Urban gas leakage risk assessment method and system based on quantitative structure-activity relationship

Abstract

The application discloses a kind of urban gas leakage risk assessment method and system based on quantitative structure-activity relationship, belong to urban gas safety technical field.For the defects of strong subjectivity of prior art, black box model is not explainable, small sample imbalance is not suitable, and risk traceability is difficult, the application constructs nonlinear leakage structure-activity activity index SAI containing discrete coefficient correction, establishes the explicit structure-activity constitutive model of " ontology-environment-working condition " ternary;Propose a hierarchical mixed sampling method based on SAI, optimize parameters using Bayesian joint fitting;Build a leakage consequence evaluation system coupled with SAI and a risk contribution decomposition formula without numerical anomalies.The application has a leakage discrimination AUC of 0.942, and a model recall rate of 92.3%, which can be directly embedded into existing gas pipeline GIS / SCADA systems and is suitable for the whole life cycle safety management of urban gas pipeline network.

Description

Technical Field

[0001] This invention relates to the field of urban gas safety risk assessment technology, specifically to a method and system for assessing urban gas leakage risk based on quantitative structure-property relationships. Background Technology

[0002] Urban gas pipeline networks are a core lifeline infrastructure of cities, and leaks can easily lead to major safety incidents such as explosions, fires, and poisoning. Currently, urban gas leak risk assessment methods are mainly divided into three categories: qualitative assessment, traditional quantitative risk assessment (QRA), and machine learning black-box assessment. All of these methods have insurmountable technical shortcomings. First, qualitative / semi-quantitative methods (such as risk matrix method and expert scoring method) rely entirely on expert experience and lack quantitative correlation between "structural parameters and risk effects". The evaluation results are highly subjective and inconsistent, and cannot guide precise operation and maintenance.

[0003] Second, the traditional QRA method uses fixed empirical formulas to calculate the probability and consequences of leakage, which severs the inherent mapping relationship between the pipeline structure, laying environment, operating conditions and leakage risk. It has poor generalization ability and cannot be adapted to complex urban scenarios.

[0004] Third, existing machine learning evaluation methods are mostly black-box models (such as ordinary LightGBM and random forest). Although they can achieve risk prediction, they have two major problems: First, they are poorly adapted to small-sample imbalanced data of gas leaks and have a low identification rate of high-risk leak samples; second, the models are not interpretable and cannot quantify the contribution of each factor to the risk, making it difficult to meet the traceability needs of engineering operation and maintenance.

[0005] Quantitative structure-activity relationship (QSAR) theory has been widely applied in the chemical and pharmaceutical fields. Its core is to construct a quantitative mapping of "structural features-effects". However, existing technologies have not deeply integrated this theory with urban gas leakage risk assessment: neither has a "volume-environment-operating condition" ternary structure-activity factor system adapted to gas scenarios been constructed, nor has a structure-activity activity index that breaks through the linear limitation been designed. At the same time, existing technologies have not disclosed reproducible ternary coefficient fitting methods and risk contribution calculation methods, which makes it impossible for QSAR theory to be applied in the gas field.

[0006] Therefore, developing a complete model, well-defined parameters, and a step-by-step reproducible method for assessing urban gas leakage risks is a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] The purpose of this invention is to overcome the above-mentioned defects of the prior art, and to solve the technical problems of existing gas leak risk assessment methods, such as strong subjectivity, uninterpretable black box models, poor adaptability due to small sample imbalance, difficulty in risk tracing, and inability to reproduce, and to provide a method and system for urban gas leak risk assessment based on quantitative structure-property relationship.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: The method for assessing urban gas leakage risk based on quantitative structure-property relationships of the present invention includes the following steps: S1. Constructing a ternary structure-activity factor system and a nonlinear leakage structure-activity index (SAI) with discrete coefficient correction: Urban gas leakage-related factors are classified into three categories: pipeline structure, environment, and operating conditions. Standardized quantitative rules conforming to national standards are formulated, and the standardized values ​​of each factor and the pipeline leakage tendency are calculated. SAI The index is used for subsequent sampling and consequence correction; S2, based on SAI Layered mixed sampling preprocessing: using the results obtained in step S1 SAI The index combines historical data of ternary structure-activity factors with leakage accident data according to... SAI The values ​​are stratified, and oversampling of leaked samples and undersampling of non-leaked samples are performed on each stratum to obtain a class-balanced standardized dataset for feature selection in step S3. S3. Structure-activity factor feature selection and optimal subset extraction: Based on the standardized dataset obtained in step S2, correlation redundancy removal, significance screening and LASSO regression dimensionality reduction are performed in sequence to remove redundant and non-significant factors and obtain the optimal core feature subset that is strongly correlated with leakage risk for use in step S4 model construction. S4. Construct a ternary explicit structure-effect constitutive model and fit the parameters: Using the optimal core feature subset obtained in step S3 as input, construct a ternary explicit structure-effect constitutive model based on the improved Weibull distribution, and use the Bayesian joint fitting method to globally optimize the model parameters to obtain the annual leakage probability of the pipeline section for use in the comprehensive risk calculation in step S6. S5, Building Coupling SAI Leakage consequence severity assessment model: using the results obtained in step S1 SAI The index-corrected benchmark leakage hole area is used to calculate the leakage consequence parameters by combining the small hole leakage model and the Gaussian diffusion model. The comprehensive consequence severity value is obtained by the combined weighting method and used for the comprehensive risk calculation in step S6. S6. Comprehensive Risk Assessment and Decomposition of Explicit Contribution Without Anomalies: Based on the probability of leakage obtained in step S4 and the comprehensive severity of consequences obtained in step S5, calculate the comprehensive risk value of the pipeline segment and classify the risk level. Construct a full-factor contribution formula without numerical anomalies to obtain the risk contribution ratio of each factor and realize the quantitative tracing of risk causes. S7. Model Validation and Dynamic Update: Divide the standardized dataset obtained in step S2 into training and test sets to validate model performance. Update model parameters using incremental training according to preset trigger conditions to ensure long-term model effectiveness.

[0009] Preferably, in step S1: The structural performance factors include seven items: pipe material, service life, nominal wall thickness, interface type, anti-corrosion coating type, weld quality grade, and wall thickness surplus rate; the environmental regulation factors include four items: soil corrosivity, population density within 50m of the pipeline, number of third-party constructions within 100m of the pipeline per year, and building density; and the operating condition regulation factors include three items: average operating pressure, monthly flow fluctuation coefficient, and cumulative duration of annual pressure exceedance. Categorical factors were standardized using the leakage propensity level assignment method, and numerical factors were standardized using the [0,1] interval min-max method. The weights of ontological structure-activity factors were determined by combining the F-significance test at a 95% confidence level with the variance contribution rate method, ωi=SSi / SST, where SSi is the sum of squares of deviations from the mean of the i-th factor, and SST is the sum of squares of deviations from the mean of all factors that passed the test, with a test threshold of p<0.05. Nonlinear SAI The calculation formula is: in, ω i Let i be the weight of the i-th ontology structure-activity factor. fi ( Xi ) is the factor standardization function, and the standardization interval is [0,1]. μ This is a nonlinear correction coefficient, with a value ranging from 0.05 to 0.2. δ ( X ) is the dispersion coefficient of the ontological structure-activity factor, which characterizes the degree of dispersion of the standardized values ​​of the ontological factor.

[0010] Preferably, in step S2: Data preprocessing includes one-hot encoding of categorical factors, filling missing values ​​with the mean, removing outliers using the 3σ criterion, and min-max normalization in the [0,1] interval; according to SAI The value divides the sample into 5 intervals: SAI <0.2, 0.2≤ SAI <0.4, 0.4≤ SAI <0.6, 0.6≤ SAI <0.8、 SAI ≥0.8; ADASYN algorithm: nearest neighbor number k=5, oversampling ratio k s=5-6× p s ,in p s This represents the interval leakage rate. k s ∈[2,5], the corresponding multipliers for the 5 intervals are 5, 4, 3, 2, and 2 respectively; The TomekLinks algorithm has a neighborhood of 3, and the proportion of redundant boundary samples removed does not exceed 15%.

[0011] Preferably, in step S3: Correlation redundancy removal was performed using Pearson analysis, with a threshold of 0.8 for the absolute value of the correlation coefficient. The significance threshold for one-way ANOVA was p < 0.05; LASSO regression uses 10-fold cross-validation between the predicted and actual leakage probabilities to determine the minimum RMSE and then eliminates factors with a regression coefficient of 0.

[0012] Preferably, in step S4: Comprehensive value of environmental regulation factors: Overall value of operating condition adjustment factor: in, η j For the first j The weights of environmental regulation factors, θ k For the first k The weights of the operating condition adjustment factors were all calculated using the random forest algorithm based on the feature importance and then normalized to the [0,1] interval using min-max. The parameters for the random forest algorithm are: 50-200 decision trees, maximum depth 6-12, and the number of features is the square root of the number of corresponding factors, rounded up. The calculation formula for the ternary explicit structure-activity constitutive model is: in, P leak This represents the annual leakage probability of the pipeline section. E adj This is the comprehensive value of environmental regulation factors. O adj This is the comprehensive value of the operating condition adjustment factor; α , β , γ The ternary structure-activity ratio, λ For shape parameters; Bayesian joint fitting method was used to... α ,β , γ , λ Perform global optimization to establish an explicit quantitative mapping from structure-activity factors to leakage probability; in P leak denoted as the annual leakage probability of the pipeline section, α, β, and γ are the ternary structure-activity coefficients, and λ is the shape parameter; The Bayesian joint fitting uses the TPE sampling method, with fitting boundaries of α∈[0.5,2.0], β∈[0.3,1.5], γ∈[0.2,1.2], and λ∈[0.8,1.5]. The termination condition is a maximum of 50 iterations or a validation set AUC convergence threshold ≤ 0.001.

[0013] Preferably, in step S5: SAI The corrected formula for the area of ​​the leakage hole is: Where k is the structure-performance correction factor, ranging from 0.8 to 1.5, and Abase is the baseline leakage orifice area, taken as 2% of the pipe diameter; The calculation formula for the orifice leakage model is: in C d =0.6 is the flow coefficient. ρ =0.717kg / m 3 The density of methane at room temperature and pressure, Δ P This is the difference between the pipeline's operating pressure and the standard atmospheric pressure. The Gaussian diffusion model uses Pasquill D-level neutral atmospheric stability, with the methane explosion limit taken as 5%~15%; The combined weighting method uses a combination of AHP and entropy weighting. ω com =0.6 ω AHP +0.4 ω entropy ,in ω AHP For AHP subjective weights, ω entropy As the objective weight of entropy, the consistency ratio CR of the judgment matrix constructed by the AHP method is <0.1.

[0014] Preferably, in step S6: Comprehensive risk value of pipeline section R= P leak ×S, where S is the overall severity of the consequences; According to the ALARP criterion of GB / T27921-2011, the risk levels are divided into: low risk R<0.1, medium risk 0.1≤R<0.3, high risk 0.3≤R<0.7, and very high risk R≥0.7; The formula for the contribution of all factors to the risk of no numerical anomalies is: ontological factors i Contribution: Environmental factors j Contribution: Operating factors k Contribution: After calculation, the absolute values ​​of all contributions are taken and normalized by min-max in the [0,1] interval to obtain the risk contribution ratio of each factor.

[0015] In the method for assessing urban gas leakage risk based on quantitative structure-property relationships, step S7 is as follows: The model evaluation metrics include AUC, recall, accuracy, MAE, and RMSE. The passing criteria are AUC ≥ 0.9 on the test set and recall ≥ 90% for leaked samples. The model update is triggered when: the fixed period does not exceed 12 months, or the number of newly leaked samples is ≥ 10% of the original number of leaked training samples; The incremental training method optimizes only four parameters: α, β, γ, and λ. It adopts mini-batch gradient descent with a batch size of 32 to 64 and a learning rate of 0.01 to 0.1. The termination condition is that the loss function convergence threshold is ≤1e-4 or the maximum number of iterations is 200. The loss function is the mean square error (MSE) between the predicted and actual leakage probability values.

[0016] Preferably, the urban gas leakage risk assessment system based on quantitative structure-property relationship of the present invention includes the following modules connected in sequence: Construction of ternary structure-activity factors and SAI The calculation module is used to divide the "body-environment-operating condition" three-element structure-property factor system, formulate standardized quantification rules, and calculate the standardized values ​​of each factor and the leakage tendency of the pipeline body. SAI The index is output to the hierarchical hybrid sampling preprocessing module and the leakage consequence assessment module; The hierarchical hybrid sampling preprocessing module is used to receive... SAI The index collects corresponding historical data and leakage accident data, performs preprocessing, and then... SAI Value stratification is performed to perform mixed sampling to obtain a class-balanced standardized dataset, which is then output to the feature selection module and the model validation and dynamic update module. The feature selection module receives a standardized dataset and sequentially performs correlation redundancy removal, significance selection, and LASSO regression dimensionality reduction to obtain the optimal core feature subset, which is then output to the ternary structure-efficacy constitutive model fitting module. The ternary structure-effect constitutive model fitting module is used to receive the optimal core feature subset, construct and fit the ternary explicit structure-effect constitutive model, obtain the annual leakage probability of the pipeline section, and output it to the comprehensive risk assessment and contribution decomposition module. The leakage consequence assessment module is used to receive... SAI The index is used to correct the leakage hole area and calculate the leakage consequence parameters. The comprehensive consequence severity value is obtained through the combined weighting method and output to the comprehensive risk assessment and contribution decomposition module. The comprehensive risk assessment and contribution decomposition module is used to receive the probability of leakage and the severity of comprehensive consequences, calculate the comprehensive risk value and classify the level, and decompose the risk contribution ratio of each factor. The model validation and dynamic update module is used to receive standardized datasets, validate model performance, and update model parameters according to preset trigger conditions.

[0017] Preferably, in the ternary structure-activity factor construction and SAI calculation module, the SAI calculation formula is: ; In the formula: ωi is the weight of the i-th ontology structure-activity factor, determined by a combination of a 95% confidence level F-significance test and the variance contribution rate method, ωi= SS i / SS T The significance threshold for the F-test was p < 0.05. fi ( Xi ) is the factor standardization function. Categorical factors are standardized using the leakage tendency level assignment method, and numerical factors are standardized using the [0,1] interval min-max standardization method, with the standardization interval being [0,1]. μ The nonlinear correction coefficient ranges from 0.05 to 0.2. It is determined by fitting historical leakage data using the nonlinear least squares method, with the goal of maximizing the AUC of the leakage sample discrimination. In the ternary structure-activity constitutive model fitting module, the model formula is: ; Bayesian joint fitting was performed using the TPE sampling method; In the leakage consequence evaluation module, the formula for correcting the leakage orifice area is: The AHP and entropy weighting method are used for weighting; where: k is the structure-activity relationship correction coefficient, with a value range of 0.8~1.5, which is determined by fitting historical leakage data using the nonlinear least squares method, with the goal of minimizing the RMSE between the predicted and actual leakage consequences. A base The baseline leakage orifice area is taken as 2% of the pipe diameter; In the comprehensive risk assessment and contribution decomposition module, the comprehensive risk value... R = P leak × S , R ∈[0,1], divided into 4 risk levels according to the ALARP criterion, the contribution formula has no denominator to avoid calculating singularities; In the model validation and dynamic update module, the model qualification criteria are AUC ≥ 0.9 on the test set and recall rate of leaked samples ≥ 90%. Incremental training only optimizes the ternary structure-activity ratio and shape parameters.

[0018] The advantages of this invention compared to the prior art are: 1. Pioneering nonlinear SAI index to improve leakage discrimination: The nonlinear SAI index, which introduces a discrete coefficient correction, solves the problem of missed detection in traditional linear indices where "a single factor is abnormal but the overall weighted value is normal". Its leakage sample discrimination AUC reaches 0.942, which is 14% higher than that of linear SAI, and can accurately characterize the leakage tendency of pipeline body.

[0019] 2. Constructing an explicit structure-effect model to overcome the limitations of the black box: A three-element explicit structure-effect constitutive model of "ontology-environment-operating conditions" is proposed. The Bayesian joint fitting method is used to globally optimize the parameters and fully consider the interaction effects between factors. The model recall rate reaches 92.3%, which is 10.3% higher than the step-by-step fitting method, and realizes the explicit quantitative mapping of "structure-effect".

[0020] 3. Design an anomaly-free contribution formula for accurate source tracing: Construct a full-factor risk contribution decomposition formula without numerical anomalies, which completely solves the calculation singularity problem of the denominator being 0 in the traditional formula. It can directly output the risk contribution ratio of each factor, providing the operation and maintenance department with a clear direction for rectification.

[0021] 4. Full process reproducibility and strong industrial applicability: All formulas, parameters and calculation steps are made public, strictly bound to relevant national standards, and standardized API interfaces are provided. It can be directly embedded into existing gas pipeline GIS / SCADA systems without adding new hardware equipment. The time for a single model incremental update is ≤2 hours, meeting the timeliness requirements of engineering operation and maintenance. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the overall process of the urban gas leakage risk assessment method based on quantitative structure-property relationship of the present invention.

[0024] Figure 2 This is a comparison curve of the leakage discrimination index ROC between the nonlinear SAI of this invention and the traditional linear SAI. Detailed Implementation

[0025] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0026] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings.

[0028] The method for assessing urban gas leakage risk based on quantitative structure-property relationship of the present invention specifically includes the following steps: S1. Constructing a ternary structure-activity factor system and a nonlinear leakage structure-activity index SAI with discrete coefficient correction. S1.1 Division of the ternary structure-activity factor system The factors related to urban gas leakage are classified into three categories, forming a ternary structure-activity factor system. The quantification rules of all factors comply with GB 50028-2006 (2020 edition) "Urban Gas Design Standard" and GB 50236-2011 "Construction Specification for Welding Engineering of On-site Equipment and Industrial Pipelines". Intrinsic structure-activity factors X(7 items): Pipe material, service life, nominal wall thickness, interface type, anti-corrosion coating type, weld quality grade, and wall thickness surplus rate are the core factors that determine the leakage tendency of the pipeline body. Environmental regulators E (4 items): Soil corrosivity, population density within 50m of the pipeline, number of third-party constructions within 100m of the pipeline per year, and building density are external factors that affect the occurrence of leaks and the spread of their consequences. Operating condition adjustment factor O (3 items): average operating pressure, monthly flow fluctuation coefficient, and annual cumulative duration of pressure exceeding limit are dynamic factors affecting pipeline operation safety.

[0029] S1.2 Construction of Nonlinear SAI Exponent The SAI index is used to quantify the leakage tendency of a pipeline. Unlike existing linear weighted indices, this invention introduces a discrete coefficient for the ontological factor. δ ( X As a nonlinear correction term, SAI addresses the issue of missed detections where "a single factor is abnormal but the overall weighted value is normal." The SAI calculation formula is as follows: (1) in: (1) ω i For the first i The weights of the structure-activity factors were determined using a combination of the 95% confidence level F-significance test and the variance contribution rate method. ω i = SS i / SS T The significance threshold for the F-test was p < 0.05. (2) f i ( X i ) is the factor standardization function. Categorical factors are standardized using the leakage tendency level assignment method, and numerical factors are standardized using the [0,1] interval min-max standardization method, with the standardization interval being [0,1]. (3) μ The nonlinear correction coefficient ranges from 0.05 to 0.2. It is determined by fitting historical leakage data using the nonlinear least squares method, with the goal of maximizing the AUC of the leakage sample discrimination. (4) δ ( X ) is the dispersion coefficient of the structure-activity relationship factor, and the calculation formula is as follows. It characterizes the degree of dispersion of the standardized value of the structure-activity relationship factor: the higher the degree of dispersion, the more likely the pipeline body has a "short-board factor" and the greater the nonlinear correction amplitude of the leakage tendency.

[0030] (3) in, - f ( X ) is the arithmetic mean of the standardized values ​​of the ontological structure-activity factor.

[0031] S2 SAI-based hierarchical hybrid sampling preprocessing To address the issue of "few and unevenly distributed" gas leak samples, historical data of ternary structure-activity factors and leak accident data were collected. After data preprocessing, the samples were divided into 5 intervals according to the SAI value. Sampling operations were performed independently for each interval to obtain a class-balanced standardized dataset, ensuring the model's ability to identify high-risk samples. (1) Data preprocessing: One-hot encoding was performed on the categorical factors, and missing values ​​were filled using the mean imputation method. 3 σ The criteria are used to remove outliers, and normalization is performed using the min-max method in the [0,1] interval. (2) Layered division: 5 SAI The interval is SAI <0.2, 0.2≤ SAI <0.4, 0.4≤ SAI <0.6, 0.6≤ SAI <0.8、 SAI ≥0.8; (3) Oversampling: For the leakage samples (minority class) within each interval, the ADASYN algorithm with a nearest neighbor number k=5 is used for oversampling. The oversampling factor is positively correlated with the interval leakage occurrence rate. k s =5-6× p s , p s This represents the interval leakage rate. k s ∈[2,5], the multipliers corresponding to the 5 intervals are 5, 4, 3, 2, and 2 respectively; (4) Undersampling: For the leak-free samples (majority class) in each interval, the Tomek Links algorithm with a neighborhood number of 3 is used for undersampling to remove redundant samples at the boundary, avoid model overfitting, and the proportion of samples removed does not exceed 15%.

[0032] S3 Structure-activity factor feature screening and optimal subset extraction To eliminate redundant factors and improve model efficiency, a three-step screening process was performed on the standardized dataset to obtain the optimal subset of core features strongly correlated with leakage risk: (1) Pearson correlation redundancy removal: Calculate the correlation coefficient between factors and remove redundant factors with an absolute value > 0.8; (2) Significance screening by one-way ANOVA: Calculate the F-statistic and p-value of the factor and the leakage label, and remove non-significant factors with p>0.05; (3) LASSO regression dimensionality reduction: the optimal regularization coefficient is determined by ten-fold cross-validation, and factors with regression coefficients of 0 are eliminated to obtain the optimal core feature subset.

[0033] S4 Construct a ternary explicit structure-activity constitutive model and fit the parameters S4.1 Calculation of Factor Composite Value Comprehensive value of environmental regulation factors E adj Comprehensive value of operating condition adjustment factor O adj The calculation formula is as follows: (4) (5) in, η j For the first j The weights of environmental regulation factors, θ k For the first k The weights of the working condition adjustment factors are all calculated by the random forest algorithm to determine the feature importance and then normalized to the [0,1] interval min-max. The parameters of the random forest algorithm are: number of decision trees 50~200, maximum depth 6~12, and the number of features is the square root of the number of corresponding factors and rounded up.

[0034] S4.2 Fitting of the Ternary Explicit Structure-Effect Constitutive Model Using the optimal core feature subset as input and the annual leakage probability of the pipeline segment as output, a ternary explicit structure-effect constitutive model based on the improved Weibull distribution is constructed, as shown in the following formula: (2) in, α , β , γ The ternary structure-activity ratio, λ The four parameters are shape parameters; a Bayesian joint fitting method is used to globally optimize the four parameters. Unlike the traditional step-by-step fitting method, the joint fitting can fully consider the interaction effect between factors and improve the model accuracy.

[0035] Specific parameters for Bayesian joint fitting: A tree-structured Parzen estimator (TPE) is used as the sampling method, implemented using the Python Hyperopt library; the fitting boundary is: α ∈[0.5,2.0]、 β ∈[0.3,1.5]、 γ∈[0.2,1.2]、 λ ∈[0.8,1.5]; the fitting termination condition is: the maximum number of iterations is 50, or the AUC of the validation set convergence threshold is ≤0.001, and the optimization objective is to maximize the AUC of the validation set.

[0036] S5 Constructs a severity assessment model for leakage consequences coupled with SAI. S5.1 Leakage Hole Area SAI Correction The leakage hole area is a core parameter for consequence calculation. Traditional methods use a fixed value. This invention corrects the leakage hole area using SAI, as shown in the following formula: Wherein, k is the structure-property correction coefficient, with a value range of 0.8 to 1.5, which is determined by fitting historical leakage data using the nonlinear least squares method, with the goal of minimizing the RMSE between the predicted and actual leakage consequences. A base The baseline leakage hole area is taken as 2% of the pipe diameter.

[0037] S5.2 Consequence Calculation and Combinatorial Weighting Based on the corrected orifice area, the leakage rate is calculated using a small orifice leakage model, and the explosion impact range is calculated using a ground continuous point source Gaussian diffusion model (the methane explosion limit is taken as 5%~15%). Sub-models for personnel casualty risk, economic loss risk, and diffusion impact risk are constructed respectively. The weights of each sub-model are determined by a combination of the analytic hierarchy process (AHP) and the entropy weight method. The weighted sums are then normalized to obtain the comprehensive consequence severity value S, where S∈[0,1].

[0038] The formula for calculating the combined weight is as follows: ω com =0.6 ω AHP +0.4 ω entropy , ω AHP For AHP subjective weights, ω entropy As the objective weight of entropy, the consistency ratio CR of the judgment matrix constructed by the AHP method is <0.1.

[0039] S6 Comprehensive Risk Assessment and Decomposition of Explicit Contributions Without Anomalies (1) Comprehensive risk calculation: Comprehensive risk value of pipeline section R = P leak × S , R ∈[0,1]; (2) Risk level classification: Based on the ALARP criteria in Section 6.4.3 of GB / T 27921-2011, it is divided into 4 levels: low risk R <0.1 (acceptable area, no additional safety measures required), medium risk 0.1≤ R <0.3 (acceptable area, requires regular monitoring), high risk 0.3≤ R <0.7 (ALARP zone, improvement measures required, meeting the minimum reasonable feasible principle), extremely high risk R ≥0.7 (Unacceptable area, immediate rectification required); (3) Contribution decomposition: Construct a full-factor risk contribution formula without singular values, and delete the traditional formula. P leak denominator terms, avoid P leak The problem of calculating singularities when =0 is solved; after calculation, the absolute value is taken and normalized by min-max in the [0,1] interval to obtain the risk contribution ratio of each factor, so as to achieve accurate tracing of the causes of risk.

[0040] S7 Model Validation and Dynamic Update Ten-fold cross-validation was used to ensure model stability, and independent test sets were used to validate model performance. Evaluation metrics included AUC, recall, accuracy, MAE, and RMSE. The model validation pass criteria were: test set AUC ≥ 0.9 and leaked sample recall ≥ 90%.

[0041] Dynamic updates to model parameters are achieved using incremental training, triggered by a fixed period not exceeding 12 months (calculated from the date of completion of the first model training) or a new number of leaked samples ≥ 10% of the original number of leaked training samples (calculated from the date of completion of the last model update). Incremental training only optimizes the ternary structure-activity ratio. α , β , γ With shape parameters λ This eliminates the need to retrain the entire model, improving update efficiency.

[0042] To more clearly illustrate the specific embodiments of the present invention, several examples are provided below: Example 1: This embodiment takes a 1186km medium-pressure gas pipeline network in the main urban area of ​​a prefecture-level city as the evaluation object, and divides it into 11860 pipeline segments of 100m in length. The samples include three types of materials: PE pipe, seamless steel pipe and gray cast iron pipe, with a service life of 1-35 years. A total of 192 leakage accidents have occurred in the past 10 years.

[0043] Step 1: Basic Data Preparation Ternary structure-property factor data were collected from all samples. Data sources included gas pipeline GIS system, SCADA system, pipeline integrity inspection report, leakage accident ledger, urban geographic information system, and construction reporting system. The data items covered 7 intrinsic structure-property factors, 4 environmental regulation factors, and 3 operating condition regulation factors.

[0044] Step 2: Ternary factorization and SAI calculation Standardization was performed on all factors: Categorical factors (pipe material, interface type, anti-corrosion coating type, weld quality grade, and soil corrosivity) were assigned values ​​based on leakage tendency, with a range of [0,1], where higher leakage tendency resulted in larger values; numerical factors were standardized using a min-max method within a [0,1] range. The weights of the structure-activity relationship factors were determined using the F-significance test combined with the variance contribution rate method. ω i The results are as follows: pipe material 0.2, service life 0.25, nominal wall thickness 0.05, interface type 0.15, anti-corrosion layer type 0.15, weld quality grade 0.1, and wall thickness residual rate 0.1. Using data from 192 historical leakage accidents as the fitting set, and aiming to maximize the AUC (discrimination value) of the leakage samples, nonlinear correction coefficients were obtained through nonlinear least squares fitting. μ =0.12.

[0045] Taking pipe segment ID: G-0815 as an example, calculate the SAI value: (1) Standardized values ​​of the structure-activity factor: material f 1=1.0, Service life f 2 = 0.853, nominal wall thickness f 3=0.9, Interface Type f 4=1.0, anti-corrosion layer f 5=1.0, weld quality f 6 = 0.8, wall thickness surplus rate f 7 = 0.3; (2) Arithmetic mean of standardized values ​​of ontological factors - f ( X ): (1.0 + 0.853 + 0.9 + 1.0 + 1.0 + 0.8 + 0.3) / 7 = 0.836; (3) Coefficients of Discrete δ ( X ): ; (4) Linear part: 0.2×1.0+0.25×0.853+...+0.1×0.3=0.868; (5) Nonlinear correction: e -0.12×0.276=0.967; (6) Final SAI value: 0.868 × 0.967 = 0.839.

[0046] Step 3: Layered Mixed Sampling Preprocessing The 11,860 samples were divided into 5 intervals based on their SAI values, and stratified pooled sampling was performed on each interval: (1) Low SAI range (SAI<0.2): Oversampling factor 5, undersampling rejection rate 5%; (2) In the interval 0.2≤SAI<0.4: oversampling factor 4, undersampling rejection rate 8%; (3) In the interval 0.4≤SAI<0.6: oversampling factor 3, undersampling rejection rate 10%; (4) In the interval 0.6≤SAI<0.8: oversampling factor 2, undersampling rejection rate 12%; (5) High SAI range (SAI≥0.8): Oversampling factor 2, undersampling rejection rate 15%; After sampling, a class-balanced standardized dataset is obtained, with a ratio of leaked samples to non-leaked samples of 1:1.2.

[0047] Step 4: Feature Selection and Optimal Subset Extraction Perform three filtering steps sequentially on the standardized dataset: (1) Pearson correlation redundancy removal: Redundant factors with an absolute correlation coefficient > 0.8 were removed. No factors were removed in this screening. (2) Significance screening by one-way ANOVA: Non-significant factors with p>0.05 were removed. All factors in this screening met the condition of p<0.05 and were not removed. (3) LASSO regression dimensionality reduction: the optimal regularization coefficient was determined to be 0.001 through ten-fold cross-validation. All factor regression coefficients were not 0, and the optimal core feature subset containing all 14 factors was finally obtained.

[0048] Step 5: Fitting the ternary explicit structure-activity constitutive model The weights of the environmental adjustment factor and the operating condition adjustment factor were calculated using the random forest algorithm, and the results are as follows: (1) Weights of environmental regulation factors η j Soil corrosivity 0.3, population density within 50m of the pipeline 0.25, annual number of third-party constructions within 100m of the pipeline 0.35, building density 0.1; (2) Weight of operating condition adjustment factor θ kAverage operating pressure: 0.4; monthly flow fluctuation coefficient: 0.3; annual cumulative pressure exceeding limit: 0.3. Taking pipe section G-0815 as an example, the comprehensive value of the factors is calculated as follows: (3) Standardized values ​​of environmental regulation factors: soil corrosivity 0.9, population density 0.8, number of third-party constructions 0.95, building density 0.85; E adj =0.3×0.9+0.25×0.8+0.35×0.95+0.1×0.85=0.888; (4) Standardized values ​​of operating condition adjustment factors: average operating pressure 0.7, monthly flow fluctuation coefficient 0.6, annual pressure over-limit cumulative duration 0.8; O adj =0.4×0.7+0.3×0.6+0.3×0.8=0.7.

[0049] The model parameters were optimized using the Bayesian joint fitting method, and the following results were obtained: α =1.25、 β =0.82、 γ =0.63、 λ =1.12.

[0050] Leakage probability calculation for pipe section G-0815: α SAI+ β · E adj +γ· O adj =1.25×0.839+0.82×0.888+0.63×0.7=2.218 P leak =1-exp(-(2.218) 1.12 =1-exp(-2.440)=0.913.

[0051] Step 6: Calculation of the severity of the leakage consequences Using historical leakage accident data as the fitting set, and aiming to minimize the RMSE between the predicted and actual leakage consequences, a structure-effect correction coefficient k=1.15 was obtained. Leakage orifice area correction for pipe section G-0815: pipe diameter 100mm, baseline leakage orifice area... A base =0.0002m 2 After correction A leak =0.0002×e 1.15×0.839 =0.000525m 2 .

[0052] Based on the corrected pore area, the calculated leakage rate is 0.28 m. 3 / s, the downwind impact distance corresponding to the lower explosive limit of methane (5%) is 82m; the severity values ​​of the three sub-models are: casualties 0.95, economic losses 0.9, and diffusion impact 0.85.

[0053] The combined weighting method was used to obtain the combined weights of the sub-models: casualties 0.52, economic losses 0.28, and diffusion impact 0.2; the overall severity of the consequences S = 0.52 × 0.95 + 0.28 × 0.9 + 0.2 × 0.85 = 0.916.

[0054] Step 7: Comprehensive Risk Assessment and Contribution Decomposition Comprehensive risk value of pipe section G-0815 R =0.913×0.916=0.836, which is an extremely high risk level and requires immediate rectification.

[0055] The contribution of each factor was calculated using the formula for contribution without singular values. After normalization, the contribution ratio of the core risk was obtained as follows: service life 24.2%, number of third-party constructions 18.5%, and pipeline material 15.3%, which is consistent with the statistical results of the dominant factors of the overall pipeline leakage risk.

[0056] Step 8: Model Validation and Dynamic Update The dataset was divided into a training set and an independent test set in an 8:2 ratio. Ten-fold cross-validation and independent test set validation were used to verify the model performance. The results are shown in the table below.

[0057] The verification results show that the method of the present invention is significantly superior to the prior art in terms of accuracy and interpretability, and meets the model qualification criteria.

[0058] The trigger conditions for dynamic model updates are set as follows: a fixed update cycle of 12 months, or ≥20 new leaked samples (10% of the original 192 leaked training samples), and parameter updates are completed using incremental training.

[0059] Example 2: Industrial applications of the method of this invention The method of this invention provides a standardized API interface that supports mainstream data formats such as CSV, JSON, and Shapefile. It can be seamlessly embedded into the existing safety management and control platform of urban gas companies and can directly read the existing data of the gas pipeline network GIS / SCADA system without the need for additional hardware acquisition equipment.

[0060] The platform outputs include: pipeline segment risk classification report, high-risk pipeline segment list, and risk contribution traceability report, which can directly guide gas companies in pipeline network operation and maintenance, hidden danger rectification and risk control. The time for a single model incremental update is ≤2 hours, which is far less than the 12 hours for full model retraining, meeting the timeliness requirements of engineering operation and maintenance.

[0061] Finally, any aspects of this invention not fully described herein utilize existing products and mature technical means.

[0062] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for assessing urban gas leakage risk based on quantitative structure-property relationships, characterized in that, Includes the following steps: S1. Constructing a ternary structure-activity factor system and a nonlinear leakage structure-activity index (SAI) with discrete coefficient correction: Urban gas leakage-related factors are classified into three categories: pipeline structure, environment, and operating conditions. Standardized quantitative rules conforming to national standards are formulated, and the standardized values ​​of each factor and the pipeline leakage tendency are calculated. SAI The index is used for subsequent sampling and consequence correction; S2, based on SAI Layered mixed sampling preprocessing: using the results obtained in step S1 SAI The index combines historical data of ternary structure-activity factors with leakage accident data according to... SAI The values ​​are stratified, and oversampling of leaked samples and undersampling of non-leaked samples are performed on each stratum to obtain a class-balanced standardized dataset for feature selection in step S3. S3. Structure-activity factor feature selection and optimal subset extraction: Based on the standardized dataset obtained in step S2, correlation redundancy removal, significance screening and LASSO regression dimensionality reduction are performed in sequence to remove redundant and non-significant factors and obtain the optimal core feature subset that is strongly correlated with leakage risk for use in step S4 model construction. S4. Construct a ternary explicit structure-effect constitutive model and fit the parameters: Using the optimal core feature subset obtained in step S3 as input, construct a ternary explicit structure-effect constitutive model based on the improved Weibull distribution, and use the Bayesian joint fitting method to globally optimize the model parameters to obtain the annual leakage probability of the pipeline section for use in the comprehensive risk calculation in step S6. S5, Building Coupling SAI Leakage consequence severity assessment model: using the results obtained in step S1 SAI The index-corrected benchmark leakage hole area is used to calculate the leakage consequence parameters by combining the small hole leakage model and the Gaussian diffusion model. The comprehensive consequence severity value is obtained by the combined weighting method and used for the comprehensive risk calculation in step S6. S6. Comprehensive Risk Assessment and Decomposition of Explicit Contribution Without Anomalies: Based on the probability of leakage obtained in step S4 and the comprehensive severity of consequences obtained in step S5, calculate the comprehensive risk value of the pipeline segment and classify the risk level. Construct a full-factor contribution formula without numerical anomalies to obtain the risk contribution ratio of each factor and realize the quantitative tracing of risk causes. S7. Model Validation and Dynamic Update: Divide the standardized dataset obtained in step S2 into training and test sets to validate model performance. Update model parameters using incremental training according to preset trigger conditions to ensure long-term model effectiveness.

2. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S1: The structural performance factors include seven items: pipe material, service life, nominal wall thickness, interface type, anti-corrosion coating type, weld quality grade, and wall thickness surplus rate; the environmental regulation factors include four items: soil corrosivity, population density within 50m of the pipeline, number of third-party constructions within 100m of the pipeline per year, and building density; and the operating condition regulation factors include three items: average operating pressure, monthly flow fluctuation coefficient, and cumulative duration of annual pressure exceedance. Categorical factors were standardized using the leakage propensity level assignment method, and numerical factors were standardized using the [0,1] interval min-max method. The weights of ontological structure-activity factors were determined by combining the F-significance test at a 95% confidence level with the variance contribution rate method, ωi=SSi / SST, where SSi is the sum of squares of deviations from the mean of the i-th factor, and SST is the sum of squares of deviations from the mean of all factors that passed the test, with a test threshold of p<0.

05. Nonlinear SAI The calculation formula is: in, ω i Let i be the weight of the i-th ontology structure-activity factor. fi ( Xi ) is the factor standardization function, with a standardization interval of [0,1]. μ This is a nonlinear correction coefficient, with a value ranging from 0.05 to 0.

2. δ ( X ) is the dispersion coefficient of the ontological structure-activity factor, which characterizes the degree of dispersion of the standardized values ​​of the ontological factor.

3. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S2: Data preprocessing includes one-hot encoding of categorical factors, filling missing values ​​with the mean, removing outliers using the 3σ criterion, and min-max normalization in the [0,1] interval; according to SAI The value divides the sample into 5 intervals: SAI <0.2, 0.2≤ SAI <0.4, 0.4≤ SAI <0.6, 0.6≤ SAI <0.8、 SAI ≥0.8; ADASYN algorithm: nearest neighbor number k=5, oversampling ratio k s =5-6× p s ,in p s This represents the interval leakage rate. k s ∈[2,5], the corresponding multipliers for the 5 intervals are 5, 4, 3, 2, and 2 respectively; The Tomek Links algorithm has a neighborhood of 3, and the proportion of redundant boundary samples removed does not exceed 15%.

4. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S3: Correlation redundancy removal was performed using Pearson analysis, with a threshold of 0.8 for the absolute value of the correlation coefficient. The significance threshold for one-way ANOVA was p < 0.05; LASSO regression uses 10-fold cross-validation between the predicted and actual leakage probabilities to determine the minimum RMSE and then eliminates factors with a regression coefficient of 0.

5. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S4: Comprehensive value of environmental regulation factors: Overall value of operating condition adjustment factor: in, η j For the first j The weights of environmental regulation factors, θ k For the first k The weights of the operating condition adjustment factors were all calculated using the random forest algorithm based on the feature importance and then normalized to the [0,1] interval using min-max. The parameters for the random forest algorithm are: 50-200 decision trees, maximum depth 6-12, and the number of features is the square root of the number of corresponding factors, rounded up. The calculation formula for the ternary explicit structure-activity constitutive model is: in, P leak This represents the annual leakage probability of the pipeline section. E adj This is the comprehensive value of environmental regulation factors. O adj This is the comprehensive value of the operating condition adjustment factor; α , β , γ The ternary structure-activity ratio, λ For shape parameters; Bayesian joint fitting method was used to... α , β , γ , λ Perform global optimization to establish an explicit quantitative mapping from structure-activity factors to leakage probability; in P leak denoted as the annual leakage probability of the pipeline section, α, β, and γ are the ternary structure-activity coefficients, and λ is the shape parameter; The Bayesian joint fitting uses the TPE sampling method, with fitting boundaries of α∈[0.5,2.0], β∈[0.3,1.5], γ∈[0.2,1.2], and λ∈[0.8,1.5]. The termination condition is a maximum of 50 iterations or a validation set AUC convergence threshold ≤ 0.

001.

6. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S5: SAI The corrected formula for the leakage hole area is: Where k is the structure-performance correction factor, ranging from 0.8 to 1.5, and Abase is the baseline leakage orifice area, taken as 2% of the pipe diameter; The calculation formula for the orifice leakage model is: in C d =0.6 is the flow coefficient. ρ =0.717kg / m 3 The density of methane at room temperature and pressure, Δ P This is the difference between the pipeline's operating pressure and the standard atmospheric pressure. The Gaussian diffusion model uses Pasquill D-level neutral atmospheric stability, with the methane explosion limit taken as 5%~15%; The combined weighting method uses a combination of AHP and entropy weighting. ω com =0.6 ω AHP +0.4 ω entropy ,in ω AHP For AHP subjective weights, ω entropy As the objective weight of entropy, the consistency ratio CR of the judgment matrix constructed by the AHP method is <0.

1.

7. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S6: Comprehensive risk value of pipeline section R= P leak ×S, where S is the overall severity of the consequences; According to the ALARP criterion of GB / T 27921-2011, the risk levels are divided into: low risk R<0.1, medium risk 0.1≤R<0.3, high risk 0.3≤R<0.7, and very high risk R≥0.7; The formula for the contribution of all factors to the risk of no numerical anomalies is: ontological factors i Contribution: Environmental factors j Contribution: Operating factors k Contribution: After calculation, the absolute values ​​of all contributions are taken and normalized by min-max in the [0,1] interval to obtain the risk contribution ratio of each factor.

8. The method for assessing urban gas leakage risk based on quantitative structure-property relationship according to claim 1, characterized in that, In step S7: The model evaluation metrics include AUC, recall, accuracy, MAE, and RMSE. The passing criteria are AUC ≥ 0.9 on the test set and recall ≥ 90% for leaked samples. The model update is triggered when: the fixed period does not exceed 12 months, or the number of newly leaked samples is ≥ 10% of the original number of leaked training samples; The incremental training method optimizes only four parameters: α, β, γ, and λ. It adopts mini-batch gradient descent with a batch size of 32 to 64 and a learning rate of 0.01 to 0.

1. The termination condition is that the loss function convergence threshold is ≤1e-4 or the maximum number of iterations is 200. The loss function is the mean square error (MSE) between the predicted and actual leakage probability values.

9. A risk assessment system for urban gas leakage based on quantitative structure-property relationships, characterized in that, Includes the following modules connected in sequence: Construction of ternary structure-activity factors and SAI The calculation module is used to divide the "body-environment-operating condition" three-element structure-property factor system, formulate standardized quantification rules, and calculate the standardized values ​​of each factor and the leakage tendency of the pipeline body. SAI The index is output to the hierarchical hybrid sampling preprocessing module and the leakage consequence assessment module; The hierarchical hybrid sampling preprocessing module is used to receive... SAI The index collects corresponding historical data and leakage accident data, performs preprocessing, and then... SAI Value stratification is performed to perform mixed sampling to obtain a class-balanced standardized dataset, which is then output to the feature selection module and the model validation and dynamic update module. The feature selection module receives a standardized dataset and sequentially performs correlation redundancy removal, significance selection, and LASSO regression dimensionality reduction to obtain the optimal core feature subset, which is then output to the ternary structure-efficacy constitutive model fitting module. The ternary structure-effect constitutive model fitting module is used to receive the optimal core feature subset, construct and fit the ternary explicit structure-effect constitutive model, obtain the annual leakage probability of the pipeline section, and output it to the comprehensive risk assessment and contribution decomposition module. The leakage consequence assessment module is used to receive... SAI The index is used to correct the leakage hole area and calculate the leakage consequence parameters. The comprehensive consequence severity value is obtained through the combined weighting method and output to the comprehensive risk assessment and contribution decomposition module. The comprehensive risk assessment and contribution decomposition module is used to receive the probability of leakage and the severity of comprehensive consequences, calculate the comprehensive risk value and classify the level, and decompose the risk contribution ratio of each factor. The model validation and dynamic update module is used to receive standardized datasets, validate model performance, and update model parameters according to preset trigger conditions.

10. The urban gas leakage risk assessment system based on quantitative structure-property relationship according to claim 9, characterized in that: In the ternary structure-activity factor construction and SAI calculation module, the SAI calculation formula is as follows: ； In the formula: ωi is the weight of the i-th ontology structure-activity factor, determined by a combination of a 95% confidence level F-significance test and the variance contribution rate method, ωi= SS i / SS T The significance threshold for the F-test was p < 0.

05. fi ( Xi ) is the factor standardization function. Categorical factors are standardized using the leakage tendency level assignment method, and numerical factors are standardized using the [0,1] interval min-max standardization method, with the standardization interval being [0,1]. μ The nonlinear correction coefficient ranges from 0.05 to 0.

2. It is determined by fitting historical leakage data using the nonlinear least squares method, with the goal of maximizing the AUC of the leakage sample discrimination. In the ternary structure-effect constitutive model fitting module, the model formula is: ； Bayesian joint fitting was performed using the TPE sampling method; In the leakage consequence evaluation module, the leakage hole area correction formula is: The AHP and entropy weighting method are used for weighting; where: k is the structure-activity relationship correction coefficient, with a value range of 0.8~1.5, which is determined by fitting historical leakage data using the nonlinear least squares method, with the goal of minimizing the RMSE between the predicted and actual leakage consequences. A base The baseline leakage orifice area is taken as 2% of the pipe diameter; In the comprehensive risk assessment and contribution decomposition module, the comprehensive risk value R = P leak × S , R ∈[0,1], classified into 4 risk levels according to the ALARP criterion, the contribution formula has no denominator term to avoid calculating singularities; In the model validation and dynamic update module, the model qualification criteria are AUC ≥ 0.9 on the test set and recall rate of leaked samples ≥ 90%. Incremental training only optimizes the ternary structure-activity ratio and shape parameters.

Images