A multi-factor coupling corrosion prediction method for municipal water supply pipelines in a coastal area

Abstract

This invention proposes a multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas, belonging to the field of pipeline fault prediction and health management technology. This invention maps multi-source heterogeneous information such as pipeline structure, operating conditions, environment, and loads onto a spatiotemporal grid unit composed of spatial grid cells and daily statistical cycles. Using the spatiotemporal grid as the unique smallest computational unit throughout the entire process, it completes differentiated preprocessing and spatiotemporal alignment of multi-source data, quantifies the coupling weight parameters of the four core corrosion systems in coastal areas, and then constructs a two-layer corrosion prediction model that integrates mechanism constraints and data-driven approaches. Combined with a two-cycle hierarchical correction mechanism, it iterative optimization of model parameters is achieved, ultimately completing pipeline corrosion level determination, remaining life quantification calculation, and graded operation and maintenance result output. This transforms the corrosion prediction of municipal water supply pipelines in coastal areas from a decentralized, single-dimensional, segment-level coarse-grained calculation to a grid-level, full-process standardized calculation under a unified spatiotemporal benchmark.

Description

Technical Field

[0001] This invention belongs to the field of pipeline fault prediction and health management technology, and in particular relates to a multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas. Background Technology

[0002] The municipal water supply pipelines in coastal areas are in a special geological environment with high salt spray, high soil chloride ion, and acid-base imbalance. At the same time, they are affected by multiple factors such as the traffic load of the road above, the fluctuation of water pressure inside the pipeline, and the aging of the material itself. The corrosion failure rate is higher than that in ordinary inland areas. Problems such as pipe wall corrosion perforation, peeling and cracking of anti-corrosion layer, and pipeline leakage and damage are prone to occur, threatening the safety of municipal water supply and the stable operation of pipeline network.

[0003] To achieve accurate prediction of corrosion status of municipal water supply pipelines in coastal areas and ensure the long-term stable operation and maintenance of the pipeline network, targeted corrosion prediction technologies are urgently needed. However, existing pipeline corrosion prediction technologies cannot meet the actual engineering needs of refined and precise corrosion prevention and early warning under the complex working conditions in coastal areas. The specific shortcomings are as follows:

[0004] (1) Existing corrosion prediction technologies mostly use single environmental factors or single structural parameters for one-sided analysis, and do not construct a full-domain coupling model for the unique corrosion conditions in coastal areas; although some technologies include environmental or mechanical parameters, they do not realize the quantification of the synergistic coupling effect between multiple factors, and cannot simultaneously include and integrate the four core influencing factors of salt spray-chloride ion coupling, temperature and humidity-acidity coupling, water pressure fluctuation-material aging coupling, and traffic load-burial depth load transfer-rainfall coupling, resulting in incomplete coupling mechanism analysis;

[0005] (2) Existing technologies generally lack accurate matching between road network traffic flow data and pipeline space data, making it impossible to quantify the damage effects of vehicle driving vibration and pipeline internal operating loads. They also fail to consider the synergistic acceleration corrosion mechanism of chemical corrosion factors and mechanical damage factors, resulting in low model prediction accuracy, poor generalization, and a large deviation between prediction results and actual corrosion conditions on site.

[0006] (3) Most existing corrosion prediction models are static unidirectional models and do not combine real-time monitoring data to build an iterative correction mechanism. After the model parameters are solidified, they are difficult to adapt to the dynamic working conditions of seasonal climate change and gradual changes in traffic flow in coastal areas. Long-term use will lead to accuracy decay. Alternatively, the parameters can be dynamically adjusted according to the actual situation and retrained to meet the latest environmental requirements. However, the model training time is long and the cost is high, resulting in low efficiency.

[0007] Existing technologies do not perform simultaneous calculations to address the dual mechanical coupling and multiple chemical couplings inside and outside the pipeline, and lack sufficient quantification of the interactive effects of multi-source corrosion factors and the synergistic acceleration of corrosion effects. Summary of the Invention

[0008] To address the above problems, this invention proposes a multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas, comprising the following steps:

[0009] S1 uses a spatiotemporal grid as the smallest computational unit. Based on the grid-deployed sensors, it collects pipeline body attribute parameters, pipe internal operation parameters, external soil and atmospheric environmental parameters, and traffic load parameters, and performs preprocessing.

[0010] S2, based on S1 data and the electrochemical mechanism of metal corrosion and municipal engineering design parameters, performs weight calculations for salt spray-chloride ion chemical coupling, temperature and humidity-acidity coupling, water pressure fluctuation-material aging coupling, traffic load-burial depth load transfer-rainfall coupling, and obtains four types of dimensionless coupling weight parameters.

[0011] S3, the coupling weight parameters obtained in S2 are combined with the pipe material and input into the constructed global coupled corrosion mechanism model covering chemical corrosion and mechanical damage, and the dimensionless basic corrosion rate is output.

[0012] Meanwhile, the coupling weight parameters obtained from S2, combined with the cumulative number of days the pipeline has been in operation, are input into the trained machine learning prediction model to output the daily average predicted corrosion rate of a single grid.

[0013] S4 calculates the final corrosion rate of a single grid cell by weighting and fusing the dimensionless basic corrosion rate with the daily average predicted corrosion rate of the single grid cell. Combined with the preset corrosion rate threshold, the final corrosion rate is mapped to the corresponding corrosion level.

[0014] Preferably, the process also includes an iterative model calibration process based on measured data: collecting true data of pipeline corrosion at the site to construct a calibration set, and using a dual-cycle hierarchical calibration mechanism of daily small calibration and monthly large calibration, calculating the error between the model prediction value and the measured value at the site with mean square error (MSE) as the indicator, and using the error backpropagation algorithm to adaptively adjust the corresponding coupling weight parameters, model coefficients, and hyperparameters; the daily small calibration only adjusts the relevant parameters of the global coupled corrosion mechanism model to adapt to the fluctuation of daily operating conditions; the monthly large calibration optimizes the global parameters of the machine learning prediction model driver class, the model fusion weights, and the cross-system interaction coupling coefficients in the global coupled corrosion mechanism model to correct the long-term prediction trend.

[0015] Preferably, it also includes the quantification of pipeline remaining life assessment, specifically:

[0016] Based on industry standards, three types of thresholds are determined, including a corrosion rate safety threshold, used to determine whether the pipeline is in a high-risk corrosion state; a critical wall thickness, used for calculating remaining service life; and a corrosion protection layer failure judgment standard, used to determine whether the corrosion protection layer has lost its protective ability.

[0017] Convert the daily corrosion rate to the annual corrosion rate using the formula. Calculate the number of days, with leap years counted as 366; then calculate the safe operating life using the formula. Calculation, where L is the safe operating life; This represents the initial wall thickness of the pipe. Critical wall thickness; Daily corrosion rate; The annual corrosion rate is given.

[0018] Preferably, the pipeline body attribute parameters include pipeline material, initial pipeline wall thickness, and anti-corrosion layer type; the pipeline internal operation parameters include pipeline burial depth, operating water pressure fluctuation amplitude, and cumulative pipeline operation days; the pipeline external soil and atmospheric environment parameters include sensor point coordinates, soil chloride ion concentration, soil pH value, atmospheric salt spray deposition flux, and ambient temperature; and the traffic load parameters include road network-pipeline topology data and daily traffic flow of road sections.

[0019] Preferably, the spatial grid adopts the CGCS2000 national geodetic coordinate system, and the study area is divided into continuous uniform rectangular grids with a side length of 5-20m. The grid side length is adjusted according to the pipe segment density in the study area. A 5-10m grid is used in densely populated pipe segment areas, and a 10-20m grid is used in sparsely populated pipe segment areas. Each grid is assigned a unique spatial number and planar coordinate range. Based on the road network-pipeline topology data, a one-to-one spatial mapping relationship of "road section-pipeline segment-spatial grid" is established. For pipe segments that cross grids, the pipe segment attributes are assigned to the corresponding grid according to the proportion of the pipe segment centerline length within the grid.

[0020] Static attribute data are directly assigned values ​​according to spatial affiliation, while time-series dynamic data are projected onto the cell grid according to the influence range. The number of discrete measurement points is achieved by using Kriging interpolation to achieve full coverage of the study area. The arithmetic mean of the same type of data in the same grid is taken as the final value.

[0021] Preferably, in step S2, the formula for calculating the salt spray-chloride ion coupling weight is:

[0022] ;

[0023] In the formula The salt spray-chloride ion coupling weights; This refers to atmospheric salt spray deposition flux; The soil chloride ion concentration is taken as α, β are taken as regional operating condition coupling coefficients; max is the maximum value in the study area.

[0024] The formula for calculating the coupling weight of temperature, humidity and pH is:

[0025] ;

[0026] In the formula Temperature, humidity, and pH are coupled weights; T represents temperature; pH represents pH value; E represents pH value. a γ is the activation energy of the metal corrosion reaction, R is the universal gas constant, and γ is the acid-base effect index, all of which are fixed indices; max is the maximum value in the study area.

[0027] The formula for calculating the coupling weight of water pressure fluctuation and material aging is as follows:

[0028] ;

[0029] In the formula The coupling weight for water pressure fluctuation and material aging; The amplitude of water pressure fluctuation during operation; t d The cumulative number of days the pipeline has been in operation; stress coefficient The aging coefficient is a fixed initial value; max is the maximum value in the study area;

[0030] The formula for calculating the coupling weight of traffic load-burial depth load transfer-rainfall is as follows:

[0031] ;

[0032] In the formula The traffic vibration coupling weight is N; N is the daily traffic volume of the road segment. The burial depth influence coefficient is determined by the collected pipeline burial depth data. is the daily rainfall correction factor, determined by the daily cumulative rainfall R in different levels; max is the maximum value for the study area.

[0033] Preferably, the global coupled corrosion mechanism model is as follows:

[0034] ;

[0035] In the formula The basic corrosion rate is dimensionless. This is a global cross-coupling correction coefficient; The cross-system interactive coupling coefficient is used to characterize the synergistic accelerated corrosion amplification effect of chemical coupling and mechanical coupling; The basic corrosion resistance coefficient of the pipeline material is determined by the collected data on the pipeline material and is a fixed value; ductile iron is an example. =1.0, carbon steel =1.2, PE pipe =0.1, Stainless Steel =0.3; ~ These are the calculated coupling weight parameters.

[0036] Preferably, the machine learning prediction model adopts a random forest regression model, using the constructed spatiotemporal computing units as the sample basis, and the input features are the cumulative number of days the pipeline runs and the coupling weight parameters. ~ Output the daily average predicted corrosion rate of a single grid. The initial hyperparameters of the model were set as follows: 100 decision trees, a maximum depth of 10 layers, and a minimum number of leaf node samples of 5.

[0037] Preferably, the formula for calculating the final corrosion rate in S4 is:

[0038] = ;

[0039] In the formula, The final dimensionless corrosion rate for a single grid cell per day; For the weights of the global coupling mechanism model, These are the weights for the machine learning model, with initial values ​​of 0.6 and 0.4 respectively, and the sum of the weights is 1.

[0040] Based on GB / T 19285-2026 and historical experience data, The corrosion is mapped to levels I through V, where level I represents weak corrosion, level II represents relatively weak corrosion, level III represents moderate corrosion, level IV represents relatively strong corrosion, and level V represents strong corrosion. Grids of level IV or higher are marked as high-risk corrosion points, and the corrosion protection layer is simultaneously mapped and output in four states: intact, slightly damaged, moderately damaged, and failed, based on the corrosion level.

[0041] Preferably, the daily minor calibration uses a single day as the calibration cycle. Based on the daily error and measured constraint data, it only adjusts the basic parameters and coupling weights of the global coupling mechanism model, specifically performing the following operations:

[0042] Basic parameter back-propagation: Based on the daily MSE error, the gradient descent error back-propagation algorithm is used to adjust the cross-coupling correction coefficient η, regional operating condition coupling coefficients α and β, stress coefficient σ0, and aging coefficient μ of the mechanism model.

[0043] Corrosion depth constraint correction: based on the actual corrosion depth measured on the day. Calculate the theoretical corrosion depth In the formula This represents the cumulative number of days the pipeline has been in operation; if there is a relative deviation... Then, deviation penalty corrections are applied to η and μ;

[0044] Anti-corrosion coating status mode switching: The status is determined based on the measured insulation resistance R of the anti-corrosion coating on the same day, and compared with the specific failure threshold determined based on the current anti-corrosion coating type of the pipe section. If the measured value is ≥ the corresponding threshold, the anti-corrosion coating is intact, and conventional coupling weight calculation is used; if the measured value is < the corresponding threshold, the anti-corrosion coating has failed, and the system switches to accelerated corrosion mode, and the salt spray-chloride ion coupling weight is adjusted. Temperature, humidity and pH coupling weights Multiply by a factor of 1.2;

[0045] Mechanical coupling weight range constraint: Based on the measured values ​​of pipeline deformation and vibration on the same day, if they exceed the pipeline operation safety control limits, the water pressure fluctuation coupling weight will be adjusted. Traffic vibration coupling weight The normalization limit is relaxed from 1.0 to 1.2; if it is within the allowable range, the limit of 1.0 remains unchanged.

[0046] Compared with the prior art, the present invention has the following innovative features:

[0047] (1) Unified calculation benchmark mechanism for corrosion prediction of water supply pipeline based on spatiotemporal grid: By dividing the study area into continuous uniform rectangular grids and assigning unique spatial numbers, the spatial grid units are bound to the daily statistical cycle of a fixed 1 day as spatiotemporal grid units. It is clear that all links of the whole process data acquisition, coupling quantization, model calculation, iterative correction and life assessment are based on the spatiotemporal grid as the smallest execution unit, realizing the binding of the whole process data with the unique grid number and timestamp, and establishing spatiotemporal consistency verification rules for the whole process.

[0048] (2) Spatiotemporal alignment and differential preprocessing method for multi-source heterogeneous pipeline data: For pipeline body, operation, environment, load and detection data with different sampling frequencies and different data types, differentiated rules for outlier removal, missing value completion and dimensionless processing are formulated respectively. Spatial coordinate unification and grid projection assignment are completed. A grid-level standardized spatiotemporal dataset with timestamp is constructed by aligning the two dimensions of "timestamp-grid number".

[0049] (3) Quantitative calculation method of multi-factor coupled corrosion in coastal areas: For the four core corrosion coupling systems in coastal areas, namely salt spray-chloride ions, temperature and humidity-acidity and alkalinity, water pressure fluctuation-material aging, traffic load-burial depth load transfer-rainfall, the calculation method of coupling intensity is clarified based on the electrochemical mechanism of metal corrosion and municipal engineering design parameters. A normalization rule is established with the maximum value of the corresponding coupling intensity in the study area as the benchmark, and the dimensionless coupling weight parameter in the interval [0,1] is output.

[0050] (4) Construction and closed-loop correction method of pipeline corrosion prediction model integrating mechanism constraint and data-driven integration: Taking standardized spatiotemporal dataset and coupling weight parameters as input, a global coupled corrosion mechanism model covering chemical corrosion and mechanical damage and a random forest regression machine learning prediction model are constructed at the same time. The weighted fusion calculation rules of the output results of the two types of models are clarified. A dual-cycle hierarchical correction mechanism of daily small correction and monthly large correction is established. The parameter adjustment objects and execution rules of different correction cycles are clarified. The error between the model prediction value and the field measured value is calculated using the mean square error (MSE) as an indicator. The corresponding model parameters are adjusted through the error backpropagation algorithm. The correction parameter backpropagation rules are established to form a closed-loop iterative process of the model. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the overall technical route of the present invention.

[0052] Figure 2 A schematic diagram of assigning values ​​to the grid projection of multi-source data.

[0053] Figure 3 This is a schematic diagram of the calculation of dimensionless coupled weight parameters.

[0054] Figure 4 This is a diagram of the overall model architecture. Detailed Implementation

[0055] In the process of corrosion prediction and operation and maintenance management of municipal water supply pipelines in coastal areas, existing technologies rely on multi-source heterogeneous data from pipeline records, online operation sensors, soil and atmospheric environmental monitoring, traffic network monitoring, and on-site corrosion detection. These data differ in time scale, spatial location, and data type, making it difficult to effectively integrate and transmit them throughout the entire process under a unified spatiotemporal benchmark. Furthermore, a quantitative calculation method for multi-factor coupled corrosion has not been established for the special service environment of coastal areas. Corrosion prediction models suffer from a disconnect between mechanism and data-driven approaches, and lack a closed-loop iterative correction mechanism. This results in insufficient matching between corrosion prediction and actual pipeline service conditions, inaccurate location of high-risk points, and a lack of unified quantitative standards for remaining life assessment, making it difficult to support refined operation and maintenance management of pipelines.

[0056] To address the aforementioned issues, this invention maps multi-source heterogeneous information, including pipeline structure, operating conditions, environment, and loads, onto a spatiotemporal grid unit composed of spatial grid cells and daily statistical cycles. This spatiotemporal grid serves as the unique smallest computational unit for the entire process, enabling differentiated preprocessing and spatiotemporal alignment of multi-source data. It quantifies the coupling weight parameters of the four core corrosion systems in the coastal area, thereby constructing a two-layer corrosion prediction model that integrates mechanistic constraints and data-driven approaches. Combined with a two-cycle hierarchical correction mechanism, iterative optimization of model parameters is achieved. Ultimately, it completes pipeline corrosion level determination, quantitative calculation of remaining life, and output of graded operation and maintenance results. This transforms the corrosion prediction of municipal water supply pipelines in the coastal area from a decentralized, single-dimensional, segment-level extensive calculation to a grid-level, standardized calculation under a unified spatiotemporal benchmark.

[0057] This invention uses a spatiotemporal grid as the smallest computational unit throughout. The spatiotemporal grid is composed of spatial grid cells and a daily statistical period. The spatial grid cells are continuous, uniform rectangular grids covering the study area, and the daily statistical period is a fixed time step of one day. All data acquisition, coupled quantization, model calculation, iterative correction, and lifetime assessment are performed based on this spatiotemporal unit, ensuring conflict-free spatial matching, seamless temporal synchronization, and uninterrupted parameter transmission. The overall process is as follows: Figure 1 As shown:

[0058] S1, Multi-dimensional and Multi-source Data Acquisition and Standardized Preprocessing: For the operating conditions of coastal municipal water supply pipelines, four main categories of raw data were collected: pipeline body attributes, internal operating parameters, external soil and atmospheric environmental parameters, and traffic load parameters. Differential preprocessing was performed on time-series dynamic data, static attribute data, and soil monitoring data, including outlier removal, missing value completion, dimensionless conversion, spatial coordinate alignment, and spatiotemporal binding preprocessing. Corrosion level labeling data was generated based on current industry standards and historical experience data. Simultaneously, the critical conditions for pipeline damage and basic information on critical wall thickness were determined, ultimately constructing a timestamped, grid-level standardized spatiotemporal dataset.

[0059] S2, Quantification of Multi-Factor Coupling Relationships: Focusing on four core corrosion impact systems in coastal areas—salt spray-chloride ion chemical coupling, temperature and humidity-acidity coupling, water pressure fluctuation-material aging coupling, and traffic load-burial depth load transfer-rainfall coupling—the coupling strength is calculated based on the electrochemical mechanism of metal corrosion and municipal engineering design parameters. The maximum value of the corresponding coupling strength in the study area is used as the benchmark for normalization, and finally, the dimensionless coupling weight parameters in the [0,1] interval are obtained.

[0060] S3, Construction of a Global Multi-Factor Coupled Corrosion Model: Using the coupling weight parameters output from S2 and the pipe material collected in S1 as inputs, a global coupled corrosion mechanism model covering chemical corrosion and mechanical damage is constructed. Simultaneously, a machine learning prediction model based on random forest regression is constructed, forming a two-layer prediction architecture that integrates mechanism constraints and data-driven approaches.

[0061] S4, Model Fusion and Corrosion Prediction: The mechanistic model calculation results output from S3 are weighted and fused with the random forest model output results to obtain the final corrosion rate for a single grid cell on a single day. Based on the corrosion rate threshold preset by current industry standards and historical experience data, the final corrosion rate is mapped to the corresponding corrosion level. Grid cells with corrosion levels higher than the preset risk threshold are marked as high-corrosion-risk points. At the same time, the protection status of the anti-corrosion layer is synchronously mapped and output according to the corrosion level.

[0062] S5, Model Iterative Correction Based on Measured Data: Collect true data of pipeline corrosion at the site to construct a verification set, and calculate the error between the model prediction value and the measured value at the site based on the dual-cycle hierarchical correction mechanism of daily small correction and monthly large correction, using the mean square error (MSE) as the indicator. Adaptively adjust the corresponding coupling weights, model coefficients and hyperparameters using the error backpropagation algorithm. The corrected parameters are synchronously fed back to the model construction step in S3 to realize the closed-loop iterative optimization of the model.

[0063] S6, Pipeline Remaining Life Assessment and Result Output: Based on the final corrosion rate corrected in S5, combined with the initial pipe wall thickness and critical failure conditions determined in S1, the safe operating life of the pipe section is calculated. High corrosion risk points are marked using spatiotemporal grid units and converted into key control sections through spatial mapping relationships. Graded operation and maintenance recommendations are output according to corrosion level, forming standardized results.

[0064] The specific implementation process of the present invention will be described in detail below with reference to specific embodiments.

[0065] S1. Data Acquisition and Preprocessing

[0066] S1-1 Multi-source data acquisition: Data acquisition was completed using a fusion method combining online sensor monitoring, pipeline ledger retrieval, meteorological statistics, manual supplementary measurements, and road network monitoring. Sensor data was acquired once per hour. Ledger data was sourced from municipal water supply project completion documents and operation and maintenance ledgers. Meteorological and road network data were sourced from publicly available official datasets. The final collected raw data included: sensor point coordinates, soil chloride ion concentration, soil pH value, atmospheric salt spray deposition flux, ambient temperature, pipeline material, initial pipeline wall thickness, anti-corrosion layer type, pipeline burial depth, operating water pressure fluctuation amplitude, cumulative pipeline operating days, road network-pipeline topology data, and daily traffic flow of road sections.

[0067] Among them, pipeline material, initial pipe wall thickness, and anti-corrosion layer type are pipeline body attribute data; pipeline burial depth, operating water pressure fluctuation amplitude, and cumulative operating days are pipeline internal operating parameter data; sensor point coordinates, soil chloride ion concentration, soil pH value, atmospheric salt spray deposition flux, and ambient temperature are external soil and atmospheric environmental parameters; road network-pipeline topology data and daily traffic flow of road sections are traffic load parameters.

[0068] S1-2 Differentiated preprocessing is applied to different types of raw data:

[0069] 1) Time-series dynamic data: including soil chloride ion concentration, soil pH, atmospheric salt spray deposition flux, ambient temperature and humidity, operating water pressure, water pressure fluctuation amplitude, daily rainfall, and daily traffic flow of road sections. Outliers were removed using the 3σ principle, and duplicate records were deleted; data with ≤3 consecutive missing records were imputed using linear interpolation, and data with >3 consecutive missing records were imputed using the average value of similar data from adjacent grids during the same period; finally, min-max standardization was used to map the data to the [0,1] interval to eliminate dimensional differences.

[0070] 2) Static attribute data: This includes pipe material, initial pipe wall thickness, anti-corrosion coating type, burial depth, road network-pipeline topology data, and sensor point coordinates. Pipe material, initial pipe wall thickness, and burial depth are uniformly coded and standardized in format. Missing values ​​are filled in using the original values ​​from the pipe ledger, without numerical normalization. For the collected anti-corrosion coating types, a differentiated threshold for the insulation resistance failure of the corresponding pipe section is determined. For example, the failure threshold for 3PE anti-corrosion coating is 5kΩ. m², the failure threshold of the epoxy coal tar anticorrosion coating is 2kΩ m², the failure threshold of ordinary asphalt anticorrosion coating is 2kΩ m² (based on GB / T19285-2026 standard). Topology and attribute checks are performed on the road network-pipeline topology data and sensor point coordinates to ensure there are no topology errors and no missing attributes.

[0071] 3) Soil monitoring data: including soil chloride ion concentration and soil pH. Outlier values ​​were removed using the Grubbs criterion, and the arithmetic mean of data from the same monitoring point over a daily statistical period was taken.

[0072] 4) Standard data: This includes corrosion level, corrosion rate safety threshold, critical wall thickness, and anti-corrosion coating failure criteria. Corrosion level labeling data is generated based on GB / T 19285-2026 "Inspection of Corrosion Protection Engineering for Buried Steel Pipelines" and historical experience data, classifying pipeline corrosion rates into levels I to V. Corrosion rate safety thresholds, critical wall thicknesses, and anti-corrosion coating failure criteria are determined based on GB / T 19285-2026 "Inspection of Corrosion Protection Engineering for Buried Steel Pipelines," GB / T 24596-2021 "Polyurethane Coating for Ductile Iron Pipes and Fittings," and historical experience data, without additional numerical transformation.

[0073] S1-3 Spatial Gridding: Using the CGCS2000 national geodetic coordinate system, the study area was divided into continuous, uniform rectangular grids with side lengths of 5-20m. The grid side length was adjusted according to the pipe segment density in the study area, with 5-10m grids used in densely populated areas and 10-20m grids used in sparsely populated areas. Each grid was assigned a unique spatial number and planar coordinate range. Based on the road network-pipeline topology data, a one-to-one spatial mapping relationship of "road section-pipeline segment-spatial grid" was established. For pipe segments that cross grids, the pipe segment attributes were assigned to the corresponding grid according to the proportion of the pipe segment's centerline length within the grid.

[0074] S1-4 Grid Projection Assignment: Static attribute data from S1-2 are directly assigned values ​​according to the spatial assignment in S1-3. Temporal dynamic data are projected onto the cell grid according to the influence range. Kriging interpolation is used to achieve full coverage of the study area, based on the number of discrete measurement points such as sensor location coordinates, atmospheric salt spray deposition flux (not covering the entire area), ambient temperature and humidity, soil chloride ion concentration, and pH value. The arithmetic mean of the same type of data within the same grid is taken as the final assignment. A schematic diagram of multi-source data grid projection assignment is shown below. Figure 2 As shown.

[0075] S1-5 Spatiotemporal Alignment: Using days as a fixed time step, spatiotemporal calculation units are constructed according to the "timestamp-grid number" dual-dimensional alignment mode; the consistency verification of the entire grid data and the spatiotemporal alignment verification are completed, and the results are stored in the geospatial database and an index is created.

[0076] S2. Calculation of the four major coupling weight parameters

[0077] The calculation of dimensionless coupling weight parameters is illustrated as follows: Figure 3 As shown:

[0078] The formula for calculating the coupling weight of salt spray-chloride ions in S2-1 is as follows:

[0079] ;

[0080] In the formula The salt spray-chloride ion coupling weights; This refers to atmospheric salt spray deposition flux; denoted as soil chloride ion concentration; α and β are regional operating condition coupling coefficients. Based on historical experience, in this embodiment, α∈[0.2,0.9] and β∈[0.3,0.8]; max is the maximum value in the study area; the results are normalized to the interval [0,1].

[0081] S2-2 Calculation of the coupling weights of temperature / humidity and pH:

[0082] ;

[0083] In the formula The temperature, humidity, and pH are coupled weights; T represents temperature; pH represents pH value. γ is the activation energy of the metal corrosion reaction, which can be taken as 45 in this embodiment; R=8.314 is the general gas constant; γ is the acid-base influence index, which is taken as 1.3 in this embodiment based on engineering experience and the preliminary verification results of a small sample; max is the maximum value in the study area; the results are normalized to the [0,1] interval.

[0084] S2-3 Water Pressure Fluctuation-Material Aging Coupling Weight Calculation:

[0085] ;

[0086] In the formula The coupling weight for water pressure fluctuation and material aging; The amplitude of water pressure fluctuation during operation; t d This represents the cumulative number of days the pipeline has been in operation. is the stress coefficient, which is taken as 0.7 in this embodiment based on engineering experience and preliminary verification results of a small sample; μ is the aging coefficient, which is taken as 0.4 in this embodiment based on engineering experience and preliminary verification results of a small sample; max is the maximum value in the study area; the results are normalized to the [0,1] interval.

[0087] S2-4 Traffic Load-Buried-Depth Load Transfer-Rainfall Coupling Weight Calculation:

[0088] ;

[0089] In the formula The traffic vibration coupling weight is N; N is the daily traffic volume of the road segment. The burial depth influence coefficient is determined by the pipeline burial depth collected by S1. When the burial depth is <1.0m... =1.2, when 1.0m ≤ burial depth ≤ 2.0m =1, when the burial depth is greater than 2.0m =0.8; This is a correction factor for the daily rainfall, determined by the daily cumulative rainfall R (in the absence of rain). =1.0, Kw=1.1 for light rain (0<R≤10mm), Kw=1.1 for moderate rain (10<R≤25mm) =1.2, during heavy rain (25<R≤50mm) =1.3, when heavy rain or above (R>50mm) =1.4); max is the maximum value in the study area; the result is normalized to the interval [0,1].

[0090] S3, Two-Layer Model Architecture Design

[0091] Overall model architecture as follows Figure 4 As shown:

[0092] Construction of the global coupling mechanism model: according to the formula Calculation. In the formula... The basic corrosion rate is dimensionless. =1.1 is the global cross-coupling correction coefficient; =1 is the cross-system interaction coupling coefficient, used to characterize the synergistic accelerated corrosion amplification effect of chemical coupling and mechanical coupling; The basic corrosion resistance coefficient of the pipeline material is determined by the pipeline material data collected in S1-1. It is a fixed value, taken from historical experience of commonly used municipal materials such as ductile iron. =1.0, carbon steel =1.2, PE pipe =0.1, Stainless Steel =0.3; ~ The coupling weights calculated for S2-1 to S2-4.

[0093] Machine learning prediction model construction: A random forest regression model is adopted, using the spatiotemporal computing units constructed by S1 as the sample basis, and the input features are the cumulative number of days the pipeline runs and the coupling weights. ~ Using the measured corrosion rate corresponding to corrosion level S1-2 as the training label, the daily average predicted corrosion rate of a single grid is output. The spatiotemporal dataset was divided into training and testing sets in a 7:3 ratio, with the mean squared error (MSE) between predicted and measured values ​​used as the loss function. Iteration training continued until the loss changed by less than 10% after 50 consecutive iterations. -6 The model reached convergence. The initial hyperparameters were set as follows: 100 decision trees, a maximum depth of 10 layers, and a minimum number of leaf node samples of 5.

[0094] Model fusion architecture configuration: Set the weighting ratio of the global coupling mechanism model and the machine learning prediction model to 0.6:0.4.

[0095] S4. Model Fusion and Corrosion Prediction

[0096] S4-1 Final corrosion rate calculation: according to formula = Calculation. Where: v_total is the final dimensionless corrosion rate per day per grid; For the weights of the global coupling mechanism model, These are the weights for the machine learning model, with initial values ​​of 0.6 and 0.4, and a sum of 1.

[0097] S4-2 Corrosion Result Judgment: Based on GB / T 19285-2026 in S1-2 and combined with historical experience data, the corrosion results will be determined... The corrosion is mapped to levels I through V, where level I represents weak corrosion, level II represents relatively weak corrosion, level III represents moderate corrosion, level IV represents relatively strong corrosion, and level V represents strong corrosion (level I: <0.1 mm / a; level II: 0.1 mm / a~0.3 mm / a; level III: 0.3 mm / a~0.6 mm / a; level IV: 0.6~0.9 mm / a; level V: >0.9 mm / a). Grids of level IV and above are marked as high-risk corrosion points, and four states of the anti-corrosion layer—intact, slightly damaged, moderately damaged, and failed—are simultaneously mapped and output according to the corrosion level.

[0098] S5, Iterative Correction Mechanism

[0099] The iterative correction mechanism employs a dual-cycle, hierarchical correction mechanism consisting of daily minor corrections and monthly major corrections. Daily minor corrections adjust only parameters related to the mechanistic model to adapt to daily operating condition fluctuations; monthly major corrections optimize only data-driven global parameters, model fusion weights, and cross-system interaction coupling coefficients to correct long-term prediction trends. The two types of corrections are separated in terms of responsibility and do not overlap. Specifically, they include:

[0100] S5-1 Verification Set Construction: Measured values ​​of corrosion depth, corrosion rate detected during excavation, insulation resistance of the anti-corrosion layer, and pipeline deformation and vibration, all with timestamps, were collected to form a measured verification dataset. Specifically, the corrosion rate detected during excavation was used to directly calculate the model prediction error; the measured corrosion depth was used to verify the conversion relationship between corrosion rate and cumulative operating days; the insulation resistance of the anti-corrosion layer was used to switch the coupled calculation mode; and the pipeline deformation and vibration data were used to constrain the range of mechanical coupling weights. All measured data were matched with corresponding grid numbers and timestamps, corresponding one-to-one with the predicted data.

[0101] S5-2 Error Calculation: Using the mean square error (MSE) as an indicator, the error between the predicted corrosion rate and the measured corrosion rate of a single grid on a given day is calculated and used for daily small corrections; the monthly average MSE is obtained by taking the arithmetic mean of the errors of all single grids on a given day for the month and is used for monthly large corrections.

[0102] S5-3 Daily Minor Calibration: Using a daily calibration cycle, based on the daily error and measured constraint data, only the basic parameters and coupling weights of the global coupling mechanism model are adjusted. Specifically, the following operations are performed:

[0103] 1) Backward correction of basic parameters: Based on the daily MSE error, the error backpropagation algorithm of gradient descent is used to adjust the cross-coupling correction coefficient η, regional operating condition coupling coefficients α and β, stress coefficient σ0, and aging coefficient μ of the mechanistic model. The adjustment step size is 0.01, and the adjustment range does not exceed ±20% of the initial value. The parameter combination with the smallest error is selected to update the model.

[0104] 2) Corrosion depth constraint correction: based on the actual measured corrosion depth of the day. Calculate the theoretical corrosion depth In the formula This represents the cumulative number of days the pipeline has been in operation. If there is a relative deviation... Then, deviation penalty corrections are applied to η and μ, with the correction magnitude being positively correlated with the deviation and not exceeding 15% of the initial value, so that the corrosion rate matches the measured depth.

[0105] 3) Anti-corrosion coating status mode switching: The status is determined based on the measured insulation resistance R of the anti-corrosion coating on the same day, and compared with the specific failure threshold determined in S1-2 based on the anti-corrosion coating type of this pipe section. If the measured value is ≥ the corresponding threshold, the anti-corrosion coating is intact, and conventional coupling weight calculation is used; if the measured value is < the corresponding threshold, the anti-corrosion coating has failed, and the system is switched to accelerated corrosion mode, and the salt spray-chloride ion coupling weight is adjusted. Temperature, humidity and pH coupling weights Multiply by a factor of 1.2.

[0106] 4) Mechanical Coupling Weight Range Constraint: Based on the measured values ​​of pipeline deformation and vibration on the same day, if they exceed the pipeline operation safety control limits, the water pressure fluctuation coupling weight will be adjusted. Traffic vibration coupling weight The normalization limit is relaxed from 1.0 to 1.2; if it is within the allowable range, the limit of 1.0 remains unchanged.

[0107] 5) Synchronous adjustment of the four coupling weights: Based on the above correction results, the weights are updated synchronously. ~ The final value of .

[0108] S5-4 Monthly Major Calibration: Using the calendar month as the calibration cycle, based on the monthly average MSE error, the global data-driven parameters of the machine learning model are optimized, as well as the model fusion weights and cross-system interaction coupling coefficient λ are optimized. The specific adjustment method is as follows:

[0109] 1) Random Forest Hyperparameter Optimization: With the goal of minimizing the monthly average MSE, within the preset applicable range of the project, the grid search method is used for traversal optimization. The traversal range is: 50-300 decision trees (step size 50), maximum depth 5-20 layers (step size 3), minimum number of leaf node samples 2-10 (step size 2). The hyperparameter combination with the smallest error is selected to update the model.

[0110] 2) Dynamic optimization of model fusion weights: The original fixed fusion ratio is updated to a dynamic weight formula for the final corrosion calculation rate. = ,exist ∈[0.5,0.7]、 ∈[0.3,0.5]、 + Iterate through the range of values ​​equal to 1, select the weight combination that minimizes the monthly average MSE, and update the global fusion rules.

[0111] 3) Adjustment of cross-system interaction coupling coefficient λ: Adjust λ in steps of 0.05 within the range of [0.9, 1.3], and select the value with the smallest monthly average MSE for updating.

[0112] S5-5 Closed-Loop Feedback: After daily minor calibration, the updated global coupling mechanism model parameters are fed back to S2 and S3 for calculation the next day; after monthly major calibration, the updated random forest model hyperparameters, model fusion weights, and cross-system interaction coupling coefficient λ are fed back to S3 for model prediction calculation the following month. This ultimately forms a complete iterative closed loop of "prediction - error calculation - hierarchical calibration - parameter feedback - re-prediction".

[0113] S6. Pipeline Remaining Life Assessment and Result Output

[0114] S6-1 Threshold Setting: Based on the three types of thresholds determined according to industry standards in S1-2, including corrosion rate safety threshold, used to determine whether the pipeline is in a high-risk corrosion state; critical wall thickness, used for remaining life calculation; and anti-corrosion layer failure judgment standard, used to determine whether the anti-corrosion layer has lost its protective ability.

[0115] Calculation of the safe operating life of S6-2 pipeline: Convert the daily corrosion rate to the annual corrosion rate using the formula. (Leap years are calculated as 366 days). Then calculate the safe operating life using the formula. Calculation, where L is the safe operating life; This represents the initial wall thickness of the pipe. Critical wall thickness; Daily corrosion rate; The annual corrosion rate is given.

[0116] S6-3 Result Output: Based on spatiotemporal grid cells, Grids exceeding the corrosion rate safety threshold are marked as high-risk corrosion points. Through the "road section-pipeline segment-spatial grid" mapping relationship established in S1-3, the pipe segments corresponding to the high-risk points are marked as key control road sections.

[0117] The method uses a spatiotemporal grid as the smallest computational unit throughout, including spatial grids and daily statistical periods. All steps are executed based on this unit. All data in the entire process carries a unique grid number and timestamp, and the output of each step undergoes spatiotemporal consistency verification to ensure that there are no conflicts in spatial matching of multi-source data, no misalignment in temporal synchronization, and no breaks in parameter transmission.

[0118] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0119] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas, characterized in that, Includes the following steps: S1 uses a spatiotemporal grid as the smallest computational unit. Based on the grid-deployed sensors, it collects pipeline body attribute parameters, pipe internal operation parameters, external soil and atmospheric environmental parameters, and traffic load parameters, and performs preprocessing. S2, based on S1 data and the electrochemical mechanism of metal corrosion and municipal engineering design parameters, performs weight calculations for salt spray-chloride ion chemical coupling, temperature and humidity-acidity coupling, water pressure fluctuation-material aging coupling, traffic load-burial depth load transfer-rainfall coupling, and obtains four types of dimensionless coupling weight parameters. S3, the coupling weight parameters obtained in S2, combined with the pipe material, are input into the constructed global coupled corrosion mechanism model covering both chemical corrosion and mechanical damage, outputting the dimensionless basic corrosion rate; the global coupled corrosion mechanism model is as follows: ； In the formula The basic corrosion rate is dimensionless. This is a global cross-coupling correction coefficient; The cross-system interactive coupling coefficient is used to characterize the synergistic accelerated corrosion amplification effect of chemical coupling and mechanical coupling. The basic corrosion resistance coefficient of the pipeline material is determined by the collected data on the pipeline material and is a fixed value; ductile iron is an example. =1.0, carbon steel =1.2, PE pipe =0.1, Stainless Steel =0.3; ~ These are the calculated coupling weight parameters; Meanwhile, the coupling weight parameters obtained from S2, combined with the cumulative number of days the pipeline has been in operation, are input into the trained machine learning prediction model to output the daily average predicted corrosion rate of a single grid. S4 calculates the final corrosion rate of a single grid cell by weighting and fusing the dimensionless basic corrosion rate with the daily average predicted corrosion rate of the single grid cell. Combined with the preset corrosion rate threshold, the final corrosion rate is mapped to the corresponding corrosion level.

2. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 1, characterized in that: It also includes a model iterative correction process based on measured data: collect true data of pipeline corrosion on site to build a verification set, and calculate the error between the model prediction value and the on-site measured value based on the two-cycle hierarchical correction mechanism of daily small correction and monthly large correction, using the mean square error (MSE) as the index, and use the error backpropagation algorithm to adaptively adjust the corresponding coupling weight parameters, model coefficients and hyperparameters. Daily minor calibrations only adjust parameters related to the global coupled corrosion mechanism model to adapt to daily operating condition fluctuations; monthly major calibrations optimize global parameters of the machine learning prediction model driver class, model fusion weights, and cross-system interaction coupling coefficients in the global coupled corrosion mechanism model to correct long-term prediction trends.

3. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 1, characterized in that: This also includes the quantification of pipeline remaining life assessment, specifically: Based on industry standards, three types of thresholds are determined, including a corrosion rate safety threshold, used to determine whether the pipeline is in a high-risk corrosion state; a critical wall thickness, used for calculating remaining service life; and a corrosion protection layer failure judgment standard, used to determine whether the corrosion protection layer has lost its protective ability. Convert the daily corrosion rate to the annual corrosion rate using the formula. Calculate the number of days, with leap years counted as 366; then calculate the safe operating life using the formula. Calculation, where L is the safe operating life; δ_limit is the initial wall thickness of the pipe; δ_limit is the critical wall thickness. Daily corrosion rate; The annual corrosion rate is given.

4. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 1, characterized in that: The pipeline body attribute parameters include pipeline material, initial pipeline wall thickness, and anti-corrosion layer type; the pipeline internal operation parameters include pipeline burial depth, operating water pressure fluctuation amplitude, and cumulative number of days of pipeline operation; the pipeline external soil and atmospheric environment parameters include sensor point coordinates, soil chloride ion concentration, soil pH value, atmospheric salt spray deposition flux, and ambient temperature; the traffic load parameters include road network-pipeline topology data and daily traffic flow of road sections.

5. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 4, characterized in that: The spatial grid adopts the CGCS2000 national geodetic coordinate system. The study area is divided into continuous uniform rectangular grids with side lengths of 5-20m. The grid side length is adjusted according to the pipe segment density in the study area. 5-10m grids are used in densely populated pipe segment areas and 10-20m grids are used in sparsely populated pipe segment areas. Each grid is assigned a unique spatial number and planar coordinate range. Based on the road network-pipeline topology data, a one-to-one spatial mapping relationship of "road section-pipeline segment-spatial grid" is established. For pipe segments that cross grids, the pipe segment attributes are assigned to the corresponding grid according to the proportion of the pipe segment centerline length within the grid. Static attribute data are directly assigned values ​​according to spatial affiliation, while time-series dynamic data are projected onto the cell grid according to the influence range. The number of discrete measurement points is achieved by using Kriging interpolation to achieve full coverage of the study area. The arithmetic mean of the same type of data in the same grid is taken as the final value.

6. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 2, characterized in that: In S2, the formula for calculating the salt spray-chloride ion coupling weight is: ； In the formula The salt spray-chloride ion coupling weights; This refers to atmospheric salt spray deposition flux; The soil chloride ion concentration is taken as α, β are taken as regional operating condition coupling coefficients; max is the maximum value in the study area. The formula for calculating the coupling weight of temperature, humidity and pH is: ； In the formula The temperature, humidity, and pH coupling weights are used; T represents temperature. pH refers to acidity or alkalinity; E a γ is the activation energy of the metal corrosion reaction, R is the universal gas constant, and γ is the acid-base effect index, all of which are fixed indices; max is the maximum value in the study area. The formula for calculating the coupling weight of water pressure fluctuation and material aging is as follows: ； In the formula The coupling weight for water pressure fluctuation and material aging; This refers to the amplitude of water pressure fluctuation during operation. t d The cumulative number of days the pipeline has been in operation; stress coefficient The aging coefficient is a fixed initial value; max is the maximum value in the study area; The formula for calculating the coupling weight of traffic load-burial depth load transfer-rainfall is as follows: ； In the formula Traffic vibration coupling weights; N represents the daily traffic volume of the road segment; The burial depth influence coefficient is determined by the collected pipeline burial depth data. is the daily rainfall correction factor, determined by the daily cumulative rainfall R in different levels; max is the maximum value for the study area.

7. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 1, characterized in that: The machine learning prediction model employs a random forest regression model, using constructed spatiotemporal computing units as the sample base, with input features including the cumulative number of days the pipeline has been running and coupling weight parameters. ~ Output the daily average predicted corrosion rate of a single grid. The initial hyperparameters of the model were set as follows: 100 decision trees, a maximum depth of 10 layers, and a minimum number of leaf node samples of 5.

8. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 2, characterized in that: The formula for calculating the final corrosion rate in S4 is as follows: = ； In the formula, v_total is the final dimensionless corrosion rate of a single grid per day; For the weights of the global coupling mechanism model, These are the weights for the machine learning model, with initial values ​​of 0.6 and 0.4 respectively, and the sum of the weights is 1. Based on the GB / T 19285-2026 standard and combined with historical experience data, The corrosion is mapped to levels I through V, where level I represents weak corrosion, level II represents relatively weak corrosion, level III represents moderate corrosion, level IV represents relatively strong corrosion, and level V represents strong corrosion. Grids of level IV or higher are marked as high-risk corrosion points, and the corrosion protection layer is simultaneously mapped and output in four states: intact, slightly damaged, moderately damaged, and failed, based on the corrosion level.

9. The multi-factor coupled corrosion prediction method for municipal water supply pipelines in coastal areas as described in claim 6, characterized in that: The daily minor calibration uses a single day as the calibration cycle. Based on the daily error and measured constraint data, it only adjusts the basic parameters and coupling weights of the global coupling mechanism model, specifically performing the following operations: Basic parameter back-propagation: Based on the daily MSE error, the gradient descent error back-propagation algorithm is used to adjust the cross-coupling correction coefficient η, regional operating condition coupling coefficients α and β, stress coefficient σ0, and aging coefficient μ of the mechanism model. Corrosion depth constraint correction: based on the actual corrosion depth measured on the day. Calculate the theoretical corrosion depth In the formula This represents the cumulative number of days the pipeline has been in operation; if there is a relative deviation... Then, deviation penalty corrections are applied to η and μ; Anti-corrosion coating status mode switching: The status is determined based on the measured insulation resistance R of the anti-corrosion coating on the same day, and compared with the specific failure threshold determined based on the current anti-corrosion coating type of the pipe section. If the measured value is ≥ the corresponding threshold, the anti-corrosion coating is intact, and conventional coupling weight calculation is used; if the measured value is < the corresponding threshold, the anti-corrosion coating has failed, and the system switches to accelerated corrosion mode, and the salt spray-chloride ion coupling weight is adjusted. Temperature, humidity and pH coupling weights Multiply by a factor of 1.2; Mechanical coupling weight range constraint: Based on the measured values ​​of pipeline deformation and vibration on the same day, if they exceed the pipeline operation safety control limits, the water pressure fluctuation coupling weight will be adjusted. Traffic vibration coupling weight The normalization limit is relaxed from 1.0 to 1.2; if it is within the allowable range, the limit of 1.0 remains unchanged.

Images