Danger source troubleshooting and early warning method and device for chemical hazardous chemical substance safety management

Abstract

The invention relates to the technical field of data processing, in particular to a dangerous source troubleshooting and early warning method and device for chemical hazardous chemical substance safety management, and the method comprises the steps: collecting field sensor data from a chemical enterprise, safety check records and accident report data; sample generation is carried out by adopting a multi-linear interpolation algorithm based on feature decoupling, and chemical production data expansion is realized; and inputting the expanded chemical production data into a pre-constructed hazard source troubleshooting and early warning model, performing hazard level classification prediction on the chemical production data through the hazard source troubleshooting and early warning model, and determining a hazard classification result. The beneficial effects of the invention are that the method can achieve the quick and accurate recognition of potential hazard sources in a chemical plant environment, improves the efficiency and accuracy of emergency response, guarantees the continuity and safety of chemical production through the real-time data processing capability, reduces the risk of human error judgment, and enhances the timeliness of prevention and control measures.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a hazard source screening and early warning method and device for the safety management of hazardous chemicals. Background Art

[0002] The chemical industry has a high level of safety risks due to the extensive use of various hazardous chemicals in its production process. Effective identification and management of hazardous sources are key measures to ensure production safety and prevent accidents. Traditional safety management methods rely on regular safety inspections and manual data analysis, which are not only time-consuming and labor-intensive, but also difficult to cope with rapidly changing production environments and potential dangerous situations due to the lack of real-time monitoring and prediction capabilities.

[0003] The existing Chinese invention patent CN202211303919.9 proposes a chemical production safety early warning system based on machine learning, which relates to the field of chemical production safety technology. The system includes: at least one data acquisition device, a cloud management device and an early warning management device, and the cloud management device is connected to the data acquisition device and the early warning management device respectively; the data acquisition device is used to collect production environment data from the chemical workshop and send the production environment data to the cloud management device; the cloud management device is used to build a prediction model based on machine learning, and determine the real-time fault early warning information of the chemical workshop based on the production environment data and the prediction model; the early warning management device is used to send early warning information to the preset personnel based on the real-time fault early warning information sent by the cloud management device. It relates to the field of communication technology. The method includes: the scheme of the present invention can realize the prediction of safety early warning of chemical workshops, and solve the problem of inaccurate human prediction of chemical workshops.

[0004] The existing Chinese invention patent CN202210618538.3 proposes a method for energy consumption diagnosis and analysis of chemical enterprises, which involves the field of energy consumption diagnosis and analysis technology, including obtaining a data set of various process parameters through a database; determining the process parameters related to the energy consumption target and the corresponding parameter interval through the data set; obtaining the weights corresponding to the various process parameters related to the energy consumption target, and using them to determine the combined parameters related to the energy consumption target and the corresponding parameter interval; comparing the obtained real-time data of the process parameters with the parameter interval of the combined parameters to determine the process parameters that affect the energy consumption target. The present invention uses machine learning to solve the problem of this industrial scenario and realize the planning of energy-saving measures.

[0005] The existing Chinese invention patent CN201810005213.1 proposes a chemical park early warning method and device based on machine learning, which obtains the historical sensor data of the target location, and then classifies the historical sensor data according to the early warning status, and finally inputs the classified data into a pre-established mathematical model for training to obtain a prediction model. Since each historical sensor data is real and has a corresponding early warning status, in addition, the model does not train a single data, so it can better reflect the correlation between the historical sensor data, making the early warning model more accurate. After obtaining the prediction model, the real-time data is input into the prediction model, and the prediction result obtained is more accurate. In addition, the calculation process of this method is relatively simple, and no manual calculation is required, saving labor costs.

[0006] However, the above technical solutions also have the following problems that need to be further solved: 1. In the task of hazard source investigation for the safety management of hazardous chemicals, traditional methods may not fully consider the imbalance and sparse areas of chemical production data sets, resulting in insufficient recognition of minority classes and limited generalization ability of the model.

[0007] 2. In the task of hazard source investigation for the safety management of hazardous chemicals, traditional feature extraction technology may not use effective optimization strategies, resulting in slow model training speed, easy to fall into local optimal solutions, and insufficient feature generalization ability.

[0008] 3. In the task of hazard source investigation for the safety management of hazardous chemicals, traditional classifiers may not integrate competitive learning mechanisms and fuzzy logic processing strategies, resulting in insufficient accuracy and robustness of the model when processing uncertain and fuzzy chemical production data. Summary of the invention

[0009] In order to address the deficiencies in the prior art, the present invention provides a hazard source investigation and early warning method and device for the safe management of hazardous chemicals in chemical industries, which can realize the rapid and accurate identification of potential hazard sources in the chemical plant environment, improve the efficiency and accuracy of emergency response, and the real-time data processing capability ensures the continuity and safety of chemical production, reduces the risk of human misjudgment, and enhances the timeliness of prevention and control measures.

[0010] In order to achieve the above object, the present invention is implemented through the following technical solutions: According to one aspect of the present invention, a method for hazard source investigation and early warning for safety management of hazardous chemicals is provided: Collect data from on-site sensors, safety inspection records and accident reports from chemical companies; The multilinear interpolation algorithm based on feature decoupling is used to generate samples and realize chemical production data expansion; Input the expanded chemical production data into the pre-built hazard source investigation and early warning model, and use the hazard source investigation and early warning model to classify and predict the hazard level of the chemical production data to determine the hazard classification result; Among them, the hazard source investigation and early warning model is constructed based on the extreme learning machine classification algorithm based on the neuron competitive learning mechanism, and the fuzzy logic quantum coding strategy is adopted to enable the extreme learning machine classification algorithm to better handle the chemical production data after the uncertainty and fuzzy features are extracted in the safety management of chemical hazardous chemicals.

[0011] Furthermore, the multilinear interpolation algorithm based on feature decoupling is specifically as follows: The features in the collected chemical production data set are decoupled according to correlation and contribution, and divided into multiple independent subspaces, each of which contains a set of relevant features that affect a specific output variable; In each decoupled subspace, multilinear interpolation is applied independently; According to the category distribution of the original chemical production data set, the number of samples generated by interpolation in each subspace is dynamically adjusted; The new samples generated by interpolation in all subspaces are recombined according to the original feature structure to form a new chemical production data set. The feature combination of each subspace represents the important information in different stages or processes of chemical production. Quality control is performed on the expanded chemical production data set to detect and eliminate possible outliers or chemical production data samples that do not conform to physical meaning.

[0012] Furthermore, the construction method of the hazard source investigation and early warning model is as follows: Obtain expanded chemical production data; Extract features from chemical production data using a pre-built feature extraction model; The chemical production data after feature extraction is input into the classifier to train the classifier model.

[0013] Furthermore, the feature extraction model is a neural network based on exploratory gradient descent, specifically: Initialize the parameters of the neural network, including the weights and bias parameters of the neural network; The input chemical production data passes through each layer of the neural network, and each layer is processed by linear transformation and nonlinear activation function, and finally the feature representation of the output layer is obtained; An exploratory gradient descent update strategy is adopted to predict possible gradient changes before each update of the neural network parameters. The step size and direction of the gradient update are adjusted according to the prediction results to optimize the parameter adjustment during the training process.

[0014] Furthermore, the extreme learning machine classification algorithm based on the neuron competition learning mechanism is specifically as follows: Initialize the weights and biases of the extreme learning machine; The feature-extracted chemical production data input into the extreme learning machine passes through the hidden layer to simulate the competitive inhibition effect in the biological nervous system; The feature conversion from the hidden layer to the output layer adopts a linear combination method, and the optimal weight is determined by minimizing the training error and regularization term; In the classification decision stage, fuzzy logic rules are used to deal with the uncertainty of classification; During the iterative training process, the weights and biases of the hidden layers, as well as the fuzzy logic rules of the output layer, are fine-tuned.

[0015] Furthermore, the hazard classification results include "safety", "caution" and "danger".

[0016] Furthermore, the optimal weight is calculated by minimizing the training error and the regularization term as follows: ; In the formula, is the weight from the hidden layer to the output layer; is the target output; is the transpose of the output of the hidden layer; is the output of the hidden layer; is the regularization parameter; is the identity matrix.

[0017] According to another aspect of the present invention, there is provided a hazard source investigation device for safety management of hazardous chemicals, comprising: Data collection module, used to collect sensor data, safety inspection records and accident report data from chemical companies; A data expansion module, used to expand the collected data; An execution module is used to input the expanded chemical production data into a pre-built hazard source investigation and early warning model, and to classify and predict the hazard level of the chemical production data through the hazard source investigation and early warning model to determine the hazard classification result; The output module is used to predict and analyze the dangerous consequences of chemical equipment based on the classification results.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. In the task of hazard source investigation for the safety management of hazardous chemicals, when processing and modeling chemical production data, a multiple linear interpolation algorithm based on feature decoupling is used to perform linear interpolation on independent subspaces to generate new samples. The interpolation step size and direction of each subspace are dynamically adjusted through adaptive resampling technology, and the interpolation strategy is automatically adjusted according to the imbalance and sparse areas in the data set.

[0019] 2. In the task of hazard source investigation for the safety management of hazardous chemicals, when processing and modeling chemical production data, a fully connected neural network with multiple hidden layers is used for feature extraction, and an optimization method based on exploratory gradient descent is used to predict the gradient change trend to adjust the learning step size and direction.

[0020] 3. In the task of hazard source investigation for the safety management of hazardous chemicals, when processing and modeling chemical production data, the extreme learning machine classification algorithm based on the neuron competitive learning mechanism is adopted, combined with the fuzzy logic quantum coding strategy, to improve the processing ability of uncertainty and fuzzy data. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Attached Figure 1 It is a flowchart of the present invention; Attached Figure 2 It is a training flow chart of the neural network algorithm based on exploratory gradient descent of the present invention. DETAILED DESCRIPTION

[0022] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the application equally.

[0023] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show the components related to the present disclosure rather than the number, shape and size of the components in actual implementation. The type, quantity and proportion of each component in actual implementation may be changed at will, and the component layout type may also be more complicated. In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that aspects can be practiced without these specific details.

[0024] Among them, a hazard source investigation and early warning method and device for the safe management of chemical hazardous chemicals provided by an embodiment of the present invention can realize the rapid and accurate identification of potential hazard sources in the chemical plant environment, improve the efficiency and accuracy of emergency response, and the real-time data processing capability ensures the continuity and safety of chemical production, reduces the risk of human misjudgment, and enhances the timeliness of prevention and control measures.

[0025] To facilitate understanding of this embodiment, a method for hazard source investigation and early warning for safety management of hazardous chemicals disclosed in an embodiment of the present invention is first introduced in detail. Figure 1 A method for checking and warning of hazardous sources for safety management of hazardous chemicals provided by an embodiment of the present invention is shown. Figure 1 As shown, the method comprises the following steps: S1. Collect data from on-site sensors, safety inspection records and accident reports from chemical companies; S2, using a multi-linear interpolation algorithm based on feature decoupling to generate samples and expand chemical production data; S3. Input the expanded chemical production data into the pre-built hazard source investigation and early warning model, and use the hazard source investigation and early warning model to classify and predict the hazard level of the chemical production data to determine the hazard classification result; Among them, the hazard source investigation and early warning model is constructed based on the extreme learning machine classification algorithm based on the neuron competitive learning mechanism, and the fuzzy logic quantum coding strategy is adopted to enable the extreme learning machine classification algorithm to better handle the chemical production data after the uncertainty and fuzzy features are extracted in the safety management of chemical hazardous chemicals.

[0026] Specifically, the data of the set is stored in a structured JSON format. In one embodiment, the attributes of the data include: is the temperature (in degrees Celsius); is the pressure (in Pascal); is the liquid level (in meters); is the chemical component concentration (in mole fraction); is the valve status (unit: open / closed); is the flow rate (in liters / minute); is the sensor status (unit: normal / fault); is the risk level (in units: low / medium / high); The device operation time (unit: hours); It is the record of emergency events (unit: yes / no).

[0027] In the task of the present invention, the collection, acquisition, labeling and preprocessing of chemical production training data are time-consuming and labor-intensive, and insufficient training samples easily lead to poor model generalization ability, while affecting the accuracy of the model. The present invention adopts a multilinear interpolation algorithm based on feature decoupling to generate samples, thereby realizing chemical production data expansion. The multilinear interpolation algorithm based on feature decoupling separates the chemical production data features into independent subspaces, applies linear interpolation to each subspace, and then reorganizes the results to generate new chemical production data samples, and adopts adaptive resampling technology to dynamically adjust the interpolation step size and direction of each subspace. By monitoring the imbalance and sparse areas in the original chemical production data set, the algorithm automatically adjusts the interpolation strategy, preferentially generates chemical production data of fewer sample categories, enhances the model's learning ability for minority classes, and thus improves the overall model generalization ability.

[0028] Specifically, the multilinear interpolation algorithm based on feature decoupling is: S21. Decouple the features in the collected chemical production data set according to correlation and contribution, and divide them into multiple independent subspaces. Each subspace contains a set of related features that affect a specific output variable. Specifically, decouple the input chemical production data set and divide it into multiple independent subspaces. Chemical production data usually has highly complex and multidimensional characteristics. For example, there may be strong correlations between features such as reaction temperature, pressure, flow rate, and concentration. Through feature decoupling technology, these features are divided into multiple independent subspaces based on principal component analysis. Each subspace contains features with high correlation. For example, the reaction temperature and pressure are assigned to one subspace, and the raw material concentration and flow rate are assigned to another subspace. This decoupling process reduces feature redundancy and highlights the independent effects of different feature combinations on the chemical process. The decoupling process is expressed as: ; In the formula, is the original chemical production data set, is the independent subspace obtained after segmentation, is the number of principal components retained, Represents the principal component analysis function.

[0029] In one embodiment, the process of performing feature decoupling using the principal component analysis function is expressed as: ; In the formula, , and The covariance matrix of chemical production data is The obtained singular value decomposition matrix is is the first decomposition matrix, is the second decomposition matrix, is the third decomposition matrix. Contains the largest singular values, and a decomposition matrix and the second decomposition matrix Contains the corresponding singular vectors.

[0030] S22. In each decoupled subspace, multiple linear interpolation is applied independently. For example, in the "reaction temperature-pressure" subspace, new temperature and pressure combination data are generated by interpolation. These combinations maintain physical rationality. In order to cope with the imbalance of different categories in chemical production data (such as too few samples under certain reaction conditions), the interpolation strategy dynamically adjusts the proportion of sparse category data. For sparse data points under high temperature and high pressure conditions, the algorithm will give priority to interpolation to generate more new samples to solve the problem of uneven sample distribution in chemical production data. The interpolation process is expressed as: ; In the formula, and It is a subspace Two chemical production data points randomly selected from ; For the random chemical production data points, For the Random chemical production data points; is the interpolation coefficient; are new chemical production data points obtained through multiple linear interpolation.

[0031] S23. According to the category distribution of the original chemical production data set, the number of samples generated by interpolation of each subspace is dynamically adjusted. The present invention automatically adjusts the interpolation strategy according to the category distribution, and preferentially generates chemical production data of a minority category. In the chemical production process, some special operating conditions (such as unconventional temperature or pressure ranges) have fewer data samples corresponding to them, which belong to sparse areas. By monitoring the data distribution, the sparsity is calculated and the number of samples generated by interpolation is dynamically adjusted. For example, for sparse samples in low temperature and low pressure areas, the algorithm increases the step size and direction of interpolation to generate more samples in sparse areas, so that the chemical model has a stronger learning ability for special operating conditions, which is expressed as: ; In the formula, Indicates the category The sample sparsity of for a specific category; It is a function that adjusts the data balance according to the sparsity level.

[0032] In one embodiment, the calculation method of the function for adjusting data balance according to the sparsity level is expressed as: ; In the formula, is the parameter that adjusts the curvature, is a parameter that determines the threshold. Preferably, Set to 2, Set to 0.1.

[0033] S24. The new samples generated by interpolation in all subspaces are recombined according to the original feature structure to form a new chemical production data set. The feature combination of each subspace represents the important information in different stages or processes of chemical production. When these subspace data are recombined, the feature structure of the original chemical data is maintained. For example, for the separated "reaction condition subspace" and "feed parameter subspace", the samples generated after recombining can reflect the effect of reaction conditions on reaction rate and retain the effect of feed parameters on product selectivity, so that the expanded data has a higher consistency with the actual chemical process, which is expressed as: ; In the formula, For the new chemical production data set after reorganization, Represents the reorganization operation of multiple subspace data, is the number of subspaces.

[0034] S25. Perform quality control on the expanded chemical production data set, detect and remove possible outliers or chemical production data samples that do not conform to physical meaning, so as to remove outliers. Chemical production data is subject to process constraints, and some interpolated data may not conform to actual production rules, such as unreasonable temperature and pressure combinations. Abnormal samples are removed based on the mean and standard deviation of each feature. If the temperature of the generated sample exceeds the allowable range of the equipment, the sample will be removed to ensure the physical rationality and engineering feasibility of the expanded data and enhance the credibility of the model training data, which is expressed as: ; In the formula, The final expanded chemical production data set, Represents the quality control function.

[0035] In one embodiment, the quality control function is implemented as: ; In the formula, The data in the reorganized new chemical production data set are: and Represents the chemical production data set The mean and standard deviation of each feature in .

[0036] When the amount of data in the final expanded chemical production data set obtained by model expansion reaches a preset upper limit, data expansion is stopped, indicating that the sample expansion task is completed.

[0037] The construction method of the hazard source investigation and early warning model is as follows: S31, obtaining expanded chemical production data; S32, extracting features from chemical production data using a pre-built feature extraction model; S33, inputting the chemical production data after feature extraction into the classifier to train the classifier model.

[0038] Specifically, Figure 2 As shown, the feature extraction model training method is as follows: The expanded chemical production data is input into the feature extraction model to train the feature extraction model. The present invention uses a 6-layer fully connected neural network for feature extraction. The structure of the 6-layer fully connected neural network is as follows: a-Input layer: receives and transmits sample chemical production data after sample expansion.

[0039] b- Hidden layers: The network contains four hidden layers, each consisting of a certain number of neurons.

[0040] b1-first hidden layer: has 128 neurons, used for preliminary feature extraction; b2-second hidden layer: has 256 neurons to further abstract features; b3-the third hidden layer: has 512 neurons, further deepening feature extraction; b4-the fourth hidden layer: has 1024 neurons, further deepening feature extraction.

[0041] b5 - output layer: has 512 neurons.

[0042] In the prior art, when some schemes use neural networks for feature extraction, in certain neural network structures, they may encounter problems such as gradient vanishing, gradient explosion, or falling into local optimal solutions, which affects the stability of training and the performance of the model. The present invention adopts a neural network based on exploratory gradient descent as a feature extraction model. In the traditional neural network training process, the exploratory gradient descent method is used to not only consider the current gradient information, but also adopt a forward exploration mechanism to predict the possible change trend of the gradient before the gradient is updated, and adjust the learning step size and direction to achieve faster convergence speed and better local minimum, thereby enhancing the generalization ability of the feature, specifically: S321, initializing the parameters of the neural network, including the weight and bias parameters of the neural network. In one embodiment, the initialization method is expressed as: ; ; In the formula, is the weight matrix of the i-th layer of the neural network; is the bias vector of the i-th layer of the neural network; is the scaling factor of the neural network weights, is the scaling factor of the bias of the neural network; represents a normal distribution with a mean of 0 and a standard deviation of 1; is a tiny noise. Preferably, is a noise vector whose elements are all 0.001.

[0043] S322. The input chemical production data passes through each layer of the neural network. Each layer is processed by linear transformation and nonlinear activation function, and finally the feature representation of the output layer is obtained. The output calculation method of each layer is: ; In the formula, Represents the neural network The activation output of the layer; The neural network The output of the layer; is the adaptive ReLU activation function.

[0044] In one embodiment, let the input of the adaptive ReLU activation function be , the calculation method of the adaptive ReLU activation function is expressed as: ; In the formula, and are the adaptive coefficients of the positive and negative parts respectively, To obtain the maximum value function, is the minimum value function. Preferably, Set to 0.3, Set to 0.7.

[0045] S323, using an exploratory gradient descent update strategy, predicting possible gradient changes before each update of the neural network parameters, adjusting the step size and direction of the gradient update according to the prediction results, optimizing the parameter adjustment during the training process, and avoiding falling into the local optimum too early. The calculation method is expressed as: ; ; In the formula, and They are the neural network Layer weights and The update amount of the layer bias; is the learning rate of the neural network; is the loss function with respect to the parameter The gradient of is the loss function with respect to the parameter The gradient of is the loss function of the neural network; is the exploration intensity coefficient; is the prediction function of the gradient. Preferably, Set to 0.03, Cross entropy loss is used.

[0046] In one embodiment, the exploratory gradient prediction function uses the historical gradient loss function to perform linear regression prediction, and the calculation method is expressed as: ; In the formula, Indicated in The loss function of the iteration is about the parameter The gradient of is the attenuation factor; To explore the strength coefficient; is the total number of iterations. Preferably, Set to 0.98, Set to 2.

[0047] Furthermore, the weight and bias parameters of the neural network are updated by using the error back propagation method.

[0048] S324, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0049] The chemical production data after feature extraction is input into the classifier to train the classifier model. The present invention adopts an extreme learning machine classification algorithm based on a neuron competitive learning mechanism as a classifier, and adopts a fuzzy logic quantum coding strategy, so that the extreme learning machine algorithm can better handle the uncertainty and fuzziness in the safety management of chemical hazardous chemicals. The chemical production data after feature extraction improves the classification accuracy and robustness of the model.

[0050] Specifically, the training process of the extreme learning machine classification algorithm based on the neuron competition learning mechanism is as follows: S331, initializing the weights and biases of the extreme learning machine. In one embodiment, the initialization method is expressed as: ; ; In the formula, is the weight of the extreme learning machine, is the bias of the extreme learning machine; Indicates the generation of a A matrix whose elements are random values ​​from a normal distribution; Generate a dimensional vector whose elements are uniformly distributed random values; is the first quantum coding coefficient, is the second quantum coding coefficient; is the number of input features of the extreme learning machine, is the number of hidden layer neurons of the extreme learning machine.

[0051] In one embodiment, the first quantum coding coefficient and the second quantum coding coefficient are set based on quantum behavior, and the calculation method is expressed as: ; ; In the formula, and are the set standard deviation and mean, is the chemical production data feature after feature extraction input into the extreme learning machine. Preferably, Set to 0.1, Set to 0.2.

[0052] S332. The chemical production data after feature extraction input into the extreme learning machine passes through the hidden layer. The activation of each neuron in the hidden layer depends not only on its weight and bias, but also on the activation state of the neighboring neurons, simulating the competitive inhibition effect in the biological nervous system. The forward propagation of the data is expressed as: ; In the formula, is the Sigmoid activation function; is the output of the hidden layer; The chemical production data matrix after input feature extraction; is the local inhibition function; is the dot product operation.

[0053] In one embodiment, the calculation method of the local suppression function is expressed as: ; In the formula, represents the average activation state of adjacent neurons, is the threshold value, To obtain the maximum value function, represents the L2 norm.

[0054] S333, the feature conversion from the hidden layer to the output layer adopts a linear combination method. In the process of determining the output weight, the present invention adopts a method based on minimizing the training error and the regularization term to quickly determine the optimal weight. The calculation method is expressed as: ; In the formula, is the weight from the hidden layer to the output layer; is the target output; is the transpose of the output of the hidden layer; is the output of the hidden layer; is the regularization parameter; is the identity matrix.

[0055] In one embodiment, the calculation method of the regularization parameter is expressed as: ; here, represents the variance of the hidden layer output, is the number of neurons in the hidden layer.

[0056] S334. In the classification decision stage, fuzzy logic rules are used to deal with the uncertainty of classification. The final classification output calculation method is expressed as: ; is a fuzzy logic function, is the final classification output.

[0057] In one embodiment, the calculation method of the fuzzy logic function is expressed as: ; In the formula, Indicates The membership function of a fuzzy rule is It is The output value of the fuzzy rule is is the total number of fuzzy rules.

[0058] S335. Through the iterative training process, the weights and biases of the hidden layer and the fuzzy logic rules of the output layer are fine-tuned to ensure that the model achieves the best classification performance on the entire chemical production data set. The adjustment method is expressed as: ; ; In the formula, is the cross entropy loss function, is the learning rate, and are the gradients of the loss function with respect to weights and biases, respectively. and To fine-tune the updated weights and biases. Preferably, Set to 0.01.

[0059] S336, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 500 times.

[0060] Further, based on the above method embodiment, the embodiment of the present invention also provides a hazard source investigation device for safety management of chemical hazardous chemicals, including a data acquisition module for collecting sensor data, safety inspection records and accident report data from on-site chemical enterprises; A data expansion module, used to expand the collected data; An execution module is used to input the expanded chemical production data into a pre-built hazard source investigation and early warning model, and to classify and predict the hazard level of the chemical production data through the hazard source investigation and early warning model to determine the hazard classification result; The output module is used to predict and analyze the dangerous consequences of chemical equipment based on the classification results.

[0061] The trained model is used to process samples to achieve real-time screening and identification of hazardous sources in the safety management of hazardous chemicals. Specifically, the collected raw data is input into the trained feature extraction model for feature processing. Further, the processed features are input into the classifier model for classifier training to obtain the classification results. The classification categories include "safety", "caution" and "danger", a total of 3 categories. It can realize the rapid and accurate identification of potential hazardous sources in the chemical plant environment, improve the efficiency and accuracy of emergency response, and the new real-time data processing capability ensures the continuity and safety of chemical production, reduces the risk of human judgment errors, and enhances the timeliness of prevention and control measures.

[0062] Finally, it should be noted that the above embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for early warning of hazardous sources for safety management of hazardous chemicals, characterized in that: Collect data from on-site sensors, safety inspection records and accident reports from chemical companies; The multilinear interpolation algorithm based on feature decoupling is used to generate samples and realize chemical production data expansion; Input the expanded chemical production data into a pre-built hazard source investigation and early warning model, and use the hazard source investigation and early warning model to classify and predict the hazard level of the chemical production data to determine the hazard classification result; Among them, the hazard source investigation and early warning model is constructed based on the extreme learning machine classification algorithm based on the neuron competitive learning mechanism, and the fuzzy logic quantum coding strategy is adopted to enable the extreme learning machine classification algorithm to better handle the chemical production data after the uncertainty and fuzzy features are extracted in the safety management of chemical hazardous chemicals.

2. A method for hazard source screening and early warning for safety management of hazardous chemicals according to claim 1, characterized in that: The multilinear interpolation algorithm based on feature decoupling is specifically as follows: The features in the collected chemical production data set are decoupled according to correlation and contribution, and divided into multiple independent subspaces, each of which contains a set of relevant features that affect a specific output variable; In each decoupled subspace, multilinear interpolation is applied independently; According to the category distribution of the original chemical production data set, the number of samples generated by interpolation in each subspace is dynamically adjusted; The new samples generated by interpolation in all subspaces are recombined according to the original feature structure to form a new chemical production data set. The feature combination of each subspace represents the important information in different stages or processes of chemical production. Quality control is performed on the expanded chemical production data set to detect and eliminate possible outliers or chemical production data samples that do not conform to physical meaning.

3. A method for hazard source investigation and early warning for safety management of hazardous chemicals according to claim 1, characterized in that: The construction method of the hazard source investigation and early warning model is: Obtain expanded chemical production data; Extracting features from the chemical production data using a pre-built feature extraction model; The chemical production data after feature extraction is input into the classifier to train the classifier model.

4. A method for hazard source screening and early warning for safety management of hazardous chemicals according to claim 3, characterized in that: The feature extraction model is a neural network based on exploratory gradient descent, specifically: Initialize the parameters of the neural network, including the weights and bias parameters of the neural network; The input chemical production data passes through each layer of the neural network, and each layer is processed by linear transformation and nonlinear activation function, and finally the feature representation of the output layer is obtained; An exploratory gradient descent update strategy is adopted to predict possible gradient changes before each update of the neural network parameters. The step size and direction of the gradient update are adjusted according to the prediction results to optimize the parameter adjustment during the training process.

5. The method for hazard source screening and early warning for safety management of hazardous chemicals according to claim 1 is characterized in that: The extreme learning machine classification algorithm based on the neuron competition learning mechanism is as follows: Initialize the weights and biases of the extreme learning machine; The feature-extracted chemical production data input into the extreme learning machine passes through the hidden layer to simulate the competitive inhibition effect in the biological nervous system; The feature conversion from the hidden layer to the output layer adopts a linear combination method, and the optimal weight is determined by minimizing the training error and regularization term; In the classification decision stage, fuzzy logic rules are used to deal with the uncertainty of classification; During the iterative training process, the weights and biases of the hidden layers, as well as the fuzzy logic rules of the output layer, are fine-tuned.

6. A method for hazard source screening and early warning for safety management of hazardous chemicals according to claim 1, characterized in that: The hazard classification results include "safety", "caution" and "danger".

7. A method for hazard source screening and early warning for safety management of hazardous chemicals according to claim 1, characterized in that: The calculation method for determining the optimal weights based on minimizing the training error and the regularization term is: ； In the formula, is the weight from the hidden layer to the output layer; is the target output; is the transpose of the output of the hidden layer; is the output of the hidden layer; is the regularization parameter; is the identity matrix.

8. A hazard source investigation device for the safety management of hazardous chemicals, characterized in that: include: Data collection module, used to collect sensor data, safety inspection records and accident report data from chemical enterprises; A data expansion module, used to expand the collected data; An execution module is used to input the expanded chemical production data into a pre-built hazard source investigation and early warning model, and to classify and predict the hazard level of the chemical production data through the hazard source investigation and early warning model to determine the hazard classification result; The output module is used to predict and analyze the dangerous results of the chemical equipment according to the classification results.

Images