An offshore oil pipeline operation and maintenance regulation method based on artificial intelligence

Abstract

The present application relates to the technical field of pipeline operation and maintenance, and discloses a marine oil pipeline operation and maintenance regulation method based on artificial intelligence, which comprises the following steps: generating a no-leakage reference trajectory of a marine oil pipeline by using a no-leakage reference transient reconstruction model, and calculating an observation residual trajectory; extracting flow imbalance features, pressure transient features and temperature coupling features in the observation residual trajectory; calculating a local wax layer peeling non-leakage confidence score and a true micro-leakage confidence score by using a double-branch gating discriminant model, determining the abnormality of a marine oil pipeline restart stage, and generating an operation and maintenance regulation instruction. The present application realizes deep coupling analysis of pressure, temperature, flow and other data based on an artificial intelligence model, accurately distinguishes between local wax layer peeling and true micro-leakage, automatically outputs leakage alarm suppression or leakage alarm interlocking instructions, reduces the intervention intensity of operation and maintenance personnel, and improves the safety, stability and economy of the restart process of a marine waxy crude oil pipeline.

Description

Technical Field

[0001] This invention relates to the field of pipeline operation and maintenance control through big data analysis, and more particularly to an artificial intelligence-based method for the operation and maintenance control of offshore oil pipelines. Background Technology

[0002] As offshore oil and gas resource development continues to expand into deep water, ultra-deep water, and marginal oil fields, offshore oil gathering and transportation systems operate under complex conditions of high pressure, low temperature, long distance, and strong disturbance. Subsea pipelines, as the core transportation infrastructure connecting wellheads, subsea production systems, offshore platforms, and onshore terminals, undertake the critical task of continuously, stably, and safely transporting oil and gas. Their operational status directly determines the reliability and economy of the entire oil and gas production system. In actual operation, the marine environment in which offshore pipelines are located is complex and changeable. Factors such as ocean currents, seabed topography changes, and platform start-up and shutdown operations can all disturb pipeline operating parameters, further increasing the difficulty of pipeline operation monitoring and anomaly identification.

[0003] Waxy crude oil, due to its high wax content, high pour point, and sensitivity to temperature changes, is highly susceptible to wax deposition on pipe walls during long-distance transportation when the pipe wall temperature falls below the wax precipitation point. This leads to a continuous reduction in the effective pipeline diameter, a sustained increase in frictional losses along the pipeline, and a significant increase in transportation pressure and energy consumption. In severe cases, it can even cause pipeline flow assurance issues. Especially under strong transient conditions such as pipeline restart after shutdown and low-flow operation, the local wax layer adhering to the pipe wall is easily peeled off, broken, and resuspended by the flow of crude oil, causing short-term abrupt changes in the local pipeline diameter. This, in turn, causes significant fluctuations in key operating parameters such as pressure, temperature, and flow rate. These fluctuations are highly similar to those caused by real micro-leakage, making it easy to misjudge the anomaly type and posing a serious challenge to the safe operation and maintenance and accurate anomaly identification of offshore oil pipelines.

[0004] Existing research in the field of liquid pipeline control and safety monitoring has yielded numerous studies on data-driven approaches and information mining. For example, patent CN121031765A discloses a method and system for data mining of liquid pipeline control information. This method collects multi-source pipeline control data and performs data preprocessing to construct a standardized process dataset. Based on data mining algorithms, it extracts frequent itemsets and forms a control knowledge base. This effectively utilizes the massive amounts of real-time operational data generated by crude oil and refined oil pipeline control systems to mine the association rules and historical operation sequences between operation records and state parameters. This provides knowledge support for scheduling command execution, functional anomaly identification, and the generation of unknown operating condition operation sequences, thereby improving the intelligence level of liquid pipeline control. However, such methods are mostly geared towards conventional steady-state transportation and macro-control scenarios, focusing on historical data statistics and rule extraction. They do not address the specific modeling and precise differentiation of transient features during the shutdown and restart of offshore waxy crude oil pipelines, wax layer peeling, and minor leaks.

[0005] To address this issue, this invention proposes an AI-based operation and maintenance control method for offshore oil pipelines. By accurately identifying non-leakage anomalies induced by localized wax layer peeling and leakage anomalies induced by actual minor leaks in the early stages of restarting, the adaptability and robustness of existing leak detection systems in complex scenarios are improved. Summary of the Invention

[0006] This invention proposes an AI-based operation and maintenance control method for offshore oil pipelines. Step S2 generates a leak-free reference trajectory and calculates the observation residual trajectory through a leak-free reference transient reconstruction model, which can effectively amplify weak abnormal signals and remove strong operating condition background interference, solving the technical problems of traditional methods such as lack of ideal comparison benchmark, difficulty in highlighting minor leaks and wax layer peeling disturbances, and easy masking of abnormal signals. Step S3 extracts three types of mechanistic features—flow imbalance features, pressure transient features, and temperature coupling features—and constructs a differential fingerprint vector for wax layer peeling leakage, which can accurately characterize the essential pattern differences between the two types of anomalies. Step S4 outputs two types of confidence scores through a dual-branch gating discrimination model, which can achieve adaptive quantitative discrimination of anomaly types, solving the technical problems of poor generalization of traditional threshold rules, difficulty in identifying weak anomalies, and insufficient intelligent discrimination capabilities. Step S5 executes anomaly judgment based on confidence scores and generates targeted operation and maintenance control instructions, which can automatically distinguish between wax layer peeling and actual leakage and output alarm suppression or alarm interlocking instructions, solving the problems of frequent false alarms and delayed leakage response during the restart phase.

[0007] To achieve the above objectives, this invention provides an artificial intelligence-based method for the operation and maintenance of offshore oil pipelines, comprising the following steps: S1: Collect multi-source time-series detection data and pipeline control data during the restart phase of offshore oil pipelines, perform time-scale alignment and cleaning and correction processing on the multi-source time-series detection data and pipeline control data, and construct the early-stage aligned observation segment of restart and the restart boundary constraint vector. S2: Based on the early alignment observation segment of the restart and the restart boundary constraint vector, generate the leak-free reference trajectory of the offshore oil pipeline using the leak-free reference transient reconstruction model, and calculate the observation residual trajectory between the early alignment observation segment of the restart and the leak-free reference trajectory. S3: Extract the flow imbalance features, pressure transient features, and temperature coupling features from the observed residual trajectory, and fuse the flow imbalance features, pressure transient features, and temperature coupling features to construct a wax layer peeling and leakage differential fingerprint vector; S4: Input the differential fingerprint vector of wax layer peeling and leakage into the dual-branch gated discrimination model to obtain the confidence score of local wax layer peeling non-leakage and the confidence score of real micro-leakage; S5: Based on the confidence score of the local wax layer peeling non-leakage and the confidence score of the real micro-leakage, anomaly determination is made during the restart phase of the offshore oil pipeline, and operation and maintenance control instructions for leakage alarm suppression or leakage alarm interlock are generated.

[0008] As a further improvement of the present invention: Further, in step S1, the multi-source time-series detection data and pipeline control data are time-scaled and cleaned / corrected to construct the early-aligned observation segment of the restart and the restart boundary constraint vector, including: S11: Resample the time-series data in the multi-source time-series detection data and pipeline control data according to a fixed sampling interval to obtain a time series where the time interval between adjacent sampling results is equal to the fixed sampling interval; S12: The resampled time series is judged for out-of-bounds outliers, the sampling results judged as out-of-bounds outliers are removed, and the removed sampling results are completed by linear interpolation. S13: Extract the maximum value, minimum value, and mean value of the resampled delivery pump speed time series, inlet regulating valve opening time series, and outlet regulating valve opening time series from the pipeline control data. Normalize and splice the extracted results with the shutdown duration, offshore oil pipeline length configuration value, offshore oil pipeline inner diameter configuration value, and offshore oil pipeline riser elevation configuration value from the pipeline control data to serve as the restart boundary constraint vector. S14: Based on the shutdown start time in the pipeline control data, filter the stable transportation multi-source time series detection data before the shutdown start time, calculate the data mean of the stable transportation multi-source time series detection data, and use the data mean as a benchmark to perform benchmark correction on the multi-source time series detection data of the offshore oil pipeline restart phase to obtain the early restart aligned observation segment.

[0009] Furthermore, in step S2, the generation of a leak-free reference trajectory for the offshore oil pipeline using a leak-free reference transient reconstruction model includes: S21: The leak-free reference transient reconstruction model includes a timing coding layer, a constraint mapping layer, a conditional modulation decoding layer, and a multi-channel reconstruction output layer; S22: The temporal coding layer receives the early-aligned restart observation segment, performs a three-layer dilated convolution operation on the early-aligned restart observation segment, and performs global attention convergence on the result of the third dilated convolution operation to obtain the transient state vector corresponding to the early-aligned restart observation segment. S23: The constraint mapping layer receives the restart boundary constraint vector and maps the restart boundary constraint vector into constraint scaling and constraint offset; Specifically, the mapping formula for the constraint scaling amount and the constraint offset amount is as follows: ; ; in, Indicates the constraint scaling amount. Indicates the constraint offset. Represents the hyperbolic tangent function. This represents the parameters of the first trainable convolutional matrix in the constraint mapping layer. This represents the parameters of the trainable second convolutional matrix in the constraint mapping layer. This represents the first trainable bias parameter in the constraint mapping layer. This represents the trainable second bias parameter in the constraint mapping layer. Indicates the restart of the boundary constraint vector; S24: The conditional modulation decoding layer combines the constraint scaling amount and constraint offset amount to modulate and decode the transient state vector, generating a decoded vector at N positions, where N represents the length of the restarted early aligned observation segment; Specifically, the modulation and decoding formula for the transient state vector is: ; ; Where z represents the concatenation result of the decoded vectors at N positions, This represents a decoded vector with N positions. This represents the decoded vector at the nth position. This represents the trainable modulation matrix parameters in the conditional modulation decoding layer. This represents the trainable bias parameters in the conditional modulation decoding layer. This represents the element-wise multiplication operator. Represents the transient state vector. Indicates the activation function; S25: The multi-channel reconstruction output layer performs multi-channel reconstruction on the decoded vectors at the N positions to obtain a multi-channel reconstruction sequence. The multi-channel reconstruction sequence is used as a leak-free reference trajectory for offshore oil pipelines. The multi-channel reconstruction sequence contains 6 reconstruction sequences, and the length of each reconstruction sequence is N.

[0010] Furthermore, the calculation of the observation residual trajectory between the early aligned observation segment and the leak-free reference trajectory in step S2 also includes: The early alignment observation segment of the restart and the leak-free reference trajectory are obtained. The difference sequence between the early alignment observation segment of the restart and the leak-free reference trajectory is calculated to form the observation residual trajectory. The observation residual trajectory contains multiple sets of residual trajectories, which are as follows: offshore platform inlet pressure residual trajectory, onshore terminal outlet pressure residual trajectory, offshore platform inlet temperature residual trajectory, onshore terminal outlet temperature residual trajectory, offshore platform outlet instantaneous flow residual trajectory, and onshore terminal inlet instantaneous flow residual trajectory.

[0011] Further, step S3 extracts the flow imbalance features, pressure transient features, and temperature coupling features from the observed residual trajectory, including: S31: Extract the instantaneous flow residual trajectory at the offshore platform outlet and the instantaneous flow residual trajectory at the onshore terminal inlet from the observed residual trajectory; calculate the cumulative imbalance offset feature and dynamic distortion feature of the flow residual between the instantaneous flow residual trajectory at the offshore platform outlet and the instantaneous flow residual trajectory at the onshore terminal inlet; and splice the cumulative imbalance offset feature and the dynamic distortion feature of the flow residual together as the flow imbalance feature of the observed residual trajectory. S32: Extract the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal from the observed residual trajectory, calculate the average pressure phase deviation and pressure oscillation entropy between the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal, and splice the average pressure phase deviation and pressure oscillation entropy together as the pressure transient feature of the observed residual trajectory. S33: Extract the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory from the observed residual trajectory, and calculate the temperature coupling evolution factor between the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory using a sliding window method, which is used as the temperature coupling feature.

[0012] Furthermore, in step S3, the flow imbalance characteristics, pressure transient characteristics, and temperature coupling characteristics are spliced ​​together, and the splicing result is used as a differential fingerprint vector of wax layer peeling and leakage.

[0013] Further, in step S4, the differential fingerprint vector of wax layer peeling and leakage is input into the dual-branch gating discrimination model, including: S41: The dual-branch gated discrimination model includes a feature embedding layer, a gated attention modulation layer, a dual-branch independent discrimination layer, and a normalized score output layer; S42: The feature embedding layer receives the differential fingerprint vector of wax layer peeling and leakage, performs high-dimensional feature mapping on the differential fingerprint vector of wax layer peeling and leakage, and obtains high-dimensional hidden features; S43: The gated attention modulation layer converts the high-dimensional latent features into gated attention weights, and uses the gated attention weights to weight the high-dimensional latent features to obtain the gated modulation enhanced features; Specifically, the formula for generating the enhanced features after gated modulation is as follows: ; in, This indicates the enhanced features after gated modulation. Indicates the gating attention weight, Represents a trainable gated attention convolution matrix. This represents the trainable gating attention bias parameters. Represents an exponential function with the natural constant as its base; S44: The dual-branch independent discrimination layer uses two independent fully connected branches. The fully connected branches perform feature discrimination on the modulated enhanced features to generate a basic score for local wax layer peeling and non-leakage and a basic score for real micro-leakage. Specifically, the feature discrimination formula for the two independent fully connected branches is as follows: ; ; in, All represent the wax layer peeling branch parameters. All of these represent parameters with minimal leakage. This indicates a partial wax layer peeling without leakage, which is the basic score. This represents the baseline score for true micro-leakage; S45: The normalized scoring output layer normalizes the local wax layer peeling non-leakage baseline score and the real micro-leakage baseline score to generate a local wax layer peeling non-leakage confidence score and a real micro-leakage confidence score.

[0014] Specifically, the formulas for generating the confidence scores for localized wax layer peeling (non-leakage) and true micro-leakage are as follows: ; ; in, This indicates a confidence score for localized wax layer peeling that is not a leakage. This represents the confidence score for a truly minor leak.

[0015] Furthermore, in step S5, based on the confidence score for the non-leakage of the localized wax layer peeling and the confidence score for the actual minor leak, an anomaly determination is made for the restart phase of the offshore oil pipeline, including: The confidence scores for the localized wax layer peeling non-leakage and the confidence scores for the actual minor leakage are obtained. A dual threshold judgment rule is used to determine the anomalies in the restart phase of the offshore oil pipeline, and the anomaly types in the restart phase of the offshore oil pipeline are obtained. The anomaly types include localized wax layer peeling non-leakage anomalies and actual minor leakage anomalies. For localized wax layer peeling that is not a leakage anomaly, a leakage alarm suppression command is generated to avoid false alarms interfering with normal operation and maintenance; for real minor leakage anomalies, a leakage alarm interlock command is generated.

[0016] Compared with existing technologies, this invention proposes an artificial intelligence-based method for the operation and maintenance of offshore oil pipelines, which has the following beneficial effects: First, this invention generates a leak-free reference trajectory for offshore oil pipelines through a leak-free reference transient reconstruction model. Specifically, it employs a structured network architecture consisting of a temporal coding layer, a constraint mapping layer, a conditional modulation decoding layer, and a multi-channel reconstruction output layer. It utilizes three layers of dilated convolution to expand the temporal receptive field, preserves the causal features of the restart transient, and accurately extracts early operating conditions by combining global attention convergence. Through constraint mapping, the restart boundary constraint vector is transformed into modulated parameters, achieving accurate trajectory fitting under multiple operating conditions. The model can quickly generate ideal leak-free pressure, temperature, and flow time-series curves that match the real operating conditions in the early stages of restart, serving as a leak-free reference trajectory and providing a clean benchmark for anomaly detection. Gradient descent optimization training is used to make the reconstructed trajectory highly approximate real leak-free samples, significantly improving the identification of weak anomaly signals. This solves the problems of computational complexity, weak adaptability, and difficulty in online deployment of traditional physical models, laying a high-precision foundation for subsequent residual calculation and anomaly identification.

[0017] Meanwhile, this invention employs a dual-branch gated discrimination model to intelligently discriminate differential fingerprint vectors. Through a feature embedding layer, multi-source mechanism features are mapped to a high-dimensional space, enhancing feature representation capabilities. Specifically, this invention utilizes a gated attention layer to adaptively weight key dimensions, strengthening the sensitivity features of leakage and wax layer peeling while suppressing noise interference. The dual-branch independent discrimination structure specifically fits the two types of anomalies, avoiding feature confusion in single-branch models and improving recognition accuracy. Softmax normalization outputs a confidence score, making the results probabilistically interpretable. This enables the stable output of high-confidence scores in the short period early in the restart phase, distinguishing wax layer peeling from genuine minor leaks. This overcomes the shortcomings of traditional threshold methods, such as easy misjudgment and poor adaptability, providing reliable, accurate, and real-time quantitative basis for automated operation and maintenance decisions. Attached Figure Description

[0018] Figure 1 A flowchart illustrating an artificial intelligence-based operation and maintenance control method for offshore oil pipelines, provided as an embodiment of the present invention; Figure 2This is a neural network architecture diagram of a leak-free reference transient reconstruction model provided in an embodiment of the present invention; Figure 3 This is an experimental comparison diagram provided for one embodiment of the present invention. Detailed Implementation

[0019] The realization of the objectives, functional characteristics, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] This invention provides an artificial intelligence-based method for the operation and maintenance control of offshore oil pipelines. The executing entity of this AI-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the AI-based method for the operation and maintenance control of offshore oil pipelines can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0021] Reference Figure 1 as well as Figure 2 Embodiment 1 of the present invention is as follows: An artificial intelligence-based method for the operation and maintenance control of offshore oil pipelines, the method comprising: S1: Collect multi-source time-series detection data and pipeline control data during the restart phase of offshore oil pipelines, perform time-scale alignment and cleaning correction on the multi-source time-series detection data and pipeline control data, and construct the early-stage aligned observation segment and restart boundary constraint vector.

[0022] Specifically, in step S1, the multi-source time-series detection data and pipeline control data are time-scaled and cleaned / corrected to construct the early-aligned observation segment of the restart and the restart boundary constraint vector, including: As an embodiment of the present invention, by deploying a pipeline SCADA system and a DCS field control station, the pipeline control data of the offshore oil pipeline, the timing detection data of the offshore platform and the onshore terminal are acquired in real time. The offshore platform is the starting end of the offshore oil pipeline and is responsible for the production, collection and pressurized transportation of waxy crude oil. The offshore platform is equipped with a delivery pump and a regulating valve. The offshore oil pipeline is used to continuously transport the waxy crude oil produced by the offshore platform to the onshore terminal. The onshore terminal is used to receive and store the waxy crude oil. Specifically, the time-series detection data of the offshore platform and the land terminal are multi-source time-series detection data collected by deploying temperature sensors, pressure sensors and flow sensors on the offshore platform and the land terminal to collect corresponding time series, including the offshore platform inlet pressure time series, the land terminal outlet pressure time series, the offshore platform inlet temperature time series, the land terminal outlet temperature time series, the offshore platform outlet instantaneous flow time series, and the land terminal inlet instantaneous flow time series. Furthermore, the offshore platform inlet is the port where the offshore platform collects waxy crude oil, the offshore platform outlet is the port where the offshore platform transports waxy crude oil to the offshore oil pipeline, the onshore terminal inlet is the port where the onshore terminal receives waxy crude oil, and the onshore terminal outlet is the port where the onshore terminal exports waxy crude oil downstream. The pipeline control data includes the time of issuance of the restart command for waxy crude oil, the start time of the shutdown, the duration of the shutdown, the time series of the transfer pump speed, the time series of the inlet regulating valve opening, the time series of the outlet regulating valve opening, the configuration value of the offshore oil pipeline length, the configuration value of the offshore oil pipeline inner diameter, and the configuration value of the offshore oil pipeline riser elevation. Among them, the time series of the transfer pump speed, the time series of the inlet regulating valve opening, and the time series of the outlet regulating valve opening are the time series of the offshore oil pipeline restart phase. The offshore oil pipeline restart phase is a time period of 900 seconds starting from the time of issuance of the restart command. Furthermore, the start time of the shutdown is the moment when the pump speed is 0, the start time of the restart command is the moment when the operator issues the restart command in the pipeline SCADA system, and the time interval between the start time of the restart command and the start time of the shutdown is taken as the duration of the shutdown. The pipeline SCADA system is used to realize the real-time acquisition, remote monitoring and restart command issuance of operating parameters such as pressure, temperature, flow rate and pump and valve status of the entire offshore oil pipeline, offshore platform and onshore terminal; the DCS field control station is used to realize the local control and status feedback of equipment such as delivery pumps and regulating valves in offshore platforms and onshore terminals. The two types of systems together constitute the data acquisition and command execution basis of the offshore oil pipeline operation and maintenance control method.

[0023] S11: Resample the time-series data in the multi-source time-series detection data and pipeline control data according to a fixed sampling interval to obtain a time series where the time interval between adjacent sampling results is equal to the fixed sampling interval; Specifically, the time-series data in the multi-source time-series detection data and pipeline control data includes the time series of offshore platform inlet pressure, onshore terminal outlet pressure, offshore platform inlet temperature, onshore terminal outlet temperature, offshore platform outlet instantaneous flow rate, onshore terminal inlet instantaneous flow rate, delivery pump speed, inlet regulating valve opening time series, and outlet regulating valve opening time series. Furthermore, the fixed sampling interval is set to 5 seconds by default; S12: The resampled time series is judged for out-of-bounds outliers, the sampling results judged as out-of-bounds outliers are removed, and the removed sampling results are completed by linear interpolation. Specifically, the method for determining the out-of-bounds anomaly is as follows: if the sampling result is not within the preset upper and lower limits of the time series to which it belongs, the sampling result is determined to be an out-of-bounds anomaly. For example, for the sampling result sample, if sample is not within the preset upper and lower limits [min, max], the sampling result sample is determined to be an out-of-bounds anomaly, where min represents the preset lower limit and max represents the preset upper limit. As an embodiment of the present invention, the default settings of the preset upper and lower limit ranges of the offshore platform inlet pressure time series, onshore terminal outlet pressure time series, offshore platform inlet temperature time series, onshore terminal outlet temperature time series, offshore platform outlet instantaneous flow rate time series, onshore terminal inlet instantaneous flow rate time series, transfer pump speed time series, inlet regulating valve opening time series, and outlet regulating valve opening time series are respectively: 0-16MPa, 0-12MPa, 0-80 degrees Celsius, 0-70 degrees Celsius, 0-1000 cubic meters / hour, 0-1000 cubic meters / hour, 0-3000 rpm, 0-100%, and 0-100%. S13: Extract the maximum value, minimum value, and mean value of the resampled delivery pump speed time series, inlet regulating valve opening time series, and outlet regulating valve opening time series from the pipeline control data. Normalize and splice the extracted results with the shutdown duration, offshore oil pipeline length configuration value, offshore oil pipeline inner diameter configuration value, and offshore oil pipeline riser elevation configuration value from the pipeline control data to serve as the restart boundary constraint vector. S14: Based on the shutdown start time in the pipeline control data, filter the stable transportation multi-source time series detection data before the shutdown start time, calculate the data mean of the stable transportation multi-source time series detection data, and use the data mean as a benchmark to perform benchmark correction on the multi-source time series detection data of the offshore oil pipeline restart phase to obtain the early restart aligned observation segment.

[0024] Specifically, the stable transport multi-source time-series detection data includes stable offshore platform inlet pressure time series, stable onshore terminal outlet pressure time series, stable offshore platform inlet temperature time series, stable onshore terminal outlet temperature time series, stable offshore platform outlet instantaneous flow time series, and stable onshore terminal inlet instantaneous flow time series. The selection rules for the stable transportation multi-source time series detection data are as follows: the sequence length, start time and end time of each time series in the stable transportation multi-source time series detection data are consistent, the sequence length of each time series is not less than the preset length threshold (the default setting is 40), the time interval between adjacent data values ​​in the time series is a fixed sampling interval, and the mean standard deviation of each time series is less than the preset standard deviation threshold (the default setting is 2). The benchmark correction formula for the multi-source time-series detection data during the restart phase of the offshore oil pipeline is as follows: ; ; ; in, This indicates the restart of the early aligned observation segment. These represent the restarted early aligned observation segments, respectively. Time series segments of offshore platform inlet pressure, onshore terminal outlet pressure, offshore platform inlet temperature, onshore terminal outlet temperature, offshore platform outlet instantaneous flow rate, and onshore terminal inlet instantaneous flow rate, all corrected to the Chinese benchmark. Indicates restarting the early aligned observation fragment The first after the standard correction Such sequence fragments, Represents sequence fragments The sequence contains N sequence values, where N represents the length of the sequence segment. Represents sequence fragments The nth sequence value in the sequence, , Indicates the first in the multi-source time series detection data The nth sampling result of the time series, where the 1st to 6th time series are, respectively, the offshore platform inlet pressure time series, the onshore terminal outlet pressure time series, the offshore platform inlet temperature time series, the onshore terminal outlet temperature time series, the offshore platform outlet instantaneous flow time series, and the onshore terminal inlet instantaneous flow time series. This indicates the first [item] in the stable transport multi-source time-series detection data. The data mean of the stable time series, where the 1st to 6th stable time series are, respectively, the stable offshore platform inlet pressure time series, the stable onshore terminal outlet pressure time series, the stable offshore platform inlet temperature time series, the stable onshore terminal outlet temperature time series, the stable offshore platform outlet instantaneous flow time series, and the stable onshore terminal inlet instantaneous flow time series.

[0025] S2: Based on the early alignment observation segment of the restart and the restart boundary constraint vector, the leak-free reference trajectory of the offshore oil pipeline is generated using the leak-free reference transient reconstruction model, and the observation residual trajectory between the early alignment observation segment of the restart and the leak-free reference trajectory is calculated.

[0026] Specifically, step S2, which uses a leak-free reference transient reconstruction model to generate a leak-free reference trajectory for an offshore oil pipeline, includes: S21: See below Figure 2 The diagram shows the neural network architecture of the leak-free reference transient reconstruction model, which includes a temporal coding layer, a constraint mapping layer, a conditional modulation decoding layer, and a multi-channel reconstruction output layer. S22: The temporal coding layer receives the early-aligned restart observation segment, performs a three-layer dilated convolution operation on the early-aligned restart observation segment, and performs global attention convergence on the result of the third dilated convolution operation to obtain the transient state vector corresponding to the early-aligned restart observation segment. Specifically, the dilation rates of the three dilated convolution operations are set to 1, 2, and 4 respectively; S23: The constraint mapping layer receives the restart boundary constraint vector and maps the restart boundary constraint vector into constraint scaling and constraint offset; Specifically, the mapping formula for the constraint scaling amount and the constraint offset amount is as follows: ; ; in, Indicates the constraint scaling amount. Indicates the constraint offset. Represents the hyperbolic tangent function. This represents the parameters of the first trainable convolutional matrix in the constraint mapping layer. This represents the parameters of the trainable second convolutional matrix in the constraint mapping layer. This represents the first trainable bias parameter in the constraint mapping layer. This represents the trainable second bias parameter in the constraint mapping layer. Indicates the restart of the boundary constraint vector; S24: The conditional modulation decoding layer combines the constraint scaling amount and constraint offset amount to modulate and decode the transient state vector, generating a decoded vector at N positions, where N represents the length of the restarted early aligned observation segment; Specifically, the modulation and decoding formula for the transient state vector is: ; ; Where z represents the concatenation result of the decoded vectors at N positions, This represents a decoded vector with N positions. This represents the decoded vector at the nth position. This represents the trainable modulation matrix parameters in the conditional modulation decoding layer. This represents the trainable bias parameters in the conditional modulation decoding layer. This represents the element-wise multiplication operator. Represents the transient state vector. This represents the activation function; optionally, the activation function can be set to the ReLU function or the GeLU function. S25: The multi-channel reconstruction output layer performs multi-channel reconstruction on the decoded vectors at the N positions to obtain a multi-channel reconstruction sequence. The multi-channel reconstruction sequence is used as a leak-free reference trajectory for offshore oil pipelines. The multi-channel reconstruction sequence contains 6 reconstruction sequences, and the length of each reconstruction sequence is N.

[0027] Specifically, the multi-channel reconstruction formula for the decoded vectors at the N positions is as follows: ; ; in, Indicates the first Road reconstruction sequence, Represents the reconstructed sequence The N sequence values ​​in the data. Indicates the trainable first The weighted vector of the nth sequence value in the path reconstruction sequence. Indicates the trainable first The bias coefficient of the nth sequence value in the path reconstruction sequence; Furthermore, the first to sixth reconstruction sequences represent, respectively, the offshore platform inlet pressure reconstruction sequence, the onshore terminal outlet pressure reconstruction sequence, the offshore platform inlet temperature reconstruction sequence, the onshore terminal outlet temperature reconstruction sequence, the offshore platform outlet instantaneous flow reconstruction sequence, and the onshore terminal inlet instantaneous flow reconstruction sequence.

[0028] In another embodiment of the present invention, by collecting early aligned observation segments of restarting conditions of offshore oil pipelines, restarting boundary constraint vectors, and corresponding leak-free true reference trajectory labels, a training loss function is constructed with the goal of minimizing the reconstruction error between the leak-free true reference trajectory labels and the leak-free reference transient trajectory output by the leak-free reference transient reconstruction model. Based on the training loss function, the trainable parameters in the leak-free reference transient reconstruction model are optimized and trained using the gradient descent algorithm. The leak-free true reference trajectory labels are multi-source time-series detection data actually collected under the restarting conditions of offshore oil pipelines where there is no leakage, no wax layer peeling, and no abnormalities.

[0029] The S2 step, which calculates the observation residual trajectory between the early alignment observation segment and the leak-free reference trajectory, also includes: The early alignment observation segment of the restart and the leak-free reference trajectory are obtained. The difference sequence between the early alignment observation segment of the restart and the leak-free reference trajectory is calculated to form the observation residual trajectory. The observation residual trajectory contains multiple sets of residual trajectories, which are as follows: offshore platform inlet pressure residual trajectory, onshore terminal outlet pressure residual trajectory, offshore platform inlet temperature residual trajectory, onshore terminal outlet temperature residual trajectory, offshore platform outlet instantaneous flow residual trajectory, and onshore terminal inlet instantaneous flow residual trajectory.

[0030] As an embodiment of the present invention, taking the residual trajectory of the inlet pressure of an offshore platform as an example, the residual trajectory is generated as follows: the reconstructed sequence of the inlet pressure of the offshore platform and the time series segment of the inlet pressure of the offshore platform in the early aligned observation segment are obtained, the difference between the time series segment of the inlet pressure of the offshore platform and the reconstructed sequence of the inlet pressure of the offshore platform is calculated point by point, and a difference sequence of length N is obtained as the residual trajectory of the inlet pressure of the offshore platform.

[0031] S3: Extract the flow imbalance features, pressure transient features, and temperature coupling features from the observed residual trajectory, and fuse the flow imbalance features, pressure transient features, and temperature coupling features to construct a differential fingerprint vector of wax layer peeling and leakage.

[0032] Specifically, step S3 extracts the flow imbalance characteristics, pressure transient characteristics, and temperature coupling characteristics from the observed residual trajectory, including: S31: Extract the instantaneous flow residual trajectory at the offshore platform outlet and the instantaneous flow residual trajectory at the onshore terminal inlet from the observed residual trajectory; calculate the cumulative imbalance offset feature and dynamic distortion feature of the flow residual between the instantaneous flow residual trajectory at the offshore platform outlet and the instantaneous flow residual trajectory at the onshore terminal inlet; and splice the cumulative imbalance offset feature and the dynamic distortion feature of the flow residual together as the flow imbalance feature of the observed residual trajectory. Specifically, the calculation formulas for the cumulative imbalance offset characteristic and the dynamic distortion characteristic of the flow residual are as follows: ; ; ; in, This indicates the characteristics of the cumulative imbalance skewness of the flow residuals. This indicates the dynamic distortion characteristics of the flow residual. The cumulative flow residual loss measure represents the nth trajectory value in the instantaneous flow residual trajectory at the offshore platform exit and the instantaneous flow residual trajectory at the onshore terminal inlet. This represents the j-th trajectory value in the residual trajectory of the instantaneous flow rate at the offshore platform's outlet. This represents the j-th trajectory value in the instantaneous flow residual trajectory at the land terminal inlet. Indicates the selection of a set The maximum value in, Indicates the selection of a set The minimum value in; S32: Extract the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal from the observed residual trajectory, calculate the average pressure phase deviation and pressure oscillation entropy between the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal, and splice the average pressure phase deviation and pressure oscillation entropy together as the pressure transient feature of the observed residual trajectory. Specifically, the formulas for calculating the average pressure phase deviation and the pressure oscillation entropy are as follows: ; ; in, This represents the average pressure phase deviation. This represents the instantaneous phase of the offshore platform inlet pressure residual trajectory at the nth trajectory value. This represents the instantaneous phase of the residual trajectory of the onshore terminal outlet pressure at the nth trajectory value. This represents the normalized probability percentage of the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal for the nth trajectory value. This represents the nth trajectory value in the residual trajectory of the pressure at the inlet of the offshore platform. This represents the nth trajectory value of the residual trajectory of the onshore terminal's outlet pressure. This represents the k-th trajectory value in the residual trajectory of the pressure at the inlet of the offshore platform. This represents the k-th trajectory value of the residual trajectory of the onshore terminal's outlet pressure. Represents a logarithmic function with the natural constant as its base; Furthermore, by performing a Hilbert transform on the residual pressure trajectory at the inlet of the offshore platform or the residual pressure trajectory at the outlet of the onshore terminal, an analytical signal is obtained, and then the instantaneous phase of the trajectory value is extracted. S33: Extract the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory from the observed residual trajectory, and calculate the temperature coupling evolution factor between the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory using a sliding window method, which is used as the temperature coupling feature.

[0033] Specifically, a sliding window of length Len and sliding step size step is constructed to generate residual trajectory sequences of the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory within the sliding window. The local coupling degree of temperature and current residuals within the sliding window is calculated, where the local coupling degree of temperature and current residuals is the cosine similarity between the two sets of residual trajectory sequences within the sliding window. The mean of the local coupling degree of temperature and current residuals of all sliding windows is calculated as the temperature coupling evolution factor. .

[0034] Specifically, the cumulative imbalance deviation characteristic of flow residuals represents the overall cumulative offset of upstream and downstream flow residuals throughout the entire restart period. Among them, real leakage will cause long-term non-conservation of flow and a large cumulative offset span, while wax layer peeling will only produce local small fluctuations. This characteristic can quantify the overall difference between the two types of abnormal flow imbalances. The dynamic distortion feature of flow residuals characterizes the overall distortion degree and bias of flow residual imbalance over time. It can distinguish between short-term irregular disturbances and continuous unidirectional flow shifts, and enhance the separability of micro-leakage and wax layer peeling in the flow time-series evolution pattern. The mean pressure phase residual represents the synchronization of the propagation time of pressure disturbances between offshore platforms and onshore terminals and the degree of phase mismatch. In actual leakage, the pressure transient propagation delay between upstream and downstream will be inconsistent and the phase deviation will be large. The pressure disturbances caused by wax layer peeling have strong synchronization and smaller phase deviation. Pressure oscillation entropy quantifies the complexity of pressure residual temporal oscillation and the dispersion of energy distribution through information entropy. Among them, the disturbance distribution of real leakage is scattered and the entropy value is high, while the disturbance of wax layer peeling is concentrated, the change law is simple, and the entropy value is low, so as to realize the quantitative characterization of pressure transient complexity. The temperature coupling evolution factor uses a sliding window and cosine similarity to characterize the local coupling correlation of temperature residuals, and uses the global mean to characterize the overall evolution characteristics of the coupling relationship with the restart time sequence. The temperature flow coupling is stable under no abnormal operating conditions, the wax layer peeling coupling changes slowly, and the actual leakage will cause the temperature flow linkage relationship to fluctuate violently. Two types of anomalies can be distinguished from the coupling evolution dimension.

[0035] It should be noted that this invention extracts three deep-level mechanistic features—flow imbalance, pressure transients, and temperature-flow coupling—from the residual trajectories observed in six channels. It abandons traditional shallow statistics such as single mean and variance, and instead constructs innovative features such as cumulative imbalance skewness, dynamic distortion, pressure phase deviation, oscillation entropy, and temperature coupling evolution factors. This fully explores the non-conservation characteristics of flow during the restart phase of offshore oil pipelines, the propagation law of pressure transient phase, and the time-series linkage evolution mechanism of temperature-flow. The extracted features are complementary and dimensionally independent, accurately depicting the inherent differences between wax layer peeling and actual leakage in residual time-series amplitude, phase synchronization, oscillation complexity, and coupling correlation characteristics. Furthermore, this invention, relying on a sliding window and cumulative time-series construction method, effectively suppresses operating condition noise interference and improves feature robustness. This provides highly discriminative and physically clear fingerprint features for subsequent anomaly confidence scoring and accurate identification of the two types of anomalies, adapting to the anomaly identification needs of complex operating conditions during short-term shutdowns and restarts of offshore oil pipelines.

[0036] In step S3, the flow imbalance characteristics, pressure transient characteristics, and temperature coupling characteristics are spliced ​​together, and the splicing result is used as the differential fingerprint vector of wax layer peeling and leakage.

[0037] S4: Input the differential fingerprint vector of wax layer peeling and leakage into the dual-branch gating discrimination model to obtain the confidence score of local wax layer peeling non-leakage and the confidence score of real micro-leakage.

[0038] Specifically, step S4 involves inputting the differential fingerprint vector of wax layer peeling and leakage into the dual-branch gating discrimination model, including: S41: The dual-branch gated discrimination model includes a feature embedding layer, a gated attention modulation layer, a dual-branch independent discrimination layer, and a normalized score output layer; S42: The feature embedding layer receives the differential fingerprint vector of wax layer peeling and leakage, performs high-dimensional feature mapping on the differential fingerprint vector of wax layer peeling and leakage, and obtains high-dimensional hidden features; Specifically, the high-dimensional feature mapping formula is: ; in, This represents high-dimensional hidden features. This represents the differential fingerprint vector indicating wax layer peeling and leakage. This represents the activation function. This represents the trainable embedding convolution matrix in the feature embedding layer. This represents the trainable embedding bias parameters in the feature embedding layer; S43: The gated attention modulation layer converts the high-dimensional latent features into gated attention weights, and uses the gated attention weights to weight the high-dimensional latent features to obtain the gated modulation enhanced features; Specifically, the formula for generating the enhanced features after gated modulation is as follows: ; in, This indicates the enhanced features after gated modulation. Indicates the gating attention weight, Represents a trainable gated attention convolution matrix. This represents the trainable gating attention bias parameters. Represents an exponential function with the natural constant as its base; S44: The dual-branch independent discrimination layer uses two independent fully connected branches. The fully connected branches perform feature discrimination on the modulated enhanced features to generate a basic score for local wax layer peeling and non-leakage and a basic score for real micro-leakage. Specifically, the feature discrimination formula for the two independent fully connected branches is as follows: ; ; in, All represent the wax layer peeling branch parameters. All of these represent parameters with minimal leakage. This indicates a partial wax layer peeling without leakage, which is the basic score. This represents the baseline score for true micro-leakage; S45: The normalized scoring output layer normalizes the local wax layer peeling non-leakage baseline score and the real micro-leakage baseline score to generate a local wax layer peeling non-leakage confidence score and a real micro-leakage confidence score.

[0039] Specifically, the formulas for generating the confidence scores for localized wax layer peeling (non-leakage) and true micro-leakage are as follows: ; ; in, This indicates a confidence score for localized wax layer peeling that is not a leakage. This represents the confidence score for a truly minor leak.

[0040] As an embodiment of the present invention, by collecting multiple sets of wax layer peeling leakage difference fingerprint vectors corresponding to the restart conditions of offshore oil pipelines, corresponding local wax layer peeling non-leakage real labels, and real labels of real micro-leakage (where the labels adopt one-hot encoding form, [1,0] represents local wax layer peeling non-leakage samples, and [0,1] represents real micro-leakage samples), a training loss function is constructed with the goal of minimizing the cross-entropy loss between the real labels and the two types of confidence scores output by the two-branch gated discriminant model. Based on the training loss function, the gradient descent algorithm is used to iteratively optimize and train all trainable parameters in the shared feature embedding layer, the gated attention modulation layer, and the two-branch independent discriminant layer of the two-branch gated discriminant model.

[0041] Furthermore, the dual-branch gated discriminant model is trained using cross-entropy loss and optimized using gradient descent, which can converge quickly and has strong generalization ability.

[0042] S5: Based on the confidence score of the local wax layer peeling non-leakage and the confidence score of the real micro-leakage, anomaly determination is made during the restart phase of the offshore oil pipeline, and operation and maintenance control instructions for leakage alarm suppression or leakage alarm interlock are generated.

[0043] In step S5, based on the confidence score for the non-leakage of the localized wax layer peeling and the confidence score for the actual minor leak, anomaly determination is made during the restart phase of the offshore oil pipeline, including: The confidence scores for the localized wax layer peeling non-leakage and the confidence scores for the actual minor leakage are obtained. A dual threshold judgment rule is used to determine the anomalies in the restart phase of the offshore oil pipeline, and the anomaly types in the restart phase of the offshore oil pipeline are obtained. The anomaly types include localized wax layer peeling non-leakage anomalies and actual minor leakage anomalies. Specifically, by setting a first confidence threshold and the second confidence threshold ,in Greater than (Default settings) It is 0.6. (0.4), if the confidence score for localized wax layer peeling and non-leakage is greater than or equal to the first confidence threshold. Furthermore, the confidence score for the true minute leakage is less than the second confidence threshold. If the anomaly is determined to be localized wax layer peeling rather than leakage, then the anomaly is determined to be a localized wax layer peeling rather than a leakage anomaly. If the confidence score for a true minor leakage is greater than or equal to the first confidence threshold... Furthermore, the confidence score for localized wax layer peeling and non-leakage is less than the second confidence threshold. If the anomaly assessment result is a true minor leak anomaly, and both the confidence score for localized wax layer peeling (non-leakage) and the confidence score for true minor leak are within a certain range... If no operation and maintenance control instructions are generated, an early warning prompt instruction will be generated, and changes in multi-source time-series detection data and pipeline control data will be continuously monitored. For localized wax layer peeling that is not a leakage anomaly, a leakage alarm suppression command is generated to avoid false alarms interfering with normal operation and maintenance; for real minor leakage anomalies, a leakage alarm interlock command is generated.

[0044] Furthermore, the operation and maintenance control commands are directly issued to the pipeline SCADA system, which then synchronously links with the DCS field control station to execute corresponding operations. Among them, the leakage alarm suppression command is used to shield false alarm signals caused by wax layer peeling, ensuring operation and maintenance efficiency; the leakage alarm interlock command is used to quickly respond to minor leakage hazards, reduce the risk of leakage expansion, and realize precise and automated control of abnormal operation and maintenance during pipeline restart, taking into account both operation and maintenance safety and economy, and adapting to the operation and maintenance needs of complex restart conditions of offshore oil pipelines.

[0045] Example 2: This invention collects 200 sets of real-world operational data on offshore oil pipeline shutdowns and restarts (covering three scenarios: normal restart, localized wax layer peeling, and minor leaks, including samples with varying shutdown durations, pump and valve speeds, and pipeline structural configurations). It compares the performance of the AI-based offshore oil pipeline operation and maintenance control method described in this invention with traditional threshold discrimination methods in four core indicators: leak identification accuracy, wax layer peeling false alarm rate, anomaly detection delay, and operation and maintenance command response accuracy. (Refer to...) Figure 3 The experimental comparison diagram shows that, compared with traditional methods, this invention, through leak-free reference trajectory reconstruction, residual mechanism feature extraction, dual-branch gating discrimination, and intelligent command generation, achieves a 28.6% improvement in leak identification accuracy, a 41.3% reduction in false alarm rate for wax layer peeling, a 67.8% reduction in average anomaly discrimination delay, and a 35.2% improvement in the accuracy of operation and maintenance command response. Experimental results demonstrate that this invention can accurately distinguish between wax layer peeling and minor leaks, effectively suppress false alarms, and achieve rapid interlocking response. It possesses significant technical advantages and engineering practical value in the highly transient, strongly coupled, and low signal-to-noise ratio scenario of restarting and maintaining offshore pipelines.

[0046] It should be noted that the terms "comprising," "including," or any other variations thereof used herein are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0047] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0048] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for operation and maintenance control of offshore oil pipelines based on artificial intelligence, characterized in that, The method includes: S1: Collect multi-source time-series detection data and pipeline control data during the restart phase of offshore oil pipelines, perform time-scale alignment and cleaning and correction processing on the multi-source time-series detection data and pipeline control data, and construct the early-stage aligned observation segment of restart and the restart boundary constraint vector. S2: Based on the early alignment observation segment of the restart and the restart boundary constraint vector, generate the leak-free reference trajectory of the offshore oil pipeline using the leak-free reference transient reconstruction model, and calculate the observation residual trajectory between the early alignment observation segment of the restart and the leak-free reference trajectory. S3: Extract the flow imbalance features, pressure transient features, and temperature coupling features from the observed residual trajectory, and fuse the flow imbalance features, pressure transient features, and temperature coupling features to construct a wax layer peeling and leakage differential fingerprint vector; S4: Input the differential fingerprint vector of wax layer peeling and leakage into the dual-branch gated discrimination model to obtain the confidence score of local wax layer peeling non-leakage and the confidence score of real micro-leakage; S5: Based on the confidence score of the local wax layer peeling non-leakage and the confidence score of the real micro-leakage, anomaly determination is made for the restart phase of the offshore oil pipeline, and operation and maintenance control instructions for leakage alarm suppression or leakage alarm interlock are generated. In step S1, the multi-source time-series detection data and pipeline control data are time-scaled and cleaned / corrected to construct the early-aligned observation segment of restart and the restart boundary constraint vector, including: S11: Resample the time-series data in the multi-source time-series detection data and pipeline control data according to a fixed sampling interval to obtain a time series where the time interval between adjacent sampling results is equal to the fixed sampling interval; Specifically, the time-series data in the multi-source time-series detection data and pipeline control data includes the time series of offshore platform inlet pressure, onshore terminal outlet pressure, offshore platform inlet temperature, onshore terminal outlet temperature, offshore platform outlet instantaneous flow rate, onshore terminal inlet instantaneous flow rate, delivery pump speed, inlet regulating valve opening time series, and outlet regulating valve opening time series. S12: The resampled time series is judged for out-of-bounds outliers, the sampling results judged as out-of-bounds outliers are removed, and the removed sampling results are completed by linear interpolation. S13: Extract the maximum value, minimum value, and mean value of the resampled delivery pump speed time series, inlet regulating valve opening time series, and outlet regulating valve opening time series from the pipeline control data. Normalize the extracted results with the shutdown duration, offshore oil pipeline length configuration value, offshore oil pipeline inner diameter configuration value, and offshore oil pipeline riser elevation configuration value from the pipeline control data, and then splice them together to form the restart boundary constraint vector. S14: Based on the shutdown start time in the pipeline control data, filter the stable transportation multi-source time series detection data before the shutdown start time, calculate the data mean of the stable transportation multi-source time series detection data, and use the data mean as a benchmark to perform benchmark correction on the multi-source time series detection data of the offshore oil pipeline restart phase to obtain the early restart aligned observation segment.

2. The method for operation and maintenance control of offshore oil pipelines based on artificial intelligence as described in claim 1, characterized in that, Step S2 involves generating a leak-free reference trajectory for the offshore oil pipeline using a leak-free reference transient reconstruction model, including: S21: The leak-free reference transient reconstruction model includes a timing coding layer, a constraint mapping layer, a conditional modulation decoding layer, and a multi-channel reconstruction output layer; S22: The temporal coding layer receives the early-aligned restart observation segment, performs a three-layer dilated convolution operation on the early-aligned restart observation segment, and performs global attention convergence on the result of the third dilated convolution operation to obtain the transient state vector corresponding to the early-aligned restart observation segment. S23: The constraint mapping layer receives the restart boundary constraint vector and maps the restart boundary constraint vector into constraint scaling and constraint offset; the mapping formulas for the constraint scaling and constraint offset are as follows: ； ； in, Indicates the constraint scaling amount. Indicates the constraint offset. Represents the hyperbolic tangent function. This represents the parameters of the first trainable convolutional matrix in the constraint mapping layer. This represents the parameters of the trainable second convolutional matrix in the constraint mapping layer. This represents the first trainable bias parameter in the constraint mapping layer. This represents the trainable second bias parameter in the constraint mapping layer. Indicates restarting the boundary constraint vector; S24: The conditional modulation decoding layer combines the constraint scaling amount and constraint offset amount to modulate and decode the transient state vector, generating a decoded vector at N positions, where N represents the length of the restarted early aligned observation segment; S25: The multi-channel reconstruction output layer performs multi-channel reconstruction on the decoded vectors at the N positions to obtain a multi-channel reconstruction sequence. The multi-channel reconstruction sequence is used as a leak-free reference trajectory for offshore oil pipelines. The multi-channel reconstruction sequence contains 6 reconstruction sequences, and the length of each reconstruction sequence is N.

3. The method for operation and maintenance control of offshore oil pipelines based on artificial intelligence as described in claim 2, characterized in that, The observation residual trajectory calculated in step S2 between the early aligned observation segment of the restart and the leak-free reference trajectory includes: The early alignment observation segment of the restart and the leak-free reference trajectory are obtained. The difference sequence between the early alignment observation segment of the restart and the leak-free reference trajectory is calculated to form the observation residual trajectory. The observation residual trajectory contains multiple sets of residual trajectories, which are as follows: offshore platform inlet pressure residual trajectory, onshore terminal outlet pressure residual trajectory, offshore platform inlet temperature residual trajectory, onshore terminal outlet temperature residual trajectory, offshore platform outlet instantaneous flow residual trajectory, and onshore terminal inlet instantaneous flow residual trajectory.

4. The method for operation and maintenance control of offshore oil pipelines based on artificial intelligence as described in claim 3, characterized in that, Step S3 extracts the flow imbalance features, pressure transient features, and temperature coupling features from the observed residual trajectory, including: S31: Extract the instantaneous flow residual trajectory at the offshore platform outlet and the instantaneous flow residual trajectory at the onshore terminal inlet from the observed residual trajectory; calculate the cumulative imbalance offset feature and dynamic distortion feature of the flow residual between the instantaneous flow residual trajectory at the offshore platform outlet and the instantaneous flow residual trajectory at the onshore terminal inlet; and splice the cumulative imbalance offset feature and the dynamic distortion feature of the flow residual together as the flow imbalance feature of the observed residual trajectory. S32: Extract the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal from the observed residual trajectory, calculate the average pressure phase deviation and pressure oscillation entropy between the pressure residual trajectory at the inlet of the offshore platform and the pressure residual trajectory at the outlet of the onshore terminal, and splice the average pressure phase deviation and pressure oscillation entropy together as the pressure transient feature of the observed residual trajectory. S33: Extract the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory from the observed residual trajectory, and calculate the temperature coupling evolution factor between the offshore platform inlet temperature residual trajectory and the onshore terminal outlet temperature residual trajectory using a sliding window method, which serves as the temperature coupling feature.

5. The method for operation and maintenance control of offshore oil pipelines based on artificial intelligence as described in claim 4, characterized in that, In step S3, the flow imbalance characteristics, pressure transient characteristics, and temperature coupling characteristics are spliced ​​together, and the splicing result is used as the differential fingerprint vector of wax layer peeling and leakage.

6. The method for operation and maintenance control of offshore oil pipelines based on artificial intelligence as described in claim 1, characterized in that, In step S4, the differential fingerprint vector of wax layer peeling and leakage is input into the dual-branch gating discrimination model, including: S41: The dual-branch gated discrimination model includes a feature embedding layer, a gated attention modulation layer, a dual-branch independent discrimination layer, and a normalized score output layer; S42: The feature embedding layer receives the differential fingerprint vector of wax layer peeling and leakage, performs high-dimensional feature mapping on the differential fingerprint vector of wax layer peeling and leakage, and obtains high-dimensional hidden features; S43: The gated attention modulation layer converts the high-dimensional latent features into gated attention weights, and uses the gated attention weights to weight the high-dimensional latent features to obtain the gated modulation enhanced features; S44: The dual-branch independent discrimination layer uses two independent fully connected branches. The fully connected branches perform feature discrimination on the gated modulation enhanced features to generate a basic score for local wax layer peeling and non-leakage and a basic score for real micro-leakage. S45: The normalized scoring output layer normalizes the local wax layer peeling non-leakage baseline score and the real micro-leakage baseline score to generate a local wax layer peeling non-leakage confidence score and a real micro-leakage confidence score.

7. The method for operation and maintenance control of offshore oil pipelines based on artificial intelligence as described in claim 1, characterized in that, In step S5, based on the confidence score for the non-leakage of the localized wax layer peeling and the confidence score for the actual minor leak, anomaly determination is made during the restart phase of the offshore oil pipeline, including: The confidence scores for the localized wax layer peeling non-leakage and the confidence scores for the actual minor leakage are obtained. A dual threshold judgment rule is used to determine the anomalies in the restart phase of the offshore oil pipeline, and the anomaly types in the restart phase of the offshore oil pipeline are obtained. The anomaly types include localized wax layer peeling non-leakage anomalies and actual minor leakage anomalies. For localized wax layer peeling that is not a leakage anomaly, a leakage alarm suppression command is generated; for real minor leakage anomalies, a leakage alarm interlock command is generated.

Images