A compressed natural gas filling station intelligent risk assessment method

Abstract

The present application relates to a kind of compressed natural gas filling station intelligent risk assessment method;Including the following construction steps: step 1 constructs wedge wave multi-core support vector machine;Step 2 determines the improved locust algorithm optimization optimization wedge wave support vector machine;Step 3 determines the risk assessment index system of compressed natural gas filling station;Step 4 gathers and is related to the data of compressed natural gas filling station risk index system and determines input sample data using expert evaluation method and questionnaire method.Step 5 the risk level of compressed natural gas filling station is divided into five grades and risk grade is used as the output sample data of evaluation model.Step 6 collects the risk assessment data of compressed natural gas filling station and determines training sample and test sample;Step 7 uses training data sample to train evaluation model and uses the evaluation model after training to evaluate the risk level of compressed natural gas filling station to be evaluated.The above-mentioned petrochemical enterprise refining unit energy management system optimization method is constructed.This application can improve the precision and efficiency of compressed natural gas filling station risk assessment, and then can improve the safety management level of compressed natural gas filling station.

Description

Technical Field

[0001] This invention relates to an intelligent risk assessment method for compressed natural gas (CNG) refueling stations, and more particularly to an intelligent risk assessment method for CNG refueling stations based on a wedge-shaped multi-kernel support vector machine optimized by an improved locust algorithm. Background Technology

[0002] Compressed natural gas (CNG) is characterized by low cost, zero pollution, and ease of use. CNG refueling stations typically store a certain amount of natural gas in storage tanks. In recent years, the number of CNG refueling stations has been increasing. The rapid development of the CNG industry has led to dispersed operations and difficulties in supervision. Furthermore, management in many areas is relatively weak, and control measures are outdated. Therefore, many new CNG refueling stations have numerous loopholes and problems in business management, equipment monitoring, and safe production. Due to corrosion or material defects, natural gas storage tanks may leak, leading to accidents such as fires and explosions. Safety accidents occur frequently, causing casualties and property damage. Therefore, the safety of CNG refueling stations is increasingly attracting public attention. Given the high-risk nature of CNG refueling stations, it is necessary to conduct hazard identification and risk analysis of potential risk factors in their production and operation. Based on accident statistical analysis results, a qualitative evaluation method is used to conduct a preliminary hazard analysis of CNG refueling station production equipment, identify the causes of common accidents at CNG refueling stations, and propose technical management measures for the safe operation of CNG refueling stations. Risk assessment research on CNG refueling stations will ensure the safety of CNG refueling stations and their surrounding facilities.

[0003] In recent years, with the rapid development of artificial intelligence (AI) technology, AI-based machine learning algorithms have become increasingly mature. AI-based machine learning algorithms can automatically acquire and disseminate information in real time, preventing major or serious damage. Furthermore, they offer advantages such as low cost, low power consumption, and reliable information. However, due to the high complexity, nonlinearity, uncertainty, and real-time nature of risk assessment for compressed natural gas (CNG) refueling stations, traditional mathematical models have certain limitations. Traditional evaluation methods are highly subjective, random, and fuzzy, relatively complex to operate, and lack self-learning capabilities. The nonlinear processing capability of Support Vector Machines (SVMs) is achieved through the "kernel mapping" method. For kernel mapping, the kernel function must satisfy the Mercer condition. For example, the Gaussian kernel function is a widely used kernel function that exhibits good mapping performance when analyzing nonlinear problems. However, for existing kernel functions, SVMs cannot approximate any function on a given L2(R) subspace because existing kernel functions cannot generate a complete basis by translation on that subspace. Summary of the Invention

[0004] This invention addresses the problems existing in the prior art by providing an intelligent risk assessment method for compressed natural gas (CNG) refueling stations using an improved locust algorithm-optimized wedge-wave multi-kernel support vector machine. This invention can improve the accuracy and efficiency of risk assessment for CNG refueling stations, thereby enhancing the safety management level of CNG refueling stations.

[0005] The technical solution of the present invention includes the following steps:

[0006] Step 1: Construct a wedge-shaped multi-kernel support vector machine

[0007] A wedge-wave support vector machine (SVM) was established by combining the wedge-wave transform, which exhibits good robustness to changes across different scales and directions, with the SVM. To address the blindness issue in evaluation of single-kernel SVMs when handling machine learning tasks with multiple feature sets, a multi-kernel SVM was generated by weighted linear addition of local and global kernel functions to classify the data, further improving classification accuracy and thus enhancing the risk assessment accuracy for compressed natural gas (CNG) refueling stations.

[0008] Step 1-1: Construct the combined kernel function, the corresponding formula is shown below:

[0009]

[0010]

[0011] In the formula, λ i k represents the weighting coefficient. i (x, y) represents a single-kernel function, as shown in the following expression:

[0012]

[0013] In the formula, c represents the scaling factor, and d i and e i This represents the translation factor.

[0014] Step 1-2: Define the regression function as follows:

[0015]

[0016] In the formula, ω represents the weight variable. Let B represent the mapping function, and let B represent the compensation factor.

[0017] Steps 1-3: Define the objective function and boundary conditions, as shown below:

[0018]

[0019]

[0020] In the formula, L represents the number of samples and D represents the penalty factor.

[0021] Steps 1-4: Construct the Lagrangian objective function and corresponding boundary conditions using Lagrangian duality, as shown below:

[0022]

[0023]

[0024] In the formula, θ i , χ i This represents the Lagrange operator.

[0025] According to equation (5), we can obtain:

[0026]

[0027] Formula (7) for w, B, ε i The partial derivatives of the derivatives are then substituted back into equation (9) to obtain the following equation:

[0028]

[0029] The minimization equation can be derived through optimization, as shown below:

[0030]

[0031]

[0032] Steps 1-5: Obtain a multi-kernel support vector machine by introducing a multi-kernel function, as shown below.

[0033]

[0034]

[0035] Steps 1-6: Determine the decision function for the wedge-shaped multi-kernel support vector machine, as shown below:

[0036]

[0037] Step 2: Determine the improved locust algorithm and optimize the wedge-shaped support vector machine

[0038] To improve the risk assessment performance of wedge-wave multi-kernel support vector machines, an improved locust optimization algorithm with better global optimization performance and higher convergence accuracy is applied to optimize the penalty factor, kernel function parameters, and weights of the wedge-wave multi-kernel support vector machine.

[0039] Step 2-1: Initialize the basic parameters of the improved locust optimization algorithm, including population size, spatial dimension, maximum number of iterations, and initial position.

[0040] Step 2-2: Update the position of the locust swarm using the following formula:

[0041]

[0042]

[0043] In the formula, T d The target location of the locust swarm is represented by η, and the attenuation coefficient is represented by ω. l η represents the weight coefficient, t represents the current iteration number, and η represents the number of iterations. max and η min represents the maximum and minimum decay factors. s(·) represents the interaction force function among locust populations, as shown below:

[0044]

[0045] In the formula, f represents the attraction intensity parameter, and r represents the attraction scale parameter.

[0046] Steps 2-3: Use the Levy flight local search strategy to adjust the position of individual locusts, as shown below:

[0047] X = X + 10 × s ts ·L·X (19)

[0048] In the formula, L represents the Levy flight stride length, s ts This represents a threshold function that can be used to control the flight method and probability of locusts.

[0049] L is calculated using the following formula:

[0050] L=μ / |v| 1 / β (20)

[0051] In the formula, β represents a constant between 0 and 2, and the parameters μ and v represent parameters that follow a normal distribution.

[0052] Steps 2-4: When an individual locust searches for the current optimal position, its original position is replaced; if no optimal solution is found, a random exit strategy with linearly decreasing parameters is used, as shown below:

[0053] P i = (2-2rand(0,1))·P i (twenty one)

[0054] In the formula, P i This represents the i-th locust. When a new position P is discovered... iIf the new position is better than the current position, the new position replaces the current position.

[0055] Steps 2-5: To ensure the decay coefficient decreases more rapidly in the early stages of the algorithm, allowing individual locusts in the population to quickly approach the optimal target and improving the algorithm's convergence speed; in the later iterations, the decay coefficient decreases at a slower rate, allowing individual locusts to carefully search the space and avoid getting trapped in local optima. The decay factor η is adjusted using a decreasing coefficient update strategy, as shown below:

[0056]

[0057] In the formula, n represents the current iteration coefficient, and N represents the maximum number of iterations.

[0058] Step 3: Determine the risk assessment indicator system for compressed natural gas refueling stations

[0059] By collecting and analyzing detailed data on the design, construction, operation, leakage, defects, personnel, social, and economic aspects of compressed natural gas (CNG) refueling stations, we systematically, comprehensively, and qualitatively identify the risk influencing factors of CNG refueling stations. Based on the risk formation mechanism, we determine a risk indicator system for CNG refueling stations. We also determine the primary and secondary indicators of the CNG refueling station risk assessment system.

[0060] Step 4: Collect data related to the risk indicator system of compressed natural gas refueling stations, determine the primary indicator values ​​through expert evaluation and questionnaire methods, and determine the input sample data.

[0061] Step 5: Divide the risk level of compressed natural gas (CNG) refueling stations into five levels: Level I (score range [0.90, 1.00]), Level II (score range [0.75, 0.90)), Level III (score range [0.60, 0.75)), Level IV (score range [0.45, 0.60)), and Level V (score range [0, 0.45)). These risk levels serve as the output sample data for the assessment model.

[0062] Step 6: Collect relevant data for the risk assessment of compressed natural gas refueling stations, and divide the data into two parts: training data samples and test data samples.

[0063] Step 7: Train the evaluation model using training data samples, and then use the trained evaluation model to assess the risk level of the compressed natural gas refueling station to be evaluated, and determine the safety status of the compressed natural gas refueling station to be evaluated.

[0064] The advantages and effects of this invention are as follows:

[0065] The improved locust algorithm-optimized wedge-wave multi-kernel support vector machine constructed in this invention can effectively improve the accuracy and efficiency of risk assessment for compressed natural gas (CNG) refueling stations, thereby enhancing the effectiveness of risk assessment and providing a strong theoretical basis for formulating risk prevention and control measures for CNG refueling stations. It has broad development prospects. Attached Figure Description

[0066] Figure 1 Training time for different risk assessment methods Detailed Implementation

[0067] Example

[0068] The technical solution of the present invention includes the following steps:

[0069] Step 1: Construct a wedge-shaped multi-kernel support vector machine

[0070] A wedge-wave support vector machine (SVM) was established by combining the wedge-wave transform, which exhibits good robustness to changes across different scales and directions, with the SVM. To address the blindness issue in evaluation of single-kernel SVMs when handling machine learning tasks with multiple feature sets, a multi-kernel SVM was generated by weighted linear addition of local and global kernel functions to classify the data, further improving classification accuracy and thus enhancing the risk assessment accuracy for compressed natural gas (CNG) refueling stations.

[0071] Step 1-1: Construct the combined kernel function, the corresponding formula is shown below:

[0072]

[0073]

[0074] In the formula, λ i k represents the weighting coefficient. i (x, y) represents a single-kernel function, as shown in the following expression:

[0075]

[0076] In the formula, c represents the scaling factor, and d i and e i This represents the translation factor.

[0077] Step 1-2: Define the regression function as follows:

[0078]

[0079] In the formula, ω represents the weight variable. Let B represent the mapping function, and let B represent the compensation factor.

[0080] Steps 1-3: Define the objective function and boundary conditions, as shown below:

[0081]

[0082]

[0083] In the formula, L represents the number of samples and D represents the penalty factor.

[0084] Steps 1-4: Construct the Lagrangian objective function and corresponding boundary conditions using Lagrangian duality, as shown below:

[0085]

[0086]

[0087] In the formula, θ i , χ i This represents the Lagrange operator.

[0088] According to equation (5), we can obtain:

[0089]

[0090] Formula (7) for w, B, ε i The partial derivatives of the derivatives are then substituted back into equation (9) to obtain the following equation:

[0091]

[0092]

[0093] The minimization equation can be derived through optimization, as shown below:

[0094]

[0095]

[0096] Steps 1-5: Obtain a multi-kernel support vector machine by introducing a multi-kernel function, as shown below.

[0097]

[0098]

[0099] Steps 1-6: Determine the decision function for the wedge-shaped multi-kernel support vector machine, as shown below:

[0100]

[0101] Step 2: Determine the improved locust algorithm and optimize the wedge-shaped support vector machine.

[0102] To improve the risk assessment performance of wedge-wave multi-kernel support vector machines, an improved locust optimization algorithm with better global optimization performance and higher convergence accuracy is applied to optimize the penalty factor, kernel function parameters, and weights of the wedge-wave multi-kernel support vector machine.

[0103] Step 2-1: Initialize the basic parameters of the improved locust optimization algorithm, including population size, spatial dimension, maximum number of iterations, and initial position.

[0104] Step 2-2: Update the position of the locust swarm using the following formula:

[0105]

[0106]

[0107] In the formula, T d The target location of the locust swarm is represented by η, and the attenuation coefficient is represented by ω. l η represents the weight coefficient, t represents the current iteration number, and η represents the number of iterations. max and η min represents the maximum and minimum decay factors. s(·) represents the interaction force function among locust populations, as shown below:

[0108]

[0109] In the formula, f represents the attraction intensity parameter, and r represents the attraction scale parameter.

[0110] Steps 2-3: Use the Levy flight local search strategy to adjust the position of individual locusts, as shown below:

[0111] X = X + 10 × s ts ·L·X (19)

[0112] In the formula, L represents the Levy flight stride length, s ts This represents a threshold function that can be used to control the flight method and probability of locusts.

[0113] L is calculated using the following formula:

[0114] L=μ / |v| 1 / β (20)

[0115] In the formula, β represents a constant between 0 and 2, and the parameters μ and v represent parameters that follow a normal distribution.

[0116] Steps 2-4: When an individual locust searches for the current optimal position, its original position is replaced; if no optimal solution is found, a random exit strategy with linearly decreasing parameters is used, as shown below:

[0117] P i = (2-2rand(0,1))·P i (twenty one)

[0118] In the formula, P i This represents the i-th locust. When a new position P is discovered... i If the new position is better than the current position, the new position replaces the current position.

[0119] Steps 2-5: To ensure the decay coefficient decreases more rapidly in the early stages of the algorithm, allowing individual locusts in the population to quickly approach the optimal target and improving the algorithm's convergence speed; in the later iterations, the decay coefficient decreases at a slower rate, allowing individual locusts to carefully search the space and avoid getting trapped in local optima. The decay factor η is adjusted using a decreasing coefficient update strategy, as shown below:

[0120]

[0121] In the formula, n represents the current iteration coefficient, and N represents the maximum number of iterations.

[0122] Step 3: Determine the risk assessment indicator system for compressed natural gas refueling stations

[0123] By collecting and analyzing detailed data on the design, construction, operation, leakage, defects, personnel, social, and economic aspects of compressed natural gas (CNG) refueling stations, we systematically, comprehensively, and qualitatively identify the risk influencing factors of CNG refueling stations. Based on the risk formation mechanism, we determine a risk indicator system for CNG refueling stations. We also determine the primary and secondary indicators of the CNG refueling station risk assessment system.

[0124] Step 4: Collect data related to the risk indicator system of compressed natural gas refueling stations, determine the primary indicator values ​​through expert evaluation and questionnaire methods, and determine the input sample data.

[0125] Step 5: Divide the risk level of compressed natural gas (CNG) refueling stations into five levels: Level I (score range [0.90, 1.00]), Level II (score range [0.75, 0.90)), Level III (score range [0.60, 0.75)), Level IV (score range [0.45, 0.60)), and Level V (score range [0, 0.45)). These risk levels serve as the output sample data for the assessment model.

[0126] Step 6: Collect relevant data for the risk assessment of compressed natural gas refueling stations, and divide the data into two parts: training data samples and test data samples.

[0127] Step 7: Train the evaluation model using training data samples, and then use the trained evaluation model to assess the risk level of the compressed natural gas refueling station to be evaluated, and determine the safety status of the compressed natural gas refueling station to be evaluated.

[0128] Specific examples are shown below:

[0129] In this embodiment, a compressed natural gas (CNG) refueling station was selected for risk assessment. The station has the following specifications: a refueling capacity of 1000 standard cubic meters per hour and a daily refueling capacity of 10000 standard cubic meters. This CNG refueling station primarily supplies gas to small and medium-sized vehicles.

[0130] The improved locust algorithm has the following parameter settings: l = 2.0, f = 1.0, b min =0.35,b max =0.90,η min =0.40,η max =0.95, population size is 350, maximum number of iterations is 400.

[0131] Fifty sets of relevant data for the risk assessment of the compressed natural gas refueling station were obtained using questionnaires and expert surveys. The first 40 sets of data were used as training samples, and the last 10 sets of data were used as test samples.

[0132] To verify the effectiveness of the proposed evaluation method, a locust algorithm optimized by B-spline wavelet support vector machine (BSWSVM-LA) and a support vector machine optimized by particle swarm optimization (SVM-PSA) were used to assess the risk of the training samples. The accuracy of different evaluation models was evaluated using mean absolute error (MAE) and mean absolute percentage error (MAPE). The risk assessment results for compressed natural gas refueling stations based on different models are shown in Table 1.

[0133] Table 1. Risk assessment results of compressed natural gas based on different models

[0134]

[0135] Training time for different models, such as Figure 1 As shown. From Figure 1 As can be seen, the training time of the evaluation method proposed in this invention is 6.7 seconds, which is the shortest among the three methods. Therefore, the risk assessment method proposed in this invention can improve the efficiency of risk assessment for compressed natural gas refueling stations.

[0136] Risk assessment of the test samples was conducted using three assessment methods trained on the training samples. The comparison results are shown in Table 2.

[0137] Table 2 Risk assessment results of test samples based on the three methods

[0138]

[0139] According to the calculation results in Table 2, the MAE of the evaluation method proposed in this invention ranges from 0.82 to 0.92, the MAE of BSWSVM-LA ranges from 3.27 to 3.68, and the MEA of SVM-PSA ranges from 4.17 to 4.57. Furthermore, the evaluation results of the evaluation method proposed in this invention are all correct, while BSWSVM-LA has 4 errors and SVM-PSA has 7 errors. Therefore, the proposed WMKSVM-ILA has the highest evaluation accuracy among the three models.

[0140] The analysis results show that the proposed risk assessment method for compressed natural gas refueling stations based on the improved locust algorithm and optimized wedge-shaped multi-core support vector machine can achieve the best results and is a risk assessment method with high practical value.

Claims

1. A method for intelligent risk assessment of a compressed natural gas refueling station, characterized in that Comprising the following steps: Step 1: Constructing wedgelet multi-kernel support vector machine The wedgelet transform with good robustness to changes in different scales and directions is combined with support vector machine to establish a wedgelet support vector machine; In view of the blindness problem of single kernel function support vector machine in evaluation when processing machine learning tasks of multiple feature sets, a multi-kernel support vector machine is generated by weighted linear addition of local and global kernel functions to classify data, so as to further improve the classification precision and improve the risk assessment precision of compressed natural gas filling station; Step 2: The improved locust optimization algorithm is used to optimize the penalty factor, kernel function parameter and weight of the wedgelet multi-kernel support vector machine; In order to improve the risk assessment effect of wedgelet multi-kernel support vector machine, the improved locust optimization algorithm with good global optimization performance and high convergence precision is applied to the optimization of the penalty factor, kernel function parameter and weight of the wedgelet multi-kernel support vector machine; The specific steps of the improved locust optimization algorithm are as follows: Step 2-1: Initialize the basic parameters of the improved locust optimization algorithm, including population size, spatial dimension, maximum iteration number and initial position; Step 2-2: Update the position of the locust group according to the following formula: where T d denotes the target position of locust population, η denotes the attenuation coefficient, ω l denotes the weight coefficient, t denotes the current iteration number, η max and η min denote the maximum and minimum attenuation factors; s(·) denotes the interaction force function between locust populations, as follows: In the formula, f represents the attraction strength parameter, and r represents the attraction force scale parameter; Step 2-3: Adjust the position of a single locust by using the Levy flight local search strategy, as follows: X = X + 10 x s ts • L • X (19) In the formula, L represents Levy flight step size, s ts represents a threshold function, which can be used to control the flying method and change probability of locusts; L is calculated by the following formula: L = μ / |v| 1 / β (20) In the formula, β represents a constant between 0 and 2, and parameters μ and v represent parameters subject to normal distribution; Step 2-4: When a locust individual searches for the current optimal position, the original position is replaced; If the optimal solution is not found, a random jump-out strategy with a linearly decreasing parameter is used, as follows: P i = (2 - 2rand(0, 1)) · P i (21) wherein P i represents the ith locust; When a new position P is found i If the new position is better than the current position, replace the current position with the new position. Step 2-5: In order to make the decay coefficient decrease at a faster speed in the early stage of algorithm execution, ensure that the locust individuals in the population can quickly approach the optimal target, and improve the convergence speed of the algorithm; In the iteration process in the later stage of the algorithm, the reduction speed of the decay coefficient is reduced, so that the locust individuals can carefully search the space and avoid falling into local optimum; The decay factor η is adjusted by using the decreasing coefficient update strategy, as follows: In the formula, n represents the current iteration coefficient, and N represents the maximum iteration number; Step 3: Determine the risk assessment index system of compressed natural gas filling station Through detailed collection and analysis of data of compressed natural gas filling station design, construction, operation, operation, leakage, defect, personnel, society and economy, the risk influencing factors of compressed natural gas filling station are systematically, comprehensively and qualitatively identified; According to the risk formation mechanism, the risk index system of compressed natural gas filling station is determined; The first-level index and the second-level index of the risk assessment system of the natural gas filling station are determined; Step 4: Collect data related to the risk index system of compressed natural gas filling station, determine the first-level index value by expert evaluation method and questionnaire method, and determine the input sample data; Step 5: The risk level of the compressed natural gas filling station is divided into five levels, namely, level I: the score interval is [0.90, 1.00], level II: the score interval is [0.75, 0.90), level III: the score interval is [0.60, 0.75), level IV: the score interval is [0.45, 0.60), and level V: the score interval is [0, 0.45); the risk level is taken as the output sample data of the evaluation model; Step 6: Collect the risk assessment related data of the compressed natural gas filling station, and divide the data into two parts, namely, training data samples and test data samples; Step 7: The training data samples are used to train the evaluation model, and then the trained evaluation model is used to evaluate the risk level of the compressed natural gas filling station to be evaluated, and the safety condition of the compressed natural gas filling station to be evaluated is determined.

2. The intelligent risk assessment method for compressed natural gas filling stations according to claim 1, characterized in that The construction steps of the wedgelet multi-kernel support vector machine are as follows: Step 1-1: Construct a combined kernel function, and the corresponding formula is as follows: where λ i denotes a weight coefficient, k i (x, y) denotes a monomial function, expressed as follows: where c represents a scale factor, d i and e i represent translation factors; Step 1-2: Define the regression function as follows: In the formula, ω represents a weight variable, represents a mapping function, and B represents a compensation factor. Step 1-3: Define the objective function and boundary conditions as follows: In the formula, L represents the number of samples, and D represents the penalty factor; Step 1-4: Construct the Lagrange objective function and the corresponding boundary conditions by using the Lagrange duality, as follows: wherein θ i , χ i denotes the Lagrangian operator; According to formula (5), we can get: The partial derivatives of equation (7) with respect to w, B, ε i and the result back-substituted into equation (9) gives the following equation: Through optimization, the minimization equation can be derived as follows: Step 1-5: Obtain the multi-kernel support vector machine by introducing the multi-kernel function as follows Step 1-6: Determine the decision function of the wedgelet multi-kernel support vector machine as follows: 。

Images