CN-119692539-B - Intelligent prediction method and system for dangerous chemical loading and unloading leakage diffusion area

CN119692539BCN 119692539 BCN119692539 BCN 119692539BCN-119692539-B

Abstract

The invention relates to the technical field of dangerous goods diffusion prediction, in particular to an intelligent prediction method and system for a dangerous chemical loading and unloading leakage diffusion area, comprising the following steps of processing dangerous goods diffusion data in a dangerous goods leakage diffusion database and dividing the dangerous goods diffusion data into a training set for multi-channel convolutional neural network training and a testing set for multi-channel convolutional neural network testing; training and testing the multichannel convolutional neural network based on the training set and the testing set to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range, and evaluating the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model by utilizing a preset evaluation index. According to the invention, the intelligent prediction of dangerous goods leakage is carried out by adopting the multichannel convolutional neural network model, and the super-parameter tuning is carried out by adopting the network searching method, so that the aim of reasonably dividing dangerous areas is fulfilled, thereby effectively developing emergency and rescue activities, reducing casualties and reducing property loss.

Inventors

CHEN RUOYU
YANG JIHONG
JIANG YUCONG
LI LING
WANG HAINING
LI HAIHANG
LIU XIN

Assignees

中国计量大学

Dates

Publication Date: 20260508
Application Date: 20241202

Claims (6)

1. The intelligent prediction method for the leakage diffusion area of the dangerous chemical loading and unloading is characterized by comprising the following steps of: Based on FLUENT software and the internal and external environments of the dangerous goods loading and unloading area, obtaining factors influencing the diffusion range of dangerous goods, and marking the factors as diffusion range influencing factors; Performing experimental simulation on the dangerous goods leakage diffusion working condition process by using FLUENT software based on the value of the diffusion range influence factor to obtain a dangerous goods diffusion distance; establishing a dangerous goods leakage diffusion database, and collecting and storing dangerous goods diffusion distance data which is formed by the values of diffusion range influence factors and dangerous goods diffusion distances together; Processing dangerous goods diffusion distance data in a dangerous goods leakage diffusion database, and dividing the dangerous goods diffusion distance data into a training set for training the multichannel convolutional neural network and a testing set for testing the multichannel convolutional neural network; training and testing the multichannel convolutional neural network based on the training set and the testing set to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range; Evaluating the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model by using a preset evaluation index; each channel in the multichannel convolution layer carries out independent convolution operation on each diffusion range influence factor of the normalized dangerous goods diffusion distance data by utilizing one-dimensional convolution, and the operation formula of the multichannel convolution layer is as follows: ; in the formula, Represent the first The diffusion range influencing factor sequence of the input of each channel is in time step Is used for the value of (a) and (b), Represent the first The convolution kernel corresponding to each output channel is at the first Positions on the individual channels Is used for the weight of the (c), Represent the first The output channels being in time steps The output value obtained after the convolution operation of (a), Indicating the number of output channels that are to be provided, An index representing an output channel; the pooling layer adopts average pooling operation, and the operation formula of the average pooling is as follows: ; in the formula, Representing pooled features at time steps The value of the above-mentioned value, Representing the size of the pooling window, Representing the average of all values within a pooling window, all values within the pooling window being derived from the convolution operation Constructing; The fusion layer adopts a splicing and fusion mode, and the operation formula of the fusion layer is as follows: ; in the formula, The output value after the splicing is represented, Representing each channel output after pooling operation Is used to determine the integrated value of (1), Representing the number of output channels of the multi-channel convolutional layer; The operation formula of the full connection layer is as follows: ; in the formula, Representing the output of the fully connected layer after the activation function, The activation function is represented as a function of the activation, Representing the output of the fully connected layer before passing through the activation function, Representing the matrix weights of the fully connected layers, The fusion layer representing the full connection layer input outputs an expanded one-dimensional vector, Is a bias vector; Activation function The method comprises the following steps: In which, in the process, Is that Is input to the computer.
2. The method for intelligently predicting a leakage diffusion area for loading and unloading hazardous chemicals according to claim 1, wherein the diffusion range influencing factors comprise leakage aperture size, natural wind speed, ground roughness, leakage height and atmospheric stability.
3. The method for intelligently predicting the leakage diffusion area of the hazardous chemical substance loading and unloading device according to claim 1, wherein the multichannel convolutional neural network is composed of a multichannel convolutional layer, a pooling layer, a fusion layer and a full-connection layer.
4. The method for intelligently predicting the hazardous chemical substance loading and unloading leakage diffusion area according to claim 1, wherein the method for processing the hazardous substance diffusion distance data in the hazardous substance leakage diffusion database comprises the following steps: when dangerous goods diffusion distance data in a dangerous goods leakage diffusion database are processed, a K-nearest neighbor filling method is adopted for a missing value to generate filling data of the missing value, and the K-nearest neighbor filling method formula is as follows: ; In the formula, A value representing the padding required for the missing value, Representing a missing value The one of the most closely adjacent observations is, Representing the number of most similar observations in space; after the missing value supplementation is completed, adopting a normalization processing mode for dangerous goods diffusion distance data, wherein the normalization formula is as follows: ; in the formula, The dangerous goods diffusion distance data after the deficiency value supplementation is completed is represented, Is the minimum value of the same attribute characteristic column of X in the diffusion distance data of dangerous goods, Is the maximum value of the same attribute characteristic column of X in the diffusion distance data of dangerous goods, Is the data value after the X normalization process.
5. The method for intelligently predicting a hazardous chemical substance loading and unloading leakage diffusion region according to claim 1, wherein the evaluation index comprises an adjusted fitting degree Mean absolute error MAE, and AIC criteria, wherein, Degree of fitting after adjustment The operation formula of (2) is as follows: ; The operation formula of the mean absolute error MAE is as follows, ; The operational formula of the AIC criterion is: ; in the formula, The degree of fit before adjustment is indicated, Representing the number of data samples in the test set, Representing the number of data parameters in the test set, Representing the actual diffusion distance of the ith data sample in the test set; Representing the predicted diffusion distance for the ith data sample in the test set, Representing the maximum likelihood function value for a given data, The number of parameters in the dangerous goods loading and unloading leakage diffusion prediction model is represented; Adjusted by the method Can be negative, indicating that the model predictive power employed is not as good as the reference model using mean variables, as Compared with the adjusted The number of model variables is changed, and the problem of deficiency and high caused by the increase of the number of dependent variables is avoided; the loss function MSE operation formula of the dangerous goods loading and unloading leakage diffusion prediction model is as follows: ; in the formula, As a function value of the nerve loss, Representing the number of data samples in the training set, Representing the actual diffusion distance of the ith data sample in the training set, Representing the predicted diffusion distance for the ith data sample in the training set.
6. An intelligent predicting system for a dangerous goods loading and unloading leakage diffusion area, which is characterized by being applied to the intelligent predicting method for the dangerous chemical loading and unloading leakage diffusion area according to any one of claims 1-5, wherein the system comprises the following steps: The influence factor determining unit is used for obtaining factors influencing the diffusion range of dangerous goods based on FLUENT software combined with the internal and external environments of the dangerous goods loading and unloading area, and marking the factors as diffusion range influence factors; the diffusion experiment simulation unit is used for carrying out experimental simulation on the dangerous goods leakage diffusion working condition process by utilizing FLUENT software based on the value of the diffusion range influence factor to obtain the dangerous goods diffusion distance; The data collection and storage unit is used for establishing a dangerous article leakage diffusion database and collecting and storing dangerous article diffusion distance data which is formed by the value of the diffusion range influence factors and the dangerous article diffusion distance together; The data processing unit is used for processing the dangerous goods diffusion distance data in the dangerous goods leakage diffusion database and dividing the dangerous goods diffusion distance data into a training set for training the multichannel convolutional neural network and a testing set for testing the multichannel convolutional neural network; The prediction model construction unit is used for training and testing the multichannel convolutional neural network based on the training set and the testing set to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range; and the prediction model evaluation unit is used for evaluating the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model by using a preset evaluation index.

Description

Intelligent prediction method and system for dangerous chemical loading and unloading leakage diffusion area Technical Field The invention relates to the technical field of diffusion region division, in particular to an intelligent prediction method and system for a leakage diffusion region of dangerous chemicals loading and unloading. Background The dangerous chemicals have dangerous characteristics such as inflammable and explosive, and reasonable division of the diffusion area after the dangerous chemicals leak can promote rescue work to be developed orderly. At present, the traditional method adopted for dividing the diffusion area of dangerous goods leakage is a numerical simulation method, and the concentration of dangerous goods in different areas is calculated by using a diffusion model, so that the division of the diffusion area is carried out. The numerical simulation method has the problems that the calculation is slower in the application process, the prediction result depends on the accuracy of the model, the nonlinearity is processed, the complex data is difficult to process and the like. Therefore, the area division speed is low when the area division is carried out, the area division is inaccurate, and the problems that personnel are not timely cured or the rescue efficiency is low when rescue activities are carried out are caused. Disclosure of Invention The invention aims to provide an intelligent prediction method for a leakage diffusion area of loading and unloading of a hazardous chemical substance, which aims to solve the technical problems that in the prior art, calculation is slow, a prediction result depends on the accuracy of a model, and the processing of nonlinear and complex data is difficult. In order to solve the technical problems, the invention specifically provides the following technical scheme: the intelligent prediction method for the leakage diffusion area of the dangerous chemical loading and unloading is characterized by comprising the following steps of: Based on FLUNT software combined with the internal and external environments of the dangerous goods loading and unloading area, obtaining factors influencing the diffusion range of dangerous goods, and marking the factors as diffusion range influencing factors; Carrying out experimental simulation on the dangerous goods leakage diffusion working condition process by utilizing FLUNT software based on the value of the diffusion range influence factor to obtain the dangerous goods diffusion distance; Establishing a dangerous goods leakage diffusion database, and collecting and storing dangerous goods diffusion distance data which is formed by the values of diffusion range influence factors and dangerous goods diffusion distances together; Processing dangerous goods diffusion data in a dangerous goods leakage diffusion database, and dividing the dangerous goods diffusion data into a training set for training the multichannel convolutional neural network and a testing set for testing the multichannel convolutional neural network; training and testing the multichannel convolutional neural network based on the training set and the testing set to obtain a dangerous goods loading and unloading leakage diffusion prediction model for predicting the dangerous goods diffusion range; And evaluating the diffusion range prediction performance of the dangerous goods loading and unloading leakage diffusion prediction model by using a preset evaluation index. As a preferred embodiment of the present invention, the diffusion range influencing factors include a leakage aperture size, a natural wind speed, a ground roughness, a leakage height, and an atmospheric stability. As a preferred scheme of the invention, the multichannel convolutional neural network is composed of a multichannel convolutional layer, a pooling layer, a fusion layer and a full connection layer. As a preferred scheme of the invention, the method for processing the dangerous goods diffusion data in the dangerous goods leakage diffusion database comprises the following steps: When dangerous goods diffusion data in a dangerous goods leakage diffusion database are processed, a K-nearest neighbor filling method is adopted for the missing values to generate filling data of the missing values, and the K-nearest neighbor filling method comprises the following formula: Wherein New Value represents the Value of the missing Value to be filled, x i represents the K nearest observations of the missing Value, K represents the number of the most similar observations in space; After the missing value supplementation is completed, adopting a normalization processing mode for dangerous goods diffusion data, wherein the normalization formula is as follows: Wherein X represents dangerous goods diffusion data after the deficiency value supplementation is completed, X min is the minimum value of X same-genus characteristic columns in the dangerous goods diffusion data, X max is the maximu