corrosion risk of drilling pipelines of an offshore drilling platform, the invention constructs a full-flow AI monitoring system of data driving-algorithm fusion-closed loop iteration, and aims to realize real-time accurate identification, grading early warning and continuous optimization of corrosion risk, ensure safe operation and maintenance of the platform and reduce operation and maintenance cost.

Inventors

HU XIAODONG
XIONG XIHUI
LIU HAIPENG
ZHONG SHILIN
Qin liufu
XU ZHEN

Assignees

上海海达通信有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. The AI monitoring method for analyzing the corrosion risk of the drilling pipeline based on the time sequence data is characterized by comprising the following steps: S1, acquiring real-time sequence data and static basic data of a drilling pipeline sensor to form an original data set; S2, preprocessing and normalizing the original data set to form a standardized structured data set; s3, extracting features by adopting double windows based on a standardized structured data set, screening the features by using a subdivision algorithm, and removing redundant features to form a fusion model feature set; S4, marking a corrosion risk label based on a standardized structured data set on the basis of a fusion model feature set, realizing sample equalization by using a time sequence SMOTE algorithm, and dividing the sample equalization into a training set, a verification set and a test set; S5, training an AI corrosion risk prediction model, after inputting a training set, extracting a feature importance weight vector through an extreme gradient lifting model, inputting the weight vector and a time sequence feature sequence of the training set into a attention mechanism to generate a focused time sequence feature sequence, inputting the focused time sequence feature sequence into a two-way long-short-period memory network model, adopting an Adam optimizer and a cross entropy loss function, and performing end-to-end joint optimization on three algorithm parameters, calculating loss by using a verification set every iteration 1 round, and stopping training without descending continuous 5 rounds of loss to form an initial AI corrosion risk prediction model; S6, inputting the test set into an initial AI corrosion risk prediction model, and performing test verification by a core index, a fusion specificity index and a working condition, and tuning the algorithm to reach the standard to form a final AI corrosion risk prediction model; S7, acquiring real-time data from a sensor, preprocessing, extracting features, inputting a final AI corrosion risk prediction model, and outputting a real-time corrosion risk reasoning result; And S8, setting an early warning threshold value based on the risk level and the probability of the real-time corrosion risk reasoning result, triggering a corresponding early warning mode and a response flow, collecting real data to form classification feedback data, and synchronizing to S2, S3 and S5.
2. The AI monitoring method based on time series data analysis well drilling pipeline corrosion risk of claim 1, further comprising S9, collecting historical full-flow data monthly, dividing algorithm multi-disc performance, supplementing training sets with classification feedback data every 3 months to conduct quarterly small iteration, annual updating sensor deployment and characteristic engineering logic to conduct annual large iteration, and replacing an original model after iteration, wherein an authentication index reaches the standard to form a monitoring closed loop.
3. The AI monitoring method based on time series data analysis drilling pipeline corrosion risk of claim 1 is characterized in that the interaction process of three algorithms in the step S5 is characterized in that a feature importance weight vector output after training of an extreme gradient lifting model is directly used as input of a feature weight weighting step in an attention mechanism and used for amplifying contribution of high importance features, the attention mechanism calculates attention weight of each time step based on the weight vector and generates a time series feature sequence after focusing, the sequence is used as the unique input of a two-way long-short-term memory model and used for capturing long-term dependency relations, parameters of the three algorithms are synchronously updated through end-to-end joint optimization, cross entropy loss of two-way long-short-term memory is used as a target in the optimization process, and tree structure parameters of extreme gradient lifting, weight matrix of the attention mechanism and hidden layer parameters of bias and two-way long-short-term memory are reversely propagated and adjusted, so that three algorithms are ensured to cooperatively lift prediction precision.
4. The AI monitoring method for analyzing the corrosion risk of a well drilling pipeline based on time series data of claim 1, wherein in the step S1, the sensors comprise a corrosion rate sensor, a temperature sensor, a pressure sensor, a medium composition sensor, an environment parameter sensor, a medium flow rate sensor and a pipeline stress sensor, and the static basic data comprise pipeline materials, wall thicknesses, service time length, historical maintenance records, pipeline working condition historical data and corrosion prevention measure data.
5. The AI monitoring method based on time series data analysis drilling pipeline corrosion risk according to claim 1, wherein in step S2, the preprocessed missing value is processed by a sliding window average method for a high-frequency sensor and a linear interpolation method for a low-frequency sensor, redundant sensor data are called or manual inspection is triggered when the missing rate exceeds a threshold value, and the preprocessed abnormal value is processed by screening according to physical constraints of the sensors, verifying by combining a3 sigma rule with an industry threshold value, smoothly replacing fault data and reserving marks for real abnormal data.
6. The AI monitoring method based on time series data analysis well drilling pipeline corrosion risk of claim 1 is characterized in that in step S3, the extracted features comprise time domain features, frequency domain features, time series trend features, two-factor interaction features, three-factor interaction features, time series mutation features and long-term accumulation features, the double window extraction comprises short-term feature extraction by a small window of 1 hour and long-term feature extraction by a large window of 24 hours, and the subdivision algorithm screening features comprise extreme gradient lifting screening nonlinear related features, attention mechanism screening time series sensitive features and bidirectional long-term memory screening long-term related features.
7. The AI monitoring method for analyzing the corrosion risk of a well drilling pipeline based on time series data of claim 1, wherein in step S4, the sample data set is divided into time series, the training set is the first 8 months data, the verification set is the 9 months data, the verification set is the 15 th month data, the test set is the 10 th month data, the verification set is the 10 th month data, and each data set covers all pipeline areas.
8. The AI monitoring method based on time series data analysis of drilling pipeline corrosion risk according to claim 1, wherein in step S6, verification core indexes comprise the accuracy rate of not less than 90%, the high risk recall rate of not less than 95%, the extremely high risk recall rate of not less than 98%, fusion specificity indexes comprise the characteristic weight contribution rate of not less than 3%, the time step focusing accuracy of not less than 90%, the long-term dependency capture rate of not less than 85%, and working conditions comprise typhoon weather, medium switching and sensor fault scenes.
9. The AI monitoring method for analyzing corrosion risk of a drilling pipeline based on time series data according to claim 1, wherein in step S7, high risk area data is scheduled preferentially when the sensor acquires the data in real time, and the real-time corrosion risk reasoning result includes pipeline ID, risk level, risk probability, key influencing factors, key time steps, and recommended measures.
10. The AI monitoring method based on time series data analysis of the corrosion risk of the drilling pipeline according to claim 1 is characterized in that in the step S8, the early warning threshold is specifically attention early warning, namely, the risk in the middle and the probability is more than or equal to 0.7, general early warning, namely, the risk in the high and the probability is more than or equal to 0.7 or the risk in the middle and the probability is more than or equal to 0.9, urgent early warning, namely, the risk in the high and the probability is more than or equal to 0.6 or the high and the probability is more than or equal to 0.9, the early warning mode comprises word prompt, audible and visual warning and short message notification, and the response flow is 24-hour attention, 1-hour inspection and 10-minute shutdown.

Description

AI monitoring method for analyzing corrosion risk of drilling pipeline based on time sequence data Technical Field The invention relates to the technical field of offshore oil engineering safety monitoring, in particular to an AI monitoring method for analyzing corrosion risk of a drilling pipeline based on time sequence data. Background The drilling pipeline of the offshore drilling platform is a core channel for oil and gas exploitation and transportation, is in a complex environment of high salt fog, strong flushing and multi-medium interaction for a long time, the corrosion problem is directly related to the safe production of the platform, once the pipeline is leaked or broken due to corrosion, huge economic loss can be caused, and serious safety accidents such as marine pollution, fire explosion and the like can be caused. There are three major core pain points in current pipeline corrosion monitoring technology in the industry: 1. the traditional monitoring is mainly based on manual regular inspection and fixed point sampling detection, has limited coverage (hidden positions such as the inner wall of a pipeline, a deep sea section and the like cannot be monitored), has long detection period (usually 1-3 months and 1 time), is difficult to capture dynamic changes of corrosion risks, and is easy to miss early risk intervention opportunities; 2. The data utilization is insufficient, although partial sensors (such as a corrosion rate sensor and a temperature sensor) are deployed in the existing monitoring, the data processing is stopped at a 'simple threshold comparison' level (such as exceeding a preset corrosion rate, namely alarming), the nonlinear interaction relation of multi-source data (such as salinity and temperature collaborative accelerated corrosion) and the time sequence key characteristics (such as the triggering effect of instantaneous pressure mutation on corrosion) are not fully excavated, so that risk misjudgment and omission rate are high; 3. The model has poor adaptability, a part of intelligent monitoring schemes adopt a single AI model (such as simple LSTM or extreme gradient lifting), the three characteristics of nonlinear interaction, time sequence key focusing and long-term accumulation dependence of corrosion data can not be solved simultaneously, the complex interaction between characteristics is difficult to capture by the single LSTM, the long-term trend of the time sequence data can not be learned by the single extreme gradient lifting, the generalization capability of the model under ocean complex working conditions (such as typhoon and medium switching) is weak, and the accuracy is easy to drop after long-term use. Disclosure of Invention The invention provides an AI monitoring method for analyzing corrosion risk of a drilling pipeline based on time sequence data, which upgrades the marine drilling pipeline corrosion monitoring from 'passive detection' to 'active intelligent prevention and control' through 'data-model-application-iteration' full-link optimization, thereby providing technical support for safety production of marine oil and gas equipment. In order to achieve the above purpose, the invention adopts the following technical scheme: the AI monitoring method for analyzing the corrosion risk of the drilling pipeline based on the time sequence data comprises the following steps: S1, acquiring real-time sequence data and static basic data of a drilling pipeline sensor to form an original data set; S2, preprocessing and normalizing the original data set to form a standardized structured data set; s3, extracting features by adopting double windows based on a standardized structured data set, screening the features by using a subdivision algorithm, and removing redundant features to form a fusion model feature set; S4, marking a corrosion risk label based on a standardized structured data set on the basis of a fusion model feature set, realizing sample equalization by using a time sequence SMOTE algorithm, and dividing the sample equalization into a training set, a verification set and a test set; S5, training an AI corrosion risk prediction model, after inputting a training set, extracting a feature importance weight vector through an extreme gradient lifting model, inputting the weight vector and a time sequence feature sequence of the training set into a attention mechanism to generate a focused time sequence feature sequence, inputting the focused time sequence feature sequence into a two-way long-short-period memory network model, adopting an Adam optimizer and a cross entropy loss function, and performing end-to-end joint optimization on three algorithm parameters, calculating loss by using a verification set every iteration 1 round, and stopping training without descending continuous 5 rounds of loss to form an initial AI corrosion risk prediction model; S6, inputting the test set into an initial AI corrosion risk prediction model, and performing test verificati