Hazardous chemical substance loading and unloading leakage diffusion area intelligent prediction method and system

Abstract

The invention relates to the technical field of hazardous chemical substance diffusion prediction, in particular to an intelligent prediction method and system for a hazardous chemical substance loading and unloading leakage diffusion area, and the method comprises the following steps: processing hazardous substance diffusion data in a hazardous substance leakage diffusion database; dividing the multi-channel convolutional neural network into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing; training and testing a multi-channel convolutional neural network based on the training set and the test set to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting a dangerous goods diffusion range; and performing diffusion range prediction performance evaluation on the dangerous goods loading and unloading leakage diffusion prediction model by using a preset evaluation index. According to the method, the multi-channel convolutional neural network model is adopted to carry out intelligent prediction of dangerous article leakage, and hyper-parameter tuning is carried out by a network search method, so that the purpose of reasonably dividing dangerous areas is achieved, emergency and rescue activities are effectively carried out, casualties are reduced, and property loss is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of diffusion area division, and in particular to an intelligent prediction method and system for diffusion areas of leakage during loading and unloading of hazardous chemicals. Background Art

[0002] Hazardous chemicals have dangerous characteristics such as being flammable and explosive. After a hazardous chemical leak occurs, the reasonable division of the diffusion area can promote the orderly development of rescue work.

[0003] At present, the traditional method used for the division of diffusion areas of hazardous material leaks is numerical simulation, which uses a diffusion model to calculate the concentration of hazardous materials in different areas, thereby dividing the diffusion area. In the application process, the numerical simulation method has problems such as slow calculation, the prediction results rely on the accuracy of the model, and there are difficulties in processing nonlinear and complex data. Therefore, when dividing the area, the area division speed will be slow and the area division will be inaccurate, resulting in problems such as personnel not receiving timely treatment or low rescue efficiency during rescue activities. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent prediction method for the leakage diffusion area of ​​hazardous chemicals loading and unloading, so as to solve the technical problems in the prior art that the calculation is slow, the prediction result depends on the accuracy of the model, and there are difficulties in processing nonlinear and complex data.

[0005] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0006] A method for intelligently predicting the leakage diffusion area of ​​hazardous chemicals loading and unloading, characterized by comprising the following steps:

[0007] Based on the FLUNT software and the internal and external environment of the dangerous goods loading and unloading area, the factors that affect the diffusion range of dangerous goods are obtained and marked as diffusion range influencing factors;

[0008] Based on the values ​​of the factors affecting the diffusion range, the FLUNT software was used to simulate the diffusion process of dangerous goods leakage to obtain the diffusion distance of dangerous goods.

[0009] Establish a dangerous goods leakage and diffusion database to collect and store dangerous goods diffusion distance data composed of the values ​​of diffusion range influencing factors and the dangerous goods diffusion distance;

[0010] The dangerous goods diffusion data in the dangerous goods leakage diffusion database is processed and divided into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing;

[0011] Based on the training set and the test set, a multi-channel convolutional neural network is trained and tested to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range;

[0012] The diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model is evaluated using preset evaluation indicators.

[0013] As a preferred solution of the present invention, the factors affecting the diffusion range include leakage aperture size, natural wind speed, ground roughness, leakage height, and atmospheric stability.

[0014] As a preferred solution of the present invention, the multi-channel convolutional neural network is composed of a multi-channel convolutional layer, a pooling layer, a fusion layer and a fully connected layer.

[0015] As a preferred solution of the present invention, a method for processing dangerous goods diffusion data in a dangerous goods leakage diffusion database includes:

[0016] When processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database, the K-nearest neighbor filling method is used to generate missing value filling data for missing values. The formula of the K-nearest neighbor filling method is:

[0017]

[0018] In the formula, New Value represents the value to be filled in for missing values, x i Represents the K nearest observations of the missing value, where K represents the number of most similar observations in space;

[0019] After the missing values ​​are supplemented, the dangerous goods diffusion data is normalized. The normalization formula is:

[0020]

[0021] In the formula, X represents the dangerous goods diffusion data after missing values ​​are supplemented, X min is the minimum value of the feature column X in the dangerous goods diffusion data. max is the maximum value of the feature column X in the dangerous goods diffusion data. norm is the data value after X normalization.

[0022] As a preferred solution of the present invention, each channel in the multi-channel convolution layer uses one-dimensional convolution to perform an independent convolution operation on each diffusion range influencing factor of the normalized dangerous goods diffusion data. The calculation formula of the multi-channel convolution layer is:

[0023]

[0024] In the formula, Ik (t) represents the value of the diffusion range influencing factor sequence of the k-th channel input at time step t, S c,k (m) represents the weight of the convolution kernel corresponding to the c-th output channel at position m on the k-th channel, (I*S)(c,t) represents the output value of the c-th output channel after the convolution operation at time step t, C in represents the number of output channels, and c represents the index of the output channel.

[0025] As a preferred solution of the present invention, the pooling layer adopts an average pooling operation, and the calculation formula of the average pooling is:

[0026]

[0027] Wherein, P(u) represents the value of the pooled feature at time step u, |window| represents the size of the pooling window, and I(u+i) represents the average value of all values ​​in the pooling window. All values ​​in the pooling window are composed of (I*S)(c,t) obtained after the convolution operation.

[0028] As a preferred solution of the present invention, the fusion layer adopts a splicing fusion method, and the calculation formula of the fusion layer is:

[0029]

[0030] In the formula, F represents the output value after splicing, F i Represents the integrated value of each channel P(u) output after the pooling operation, C in Indicates the number of output channels of a multi-channel convolutional layer.

[0031] As a preferred solution of the present invention, the operation formula of the fully connected layer is:

[0032] A=f(Z)=f(W*H+b);

[0033] In the formula, A represents the output of the fully connected layer after the activation function, f represents the activation function, Z represents the output of the fully connected layer before the activation function, W represents the matrix weight of the fully connected layer, H represents the one-dimensional vector after the fusion layer output of the fully connected layer input is expanded, and b is the bias vector;

[0034] The activation function f is: Where x is the input of ReLU.

[0035] As a preferred solution of the present invention, the evaluation index includes the adjusted fit R 2 best , mean absolute error MAE and AIC criteria, where

[0036] Adjusted goodness of fit R 2 best The calculation formula is:

[0037] The calculation formula of mean absolute error MAE is: The calculation formula of AIC criterion is: AIC = -2In(L) + 2a;

[0038] In the formula, R 2 represents the fit before adjustment, v represents the number of data samples in the test set, w represents the number of data parameters in the test set, and q i represents the actual diffusion distance of the i-th data sample in the test set; represents the predicted diffusion distance of the i-th data sample in the test set, L represents the maximum likelihood function value under the given data, and a represents the number of parameters in the dangerous goods loading and unloading leakage diffusion prediction model;

[0039] Adjusted R 2 Can be negative, indicating that the adopted model is less predictive than the baseline model using the mean variable. 2 Compared with the adjusted R 2 The number of model variables was changed to avoid the problem of artificial inflated values ​​caused by the increase in the number of variables.

[0040] The loss function MSE calculation formula of the dangerous goods loading and unloading leakage diffusion prediction model is:

[0041]

[0042] In the formula, MSE is the value of the neural loss function, n represents the number of data samples in the training set, and y i Represents the actual diffusion distance of the i-th data sample in the training set. Represents the predicted diffusion distance of the i-th data sample in the training set.

[0043] As a preferred solution of the present invention, the present invention provides an intelligent prediction system for leakage diffusion area of ​​hazardous chemicals loading and unloading, which is applied to an intelligent prediction method for leakage diffusion area of ​​hazardous chemicals loading and unloading, and the system includes:

[0044] The influencing factor determination unit is used to obtain the factors that affect the spread of dangerous goods based on the FLUNT software and the internal and external environment of the dangerous goods loading and unloading area, and mark them as the factors affecting the spread of dangerous goods;

[0045] The diffusion experiment simulation unit is used to simulate the diffusion process of dangerous goods leakage based on the values ​​of the factors affecting the diffusion range using the FLUNT software to obtain the diffusion distance of dangerous goods;

[0046] A data collection and storage unit is used to establish a dangerous goods leakage and diffusion database, collect and store dangerous goods diffusion distance data composed of the values ​​of diffusion range influencing factors and the dangerous goods diffusion distance;

[0047] A data processing unit, used for processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database, and dividing it into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing;

[0048] A prediction model building unit, used for training and testing the multi-channel convolutional neural network based on the training set and the test set, to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range;

[0049] The prediction model evaluation unit is used to evaluate the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model using preset evaluation indicators.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention simulates the accident conditions of hazardous materials leakage through FLUNT software, establishes data sets and test sets for training models, adopts a multi-channel convolutional neural network model for intelligent prediction of hazardous materials leakage, and uses a network search method to perform hyperparameter tuning to achieve the purpose of reasonably dividing dangerous areas, thereby effectively carrying out emergency and rescue activities, reducing casualties and reducing property losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0053] Figure 1 A flow chart of an intelligent prediction method for leakage and diffusion areas of hazardous chemicals loading and unloading provided by an embodiment of the present invention;

[0054] Figure 2 A block diagram of an intelligent prediction system for leakage and diffusion areas of hazardous chemicals loading and unloading provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] like Figure 1 As shown, the present invention provides an intelligent prediction method for the leakage diffusion area of ​​hazardous chemicals loading and unloading, which is characterized by comprising the following steps:

[0057] Based on the FLUNT software and the internal and external environment of the dangerous goods loading and unloading area, the factors that affect the diffusion range of dangerous goods are obtained and marked as diffusion range influencing factors;

[0058] Based on the values ​​of the factors affecting the diffusion range, the FLUNT software was used to simulate the diffusion process of dangerous goods leakage to obtain the diffusion distance of dangerous goods.

[0059] Establish a dangerous goods leakage and diffusion database to collect and store dangerous goods diffusion distance data composed of the values ​​of diffusion range influencing factors and the dangerous goods diffusion distance;

[0060] The dangerous goods diffusion data in the dangerous goods leakage diffusion database is processed and divided into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing;

[0061] The multi-channel convolutional neural network is trained and tested based on the training set and the test set to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range;

[0062] The diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model is evaluated using preset evaluation indicators.

[0063] In the present invention, the values ​​of the diffusion range influencing factors determine the diffusion distance of dangerous goods. When performing multi-channel convolutional neural network learning, it is necessary to use the values ​​of the diffusion range influencing factors and the diffusion distance of dangerous goods. Using a multi-channel convolutional neural network for learning can ultimately achieve the purpose of predicting the diffusion distance of dangerous goods based on the values ​​of the diffusion range influencing factors, thereby realizing the modeling and objective prediction of the diffusion distance of dangerous goods, efficiently obtaining the diffusion range of dangerous goods, and meeting the timeliness requirements of dangerous goods rescue.

[0064] The factors affecting the diffusion range include the size of the leakage aperture, natural wind speed, ground roughness, leakage height, and atmospheric stability, which can be added or deleted as needed in actual use.

[0065] The multi-channel convolutional neural network consists of a multi-channel convolutional layer, a pooling layer, a fusion layer and a fully connected layer.

[0066] The method for processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database includes:

[0067] When processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database, the K-nearest neighbor filling method is used to generate missing value filling data for missing values. The formula of the K-nearest neighbor filling method is:

[0068]

[0069] In the formula, New Value represents the value to be filled in for missing values, x i Represents the K nearest observations of the missing value, where K represents the number of most similar observations in space;

[0070] After the missing values ​​are supplemented, the dangerous goods diffusion data is normalized. The normalization formula is:

[0071]

[0072] In the formula, X represents the dangerous goods diffusion data after missing values ​​are supplemented, X min is the minimum value of the feature column X in the dangerous goods diffusion data. max is the maximum value of the feature column X in the dangerous goods diffusion data. norm is the data value after X normalization.

[0073] Each channel in the multi-channel convolution layer uses one-dimensional convolution to perform an independent convolution operation on each diffusion range influencing factor of the normalized dangerous goods diffusion data. The calculation formula of the multi-channel convolution layer is:

[0074]

[0075] In the formula, I k (t) represents the value of the diffusion range influencing factor sequence of the k-th channel input at time step t, S c,k (m) represents the weight of the convolution kernel corresponding to the c-th output channel at position m on the k-th channel, (I*S)(c,t) represents the output value of the c-th output channel after the convolution operation at time step t, C in represents the number of output channels, and c represents the index of the output channel.

[0076] The pooling layer adopts the average pooling operation, and the calculation formula of the average pooling is:

[0077]

[0078] Wherein, P(u) represents the value of the pooled feature at time step u, |window| represents the size of the pooling window, and I(u+i) represents the average value of all values ​​in the pooling window. All values ​​in the pooling window are composed of (I*S)(c,t) obtained after the convolution operation.

[0079] The fusion layer adopts the splicing fusion method, and the calculation formula of the fusion layer is:

[0080]

[0081] In the formula, F represents the output value after splicing, F i Represents the integrated value of each channel P(u) output after the pooling operation, C in Indicates the number of output channels of a multi-channel convolutional layer.

[0082] The calculation formula of the fully connected layer is:

[0083] A=f(Z)=f(W*H+b);

[0084] In the formula, A represents the output of the fully connected layer after the activation function, f represents the activation function, Z represents the output of the fully connected layer before the activation function, W represents the matrix weight of the fully connected layer, H represents the one-dimensional vector after the fusion layer output of the fully connected layer input is expanded, and b is the bias vector;

[0085] The activation function f is: Where x is the input of ReLU.

[0086] The evaluation indicators include the adjusted goodness of fit R 2 best , mean absolute error MAE and AIC criteria, where

[0087] Adjusted goodness of fit R 2 best The calculation formula is:

[0088] The calculation formula of mean absolute error MAE is:

[0089] The calculation formula of AIC criterion is: AIC = -2In(L) + 2a;

[0090] In the formula, R 2 represents the fit before adjustment, v represents the number of data samples in the test set, w represents the number of data parameters in the test set, and q i represents the actual diffusion distance of the i-th data sample in the test set; represents the predicted diffusion distance of the i-th data sample in the test set, L represents the maximum likelihood function value under the given data, and a represents the number of parameters in the dangerous goods loading and unloading leakage diffusion prediction model;

[0091] The adjusted fitting degree R in the present invention 2 best It can be negative, indicating that the model used is less predictive than the baseline model using the mean variable, which is different from the adjusted R 2 Compared with the adjusted goodness of fit R 2 best Changing the number of model variables avoids the problem of false high values ​​caused by the increase in the number of variables, and can more accurately feedback the performance of the model, so as to provide more accurate tuning guidance in model parameter tuning, and finally obtain the optimal model. Therefore, the adjusted fit R 2 best It can ensure that the leakage and diffusion prediction model for dangerous goods loading and unloading is at its best performance.

[0092] In the present invention, MAE is an indicator used to measure the accuracy of the prediction model. It is the average of the absolute values ​​of the difference between the predicted value and the true value. The smaller the value of MAE, the more accurate the prediction result of the model. MAE can reflect the performance of the model in the presence of noisy data because it is less sensitive to outliers and will not be disturbed by noisy data. In addition, MAE directly reflects the prediction accuracy of the model and will not be masked by the positive and negative values ​​of the prediction error.

[0093] The AIC criterion in the present invention is a standard for measuring the goodness of statistical model fitting, which is based on the concept of information entropy and aims to weigh the complexity of the model and the quality of the model fitting data.

[0094] Adjusted goodness of fit R 2 best Combining the mean absolute error (MAE) and the AIC criterion, it is possible to avoid the problem of inflated indicators and restore the most objective evaluation index values ​​while ensuring the accuracy and complexity performance of the MAE and AIC criterion evaluation models, thereby achieving the most objective evaluation of the model. This can be used as a tuning benchmark to ultimately ensure that the hazardous goods loading and unloading leakage and diffusion prediction model obtains the best guidance and achieves the best performance.

[0095] The loss function MSE calculation formula of the dangerous goods loading and unloading leakage diffusion prediction model is:

[0096]

[0097] In the formula, MSE is the value of the neural loss function, n represents the number of data samples in the training set, and y i Represents the actual diffusion distance of the i-th data sample in the training set. Represents the predicted diffusion distance of the i-th data sample in the training set.

[0098] like Figure 2 As shown, the present invention provides an intelligent prediction system for leakage diffusion area of ​​hazardous chemicals loading and unloading, which is applied to an intelligent prediction method for leakage diffusion area of ​​hazardous chemicals loading and unloading. The system includes:

[0099] The influencing factor determination unit is used to obtain the factors that affect the spread of dangerous goods based on the FLUNT software and the internal and external environment of the dangerous goods loading and unloading area, and mark them as the factors affecting the spread of dangerous goods;

[0100] The diffusion experiment simulation unit is used to simulate the diffusion process of dangerous goods leakage based on the values ​​of the factors affecting the diffusion range using the FLUNT software to obtain the diffusion distance of dangerous goods;

[0101] A data collection and storage unit is used to establish a dangerous goods leakage and diffusion database, collect and store dangerous goods diffusion distance data composed of the values ​​of diffusion range influencing factors and the dangerous goods diffusion distance;

[0102] A data processing unit, used for processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database, and dividing it into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing;

[0103] A prediction model building unit, used for training and testing the multi-channel convolutional neural network based on the training set and the test set, to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range;

[0104] The prediction model evaluation unit is used to evaluate the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model using preset evaluation indicators.

[0105] The present invention simulates the accident conditions of hazardous materials leakage through FLUNT software, establishes data sets and test sets for training models, adopts a multi-channel convolutional neural network model for intelligent prediction of hazardous materials leakage, and uses a network search method to perform hyperparameter tuning to achieve the purpose of reasonably dividing dangerous areas, thereby effectively carrying out emergency and rescue activities, reducing casualties and reducing property losses.

[0106] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. An intelligent prediction method for the leakage diffusion area of ​​hazardous chemicals loading and unloading, characterized in that: The following steps are involved: Based on the FLUNT software and the internal and external environment of the dangerous goods loading and unloading area, the factors that affect the diffusion range of dangerous goods are obtained and marked as diffusion range influencing factors; Based on the values ​​of the factors affecting the diffusion range, the FLUNT software was used to simulate the diffusion process of dangerous goods leakage to obtain the diffusion distance of dangerous goods. Establish a dangerous goods leakage and diffusion database to collect and store dangerous goods diffusion distance data composed of the values ​​of diffusion range influencing factors and the dangerous goods diffusion distance; The dangerous goods diffusion data in the dangerous goods leakage diffusion database is processed and divided into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing; Based on the training set and the test set, a multi-channel convolutional neural network is trained and tested to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range; The diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model is evaluated using preset evaluation indicators.

2. According to claim 1, a method for intelligent prediction of leakage diffusion area of ​​hazardous chemicals loading and unloading, characterized in that: The factors affecting the diffusion range include the size of the leakage aperture, natural wind speed, ground roughness, leakage height, and atmospheric stability.

3. According to claim 1, a method for intelligent prediction of leakage diffusion area of ​​hazardous chemicals loading and unloading, characterized in that: The multi-channel convolutional neural network consists of a multi-channel convolutional layer, a pooling layer, a fusion layer and a fully connected layer.

4. According to claim 1, a method for intelligent prediction of leakage diffusion area of ​​hazardous chemicals loading and unloading, characterized in that: The method for processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database includes: When processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database, the K-nearest neighbor filling method is used to generate missing value filling data for missing values. The formula of the K-nearest neighbor filling method is: In the formula, New Value represents the value to be filled in for missing values, x i Represents the K nearest observations of the missing value, where K represents the number of most similar observations in space; After the missing values ​​are supplemented, the dangerous goods diffusion data is normalized. The normalization formula is: In the formula, X represents the dangerous goods diffusion data after missing values ​​are supplemented, X min is the minimum value of the feature column X in the dangerous goods diffusion data. max is the maximum value of the feature column X in the dangerous goods diffusion data. norm is the data value after X normalization.

5. According to claim 4, a method for intelligent prediction of leakage diffusion area during loading and unloading of hazardous chemicals is characterized by: Each channel in the multi-channel convolution layer uses one-dimensional convolution to perform an independent convolution operation on each diffusion range influencing factor of the normalized dangerous goods diffusion data. The calculation formula of the multi-channel convolution layer is: In the formula, I k (t) represents the value of the diffusion range influencing factor sequence of the k-th channel input at time step t, S c,k (m) represents the weight of the convolution kernel corresponding to the c-th output channel at position m on the k-th channel, (I*S)(c,t) represents the output value of the c-th output channel after the convolution operation at time step t, C in represents the number of output channels, and c represents the index of the output channel.

6. The method for intelligent prediction of leakage diffusion area during loading and unloading of hazardous chemicals according to claim 5 is characterized by: The pooling layer adopts the average pooling operation, and the calculation formula of the average pooling is: Wherein, P(u) represents the value of the pooled feature at time step u, |window| represents the size of the pooling window, and I(u+i) represents the average value of all values ​​in the pooling window. All values ​​in the pooling window are composed of (I*S)(c,t) obtained after the convolution operation.

7. The method for intelligent prediction of leakage diffusion area during loading and unloading of hazardous chemicals according to claim 6 is characterized by: The fusion layer adopts the splicing fusion method, and the calculation formula of the fusion layer is: In the formula, F represents the output value after splicing, F i Represents the integrated value of each channel P(u) output after the pooling operation, Ci n Indicates the number of output channels of a multi-channel convolutional layer.

8. The method for intelligent prediction of leakage diffusion area during loading and unloading of hazardous chemicals according to claim 7 is characterized by: The calculation formula of the fully connected layer is: A=f(Z)=f(W*H+b); In the formula, A represents the output of the fully connected layer after the activation function, f represents the activation function, Z represents the output of the fully connected layer before the activation function, W represents the matrix weight of the fully connected layer, H represents the one-dimensional vector after the fusion layer output of the fully connected layer input is expanded, and b is the bias vector; The activation function f is: Where x is the input of ReLU.

9. The method for intelligent prediction of leakage diffusion area during loading and unloading of hazardous chemicals according to claim 1 is characterized by: The evaluation indicators include the adjusted goodness of fit R 2 best , mean absolute error MAE and AIC criteria, where Adjusted goodness of fit R 2 The calculation formula of best is: The calculation formula of mean absolute error MAE is: The calculation formula of AIC criterion is: AIC = -2In(L) + 2a; In the formula, R 2 represents the fit before adjustment, v represents the number of data samples in the test set, w represents the number of data parameters in the test set, and q i represents the actual diffusion distance of the i-th data sample in the test set; represents the predicted diffusion distance of the i-th data sample in the test set, L represents the maximum likelihood function value under the given data, and a represents the number of parameters in the dangerous goods loading and unloading leakage diffusion prediction model; Adjusted R 2 Can be negative, indicating that the adopted model is less predictive than the baseline model using the mean variable. 2 Compared with the adjusted R 2 The number of model variables was changed to avoid the problem of artificial inflated values ​​caused by the increase in the number of variables. The loss function MSE calculation formula of the dangerous goods loading and unloading leakage diffusion prediction model is: In the formula, MSE is the value of the neural loss function, n represents the number of data samples in the training set, and y i Represents the actual diffusion distance of the i-th data sample in the training set. Represents the predicted diffusion distance of the i-th data sample in the training set.

10. An intelligent prediction system for leakage and diffusion areas during dangerous goods loading and unloading, characterized in that: A method for intelligently predicting the leakage diffusion area of ​​hazardous chemicals loading and unloading as described in any one of claims 1 to 9, the system comprising: The influencing factor determination unit is used to obtain the factors that affect the spread of dangerous goods based on the FLUNT software and the internal and external environment of the dangerous goods loading and unloading area, and mark them as the factors affecting the spread of dangerous goods; The diffusion experiment simulation unit is used to simulate the diffusion process of dangerous goods leakage based on the values ​​of the factors affecting the diffusion range using the FLUNT software to obtain the diffusion distance of dangerous goods; A data collection and storage unit is used to establish a dangerous goods leakage and diffusion database, collect and store dangerous goods diffusion distance data composed of the values ​​of diffusion range influencing factors and the dangerous goods diffusion distance; A data processing unit, used for processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database, and dividing it into a training set for multi-channel convolutional neural network training and a test set for multi-channel convolutional neural network testing; A prediction model building unit, used for training and testing the multi-channel convolutional neural network based on the training set and the test set, to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range; The prediction model evaluation unit is used to evaluate the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model using preset evaluation indicators.

Images