A continuous leakage coefficient C for atmospheric pressure vertical storage tanks based on feedforward neural networks s Prediction methods

Abstract

This invention discloses a continuous leakage coefficient C for atmospheric pressure vertical storage tanks based on a feedforward neural network. s The prediction method relates to the field of storage tank leakage safety technology. This invention employs this model to predict leakage characteristics of known atmospheric pressure vertical storage tanks (liquid level h above the leak orifice, liquid level drop rate v). h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T Learning this method allows for the rapid and accurate calculation of the continuous leakage coefficient C under different orifice conditions. s This invention solves the problem of leakage coefficient C during continuous leakage in atmospheric pressure vertical storage tanks. s The problem of real-time changes and difficulty in direct mathematical modeling has improved the calculation accuracy of mass flow rate and leakage volume in continuous leakage, thus providing necessary theoretical support for the calculation accuracy of evaporation diffusion concentration in continuous leakage pools of atmospheric pressure vertical storage tanks and its risk assessment.

Description

Technical Field

[0001] This invention belongs to the field of storage tank leakage safety technology, specifically relating to a continuous leakage coefficient C for atmospheric pressure vertical storage tanks based on a feedforward neural network. s Prediction methods. Background Technology

[0002] Light oil storage tanks that have been in service for a long time or beyond their service life may leak due to internal corrosion, aging, or other reasons. The resulting continuous, expanding liquid pool can evaporate due to the temperature difference between the ground and the environment, mixing with air to form explosive gases. When the gas concentration reaches the fire and explosion range, it can easily trigger a fire and explosion accident. Determining the gas concentration depends on the accurate calculation of the liquid leakage rate. In the calculation model for instantaneous or continuous leakage, the determination of the leakage coefficient is often involved. For example, in the "Guideline for Quantitative Risk Assessment of Chemical Enterprises" (AQ / T 3046-2013), the calculation model for the instantaneous mass flow rate of liquid flowing out of the tank orifice is as follows: C0 is the leakage coefficient. Furthermore, the empirical constant value of C0 needs to be determined based on the orifice conditions and the Reynolds number (Re) range. Therefore, the leakage coefficient is one of the key factors affecting the accuracy of leakage calculation.

[0003] The leakage coefficient is a dimensionless parameter characterizing orifice flow loss, and its value can be determined using two methods: empirical constants and predictive models. The continuous leakage rate calculated using empirical constants of the leakage coefficient usually deviates significantly from the actual value, and empirical or semi-empirical models of the leakage coefficient rely on the established experimental conditions, exhibiting significant limitations in practical applications. The leakage coefficient is influenced by factors such as the geometric parameters of the leakage orifice, the pressure difference between the inside and outside of the orifice, and the liquid flow characteristics, making it difficult to establish an accurate model directly through fluid mechanics and mathematical derivation. Therefore, this paper considers leveraging the ease of implementation, high predictive performance, and nonlinear modeling capabilities of feedforward neural networks to evaluate the continuous leakage coefficient C of an atmospheric pressure vertical storage tank. s Make predictions. Summary of the Invention

[0004] To address the above problems, this invention provides a continuous leakage coefficient C for atmospheric pressure vertical storage tanks based on a feedforward neural network. s The prediction method uses the continuous leakage coefficient C for atmospheric pressure vertical storage tanks. s Rapid and accurate calculations are performed to obtain the continuous leakage coefficient C of the atmospheric pressure vertical storage tank. s The real-time trend of liquid changes. This invention can be used to calculate the mass flow rate and leakage volume of continuous liquid leakage in atmospheric pressure vertical storage tanks, thereby providing necessary theoretical data support for tank diffusion range prediction and risk assessment.

[0005] This invention is achieved through the following technical solution:

[0006] Continuous leakage coefficient C of atmospheric pressure vertical storage tank based on feedforward neural network s The prediction method includes the following steps:

[0007] (1) Obtain monitoring data of continuous leakage characteristic parameters of atmospheric pressure vertical storage tanks, and construct the continuous leakage coefficient C of atmospheric pressure vertical storage tanks based on feedforward neural network. s The predicted dataset;

[0008] (2) Based on the monitoring data from step (1), and in accordance with the conservation of energy and mass, establish the Bernoulli equation for the continuous leakage process of the storage tank, and calculate the continuous leakage coefficient C of the atmospheric pressure vertical storage tank. s The experimental values;

[0009] (3) Normalize the continuous leakage characteristic parameter data of the atmospheric pressure vertical storage tank obtained in step (1), and then divide the dataset into 80% training set and 20% test set according to the time sequence corresponding to the data samples.

[0010] (4) The feedforward neural network method is used to analyze the continuous leakage coefficient C. s The predicted value is calculated by taking the normalized result of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank in step (3) as the input data of the feedforward neural network for iterative calculation, and finally generating the continuous leakage coefficient C in the output layer. s Predicted value;

[0011] (5) Test the feedforward neural network trained in step (4) using the test set from step (3) and obtain the continuous leakage coefficient C of the atmospheric pressure vertical storage tank. s Prediction results; using mean absolute error, explained variance, and coefficient of determination as evaluation indicators, the continuous leakage coefficient C of atmospheric pressure vertical storage tanks is analyzed. s The performance of the prediction model was evaluated; mean absolute error (MAE) ≤ 0.1, explained variance (EVS) ≥ 0.7, and coefficient of determination (R²) was [value missing]. 2 A value ≥0.7 indicates that the predicted result tends towards the actual value; the continuous leakage coefficient C for atmospheric pressure vertical storage tanks. s The prediction results are considered valid;

[0012] (6) Based on the continuous leakage coefficient C of the atmospheric pressure vertical storage tank obtained in step (5) s To calculate the mass flow rate Q of continuous leakage from the storage tank. m and leakage amount M leak .

[0013] The specific operation method of this invention is as follows:

[0014] This invention can be controlled by the liquid level h above the leak hole and the liquid level drop rate v. h Leakage time t, equivalent diameter of leakage hole d oDynamic viscosity μ and tank diameter d T These six characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks allow for the rapid calculation of the continuous leakage coefficient C throughout the entire leakage process of the atmospheric pressure vertical storage tank. s It includes the following steps:

[0015] (1) Obtain monitoring data of continuous leakage characteristic parameters of atmospheric pressure vertical storage tanks, and construct the continuous leakage coefficient C of atmospheric pressure vertical storage tanks based on feedforward neural network. s The dataset for the prediction model. Among them, the continuous leakage coefficient C... s The dataset consists of characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks and the continuous leakage coefficient C. s Composition. The continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank include the liquid level h above the leakage hole and the liquid level drop rate v. h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T .

[0016] (2) Based on the conservation of energy and mass, Bernoulli's equation is established for the continuous leakage process of the storage tank, and the continuous leakage coefficient C of the atmospheric pressure vertical storage tank is calculated. s The experimental value. This experimental value will be used as the continuous leakage coefficient C. s The true values ​​are used for training and learning of the feedforward neural network and for evaluating model performance:

[0017]

[0018] in,

[0019] d T Where is the diameter of the storage tank, in meters (m);

[0020] d o The equivalent diameter of the leakage hole, in meters (m).

[0021] v h The rate at which the liquid level falls is expressed in m / s;

[0022] g is the acceleration due to gravity, taken as g = 9.8 m / s². 2 ;

[0023] h represents the liquid level height above the leak hole, in meters (m).

[0024] (3) Normalize the continuous leakage characteristic parameter data of the atmospheric pressure vertical storage tank:

[0025]

[0026] in,

[0027] x norm The normalized results are for the characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks.

[0028] x represents the raw data of the continuous leakage characteristic parameters of an atmospheric pressure vertical storage tank;

[0029] x min This represents the minimum value of the characteristic parameter for continuous leakage in an atmospheric pressure vertical storage tank.

[0030] x max This represents the maximum value of the characteristic parameter for continuous leakage in an atmospheric pressure vertical storage tank.

[0031] The dataset was divided into a training set (80%) and a test set (20%) based on the time sequence of the data samples. The training set was used to learn the leakage characteristic parameters of the atmospheric pressure vertical storage tank (liquid level h above the leak orifice, liquid level drop rate v). h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T ) and continuous leakage coefficient C s The correlation between them, the test set is used to predict the continuous leakage coefficient C s To verify the effectiveness of the model.

[0032] (4) The feedforward neural network includes an input layer, a hidden layer, and an output layer; the normalized result of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank is used as the input feature of the input layer and passed from the input layer to the hidden layer; the hidden layer is composed of multiple neurons, and its function is to extract and transform features from the input data; finally, the continuous leakage coefficient C is generated in the output layer. s Predicted value.

[0033] The inputs and outputs of the hidden layer are as follows:

[0034] z (l) =W (l) f l-1 (z (l-1) )+b (l) (3);

[0035] a (l) =f l (W (l) a (l-1) +b (l) (4);

[0036] in,

[0037] z represents the static input of the neuron;

[0038] l represents the number of layers in the FNN;

[0039] W is the weight matrix;

[0040] f is the activation function;

[0041] b is the bias parameter;

[0042] 'a' represents the output of the neuron.

[0043] The activation function is the ReLU function, which aims to improve network training speed, help the network adjust parameters more stably, and enable the feedforward neural network model to learn C. s and the nonlinear relationships between its influencing factors:

[0044] f(x) = max(0,x) (5);

[0045] in,

[0046] f(x) is the output value of the neuron;

[0047] x is the input value of the neuron.

[0048] The weight matrix W and bias parameter b are continuously adjusted during the training of the feedforward neural network to make the training output closer to the true value. The degree of difference between the training output and the true value is calculated by the loss function Loss. The mean squared error (MSE) function is used as the loss function:

[0049]

[0050] in,

[0051] n represents the number of samples in the training set;

[0052] y i For training set C s The true value;

[0053] For training set C s The predicted value.

[0054] During model training, to avoid the loss function from getting stuck in a local minimum, the Adam optimizer is used to adaptively adjust the learning rate α:

[0055]

[0056] in,

[0057] θ t Indicates the current time step;

[0058] θ t-1 Indicates the time step of the previous iteration;

[0059] m t This represents the first-moment estimate of the corrected gradient;

[0060] This represents the second-order moment estimate of the corrected gradient;

[0061] ε represents a non-zero constant.

[0062] (5) The trained feedforward neural network is tested using a test set, and the continuous leakage coefficient C of the atmospheric pressure vertical storage tank is obtained. s Prediction results. The mean absolute error (MAE), explained variance (EVS), and coefficient of determination (R²) will be used to calculate the results. 2 As an evaluation indicator, the continuous leakage coefficient C of atmospheric pressure vertical storage tanks is used. s The performance of the prediction model was evaluated:

[0063]

[0064]

[0065]

[0066] in,

[0067] n represents the number of samples in the test set;

[0068] y i C s The true value;

[0069] C s The predicted value;

[0070] C s The average of the true values;

[0071] Var{y} is C s The variance of the true value.

[0072] Mean absolute error (MAE) ≤ 0.1, explained variance (EVS) ≥ 0.7, coefficient of determination R0.05 2 A value ≥0.7 indicates that the predicted result tends towards the actual value. The continuous leakage coefficient C for atmospheric pressure vertical storage tanks... s The prediction results are considered valid;

[0073] (6) Based on the continuous leakage coefficient C of the atmospheric pressure vertical storage tank obtained in step (5) s To calculate the mass flow rate Q of continuous leakage from the storage tank. m and leakage amount M leak .

[0074] As a preferred technical solution, when continuous leakage occurs in the storage tank body, given the initial liquid level and orifice, the continuous leakage coefficient C sIt varies with leakage time. For leakage orifices of different sizes and shapes, the continuous leakage coefficient C varies. s It will show a non-linear downward trend to varying degrees.

[0075] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0076] (1) This invention addresses the problem that the leakage coefficient changes in real time under continuous leakage in atmospheric pressure vertical storage tanks and is difficult to model mathematically directly. Considering the differences in geometric parameters between different leakage orifices, it provides a continuous leakage coefficient C applicable to leakage orifices of various shapes. s Predictive model; obtained continuous leakage coefficient C of atmospheric pressure vertical storage tank s The predictive model can accurately calculate the real-time leakage coefficient of atmospheric pressure vertical storage tanks, improving the accuracy of calculating the mass flow rate and leakage volume of continuous leakage in atmospheric pressure vertical storage tanks. This effectively solves the problem of significant deviations in calculation results due to inaccurate leakage coefficient values ​​for continuous leakage in atmospheric pressure vertical storage tanks. Applying this invention to industrial production, it can be used to calculate the diffusion range of continuous leakage from atmospheric pressure vertical storage tanks and to quantitatively analyze accident risks, providing effective technical support for accident consequence prediction and prevention, and further reducing the risk of tank farm leakage accidents.

[0077] (2) The continuous leakage coefficient C of the atmospheric pressure vertical storage tank of the present invention s The prediction method uses known characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks (liquid level h above the leak orifice, rate of liquid level drop v). h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T The continuous leakage coefficient C of an atmospheric pressure vertical storage tank can be quickly calculated. s And plot the continuous leakage coefficient C of the atmospheric pressure vertical storage tank. s The curve showing the relationship between leakage time t and the leakage rate improves the accuracy of calculating the mass flow rate and leakage volume of continuous leakage in atmospheric pressure vertical storage tanks. This provides effective theoretical calculation reference data for quantitative risk assessment and consequence prediction of continuous leakage accidents in atmospheric pressure vertical storage tanks, and also provides theoretical data basis for the formulation of accident emergency plans. This helps to minimize the consequences of accidents and provides quantitative analysis of accident risks in tank areas. Attached Figure Description

[0078] Figure 1 The continuous leakage coefficient C of an atmospheric pressure vertical storage tank based on a feedforward neural network. s Flowchart of the prediction method implementation.

[0079] Figure 2 This is a schematic diagram of a feedforward neural network structure.

[0080] Figure 3Train an iterative loss graph for a feedforward neural network dataset.

[0081] Figure 4 The continuous leakage coefficient C of an atmospheric pressure vertical storage tank based on a feedforward neural network. s Prediction results chart.

[0082] Figure 5 The continuous leakage coefficient C of the atmospheric pressure vertical storage tank in Examples 1-16 s Graph showing the relationship between predicted values ​​and leakage time t.

[0083] Figure 6 For Examples 1-16, the continuous leakage coefficient C is applied. s The results of the calculation of the predicted continuous mass flow rate and leakage. Detailed Implementation

[0084] The present invention will be further described in detail below with reference to specific embodiments, but is not limited to the scope of protection of the present invention.

[0085] Taking a vertical atmospheric pressure storage tank (diameter 0.44m, initial liquid level approximately 0.53m) as an example, calculate the continuous leakage coefficient C of this vertical atmospheric pressure storage tank. s :

[0086] (I) First, collect relevant parameters for the atmospheric pressure vertical storage tank, as follows:

[0087] Storage tank volume 0.1m³ 3 (Ф0.44×0.66m), filling coefficient 80%, initial liquid level in tank 0.53m, storage temperature 25℃, storage pressure 0.1MPa, ambient pressure 0.1MPa; its specific related parameters are shown in Table 1.

[0088] Table 1. Relevant Parameters for Liquid Ammonia Horizontal Tanks

[0089]

[0090] According to the "Guideline for Quantitative Risk Assessment of Chemical Enterprises" (AQ / T3046-2013) 8.1.1, leakage scenarios are specified. The leakage diameter can be divided into two categories: complete rupture and hole leakage. Representative leakage scenarios and hole diameter ranges are shown in Table 2.

[0091] Table 2 Leakage Scenarios and Leakage Diameter Values

[0092]

[0093] According to Appendix F.1 of the "Technical Guidelines for Environmental Risk Assessment of Construction Projects" (HJ 169-2018), the shape of the leak in an atmospheric pressure vertical storage tank can be circular, polygonal, triangular, or rectangular. Therefore, the continuous leakage coefficient C for different orifices shown in Table 3 will be used...s Perform the calculation.

[0094] Table 3. Parameters of Leakage Holes in Atmospheric Pressure Vertical Storage Tanks

[0095]

[0096]

[0097] (II) To further clarify the objectives, technical solutions, and advantages of the present invention, the embodiments of the present invention will be described in further detail with reference to the accompanying drawings, such as... Figure 1 As shown, the specific steps are as follows:

[0098] Step 1: Establish the continuous leakage coefficient C s Dataset

[0099] Monitoring data of continuous leakage characteristic parameters of atmospheric pressure vertical storage tanks are obtained, and a continuous leakage coefficient C of atmospheric pressure vertical storage tanks based on a feedforward neural network is constructed. s The dataset for the prediction model. Among them, the continuous leakage coefficient C... s The dataset consists of characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks and the continuous leakage coefficient C. s Composition. The continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank include the liquid level h above the leakage hole and the liquid level drop rate v. h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T .

[0100] Furthermore, based on the principles of energy conservation and mass conservation, Bernoulli's equation is established for the continuous leakage process of the storage tank, and the continuous leakage coefficient C of the atmospheric pressure vertical storage tank is calculated. s The experimental value. This experimental value will be used as the continuous leakage coefficient C. s The true values ​​are used for training and learning of the feedforward neural network and for evaluating model performance:

[0101]

[0102] Where, d T Where d is the diameter of the storage tank, in meters (m); o v is the equivalent diameter of the leakage hole, in meters. h Let g be the rate of descent of the liquid surface, in m / s; g is the acceleration due to gravity, taken as g = 9.8 m / s². 2 h represents the liquid level height above the leak hole, in meters (m).

[0103] The continuous leakage characteristic parameter data of the atmospheric pressure vertical storage tank were normalized:

[0104]

[0105] Where, xnorm is the normalized result of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank; x is the original data of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank; x min x represents the minimum value of the characteristic parameter for continuous leakage in an atmospheric pressure vertical storage tank. max This represents the maximum value of the characteristic parameter for continuous leakage in an atmospheric pressure vertical storage tank.

[0106] The dataset was divided into a training set (80%) and a test set (20%) based on the time sequence of the data samples. The training set was used to learn the leakage characteristic parameters of the atmospheric pressure vertical storage tank (liquid level h above the leak orifice, liquid level drop rate v). h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T ) and continuous leakage coefficient C s The correlation between them, the test set is used to predict the continuous leakage coefficient C s To verify the effectiveness of the model.

[0107] Step 2: Train the feedforward neural network using the training set.

[0108] like Figure 2 The feedforward neural network comprises an input layer, a hidden layer, and an output layer. The normalized result of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank is used as the input feature of the input layer and passed from the input layer to the hidden layer. The hidden layer consists of multiple neurons and its function is to extract and transform features from the input data. Finally, the continuous leakage coefficient C is generated in the output layer. s Predicted value.

[0109] The inputs and outputs of the hidden layer are as follows:

[0110] z (l) =W (l) f l-1 (z (l-1) )+b (l) (3);

[0111] a (l) =f l (W (l) a (l-1) +b (l) (4);

[0112] Where z is the static input of the neuron; l represents the number of layers in the FNN; W is the weight matrix; f is the activation function; b is the bias parameter; and a is the output of the neuron.

[0113] The activation function is the ReLU function, which aims to improve network training speed, help the network adjust parameters more stably, and enable the feedforward neural network model to learn C. sand the nonlinear relationships between its influencing factors:

[0114] f(x) = max(0,x) (5);

[0115] Where f(x) is the output value of the neuron; x is the input value of the neuron.

[0116] The weight matrix W and bias parameter b are continuously adjusted during the training of the feedforward neural network to make the training output closer to the true value. The degree of difference between the training output and the true value is calculated by the loss function Loss. The mean squared error (MSE) function is used as the loss function:

[0117]

[0118] Where n represents the number of samples in the training set; y i For training set C s The true value; For training set C s The predicted value.

[0119] During model training, to avoid the loss function from getting stuck in a local minimum, the Adam optimizer is used to adaptively adjust the learning rate α:

[0120]

[0121] Where, θ t Indicates the current time step; θ t-1 Indicates the time step of the previous iteration; m t This represents the first-moment estimate of the corrected gradient; ε represents the second moment estimate of the corrected gradient; ε represents a non-zero constant.

[0122] Step 3: Utilize the trained feedforward neural network to calculate the continuous leakage coefficient C of the atmospheric pressure vertical storage tank. s predict

[0123] The trained feedforward neural network was tested using a test set, and the continuous leakage coefficient C of the atmospheric pressure vertical storage tank was obtained. s Prediction results. The mean absolute error (MAE), explained variance (EVS), and coefficient of determination (R²) will be used to calculate the results. 2 As an evaluation indicator, the continuous leakage coefficient C of atmospheric pressure vertical storage tanks is used. s The performance of the prediction model was evaluated:

[0124]

[0125]

[0126]

[0127] Where n represents the number of samples in the test set; y i C s The true value; C s The predicted value; C s The average of the true values; Var{y} is C s The variance of the true value.

[0128] The continuous leakage coefficient C of the atmospheric pressure vertical storage tank based on the feedforward neural network is utilized. s The prediction model calculates the continuous leakage coefficient C for circular, elliptical (1:2), elliptical (1:3), square, rectangular (1:2), rectangular (1:3), equilateral triangle, and isosceles triangle leaking holes under the continuous leakage scenarios of small and medium-sized holes in Examples 1-16. s Prediction was performed. Table 4 shows the monitoring data of continuous leakage characteristic parameters of the atmospheric pressure vertical storage tanks in Examples 1-16. The continuous leakage coefficient C of the atmospheric pressure vertical storage tank based on a feedforward neural network... s The training process of the prediction model is as follows Figure 3 As shown, the training iterations were performed 500 times, and the overall iteration loss showed a fluctuating downward trend. Within the first 62 iterations, the loss value rapidly decreased from 0.21 to 4.62 × 10⁻⁶. -3 In subsequent iterations, the loss function value decreased slowly. When the algorithm iterated to the 226th iteration, the loss value was 1.23 × 10⁻⁶. -3 Furthermore, the loss value tends to stabilize in subsequent iterations. This indicates that the feedforward neural network has learned the continuous leakage coefficient C of the atmospheric pressure vertical storage tank. s The coupling relationship between the continuous leakage characteristic parameters and the continuous leakage coefficient C can be used to determine the continuous leakage coefficient C. s Make effective predictions.

[0129] Table 4 Characteristic parameters of continuous leakage from atmospheric pressure vertical storage tanks

[0130]

[0131]

[0132] Feedforward neural network for the continuous leakage coefficient C in Examples 1-16 s The prediction results can be found in Figure 4 The leakage hole conditions and continuous leakage coefficient C of Examples 1-16 are described. s The predicted values ​​were statistically analyzed, and the results are shown in Table 5. The continuous leakage coefficient C of the atmospheric pressure vertical storage tank is calculated with leakage time t as the horizontal axis. sPlotting the leakage time t on the ordinate and the corresponding continuous leakage coefficient C of the atmospheric pressure vertical storage tank... s By plotting the predicted values, the continuous leakage coefficient C of the atmospheric pressure vertical storage tanks in Examples 1-16 was obtained. s The curve showing the relationship between leakage time t and the time t is as follows: Figure 5 As shown. (Through) Figure 5 It is known that the continuous leakage coefficient C of an atmospheric pressure vertical storage tank is... s The leakage coefficient C exhibits a non-linear decreasing trend with increasing leakage time t, and the continuous leakage coefficient C varies with different orifices. s The nonlinear downward trends differ, and can be broadly categorized into two types: global downward trends and fluctuating downward trends. Among them, the C values ​​of all small holes... s The -t curves all show a global downward trend, transitioning from a slow decrease to a rapid decrease. In the scenario of a central hole leak, the C values ​​for circular, elliptical (1:2), elliptical (1:3), rectangular (12), and rectangular (1:3) holes are... s The -t curve exhibits a fluctuating downward trend, slowly increasing to a peak and then rapidly decreasing, while the C of the hole in the square, equilateral triangle, and isosceles triangle... s The -t curve exhibits a similar global downward trend to that of the small hole.

[0133] The calculated MAE value of the model is 0.0154, indicating that C s The smaller the prediction error, the better. An EVS value of 0.9492 indicates a high degree of fit for this nonlinear regression model. R 2 The value of 0.9482 indicates that the evaluation model has good predictive performance.

[0134] According to Bernoulli's equation, the mass flow rate Q of liquid leakage from the storage tank is... m Calculation formula (Equation (11)):

[0135]

[0136] Where ρ is the liquid density, and since the leaking medium is water, ρ is taken as 1000 kg / m³. 3 ;

[0137] Further analysis of the mass flow rate Q m The continuous leakage rate M of the liquid in the storage tank is obtained by integration. leak Using the above C s mass flow rate Q for continuous leakage from storage tank m and leakage amount M leak Perform the calculation and compare the calculated value with the corresponding true value. Figure 6 Mass flow rate Q in Examples 1-16 m and leakage amount M leak The mean absolute deviation of the calculation results is shown in Table 5. Deviation analysis results indicate that the continuous leakage coefficient C of the atmospheric pressure vertical storage tank based on the feedforward neural network...s The predictive model can be effectively applied to calculate the mass flow rate Q of continuous leakage in storage tanks. m and leakage amount M leak .

[0138] Table 5 Mass flow rate Q in Examples 1-16 m and leakage amount M leak Mean absolute deviation of the calculation results

[0139] Example 1 9.87 4.40 Example 2 8.06 0.16 Example 3 5.53 4.34 Example 4 3.20 0.95 Example 5 7.64 0.29 Example 6 5.56 1.73 Example 7 6.54 3.29 Example 8 5.65 1.09 Example 9 2.06 1.37 Example 10 5.81 0.93 Example 11 6.95 0.66 Example 12 5.67 0.73 Example 13 4.35 0.29 Example 14 2.54 0.57 Example 15 6.58 1.14 Example 16 2.82 1.01

[0140] For other types of atmospheric pressure vertical storage tanks, after obtaining the data on the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank, the continuous leakage coefficient C of the atmospheric pressure vertical storage tank can be determined according to the above method. s The relationship between leakage time t and the leakage time t. Based on the characteristic parameters of continuous leakage in an atmospheric pressure vertical storage tank (liquid level h above the leakage orifice, liquid level drop rate v) h Leakage time t, equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T It can quickly and accurately determine the continuous leakage coefficient C of an atmospheric pressure vertical storage tank. s Based on the information and its changing trends, this invention can be used to calculate the mass flow rate and leakage volume of continuous liquid leakage in atmospheric pressure vertical storage tanks, thereby providing necessary theoretical data support for predicting the diffusion range of storage tanks and assessing risks.

Claims

1. Continuous leakage coefficient of atmospheric pressure vertical storage tank based on feedforward neural network C s The prediction method is characterized by, Includes the following steps: (1) Obtain monitoring data of continuous leakage characteristic parameters of atmospheric pressure vertical storage tank, and construct the continuous leakage coefficient of atmospheric pressure vertical storage tank based on feedforward neural network. C s The predicted dataset; (2) Based on the monitoring data from step (1), and in accordance with the conservation of energy and mass, establish the Bernoulli equation for the continuous leakage process of the storage tank, and calculate the continuous leakage coefficient of the atmospheric pressure vertical storage tank. C s The experimental values; (3) Normalize the continuous leakage characteristic parameter data of the atmospheric pressure vertical storage tank obtained in step (1), and then divide the dataset into 80% training set and 20% test set according to the time sequence corresponding to the data samples. (4) The feedforward neural network method is used to measure the continuous leakage coefficient. C s The predicted value is calculated by taking the normalized result of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank in step (3) as the input data of the feedforward neural network for iterative calculation, and finally generating the continuous leakage coefficient in the output layer. C s Predicted value; (5) Test the feedforward neural network trained in step (4) using the test set from step (3) and obtain the continuous leakage coefficient of the atmospheric pressure vertical storage tank. C s Prediction results; using mean absolute error, explained variance, and coefficient of determination as evaluation indicators, the continuous leakage coefficient of atmospheric pressure vertical storage tanks is assessed. C s The performance of the prediction model was evaluated; mean absolute error (MAE) ≤ 0.1, explained variance (EVS) ≥ 0.7, and coefficient of determination (R²) was [value missing]. 2 A value ≥0.7 indicates that the predicted result tends towards the actual value; this represents the continuous leakage coefficient for atmospheric pressure vertical storage tanks. C s The prediction results are considered valid; (6) Based on the continuous leakage coefficient of the atmospheric pressure vertical storage tank obtained in step (5) C s To calculate the mass flow rate of continuous leakage from the storage tank. Q m and leakage amount M leak ; The feedforward neural network includes an input layer, a hidden layer, and an output layer; the normalized result of the continuous leakage characteristic parameters of the atmospheric pressure vertical storage tank obtained in step (3) is used as the input feature of the input layer and passed from the input layer to the hidden layer; The hidden layer consists of multiple neurons and its function is to extract and transform features from the input data; ultimately, it generates a continuous leakage coefficient in the output layer. C s Predicted value; The inputs and outputs of the hidden layer are as follows: ； ； in, z This is the static input to the neuron; l Indicates the number of layers in the FNN; W This is the weight matrix; f For activation functions; b These are bias parameters; a The output of the neuron; The activation function is the ReLU function, which aims to improve network training speed, help the network adjust parameters more stably, and enable the feedforward neural network model to learn. C s and the nonlinear relationships between its influencing factors: ； in, f ( x ) represents the output value of the neuron; x The input value for the neuron; The weight matrix W and bias parameters b During the training of the feedforward neural network, continuous adjustments are made to make the training output closer to the true value; the degree of difference between the training output and the true value is calculated by the loss function Loss; the mean squared error function is used as the loss function Loss. ； in, During model training, to avoid the loss function getting stuck in a local minimum, the Adam optimizer is used to adjust the learning rate. α Perform adaptive adjustments: in, 。 2. The continuous leakage coefficient of an atmospheric pressure vertical storage tank based on a feedforward neural network as described in claim 1. C s The prediction method is characterized by, The characteristic parameters of continuous leakage in the atmospheric pressure vertical storage tank include the liquid level above the leakage hole. h、 Liquid level drop rate v h Leakage time t Equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T .

3. The continuous leakage coefficient of an atmospheric pressure vertical storage tank based on a feedforward neural network as described in claim 1. C s The prediction method is characterized by, The continuous leakage coefficient of the atmospheric pressure vertical storage tank in step (2) C s The experimental values ​​were used for training and learning the feedforward neural network and for evaluating model performance, as follows: in, 。 4. The continuous leakage coefficient of an atmospheric pressure vertical storage tank based on a feedforward neural network as described in claim 1. C s The prediction method is characterized by, The data normalization process in step (3) is specifically performed as follows: in, x norm The normalized results are for the characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks. x The raw data are for the characteristic parameters of continuous leakage in atmospheric pressure vertical storage tanks; x min This represents the minimum value of the characteristic parameter for continuous leakage in an atmospheric pressure vertical storage tank. x max This represents the maximum value of the characteristic parameter for continuous leakage in an atmospheric pressure vertical storage tank.

5. The continuous leakage coefficient of an atmospheric pressure vertical storage tank based on a feedforward neural network as described in claim 1. C s The prediction method is characterized by, The training set in step (3) is used to learn the leakage characteristic parameters of the atmospheric pressure vertical storage tank and establish the liquid level above the leakage hole. h、 Liquid level drop rate v h Leakage time t Equivalent diameter of leakage hole d o Dynamic viscosity μ and tank diameter d T With continuous leakage coefficient C s The correlation between them; the test set is used to predict the continuous leakage coefficient. C s To verify the effectiveness of the model.

6. The continuous leakage coefficient of an atmospheric pressure vertical storage tank based on a feedforward neural network as described in claim 1. C s The prediction method is characterized by, The mean absolute error: in, 。 7. The continuous leakage coefficient of an atmospheric pressure vertical storage tank based on a feedforward neural network as described in claim 1. C s The prediction method is characterized by, The continuous leakage coefficient C s In the prediction method, when a storage tank experiences continuous leakage, the continuous leakage coefficient is... C s With the time of leakage t It changes with the shape of the leakage hole; the continuous leakage coefficient varies for different shapes of leakage holes. C s It will show a non-linear downward trend to varying degrees.

Images