A buried liquefied natural gas pipeline stress prediction method, system and device

Abstract

This invention discloses a method, system, and equipment for stress prediction of buried liquefied natural gas (LNG) pipelines, belonging to the field of pipeline structural health monitoring technology. The method first acquires pipeline time-series data and target stress sequences; delineates historical time windows based on historical training moments, extracts historical temperature data, and sorts it according to lag order; generates normalized weights based on the power-law decay relationship of fractional integral operators, and weighted fuses historical temperatures to obtain thermal memory features; embeds the thermal memory feature sequence into a time-series prediction model to train a stress prediction model; finally, outputs the stress prediction result based on the thermal memory features of the moment to be predicted. This invention characterizes the soil thermal hysteresis effect through fractional thermal memory features, while simultaneously supporting independent prediction of tensile and compressive stresses, significantly improving prediction accuracy and physical interpretability, and can be used for structural health monitoring and safety early warning of buried LNG pipelines.

Description

Technical Field

[0001] This invention relates to the field of pipeline structural health monitoring technology, and specifically discloses a method, system and equipment for predicting stress in buried liquefied natural gas pipelines. Background Technology

[0002] Buried liquefied natural gas (LNG) pipelines operate under extremely low temperatures (approximately -162°C) and high pressure conditions. The pipeline walls are subjected to a combination of axial thermal stress, circumferential stress caused by internal pressure, and bending stress caused by ground settlement. Abnormal changes in the pipeline stress state are often precursors to catastrophic failure. Therefore, continuous prediction of pipeline stress is a core requirement of structural health monitoring (SHM).

[0003] Existing pipeline stress prediction methods are mainly time series prediction methods: SARIMA, LSTM and other models are widely used to capture the periodic fluctuations of energy systems; at the same time, the Prophet model based on the additive decomposition framework supports exogenous regression inputs, performs robustly on complex time series data, and has been gradually applied to engineering time series prediction scenarios in recent years.

[0004] The aforementioned prior art suffers from the following fundamental defects:

[0005] Ignoring the soil thermal hysteresis effect: Existing methods all assume that ambient temperature has an immediate effect on pipeline stress, that is, the temperature of a day determines the stress of that day. However, the thermal diffusivity of soil is extremely low, and the heat exchange process between LNG pipelines and the surrounding soil is very slow. The current stress state is not only determined by the temperature of the day, but also by the cumulative heat history of the past few days or even weeks. This "underground thermal memory effect" is completely ignored by existing methods, resulting in systematic errors in prediction accuracy.

[0006] Limitations of integer hysteresis models: If historical temperature information is to be introduced, the conventional approach is to use integer hysteresis features, which requires assigning independent coefficients to each hysteresis term. This results in high-dimensional sparsity problems and a serious risk of overfitting, and it cannot reflect the inherent power-law decay characteristics in the heat conduction process.

[0007] The shortcomings of mixed tensile and compressive stress prediction: Existing methods treat pipeline stress as a single prediction target. However, tensile stress in buried pipelines mainly leads to crack initiation and propagation, while compressive stress mainly leads to local buckling. The two have different failure modes and significantly different response characteristics to thermal memory. Unified prediction masks the mechanical differences between the two types of stress, which is not conducive to targeted safety management. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention provides a method, system, and equipment for predicting stress in buried liquefied natural gas pipelines. It characterizes the soil thermal hysteresis effect through fractional-order thermal memory features and supports independent prediction of tensile and compressive stresses, thereby significantly improving prediction accuracy and physical interpretability.

[0009] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting stress in buried liquefied natural gas pipelines, the method comprising: Obtain time-series data and target stress sequences corresponding to buried liquefied natural gas pipelines; A historical time window is defined based on the historical training time. Historical temperature data located within the historical time window is extracted from the time series data and sorted according to the lag order of the historical temperature data relative to the historical training time. Based on the power-law decay relationship corresponding to the fractional integral operator, normalized weights are generated according to the lag order; the normalized weights are used to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time. A thermal memory feature sequence consisting of thermal memory features corresponding to multiple historical training moments is obtained. The thermal memory feature sequence is input into a time series prediction model, and the time series prediction model is trained in combination with the target stress sequence to obtain a stress prediction model. Using the time to be predicted as the benchmark, thermal memory features corresponding to the time to be predicted are constructed based on time series data prior to the time to be predicted, and the thermal memory features corresponding to the time to be predicted are input into the stress prediction model to output the stress prediction results of the buried liquefied natural gas pipeline.

[0010] Optionally, the step of defining a historical time window based on a historical training time, extracting historical temperature data within the historical time window from the time-series data, and sorting the historical temperature data according to their lag order relative to the historical training time includes: Based on the aforementioned historical training time, a historical time window for constructing hot memory features is determined; Extract the temperature values ​​located within the historical time window from the time series data; The lag order corresponding to each temperature value is determined according to the temporal relationship between each temperature value and the historical training time.

[0011] Optionally, generating normalized weights according to the lag order based on the power-law decay relationship corresponding to the fractional integral operator includes: Based on the hysteresis order corresponding to each temperature value and the preset attenuation parameter, calculate the power-law attenuation value corresponding to each hysteresis order; The normalized reference value is obtained by summing the power-law decay values ​​described above. The normalized weight corresponding to each lag order is obtained by comparing the power-law decay value corresponding to each lag order with the normalized benchmark value.

[0012] Optionally, the fractional integral operator is the Riemann-Liouville fractional integral operator.

[0013] Optionally, the step of using the normalized weights to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time includes: Multiply the temperature value corresponding to each lag order by the normalized weight corresponding to that lag order to obtain the weighted temperature value; The weighted temperature values ​​are summed to obtain the thermal memory features corresponding to the historical training time. The weighted fusion is performed on each historical training moment in chronological order to obtain the hot memory feature sequence.

[0014] Optionally, the step of obtaining a thermal memory feature sequence composed of thermal memory features corresponding to multiple historical training times, inputting the thermal memory feature sequence into a time series prediction model, and training the time series prediction model in conjunction with the target stress sequence to obtain a stress prediction model includes: Time information is input into the time series forecasting model to establish trend and seasonal terms; The hot memory feature sequence is input into the time series prediction model to establish a hot memory exogenous regression term; The current temperature characteristics and the temperature nonlinear characteristics transformed from the current temperature characteristics are input into the time series prediction model to establish an exogenous temperature regression term; The target stress sequence is fitted based on the trend term, the seasonal term, the thermal memory exogenous regression term, and the temperature exogenous regression term to obtain the stress prediction model.

[0015] Optionally, before obtaining the stress prediction model, the method further includes: Set the candidate decay parameter set and the candidate historical time window set; Multiple candidate parameter combinations are formed based on the candidate attenuation parameter set and the candidate historical time window set; Thermal memory features are constructed according to each of the candidate parameter combinations, and corresponding candidate stress prediction models are trained. The candidate stress prediction models were validated in chronological order. The target parameter combination for constructing the thermal memory feature is determined from the plurality of candidate parameter combinations based on the verification error.

[0016] Optionally, the step of constructing thermal memory features corresponding to the time to be predicted based on time-series data prior to the time to be predicted, using the time to be predicted as a reference, and inputting the thermal memory features corresponding to the time to be predicted into the stress prediction model to output the stress prediction results of the buried liquefied natural gas pipeline includes: Obtain time series data for the period to be predicted and its preceding periods; Using the attenuation parameter and historical time window in the target parameter combination, normalized weights are generated for each time point to be predicted within the time period to be predicted; The historical temperature data before the corresponding time to be predicted are weighted and fused using the normalized weights to obtain the thermal memory features corresponding to each time to be predicted within the time period to be predicted. The thermal memory features corresponding to each predicted time within the predicted time period are input into the stress prediction model, and the stress prediction sequence within the predicted time period is output. The stress prediction result is obtained based on the stress prediction sequence within the predicted time period.

[0017] Secondly, the present invention provides a stress prediction system for buried liquefied natural gas pipelines, comprising: The acquisition module is used to acquire time-series data and target stress sequences corresponding to buried liquefied natural gas pipelines; The data extraction module is used to define a historical time window based on the historical training time, extract historical temperature data located within the historical time window from the time series data, and sort the historical temperature data according to the lag order relative to the historical training time. The processing module is used to generate normalized weights according to the lag order based on the power-law decay relationship corresponding to the fractional integral operator; and to use the normalized weights to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time. The model training module is used to obtain a thermal memory feature sequence composed of thermal memory features corresponding to multiple historical training moments, input the thermal memory feature sequence into the time series prediction model, and train the time series prediction model in combination with the target stress sequence to obtain a stress prediction model. The stress prediction module is used to construct the thermal memory features corresponding to the time to be predicted based on the time series data before the time to be predicted, and input the thermal memory features corresponding to the time to be predicted into the stress prediction model to output the stress prediction results of the buried liquefied natural gas pipeline.

[0018] Thirdly, the present invention provides an electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method described in any one of the first aspects.

[0019] Compared with the closest existing technology, the present invention has the following advantages: This invention proposes a method, system, and equipment for predicting stress in buried liquefied natural gas pipelines. Based on the power-law decay relationship corresponding to the fractional integral operator, normalized weights are constructed. Historical temperature data is weighted and fused to obtain thermal memory features. The thermal memory effect caused by the low thermal diffusivity of the soil around the buried pipeline is introduced into the prediction process in a mathematically rigorous and physically matched manner. This invention breaks through the unreasonable assumption of existing technologies that only use temperature as an immediate input variable. It can accurately characterize the continuous impact of historical temperature accumulation on the current pipeline stress and fundamentally correct the systematic prediction bias of existing methods.

[0020] This invention controls the weight decay pattern throughout the entire historical time window using a single decay parameter, compressing temperature data from multiple lag times into a one-dimensional thermal memory feature without requiring independent coefficient fitting for each lag term. Compared to traditional integer-order lag feature schemes, this method significantly reduces the dimensionality of the parameter space, effectively solving the sparsity problem and overfitting risk caused by high-dimensional lag features while fully preserving historical thermal input information, thus improving the model's generalization ability.

[0021] This invention uses thermal memory feature sequences with clear physical meaning as exogenous input into a time series prediction model. This allows the model to retain its trend fitting and periodic fluctuation capture capabilities while adding the ability to characterize the dynamic processes of thermal history. This approach leverages the adaptability of time series prediction models to engineering time series data, and enhances the rationality and interpretability of the prediction results through physical constraints. The core parameters have clear meanings, facilitating understanding and optimization by engineers.

[0022] The method of this invention has clear steps and lightweight computational logic. It does not require complex deep learning architecture and large-scale hyperparameter tuning. It can directly perform training and prediction based on time-series monitoring data collected on-site in pipelines. It has low computational overhead and can be embedded into existing structural health monitoring systems to achieve continuous rolling prediction of pipeline stress, providing stable and reliable technical support for the safety status assessment of buried liquefied natural gas pipelines. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0024] Figure 1 This is a flowchart of a method for predicting stress in buried liquefied natural gas pipelines provided by the present invention; Figure 2(a) is a sensitivity analysis diagram of the attenuation parameters of compressive stress and the historical time window provided in the embodiment of the present invention; Figure 2(b) is a sensitivity analysis diagram of the attenuation parameter of tensile stress and the historical time window provided in the embodiment of the present invention.

[0025] Figure 3(a) is a comparison chart of the historical trend and 90-day prediction of compressive stress provided in the embodiment of the present invention; Figure 3(b) is an enlarged comparison diagram of the 90-day prediction period of compressive stress provided in the embodiment of the present invention; Figure 4(a) is a comparison chart of the historical trend and 90-day prediction of tensile stress provided by the embodiment of the present invention; Figure 4(b) is an enlarged comparison diagram of the 90-day prediction period of tensile stress provided in the embodiment of the present invention; Figure 5 This is a schematic diagram of stress superposition in the cross section of a buried liquefied natural gas pipeline provided in an embodiment of the present invention; wherein, (a) is a uniform axial thermal stress distribution, (b) is a bending stress gradient distribution, and (c) is a superimposed tensile and compressive extreme stress distribution; Figure 6 A structural block diagram of a buried liquefied natural gas pipeline stress prediction system provided in an embodiment of the present invention; Figure 7 This is a diagram of the internal structure of the electronic device provided by the present invention. Detailed Implementation

[0026] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore merely examples, and should not be construed as limiting the scope of protection of the present invention.

[0027] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0028] This invention provides a method, system, and equipment for predicting the stress of buried liquefied natural gas pipelines; it is particularly suitable for predicting the stress state of long-distance buried LNG pipelines considering the soil thermal memory effect. The embodiments of this invention are described below with reference to the accompanying drawings.

[0029] Example 1: As Figure 1 As shown, this invention provides a method for predicting stress in buried liquefied natural gas pipelines. This method specifically includes the following steps: S101 acquires the time-series data and target stress sequence corresponding to the buried liquefied natural gas pipeline; S102 defines a historical time window based on the historical training time, extracts historical temperature data within the historical time window from the time series data, and sorts the historical temperature data according to the lag order relative to the historical training time. S103 generates normalized weights based on the power-law decay relationship corresponding to the fractional integral operator and according to the lag order; the normalized weights are used to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time. S104 Obtains a thermal memory feature sequence composed of thermal memory features corresponding to multiple historical training times, inputs the thermal memory feature sequence into the time series prediction model, and trains the time series prediction model in combination with the target stress sequence to obtain a stress prediction model. S105 uses the time to be predicted as a reference, constructs the thermal memory feature corresponding to the time to be predicted based on the time series data before the time to be predicted, inputs the thermal memory feature corresponding to the time to be predicted into the stress prediction model, and outputs the stress prediction result of the buried liquefied natural gas pipeline.

[0030] In step S101, the time series data includes temperature time series data, as well as relevant parameters affecting the stress state of the pipeline, such as internal pressure data and settlement monitoring data.

[0031] Preferably, high-frequency raw data is collected from sensors installed on buried LNG pipelines and aggregated into daily average time series. Missing values ​​due to sensor malfunction or transmission delays are filled using time-weighted linear interpolation. ; in, , For missing moments Effective observations on both sides , Its corresponding timestamp.

[0032] In step S101, the target stress sequence corresponding to the buried liquefied natural gas pipeline is obtained, and the target stress sequence includes a tensile stress sequence and / or a compressive stress sequence.

[0033] Furthermore, based on the stress-related data and pipeline mechanical relationships obtained for the buried liquefied natural gas pipeline, the target stress sequence is determined as follows: Calculate the axial stress based on the pipe internal pressure, pipe material parameters, and pipe temperature; Calculate the bending stress based on the pipe bending moment and pipe cross-sectional parameters; The target stress sequence is determined based on the axial stress and the bending stress.

[0034] The step of determining the target stress sequence based on the axial stress and the bending stress includes: The axial stress and the bending stress are superimposed to obtain a tensile stress sequence; Subtracting the axial stress from the bending stress yields the compressive stress sequence; The target stress sequence is defined as at least one of the tensile stress sequence and the compressive stress sequence.

[0035] In the above embodiments, the pipe bending moment in the bending stress can be obtained by any of the following methods: (1) Direct monitoring back calculation: An inclination sensor or strain rose sensor is arranged along the pipe axis, and the section bending moment is calculated based on the correspondence between the bending strain and the bending moment of the beam; (2) Settlement data calculation: Based on the foundation settlement monitoring data, the pipe section is equivalent to an elastic foundation beam, and the axial bending moment of each section is calculated using the material mechanics method.

[0036] Step S102 specifically includes: S21. Based on the aforementioned historical training time, determine the historical time window used to construct the hot memory feature (denoted as...). (Unit: days) S22. Extract the temperature value located within the historical time window from the time series data; S23. Determine the lag order of each temperature value according to its temporal relationship relative to the historical training time. The lag order... The first time before the corresponding historical training moment Each sampling time, It is a positive integer.

[0037] Furthermore, step S103 specifically includes: S31. Calculate the power-law decay value corresponding to each of the temperature values ​​according to the hysteresis order and the preset decay parameter. S32. Sum the power-law decay values ​​to obtain the normalized reference value; S33. Ratio the power-law decay value corresponding to each lag order with the normalized reference value to obtain the normalized weight corresponding to each lag order (denoted as). Where k=1 represents the day before the historical training time. Indicates the number of training moments prior to the previous training moment. sky).

[0038] In step S103, the fractional integral operator is the Riemann-Liouville (RL) fractional integral operator. For continuous functions... of The fractional integral of order RL is defined as: ; in, ∈(0,1) represents the fractional order. Let Gamma be the function, and τ be a historical moment. Let τ be the temperature function at a historical moment. It is the power-law weight kernel.

[0039] make For daily discrete data (sampling interval Δ =1 day) Riemann summation approximation:

[0040] make ,but That is, the weights lag behind. Decreasing in a power-law manner. Reparameterization. And omit with Irrelevant constants The weights are normalized to eliminate the truncation length. Impact:

[0041] This expression satisfies That is, the sum of all normalized weights is 1, which constitutes the history. The convex combination coefficient of temperature at each time point.

[0042] Furthermore, the phrase "generating normalized weights according to the lag order based on the power-law decay relationship corresponding to the fractional integral operator" specifically includes: according to the lag order corresponding to each temperature value. and preset attenuation parameters Calculate the power-law decay value corresponding to each of the aforementioned lag orders. Summing the power-law decay values ​​yields the normalized reference value. ; the power-law decay value corresponding to each of the aforementioned lag orders The normalized weight is obtained by performing a ratio calculation with the normalized benchmark value to obtain the normalized weight corresponding to each lag order. .

[0043] Step S103 further includes: S301. Multiply the temperature value corresponding to each lag order by the normalized weight corresponding to that lag order to obtain a weighted temperature value. S302. Sum the weighted temperature values ​​to obtain the thermal memory features corresponding to the historical training time. S303. Perform the weighted fusion on each historical training moment in chronological order to obtain the hot memory feature sequence.

[0044] Specifically, step S301 involves setting the temperature value corresponding to each of the aforementioned hysteresis sequences. Normalized weights corresponding to this lag order Multiply to obtain a weighted temperature value; sum the weighted temperature values ​​to obtain the thermal memory feature corresponding to the historical training time. : ; in, Indicates historical training moments The previous The temperature value of the day. Historical temperatures are weighted using a power-law decay method: recent temperatures have higher weights and older temperatures have lower weights. Physically, this corresponds to the fact that the more recent the thermal state in soil heat conduction, the greater its contribution to the current stress. This weighted fusion is performed on multiple historical training moments in chronological order to obtain a thermal memory feature sequence.

[0045] Furthermore, step S104 specifically includes: S41. Input the time information into the time series prediction model to establish the trend term and the seasonal term; S42. Input the hot memory feature sequence into the time series prediction model to establish a hot memory exogenous regression term; S43. Input the current temperature characteristics and the temperature nonlinear characteristics transformed from the current temperature characteristics into the time series prediction model to establish a temperature exogenous regression term; S44. Fit the target stress sequence according to the trend term, the seasonal term, the thermal memory exogenous regression term and the temperature exogenous regression term to obtain the stress prediction model.

[0046] Preferably, the time series forecasting model is the Prophet time series forecasting model. This model can be decomposed into a trend term, a seasonal term, a holiday term, and an exogenous regression term. It has strong physical interpretability and is suitable for engineering time series forecasting scenarios.

[0047] The thermal memory feature F(t) is embedded as an exogenous regression quantity of physical information into the Prophet additive framework, and the model equation is: ; The terms in the above formula have the following meanings: g(t) represents the nonlinear trend term, which captures the long-term structural evolution of pipeline stress; s(t) represents the multi-timescale seasonal term (year, week, day), which captures the periodic fluctuations driven by temperature. Indicates the effect of holidays / special events; Indicates hot memory terms, This is the system's sensitivity coefficient to accumulated historical temperatures; Indicates other exogenous regression quantities; This represents the residual error term.

[0048] Optionally, the other exogenous regressors include a linear temperature term. Sum of squares temperature term It is used to capture nonlinear thermal responses.

[0049] Furthermore, before obtaining the stress prediction model in step S105, a hyperparameter optimization step is also included: S501, Set the candidate attenuation parameter set and the candidate historical time window set; S502. Multiple candidate parameter combinations are formed based on the candidate attenuation parameter set and the candidate historical time window set; S503. Construct thermal memory features according to each of the candidate parameter combinations, and train the corresponding candidate stress prediction models. S504. Verify each of the candidate stress prediction models in chronological order; S505. Determine the target parameter combination for constructing the thermal memory feature from the plurality of candidate parameter combinations based on the verification error.

[0050] Step S504 above, which verifies each of the candidate stress prediction models in chronological order, includes: The training and validation intervals are divided according to their chronological order. The candidate stress prediction model is trained using the training interval; The prediction error of the candidate stress prediction model is calculated using the verification interval; The training interval and the validation interval are moved along the time direction, and the training and error calculation process is repeated; The verification error corresponding to each candidate parameter combination is determined based on the results of multiple error calculations.

[0051] In this embodiment, the specific parameters for rolling window cross-validation are set as follows: the training window length is 730 days (2 years), the validation window length is 90 days, and the window sliding step size is 30 days; the root mean square error (RMSE) of the validation set is used as the evaluation index, and the parameter combination with the smallest average RMSE of all rolling windows is taken as the target parameter combination.

[0052] In this embodiment, the optimal attenuation parameter corresponding to the compressive stress is obtained based on measured data from an in-service buried LNG pipeline in Guangdong. =0.2, Optimal historical time window =60 days, exhibiting slow decay and long memory characteristics, corresponding to the dominant role of long-term soil heat accumulation on compressive stress; the optimal decay parameter corresponding to tensile stress. =0.8, optimal historical time window =120 days, exhibiting rapid decay characteristics, corresponding to the high sensitivity of tensile stress to recent thermal shock. The sensitivity analysis results of decay parameters and historical time windows on the root mean square error of prediction are shown in Figure 2(a) and Figure 2(b), where Figure 2(a) corresponds to compressive stress and Figure 2(b) corresponds to tensile stress.

[0053] Furthermore, step S105 specifically includes: S51. Obtain the time series data of the period to be predicted and its preceding periods; S52. Using the attenuation parameter and historical time window in the target parameter combination, generate the normalized weights corresponding to each time point to be predicted within the time period to be predicted; S53. Using the normalized weights, the historical temperature data before the corresponding time to be predicted are weighted and fused to obtain the thermal memory features corresponding to each time to be predicted within the time period to be predicted. S54. Input the thermal memory features corresponding to each predicted time within the predicted time period into the stress prediction model, output the stress prediction sequence within the predicted time period, and obtain the stress prediction result based on the stress prediction sequence within the predicted time period.

[0054] Specifically, the time series data of the period to be predicted and its preceding time are obtained; using the attenuation parameter α and the historical time window K in the target parameter combination, normalized weights are generated for each predicted moment within the period to be predicted; the normalized weights are used to perform weighted fusion of the historical temperature data before the corresponding predicted moment to obtain the thermal memory features corresponding to each predicted moment within the period to be predicted; the thermal memory features corresponding to each predicted moment within the period to be predicted are input into the stress prediction model to output the stress prediction sequence within the period to be predicted.

[0055] Examples of the prediction results are shown in Figures 3(a) and 3(b) (compressive stress) and Figures 4(a) and 4(b) (tensile stress). The prediction results of the method of the present invention are significantly better than those of the measured values ​​than those of the comparative model.

[0056] In one embodiment, stress component calculation and bi-objective decomposition are performed: The total axial stress of a buried LNG pipeline consists of the superposition of axial stress and bending stress. For example... Figure 5 Based on the superposition principle, the tensile stress on the convex side and the compressive stress on the concave side are treated as independent prediction targets:

[0057]

[0058] For tensile stress and compressive stress The stress prediction models described above are trained independently, with hyperparameters... Each component is optimized independently, outputting an independent prediction sequence. Therefore, independent stress prediction is performed for both tensile and compressive failure modes.

[0059] Example 2: To verify the effectiveness of Example 1 of the present invention, Example 2 uses the measured daily average data (approximately 1160 days) of an in-service buried LNG pipeline in Guangdong Province from December 2020 to February 2024 for verification. A 90-day rolling window cross-validation was used, with RMSE, MAE, and MAPE as evaluation indicators. The rolling window was set as follows: training window length 730 days, validation window length 90 days, and sliding step size 30 days. The average evaluation indicator of all rolling windows was taken as the final result. The results are compared with three existing methods: Prophet baseline, XGBoost, and LSTM, as shown in the table below:

[0060] The results show that the method of the present invention achieves the best accuracy in predicting both compressive and tensile stresses, and the accuracy is significantly improved compared with the existing methods, thus verifying the effectiveness of the present invention.

[0061] Example 3: Based on the same technical concept, Example 3 of this invention also provides a stress prediction system for buried liquefied natural gas pipelines, which implements a stress prediction method for buried liquefied natural gas pipelines, such as... Figure 6 As shown, it includes: an acquisition module 210, a data extraction module 220, a processing module 230, a model training module 240, and a stress prediction module 250; Among them, the acquisition module 210 is used to acquire the time series data and target stress sequence corresponding to the buried liquefied natural gas pipeline; The data extraction module 220 is used to define a historical time window based on the historical training time, extract historical temperature data located within the historical time window from the time series data, and sort the historical temperature data according to the lag order of the historical temperature data relative to the historical training time. Processing module 230 is used to generate normalized weights according to the lag order based on the power-law decay relationship corresponding to the fractional integral operator; and to use the normalized weights to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time. The model training module 240 is used to obtain a thermal memory feature sequence composed of thermal memory features corresponding to multiple historical training times, input the thermal memory feature sequence into the time series prediction model, and train the time series prediction model in combination with the target stress sequence to obtain a stress prediction model. The stress prediction module 250 is used to construct the thermal memory features corresponding to the time to be predicted based on the time series data before the time to be predicted, and input the thermal memory features corresponding to the time to be predicted into the stress prediction model, and output the stress prediction results of the buried liquefied natural gas pipeline.

[0062] Example 4: In one embodiment, Example 4 of the present invention also provides an electronic device; the electronic device may be a terminal, and its internal structure diagram may be as follows. Figure 7 As shown. The electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the stress prediction method for buried liquefied natural gas pipelines as described in any one of steps S101 to S105. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0063] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0064] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0065] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for predicting stress in buried liquefied natural gas pipelines, characterized in that, The method includes: Obtain time-series data and target stress sequences corresponding to buried liquefied natural gas pipelines; A historical time window is defined based on the historical training time. Historical temperature data located within the historical time window is extracted from the time series data and sorted according to the lag order of the historical temperature data relative to the historical training time. Based on the power-law decay relationship corresponding to the fractional integral operator, normalized weights are generated according to the lag order; the normalized weights are used to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time. A thermal memory feature sequence consisting of thermal memory features corresponding to multiple historical training moments is obtained. The thermal memory feature sequence is input into a time series prediction model, and the time series prediction model is trained in combination with the target stress sequence to obtain a stress prediction model. Using the time to be predicted as a benchmark, thermal memory features corresponding to the time to be predicted are constructed based on time series data prior to the time to be predicted. These thermal memory features are then input into the stress prediction model, which outputs the stress prediction results for the buried liquefied natural gas pipeline.

2. The method according to claim 1, characterized in that, The step of defining a historical time window based on a historical training time, extracting historical temperature data within the historical time window from the time-series data, and sorting the historical temperature data according to their lag order relative to the historical training time includes: Based on the aforementioned historical training time, a historical time window for constructing hot memory features is determined; Extract the temperature values ​​located within the historical time window from the time series data; The lag order corresponding to each temperature value is determined according to the temporal relationship between each temperature value and the historical training time.

3. The method according to claim 2, characterized in that, The generation of normalized weights based on the power-law decay relationship corresponding to the fractional integral operator, according to the lag order, includes: Based on the hysteresis order corresponding to each temperature value and the preset attenuation parameter, calculate the power-law attenuation value corresponding to each hysteresis order; The normalized reference value is obtained by summing the power-law decay values ​​described above. The normalized weight corresponding to each lag order is obtained by comparing the power-law decay value corresponding to each lag order with the normalized benchmark value.

4. The method according to claim 3, characterized in that, The fractional integral operator is the Riemann-Liouville fractional integral operator.

5. The method according to claim 3, characterized in that, The step of using the normalized weights to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time includes: Multiply the temperature value corresponding to each lag order by the normalized weight corresponding to that lag order to obtain the weighted temperature value; The weighted temperature values ​​are summed to obtain the thermal memory features corresponding to the historical training time. The weighted fusion is performed on each historical training moment in chronological order to obtain the hot memory feature sequence.

6. The method according to claim 5, characterized in that, The process of obtaining a thermal memory feature sequence composed of thermal memory features corresponding to multiple historical training moments, inputting the thermal memory feature sequence into a time series prediction model, and training the time series prediction model in conjunction with the target stress sequence to obtain a stress prediction model includes: Time information is input into the time series forecasting model to establish trend and seasonal terms; The hot memory feature sequence is input into the time series prediction model to establish a hot memory exogenous regression term; The current temperature characteristics and the temperature nonlinear characteristics transformed from the current temperature characteristics are input into the time series prediction model to establish an exogenous temperature regression term; The target stress sequence is fitted based on the trend term, the seasonal term, the thermal memory exogenous regression term, and the temperature exogenous regression term to obtain the stress prediction model.

7. The method according to claim 6, characterized in that, Before obtaining the stress prediction model, the following steps are also included: Set the candidate decay parameter set and the candidate historical time window set; Multiple candidate parameter combinations are formed based on the candidate attenuation parameter set and the candidate historical time window set; Thermal memory features are constructed according to each of the candidate parameter combinations, and corresponding candidate stress prediction models are trained. The candidate stress prediction models were validated in chronological order. The target parameter combination for constructing the thermal memory feature is determined from the plurality of candidate parameter combinations based on the verification error.

8. The method according to claim 7, characterized in that, The process of constructing thermal memory features corresponding to the time to be predicted based on time-series data prior to the time to be predicted, and inputting these thermal memory features into the stress prediction model to output stress prediction results for buried liquefied natural gas pipelines includes: Obtain time series data for the period to be predicted and its preceding periods; Using the attenuation parameter and historical time window in the target parameter combination, normalized weights are generated for each time point to be predicted within the time period to be predicted; The historical temperature data before the corresponding time to be predicted are weighted and fused using the normalized weights to obtain the thermal memory features corresponding to each time to be predicted within the time period to be predicted. The thermal memory features corresponding to each predicted time within the predicted time period are input into the stress prediction model, and the stress prediction sequence within the predicted time period is output. The stress prediction result is obtained based on the stress prediction sequence within the predicted time period.

9. A stress prediction system for buried liquefied natural gas pipelines, characterized in that, include: The acquisition module is used to acquire time-series data and target stress sequences corresponding to buried liquefied natural gas pipelines; The data extraction module is used to define a historical time window based on the historical training time, extract historical temperature data located within the historical time window from the time series data, and sort the historical temperature data according to the lag order relative to the historical training time. The processing module is used to generate normalized weights according to the lag order based on the power-law decay relationship corresponding to the fractional integral operator. The normalized weights are used to perform weighted fusion on the sorted historical temperature data to obtain the thermal memory features corresponding to the historical training time. The model training module is used to obtain a thermal memory feature sequence composed of thermal memory features corresponding to multiple historical training moments, input the thermal memory feature sequence into the time series prediction model, and train the time series prediction model in combination with the target stress sequence to obtain a stress prediction model. The stress prediction module is used to construct the thermal memory features corresponding to the time to be predicted based on the time series data before the time to be predicted, and input the thermal memory features corresponding to the time to be predicted into the stress prediction model to output the stress prediction results of the buried liquefied natural gas pipeline.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-8.

Images