Machine learning based safety assessment method for new expansion site of underground steel tank

Abstract

The application discloses a new expansion site safety evaluation method for machine learning of underground steel storage tanks, and comprises the following steps: establishing a seepage and leakage risk evaluation case library of underground steel storage tanks, screening seepage and leakage risk evaluation indexes, assigning indexes and optimizing weights, constructing a seepage and leakage risk evaluation model to calculate risk indexes of each storage tank, and screening out storage tanks greater than or equal to a first preset threshold; screening underground water pollution risk evaluation indexes, assigning scores, determining weights, constructing an underground water pollution risk evaluation model, and calculating water pollution risk indexes of underground steel storage tanks with seepage and leakage risk indexes greater than or equal to the first preset threshold; judging whether the water pollution risk indexes of the underground steel storage tanks with the water pollution risk indexes greater than or equal to the first preset threshold are greater than or equal to a second preset threshold; if yes, the site is unsafe; and if no, the site is safe; and the application realizes the determination of the safety of the site related to the underground steel storage tanks by constructing two-level risk evaluation models of seepage and leakage and underground water pollution and setting thresholds.

Description

Technical Field

[0001] This invention relates to the field of groundwater pollution prevention and control technology, and more specifically to a method for safety assessment of new, renovated and expanded land use for underground steel storage tanks using machine learning. Background Technology

[0002] Newly built, renovated, or expanded industrial and mining land is crucial for preventing new groundwater pollution, controlling the risk of pollution spread, and eliminating risks to human health. These enterprises span a wide range of industries, are numerous, and employ complex production processes. Significant differences exist between these enterprises in their processes, raw materials, and the types of toxic and hazardous substances they contain. The highly concealed nature of leaks from underground steel storage tanks, coupled with varying levels of pollution prevention measures among these enterprises, creates a substantial risk of groundwater pollution spread and human health risks, making it difficult to guarantee the safety of newly built, renovated, or expanded industrial and mining land. Existing methods cannot conduct leak risk assessments based on the conditions under which leaks from underground steel storage tanks occur, nor can they conduct groundwater pollution risk assessments based on the risk source-material migration pathway-sensitive receptor risk formation factors. Consequently, it is impossible to couple leak risk assessment and pollution risk assessment to achieve a safety assessment of newly built, renovated, or expanded land. Therefore, there is an urgent need to establish a safety assessment method for newly built, renovated, or expanded land involving underground steel storage tanks, in order to comprehensively support enterprises in accurately formulating groundwater pollution prevention and control measures and to facilitate the scientific supervision of newly built, renovated, or expanded industrial and mining land by environmental protection departments. Summary of the Invention

[0003] The purpose of this invention is to provide a machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks. By constructing a two-level risk assessment model for seepage and groundwater pollution and setting thresholds, the safety of land use involving underground steel storage tanks can be accurately determined.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A machine learning-based safety assessment method for new construction, renovation, and expansion projects involving underground steel storage tanks, comprising: S1. Construct a case library for leakage risk assessment of underground steel storage tanks, screen out leakage risk assessment indicators and assign scores and grades them, determine the initial weights of the leakage risk assessment indicators and optimize them. Using leakage risk assessment indicators and optimized weights, a leakage risk assessment model is constructed, and a first preset threshold is set. S2. Using the leakage risk assessment model, calculate the leakage risk index of each underground steel storage tank one by one, and screen out the underground steel storage tanks whose leakage risk index is greater than or equal to the first preset threshold. S3. Select groundwater pollution risk assessment indicators from the leakage risk assessment case library, assign scores and classify them, and determine the weight of the groundwater pollution risk assessment indicators. A groundwater pollution risk assessment model was constructed using groundwater pollution risk assessment indicators and weights, and a second preset threshold was determined. S4. Using the groundwater pollution risk assessment model, calculate the water pollution risk index of each underground steel storage tank whose leakage risk index is greater than or equal to the first preset threshold. S5. Determine whether the water pollution risk index of underground steel storage tanks that are greater than or equal to the first preset threshold is greater than or equal to the second preset threshold; if so, the land use is unsafe; otherwise, the land use is safe.

[0005] Furthermore, in S1, the leakage risk assessment case library includes: multiple information items; The leakage risk assessment indicators include: Properties of stored toxic and hazardous substances: type of substance, quantity of substance, toxicity of substance, corrosiveness of substance; Characteristics of storage tank construction and operation: number of tank layers, service life, cathodic protection system, anti-corrosion materials, and anti-seepage materials; Groundwater pollution prevention measures: leakage monitoring devices, remote monitoring systems, routine inspection and maintenance, environmental emergency plans, environmental emergency supplies, daily inspection records, technical training, and surrounding monitoring wells; External environment: groundwater level, groundwater corrosivity, soil corrosivity.

[0006] Furthermore, in step S1, a leakage risk assessment model is constructed using leakage risk assessment indicators and optimized weights; specifically, the PSO-BPNN algorithm is used to construct the leakage risk assessment model. The optimization process of the PSO-BPNN algorithm includes: Using fitting coefficient, mean absolute error, and root mean square error as performance evaluation indicators, we optimized and adjusted the inertia weight, individual learning factor, social learning factor of PSO, and the training function, number of hidden layer neurons, and learning rate hyperparameter of BPNN.

[0007] Furthermore, the optimization parameters of the PSO-BPNN algorithm are specifically as follows: The training function is trainlm; the number of hidden layer neurons is 10; the learning rate is 0.1; the inertia weight is 0.6; the individual learning factor is 2; and the social learning factor is 2.

[0008] Furthermore, in S1, the expression for the leakage risk assessment model is: LRI =0.078 SC +0.072 SQ +0.125 ST +0.127 SCOR +0.054TL +0.044 SY +0.034 CPS +0.074 ACM +0.061 ASM +0.051 LMD +0.035 RMS +0.034 RM +0.018 EEP +0.016 EES +0.021 DIR +0.023 TT +0.038 PMW +0.035 GWD +0.029 GWCOR +0.031 SOCOR in, LRI This indicates the risk index of leakage. SC Indicates the type of substance. SQ Indicates the quantity of matter. ST Indicates the toxicity of a substance. SCOR Indicates the corrosiveness of a substance. TL Indicates the number of tank layers. SY Indicates the service life. CPS Indicates cathodic protection system, ACM Indicates anti-corrosion materials. ASM Indicates waterproof material, LMD Indicates a leakage monitoring device. RMS Indicates a remote monitoring system. RM This indicates routine inspection and maintenance. EEP This indicates the environmental emergency response plan. EES Indicates environmental emergency supplies. DIR This indicates daily inspection records. TT Indicates technical training, PMW Indicates surrounding monitoring wells, GWD Indicates the groundwater level. GWCOR Indicates the corrosiveness of groundwater. SOCOR It indicates soil corrosivity.

[0009] Furthermore, in S3, the water pollution risk indicators include: Risk sources: leakage risk index, solubility of substances, volatility of substances; Material migration pathways: wet-dry index, vadose zone lithology, groundwater depth, aquifer lithology; Sensitive receptors: People who come into contact with the storage tank, and water sources within 1 km downstream.

[0010] Furthermore, in step S3, the weights of the groundwater pollution risk assessment indicators are determined using the analytic hierarchy process (AHP), specifically including: The judgment matrix is ​​constructed using the 1-9 scaling method; Perform a consistency check on the judgment matrix; The weights of the groundwater pollution risk assessment indicators were obtained through normalization.

[0011] Furthermore, the first preset threshold is 4.1 points, and the second preset threshold is 7.5 points.

[0012] Furthermore, in S3, the expression for the groundwater pollution risk assessment model is: PRI =0.223 LRI +0.081 SS +0.082 SV +0.093 DWI +0.059 VZL +0.076 GD +0.115 AL +0.141 CSTP+0.130WS in, PRI Indicates the pollution risk index; LRI Indicates the risk index of leakage; SS Indicates the solubility of a substance; SV Indicates the volatility of a substance; DWI Indicates the dryness index; VZL Indicates the lithology of the vadose zone; GD Indicates the depth of groundwater; AL Indicates the lithology of the aquifer; CSTP This refers to the group of people who come into contact with the storage tank; WS This indicates a water source within a 1km downstream radius.

[0013] According to specific embodiments provided by the present invention, the present invention has the following technical effects compared to the prior art: This invention utilizes the PSO-BPNN algorithm to objectively establish a leakage risk assessment model and set a first preset threshold. This allows for rapid screening of high-risk objects with a high probability of leakage from a large number of storage tanks, achieving initial risk focus and narrowing the scope of investigation, effectively improving assessment efficiency. Furthermore, by constructing a groundwater pollution risk assessment model and setting a second preset threshold, it provides a deeper assessment of the actual severity of groundwater pollution caused by a leak. When establishing the groundwater pollution risk assessment model, both human health risks and pollution spread risks are considered simultaneously, selecting individuals who come into contact with the storage tanks and those within a 1km downstream radius. The two indicators for internal water sources compensate for the shortcomings of existing methods that only consider single risks to human health or pollution spread, and can be used for groundwater pollution risk assessment of pore water in loose rocks. It not only scientifically couples the two core risk dimensions of "leakage probability" and "pollution hazard" to avoid the one-sidedness of single assessment, but also transforms complex ecological risk issues into clear and operable decision-making basis through quantitative models and clear threshold standards. Ultimately, it can accurately identify storage tanks that truly threaten land use safety, and provide strong technical support for environmental supervision, risk early warning, and rational planning and safe use of land. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] The following description, in conjunction with the accompanying drawings, further illustrates the machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks provided by this invention. Figure 1 This is a schematic diagram of the overall process of the safety assessment method for new, renovated and expanded land involving underground steel storage tanks provided by the present invention using machine learning; Figure 2 This is a graph showing the optimization curve of the training function of the PSO-BPNN hybrid machine learning algorithm in this invention; Figure 3 This is a curve showing the optimization of the number of hidden layer nodes in the PSO-BPNN hybrid machine learning algorithm of this invention; Figure 4 This is a learning rate optimization curve of the PSO-BPNN hybrid machine learning algorithm in this invention; Figure 5 This is a comparison chart of the optimal results of the PSO-BPNN hybrid machine learning algorithm in this invention. Detailed Implementation

[0016] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0017] To better understand the purpose, structure, and function of this invention, the invention will be described in further detail below with reference to the accompanying drawings.

[0018] Example 1 like Figure 1 As shown, this invention provides a machine learning-based safety assessment method for new, renovated, or expanded land use involving underground steel storage tanks, comprising: I. Construct a case library of leakage risk assessments involving underground steel storage tanks, screen out leakage risk assessment indicators and assign scores and grades them, determine the initial weights of the leakage risk assessment indicators and optimize them. Using leakage risk assessment indicators and optimized weights, a leakage risk assessment model is constructed, and a first preset threshold is set. 1. The process of constructing the leakage risk assessment case library is as follows: By collecting soil pollution hazard investigation reports and enterprise operation and management information from nine key industry enterprises (involving four industries: smelting, coking, chemical and oil and gas extraction), and taking the production equipment in each enterprise as the research object, a case library for seepage and leakage risk assessment was constructed. The case library contains 48 information items for 397 storage tanks, including equipment name, company name, plant location, substance type, substance molecular weight, substance color, substance odor, substance quantity, substance toxicity, substance corrosivity, substance thermal conductivity, number of tank layers, service life, installation type, cathodic protection system, anti-corrosion materials, anti-seepage materials, leakage monitoring devices, remote monitoring system, daily inspection and maintenance, ground hardening, designated maintenance personnel, environmental emergency plan, environmental emergency supplies, daily inspection records, technical training, surrounding monitoring wells, groundwater level, groundwater hydrochemical type, groundwater flow velocity, groundwater hydraulic gradient, groundwater corrosivity, soil corrosivity, sub-category of land use, environmental pollution accidents, environmental violations, substance solubility, substance volatility, wet / dry index, vadose zone lithology, groundwater flow velocity, microbial activity, groundwater depth, plant root action, aquifer lithology, specific surface area of ​​aquifer medium, people who come into contact with the storage tank, and water sources within 1km downstream.

[0019] 2. The specific process for selecting leakage risk assessment indicators is as follows: Based on the characteristics of underground steel storage tanks and the conditions under which leakage occurs, and using expert consultation and scoring methods, 15 industry experts were required to select 20 indicators from 30 potential indicators and rank them by importance. This aimed to reduce individual expert bias towards potential indicators, enhance the authority and rationality of the selected indicators, reduce the workload of indicator information collection, and increase the efficiency of leakage risk assessment. Specifically, the most important indicator was assigned 20 points, the second most important indicator 19 points, the third most important indicator 18 points, the fourth most important indicator 17 points, the fifth most important indicator 16 points, the sixth most important indicator 15 points, the seventh most important indicator 14 points, the eighth most important indicator 13 points, the ninth most important indicator 12 points, the tenth most important indicator 11 points, the eleventh most important indicator 10 points, the twelfth most important indicator 9 points, the thirteenth most important indicator 8 points, the fourteenth most important indicator 7 points, the fifteenth most important indicator 6 points, the sixteenth most important indicator 5 points, the seventeenth most important indicator 4 points, the eighteenth most important indicator 3 points, the nineteenth most important indicator 2 points, and the twentieth most important indicator 1 point. As shown in Table 1, 20 potential indicators were selected as leakage risk assessment indicators, including substance type, substance quantity, substance toxicity, substance corrosivity, number of tank layers, service life, cathodic protection system, anti-corrosion materials, anti-seepage materials, leakage monitoring devices, remote monitoring system, daily maintenance, environmental emergency plan, environmental emergency supplies, daily inspection records, technical training, surrounding monitoring wells, groundwater level, groundwater corrosivity, and soil corrosivity (Table 1).

[0020] Table 1 Importance of Potential Indicators for Leakage Risk Assessment

[0021] Note: Importance = S i / [(1+2+3+4+5+6+7+8+9+10+11+12+13+14+15+16+17+18+19+20)×15], where S i For 15 experts on the first i The sum of the scores for each potential indicator.

[0022] 3. The specific process for constructing the leakage risk assessment index system is as follows: Based on four aspects—the nature of the stored toxic and hazardous substances, the characteristics of the tank construction and operation, groundwater pollution prevention measures, and the external environment—the aforementioned 20 leakage risk assessment indicators are classified to form a tank leakage risk assessment indicator system (Table 2).

[0023] Table 2. Risk Assessment Index System for Storage Tank Leakage

[0024] (1) The properties of stored toxic and harmful substances include four indicators: substance type, substance quantity, substance toxicity, and substance corrosivity.

[0025] ① Substance type: refers to the category to which the stored toxic and harmful substances belong, such as organic matter or inorganic matter.

[0026] ② Quantity of substances: refers to the quantity of toxic and harmful substances stored, usually measured in units of mass (kilograms) or volume (cubic meters).

[0027] ③ Toxicity of substances: refers to the degree of toxicity of a substance to an organism (such as a human, animal or plant), usually measured by indicators such as toxicity level, median lethal dose (LD50) or median lethal concentration (LC50).

[0028] ④ Corrosiveness of substances: refers to the ability of toxic and harmful substances to cause chemical damage and erosion from the inside of the storage tank.

[0029] (2) The characteristics of the construction and operation of storage tanks include five indicators: number of tank layers, service life, cathodic protection system, anti-corrosion materials, and anti-seepage materials.

[0030] ① Number of tank layers: refers to the number of structural layers of a tank designed and constructed in the vertical direction, such as single-layer tanks and double-layer tanks.

[0031] ② Service life: refers to the actual time when the storage tank begins to be used, that is, the duration of operation or use, in years.

[0032] ③ Cathodic protection system: refers to a protection system established through electrochemical principles (such as sacrificial anode method and impressed current method) to mitigate the corrosion of storage tanks.

[0033] ④ Anti-corrosion materials: These are materials that are coated or lined on the inner and outer walls of storage tanks to protect the surface of the tank from corrosion by the medium and environmental erosion, such as anti-corrosion coatings and fiberglass linings.

[0034] ⑤ Impermeable materials: These are materials laid at the bottom of the storage tank foundation or in the impermeable pool to prevent toxic and harmful substances leaking from the storage tank from seeping into the soil or groundwater, such as high-density polyethylene film and bentonite waterproof blanket.

[0035] (3) Groundwater pollution prevention measures include eight indicators such as leakage monitoring devices, remote monitoring systems, daily inspection and maintenance, environmental emergency plans, environmental emergency supplies, daily inspection records, technical training, and surrounding monitoring wells.

[0036] ① Leakage monitoring device: refers to equipment installed at the bottom of the storage tank, the seepage prevention layer or the surrounding area, used to detect the leakage of toxic and harmful substances in real time or periodically, such as leakage sensors and detection wells.

[0037] ② Remote monitoring system: refers to a system that uses the Internet of Things and communication technology to remotely monitor and transmit data on the operating status of storage tanks, leakage monitoring data, and surrounding environmental parameters in real time.

[0038] ③ Routine inspection and maintenance: refers to the daily work of inspecting, repairing, cleaning and maintaining the storage tank body, valves, monitoring devices and other facilities in accordance with the prescribed cycle to ensure their normal operation.

[0039] ④ Environmental emergency response plan: refers to a pre-formulated plan that includes emergency organization, response procedures, and disposal measures to deal with sudden environmental incidents such as tank leaks and fires that may cause soil or groundwater pollution.

[0040] ⑤ Environmental emergency supplies: refers to supplies that are stockpiled in advance to deal with sudden environmental incidents such as tank leaks, such as adsorbents, protective equipment, and emergency monitoring instruments.

[0041] ⑥ Daily inspection records: These are written or electronic documents recorded by staff during daily inspections, documenting the status of storage tanks, the operation of monitoring devices, and the surrounding environment.

[0042] ⑦ Technical training: refers to training conducted to improve the professional capabilities of tank operators, managers, and emergency personnel, covering topics such as the characteristics of toxic and hazardous substances, tank operation and maintenance, pollution prevention and control, and emergency response.

[0043] ⑧ Surrounding monitoring wells: These are well facilities excavated or constructed in the area surrounding the storage tank according to certain layout principles, used to periodically collect soil or groundwater samples to monitor whether they are contaminated.

[0044] (4) The external environment includes three indicators: groundwater level, groundwater corrosivity, and soil corrosivity.

[0045] ① Groundwater level: refers to the elevation of the groundwater level in the area where the storage tank is located relative to the local ground reference level, reflecting the depth of groundwater burial.

[0046] ② Groundwater corrosivity: refers to the ability of groundwater surrounding the storage tank to chemically corrode the metal components of the tank due to the different chemical compositions it contains.

[0047] ③ Soil corrosivity: refers to the ability of the soil in the area where the storage tank is located to corrode the metal components of the storage tank due to the characteristics of pH value, salt content, and moisture content.

[0048] 4. Scoring and grading of leakage risk assessment indicators Primary indicator: Leakage risk (200 points); Secondary indicators: nature of stored toxic and hazardous substances (40 points), characteristics of storage tank construction and operation (50 points), groundwater pollution prevention measures (80 points), external environment (30 points); Level 3 indicators: Level 3 indicators are divided into categories and assigned corresponding scores (Table 3).

[0049] Table 3. Scoring and Classification of Leakage Risk Assessment Indicators

[0050] 5. Determination of initial weights for leakage risk assessment indicators Based on the analytic hierarchy process (AHP), pairwise comparisons of each indicator are performed using the 1-9 scale to determine the importance scores of each indicator, thereby constructing a judgment matrix. The consistency of this judgment matrix is ​​then tested. Consistency indicators are determined (…). The value was 0.0779, and the random consistency index ( The consistency ratio was 1.630. The value of 0.0394 < 0.1, indicating acceptable consistency. Finally, the weights of the leakage risk assessment indicators were obtained through normalization (Table 4).

[0051] Table 4 Initial Weights of Leakage Risk Assessment Indicators

[0052] 6. Performance Comparison of Different Basic Machine Learning Algorithms Using the fitting coefficient (R) 2 The root mean square error (RMSE) and root mean square error (RMSE) were used as performance evaluation metrics to compare the accuracy of five basic machine learning algorithms—Genetic Algorithm (GA), Backpropagation Neural Network (BPNN), Optimized Particle Swarm Optimization (PSO), GA-BPNN, and PSO-BPNN—in predicting the leakage risk of the test centralized storage tanks (Table 5). Typically, R... 2 The larger the value and the smaller the RMSE, the higher the prediction accuracy of the basic machine learning algorithm. Table 5 shows that, compared with the other four basic algorithms, PSO-BPNN has the highest prediction accuracy (R²). 2 The value was 0.980, and the RMSE was 0.186.

[0053] Table 5 Performance of different basic machine learning algorithms

[0054] PSO (Programmatic Search) is a swarm intelligence-based stochastic search method (Equation 1). It overcomes the limitation of BPNN (Background-Based Neural Network) in easily getting trapped in local optima using gradient descent. Through cooperation and information sharing among particles, it efficiently handles complex, multivariable, and nonlinear problems, and has the advantages of simple computation, fast convergence, and good robustness. PSO uses the weights and thresholds of the BPNN neural network as particles. After initializing the population, calculating fitness values, and updating particle velocity and position, it outputs the optimal solutions of the particles, which are then used as the optimal individual weights and thresholds of the BPNN, thus obtaining the output results of the BPNN.

[0055] (1) In the formula: w It is a weight matrix; δ The threshold for each layer of neurons; f It is called an activation function.

[0056] 7. Optimization of PSO-BPNN hybrid machine learning algorithm The PSO-BPNN algorithm was optimized by adjusting three hyperparameters: training functions (trainlm, trainingd, trainingdm, trainingda, trainingdx, trainingrp, trainingcgf, trainingcgp, trainingcgb, trainingscg, trainingbfg, trainingoss), the number of hidden layer neurons (6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16), and the learning rate (0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.10, 0.11, 0.12, 0.13, 0.14, 0.15). The data was divided into a training set:test set ratio of 7:3, with an inertia weight of 0.6 and individual learning factors and social learning factors of 2. R0 was used as the training set. 2 The prediction accuracy of GA-BPNN is evaluated using mean absolute error (MAE) and RMSE as evaluation metrics to determine the optimal hyperparameters of the GA-BPNN algorithm.

[0057] R of the training function "trainlm" 2 R0 (0.964) is higher than other training functions, while MAE (0.112) and RMSE (0.162) are lower than other training functions; R0 for a hidden layer with 10 nodes is higher. 2 The R-value (0.981) is higher than other values, while the MAE (0.071) and RMSE (0.118) are lower; the R-value with a learning rate of 0.1 is lower. 2 (0.972) is higher than other learning rates, while MAE (0.071) and RMSE (0.143) are lower than other learning rates. Figures 2-4Therefore, the training function was ultimately determined to be trainlm, the number of hidden layer nodes was 10, and the learning rate was 0.1.

[0058] In MATLAB R2019b, run the PSO-BPNN hybrid machine learning algorithm with trained hyperparameters until the optimal result is obtained. Figure 5 The optimal training results of the algorithm show that R0 is the optimal value for the test set data. 2 The value was 0.991, the MAE was 0.044, and the RMSE was 0.061.

[0059] 8. Optimization of the weighting of leakage risk assessment indicators Extract the weight coefficients between hidden layer neurons and input layer neurons from the obtained optimal running results. and the weight coefficients between output layer neurons and hidden layer neurons The significance coefficient, correlation index and absolute influence coefficient are calculated (Formula 2-Formula 4). The absolute influence coefficient is the weight of each indicator after optimization (Table 6).

[0060] (2) (3) (4) In the formula, denoted as the significance coefficient; i represents the input unit of the neural network, i=1,…m; j represents the output unit of the neural network, j=1,…n; k represents the hidden layer unit of the neural network, k=1,…p; These are the weighting coefficients between hidden layer neurons and input layer neurons; These are the weighting coefficients between output layer neurons and hidden layer neurons; For related indices; This represents the absolute influence coefficient.

[0061] Table 6. Weights of Optimized Leakage Risk Assessment Indicators

[0062] 9. Construction of Leakage Risk Assessment Model Based on the aforementioned assessment indicators and their weights, a leakage risk assessment model (Formula 5) is constructed, and a leakage risk index score of 4.1 is set as a preset threshold. When the leakage risk index of a storage tank is greater than or equal to 4.1, it is considered to have a high leakage risk; when the leakage risk index of a storage tank is less than 4.1, it is considered to have a low leakage risk.

[0063] LRI=0.078SC+0.072SQ+0.125ST+0.127SCOR+0.054TL+0.044SY+0.034CPS+ 0.074ACM+0.061ASM+0.051LMD+0.035RMS+0.034RM+0.018EEP+0.016EES+0.021DIR+ 0.023TT+0.038PMW+0.035GWD+0.029GWCOR+0.031SOCOR (5) in,LRI Indicates the risk index of leakage; SC Indicates the type of substance; SQ Indicates the quantity of matter; ST Indicates the toxicity of a substance; SCOR Indicates the corrosiveness of a substance; TL Indicates the number of tank layers; SY Indicates the service life; CPS Indicates a cathodic protection system; ACM Indicates anti-corrosion materials; ASM Indicates waterproof material; LMD Indicates a leakage monitoring device; RMS Indicates a remote monitoring system; RM This indicates routine inspection and maintenance; EEP This indicates the environmental emergency response plan; EES Indicates environmental emergency supplies; DIR This indicates daily inspection records; TT Indicates technical training; PMW Indicates surrounding monitoring wells; GWD Indicates the groundwater level; GWCOR Indicates the corrosiveness of groundwater; SOCOR It indicates soil corrosivity.

[0064] 2. Using the leakage risk assessment model, calculate the leakage risk index of each underground steel storage tank, and screen out the underground steel storage tanks whose leakage risk index is greater than or equal to the first preset threshold. This embodiment specifically targets 20 underground steel storage tanks in a newly constructed or expanded land area. A tank leakage risk assessment model is used to calculate the leakage risk index for each tank (Table 7). According to Table 7, the leakage risk index for tanks #2 (sulfuric acid), #1 (hydrochloric acid), #2 (hydrochloric acid), #1 (ethanol), #2 (ethanol), #1 (benzene), #2 (benzene), #3 (benzene), #1 (methanol), #3 (gasoline), and #1 (diesel) is greater than or equal to 4.1. Therefore, these tanks pose a leakage risk, and the company should strengthen its supervision of them.

[0065] Table 7. Risk Assessment Scores for Leakage of the Storage Tanks to be Tested

[0066] Third, from the case library of seepage risk assessment, groundwater pollution risk assessment indicators are selected, scored, and categorized to determine the weight of the groundwater pollution risk assessment indicators. A groundwater pollution risk assessment model was constructed using groundwater pollution risk assessment indicators and weights, and a second preset threshold was determined. 1. Screening of groundwater pollution risk assessment indicators Based on the characteristics of underground steel storage tanks, and using expert consultation and scoring methods, 20 industry experts were asked to select 9 indicators from 16 potential indicators and rank them by importance (Table 8). This aimed to reduce individual expert biases regarding potential indicators, enhance the authority and rationality of the selected indicators, reduce the workload of indicator information collection, and increase the efficiency of pollution risk assessment. The most important indicator was assigned 9 points, the second most important indicator 8 points, the third most important indicator 7 points, the fourth most important indicator 6 points, the fifth most important indicator 5 points, the sixth most important indicator 4 points, the seventh most important indicator 3 points, the eighth most important indicator 2 points, and the ninth most important indicator 1 point. Table 8 shows that the following 9 potential indicators were selected as pollution risk assessment indicators: leakage risk index, substance solubility, substance volatility, wet-dry index, vadose zone lithology, groundwater depth, aquifer lithology, population in contact with the storage tank, and water sources within 1 km downstream.

[0067] Table 8 Potential Indicators for Groundwater Pollution Risk Assessment

[0068] Note: Importance = S i / [(1+2+3+4+5+6+7+8+9)×20], where S i For 20 experts on the first i The sum of the scores of each potential indicator 2. Construction of Groundwater Pollution Risk Assessment Index System The nine pollution risk assessment indicators selected above are classified into three aspects: risk sources, material migration pathways, and sensitive receptors, and a groundwater pollution risk assessment indicator system for storage tanks is constructed (Table 9).

[0069] Table 9 Groundwater Pollution Risk Assessment Index System

[0070] (1) Risk sources include three indicators: leakage risk index, substance solubility, and substance volatility.

[0071] ①Leakage Risk Index: This refers to the level of leakage risk of the storage tank.

[0072] ② Solubility of a substance: refers to the mass of a substance that can be dissolved per unit volume of solvent under certain temperature and pressure. It is usually expressed as the ratio of the maximum mass of a substance that can be dissolved to the mass of the solvent.

[0073] ③ Volatility of a substance: refers to the ability of a liquid substance to transform into a gaseous state at temperatures below its boiling point. It is used to characterize whether a substance is easily volatile and whether it is easily inhaled or comes into contact with the skin.

[0074] (2) The material migration pathways include four indicators: wet-dry index, vadose zone lithology, groundwater depth, and aquifer lithology.

[0075] ①Dryness index: refers to the ratio of rainfall to evaporation (Formula 6).

[0076] (6) in, DWI Indicates the dryness index; This represents the average annual precipitation over many years. This represents the average annual evaporation over many years.

[0077] ② Vadose zone lithology: refers to the lithology with the highest permeability coefficient in the vadose zone.

[0078] ③Groundwater depth: refers to the distance from the ground surface to the water table.

[0079] ④ Aquifer lithology: refers to the lithology with the highest permeability coefficient in loose rock pore aquifers.

[0080] (3) Sensitive receptors include two indicators: people who come into contact with the storage tank and water sources within 1 km downstream.

[0081] ① People who come into contact with the storage tanks: refers to the number of employees and visitors who may come into contact with the storage tanks within the newly built, renovated or expanded land area.

[0082] ②Water sources within 1km downstream: refers to the number of surface water or groundwater sources within 1km of newly built, renovated or expanded land.

[0083] 3. Scoring and grading of groundwater pollution risk assessment indicators Primary indicator: Groundwater pollution risk (90 points); Secondary indicators: Risk source (30 points), substance migration pathway (40 points), sensitive receptor (20 points); Level 3 indicators: Level 3 indicators are categorized and assigned corresponding scores (Table 10).

[0084] Table 10. Scoring and Classification of Groundwater Pollution Risk Assessment Indicators

[0085] 4. The specific process for determining the weights of groundwater pollution risk assessment indicators is as follows: Based on the analytic hierarchy process (AHP), pairwise comparisons of each indicator are performed using the 1-9 scale to determine the importance scores of each indicator, thereby constructing a judgment matrix. The consistency of this judgment matrix is ​​then tested. Consistency indicators are determined (…). The value is 0.011, and the random consistency index ( The consistency ratio was 1.654. The value of 0.0067 < 0.1, indicating acceptable consistency. Finally, the weights of the pollution risk assessment indicators were obtained through normalization (Table 11).

[0086] Table 11 Weights of Groundwater Pollution Risk Assessment Indicators

[0087] 5. Construction of Groundwater Pollution Risk Assessment Model Based on the aforementioned assessment indicators and their weights, a pollution risk assessment model (Formula 7) is constructed, with a pollution risk index score of 7.5 as a preset threshold. If the pollution risk index of any storage tank is greater than or equal to 7.5, the newly constructed or expanded land is considered unsafe; if the pollution risk index of all storage tanks is less than 7.5, the newly constructed or expanded land is considered safe.

[0088] PRI =0.223 LRI +0.081 SS +0.082 SV +0.093 DWI +0.059 VZL +0.076 GD +0.115 AL +0.141 CSTP+0.130WS (7) in, PRI Indicates the pollution risk index; LRI Indicates the risk index of leakage; SS Indicates the solubility of a substance; SV Indicates the volatility of a substance; DWI Indicates the dryness index; VZL Indicates the lithology of the vadose zone; GD Indicates the depth of groundwater; AL Indicates the lithology of the aquifer; CSTP This refers to the group of people who come into contact with the storage tank; WS This indicates a water source within 1km downstream.

[0089] (4) Using the groundwater pollution risk assessment model, calculate the pollution risk index of each storage tank that is greater than or equal to the first preset threshold; For 11 underground steel storage tanks in a newly constructed or expanded land area, a pollution risk assessment model was used to calculate the pollution risk index of each tank (Table 12). According to Table 12, the pollution risk index of the tanks ranges from 3.8 to 8.2.

[0090] Table 12 Safety Assessment Scores for Newly Built, Renovated, and Expanded Land Use

[0091] (5) Determine whether the newly built or expanded land is safe based on whether there is a tank with a pollution risk index greater than or equal to the second preset threshold among all tanks with a pollution risk index greater than or equal to the second preset threshold (i.e., if there is any tank with a pollution risk index greater than or equal to the second preset threshold, the newly built or expanded land is not safe; if there is no tank with a pollution risk index greater than or equal to the second preset threshold, the newly built or expanded land is safe).

[0092] In this specific embodiment, the pollution risk index of storage tank #1 (hydrochloric acid) and storage tank #3 (benzene) is greater than or equal to 7.5 points (Table 12). Therefore, the newly constructed or expanded land to be tested is unsafe, and the land user should investigate the causes of pollution risks to avoid risks to human health and downstream water sources.

[0093] Compared with the prior art, the present invention also has the following beneficial effects: (1) The safety assessment method for new, renovated and expanded land involving underground steel storage tanks provided by the present invention addresses the technical requirements for safety assessment of new, renovated and expanded land, and is geared towards underground steel storage tanks. It constructs a leakage risk assessment index system based on the leakage conditions of the storage tank, and constructs a groundwater pollution risk assessment index system based on the risk formation elements of risk source-material migration pathway-sensitive receptor. This significantly reduces the number of two types of risk assessment indicators and saves the cost of collecting assessment indicator information.

[0094] (2) The safety assessment method for new and expanded construction of underground steel storage tanks based on machine learning provided by this invention addresses the management needs of safety assessment for new and expanded construction of land. It successively carried out leakage risk assessment and groundwater pollution risk assessment based on the summation method, thereby realizing a rapid qualitative assessment of the safety of new and expanded construction of land coupled with leakage and pollution migration. This greatly reduces the workload of traditional risk assessment based on numerical simulation. The calculation process is simple and practical, the technology is reliable, and it is easy to implement in engineering. It can provide a reliable reference for scientific and accurate groundwater pollution prevention and control, and has important environmental and social significance.

[0095] (3) The safety assessment method for new, renovated and expanded land involving underground steel storage tanks provided by the present invention selects two indicators: the population in contact with the storage tank and the water source within 1km downstream. When establishing the groundwater pollution risk assessment model, the risk to human health and the risk of pollution spread are considered simultaneously, which makes up for the shortcomings of existing methods that only consider a single risk to human health or the risk of pollution spread. It can be used for groundwater pollution risk assessment of pore water in loose rock.

[0096] (4) The safety assessment method for new, renovated and expanded land use involving underground steel storage tanks provided by the present invention, based on the preliminary determination of the weight of leakage risk assessment indicators by human subjective judgment based on expert judgment, uses the optimal hybrid machine learning algorithm to objectively optimize and determine the weight of leakage risk assessment indicators, realizes the objective assignment of indicator weights, solves the problem of large ambiguity when human subjectively assigns weights, and enhances the scientificity and accuracy of leakage risk assessment model prediction.

[0097] (5) The safety assessment method for new, renovated and expanded land use involving underground steel storage tanks provided by the present invention couples PSO and BPNN to form a PSO-BPNN hybrid machine learning algorithm, which overcomes the defects of existing BPNN such as passive learning, instability during training, and easy getting stuck in local extreme points, and further enhances the scientificity and accuracy of the leakage risk assessment model prediction.

[0098] The above description of the disclosed embodiments enables those skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A machine learning-based safety assessment method for new construction, renovation, and expansion projects involving underground steel storage tanks, characterized in that... include: S1. Construct a case library for leakage risk assessment of underground steel storage tanks, screen out leakage risk assessment indicators and assign scores and grades them, determine the initial weights of the leakage risk assessment indicators and optimize them. Using leakage risk assessment indicators and optimized weights, a leakage risk assessment model is constructed, and a first preset threshold is set. The construction process of the leakage risk assessment model in S1 is as follows: A hybrid PSO-BPNN machine learning algorithm was constructed based on optimized particle swarm optimization and backpropagation neural network. The inertia weight, individual learning factor and social learning factor of the PSO algorithm were set in the hybrid PSO-BPNN machine learning algorithm. The fitting coefficient, mean absolute error and root mean square error were used as performance evaluation indicators. The training function, number of hidden layer neurons and learning rate hyperparameter of BPNN were optimized and adjusted. In the optimized and adjusted PSO-BPNN hybrid machine learning algorithm, the weight coefficients between the hidden layer neurons and the input layer neurons, and the weight coefficients between the output layer neurons and the hidden layer neurons are extracted. The absolute influence coefficients of each leakage risk assessment index are calculated and used as the weights of the optimized leakage risk assessment indexes. Based on the summation method, the leakage risk assessment model is obtained by multiplying each leakage risk assessment index by its corresponding optimized leakage risk assessment index weight and then summing the results. S2. Using the leakage risk assessment model, calculate the leakage risk index of each underground steel storage tank one by one, and screen out the underground steel storage tanks whose leakage risk index is greater than or equal to the first preset threshold. S3. Select groundwater pollution risk assessment indicators from the leakage risk assessment case library, assign scores and classify them, and determine the weight of the groundwater pollution risk assessment indicators. A groundwater pollution risk assessment model was constructed using groundwater pollution risk assessment indicators and weights, and a second preset threshold was determined. S4. Using the groundwater pollution risk assessment model, calculate the water pollution risk index of each underground steel storage tank whose leakage risk index is greater than or equal to the first preset threshold. S5. Determine whether the water pollution risk index of underground steel storage tanks with a leakage risk index greater than or equal to the first preset threshold is greater than or equal to the second preset threshold; if so, the land use is unsafe; otherwise, the land use is safe. The PSO-BPNN hybrid machine learning algorithm was used to evaluate the predictive accuracy of leakage risks from underground steel storage tanks in the test set.

2. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, In S1, the leakage risk assessment case library includes: multiple information items; The leakage risk assessment indicators include: the nature of the stored toxic and hazardous substances, the characteristics of the construction and operation of the storage tank, groundwater pollution prevention measures, and the external environment; The properties of the stored toxic and hazardous substances include: substance type, substance quantity, substance toxicity, and substance corrosivity; The characteristics of the construction and operation of the storage tanks include: number of tank layers, service life, cathodic protection system, anti-corrosion materials, and anti-seepage materials; The groundwater pollution prevention measures include: leakage monitoring devices, remote monitoring systems, routine inspection and maintenance, environmental emergency plans, environmental emergency supplies, routine inspection records, technical training, and surrounding monitoring wells; The external environment includes: groundwater level, groundwater corrosivity, and soil corrosivity.

3. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, The specific optimization parameters of the PSO-BPNN hybrid machine learning algorithm are as follows: The training function is trainlm; the number of hidden layer neurons is 10; the learning rate is 0.1; the inertia weight is 0.6; the individual learning factor is 2; and the social learning factor is 2.

4. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, In S1, the expression for the leakage risk assessment model is: LRI =0.078 SC +0.072 SQ +0.125 ST +0.127 SCOR +0.054 TL +0.044 SY +0.034 CPS +0.074 ACM +0.061 ASM +0.051 LMD +0.035 RMS +0.034 RM +0.018 EEP +0.016 EES +0.021 DIR +0.023 TT +0.038 PMW +0.035 GWD +0.029 GWCOR +0.031 SOCOR in, LRI This indicates the risk index of leakage. SC Indicates the type of substance. SQ Indicates the quantity of matter. ST Indicates the toxicity of a substance. SCOR Indicates the corrosiveness of a substance. TL Indicates the number of tank layers. SY Indicates the service life. CPS Indicates cathodic protection system, ACM Indicates anti-corrosion materials. ASM LMD indicates a leak-proof material, and LMD indicates a leak monitoring device. RMS Indicates a remote monitoring system. RM This indicates routine inspection and maintenance. EEP This indicates the environmental emergency response plan. EES Indicates environmental emergency supplies. DIR This indicates daily inspection records. TT Indicates technical training, PMW Indicates surrounding monitoring wells, GWD Indicates the groundwater level. GWCOR Indicates the corrosiveness of groundwater. SOCOR It indicates soil corrosivity.

5. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, In S3, the water pollution risk indicators include: risk sources, material migration pathways, and sensitive receptors; The risk sources include: leakage risk index, solubility of substances, and volatility of substances; The material migration pathways include: wet-dry index, vadose zone lithology, groundwater depth, and aquifer lithology; The sensitive receptors include: people who come into contact with the storage tank and water sources within 1 km downstream.

6. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, In step S3, the weights of the groundwater pollution risk assessment indicators are determined using the analytic hierarchy process (AHP), specifically including: The judgment matrix is ​​constructed using the 1-9 scaling method; Perform a consistency check on the judgment matrix; The weights of the groundwater pollution risk assessment indicators were obtained through normalization.

7. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, The first preset threshold is 4.1 points, and the second preset threshold is 7.5 points.

8. The machine learning-based safety assessment method for new, renovated, and expanded land use involving underground steel storage tanks according to claim 1, characterized in that, In S3, the expression for the groundwater pollution risk assessment model is: PRI =0.223 LRI +0.081 SS +0.082 SV +0.093 DWI +0.059 VZL +0.076 GD +0.115 AL +0.141 CSTP+ 0.130WS in, PRI Indicates the pollution risk index; LRI Indicates the risk index of leakage; SS Indicates the solubility of a substance; SV Indicates the volatility of a substance; DWI Indicates the dryness index; VZL Indicates the lithology of the vadose zone; GD Indicates the depth of groundwater; AL Indicates the lithology of the aquifer; CSTP This refers to the group of people who come into contact with the storage tank; WS This indicates a water source within a 1km downstream radius.

Images