Industrial product risk early warning method and related device

Abstract

The embodiment of the invention provides an industrial product risk early warning method and a related device, and the method comprises the steps: carrying out the feature extraction of industrial product data, obtaining the feature data of the industrial product data, inputting the feature data of the industrial product data into a grading early warning model, and obtaining the risk grade of the industrial product data; and performing risk early warning on the industrial product data according to the risk level of the industrial product data. In the embodiment of the invention, the hierarchical early warning model is obtained by taking the feature data set and the label data set as training data and training based on the least square support vector machine (LSSVM) algorithm, the LSSVM algorithm can process classification and regression tasks in a high-dimensional input space, risk level division is carried out on the industrial product data based on the hierarchical early warning model, and the risk level of the industrial product data is obtained. The risk level division accuracy can be improved, risk early warning is performed on the industrial product based on the accurate risk level, and the risk early warning accuracy can be effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of risk warning, and in particular to a risk warning method and related devices for industrial products. Background Art

[0002] With the development of global industrialization, the safety of industrial products has attracted much attention. Accidents caused by industrial products may cause damage to people's safety and property.

[0003] In order to avoid industrial product accidents and protect people's life and property safety, there is an urgent need for a solution that can provide early warning of industrial product risks before industrial product accidents occur. Summary of the invention

[0004] Based on the above problems, the present application provides a risk warning method and related devices for industrial products.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In the first aspect, the present application provides a risk warning method for industrial products, including:

[0007] Extracting features from industrial product data to obtain feature data of the industrial product data;

[0008] Inputting the characteristic data of the industrial product data into the hierarchical early warning model to obtain the risk level of the industrial product data; the hierarchical early warning model is obtained by training based on the least squares support vector machine LSSVM algorithm using the characteristic data set and the label data set as training data, the characteristic data set includes the characteristic data of the industrial product data in the training data set, and the label data set includes the risk level of the characteristic data of the industrial product data in the training data set;

[0009] According to the risk level of the industrial product data, a risk warning is issued for the industrial product data.

[0010] In a possible implementation, the step of training the hierarchical early warning model includes:

[0011] Extract features of industrial product data in the training data set to obtain a feature data set;

[0012] Based on the industrial product quality standard, marking the risk level of the feature data in the feature data set, and obtaining a label data set corresponding to the feature data set;

[0013] The feature data set and the label data set are used as training data, and training is performed based on the LSSVM algorithm to obtain the hierarchical warning model.

[0014] In a possible implementation, the step of marking the risk level of the feature data in the feature data set based on the industrial product quality standard to obtain a label data set corresponding to the feature data set includes:

[0015] Based on the LSSVM algorithm, a separation hyperplane of the feature data set is generated;

[0016] Based on the segmentation hyperplane, the risk level of the feature data in the feature data set is determined, the risk level of the feature data in the feature data set is marked, and a label data set corresponding to the feature data set is obtained.

[0017] In a possible implementation, generating a separation hyperplane of the feature data set based on the LSSVM algorithm includes:

[0018] Based on the ISSA algorithm, the penalty factor and kernel function parameters are optimized to obtain the optimized penalty factor and kernel function parameters;

[0019] Based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, a separation hyperplane of the feature data set is generated.

[0020] In a possible implementation, the step of performing parameter optimization on the penalty factor and the kernel function parameter based on the ISSA algorithm to obtain the optimized penalty factor and the optimized kernel function parameter includes:

[0021] Using the ISSA algorithm, based on the position update function and the follower's local search formula, the penalty factor and kernel function parameters are optimized to obtain the optimized penalty factor and kernel function parameters.

[0022] The position update function represents the relationship between the current global optimal position information, the product of the first product and the second product, and the updated position information; the first product is the product of the flag information for changing the optimal direction, the foraging ability information, the food finding ability information, and the current global optimal position; the second product is the product of the flag information for changing the optimal direction, the first random number, the step scaling factor, the distance information to the prey, the absolute value of the cosine value of the product between the second random number and 2π, and the absolute value of the difference between the cosine value of the product between the third random number and 2π minus 1;

[0023] The first sub-formula in the local search formula of the follower represents the relationship between the current disturbance and the fitness value of the current position, the fitness value of the best position and the fitness value of the worst position, and the second sub-formula in the local search formula of the follower represents the relationship between the flight step size and the step scaling factor, the fourth random number and the fifth random number of the normal distribution.

[0024] In a possible implementation, generating a separation hyperplane of the feature data set based on the optimized penalty factor, the optimized kernel function parameter, and the LSSVM algorithm includes:

[0025] Based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, a number of separation hyperplanes corresponding to the number of risk levels is generated.

[0026] In a possible implementation, extracting features from industrial product data to obtain feature data of the industrial product data includes:

[0027] Extract features from industrial product data to obtain risk feature data of the industrial product data;

[0028] Based on the BERT network, the risk feature data of the industrial product data is converted into multi-dimensional feature data, and the number of the dimensions corresponds to the number of risk levels.

[0029] In a possible implementation, before extracting features from the industrial product data to obtain feature data of the industrial product data, the method further includes:

[0030] Obtain unstructured industrial product data;

[0031] The unstructured industrial product data is preprocessed to obtain the preprocessed industrial product data.

[0032] Second aspect: The embodiment of the present application provides a risk warning device for industrial products, including:

[0033] Extraction unit, classification unit and early warning unit;

[0034] The extraction unit is used to extract features from the industrial product data to obtain feature data of the industrial product data;

[0035] The grading unit is used to input the characteristic data of the industrial product data into a grading warning model to obtain the risk level of the industrial product data; the grading warning model is obtained by training based on a least squares support vector machine (LSSVM) algorithm using a characteristic data set and a label data set as training data, wherein the characteristic data set includes the characteristic data of the industrial product data in the training data set, and the label data set includes the risk level of the characteristic data of the industrial product data in the training data set;

[0036] The early warning unit is used to issue a risk early warning to the industrial product data according to the risk level of the industrial product data.

[0037] A third aspect: An embodiment of the present application provides a computing device, the computing device comprising: a processor and a memory;

[0038] The memory is used to store program code and transmit the program code to the processor;

[0039] The processor is used to execute the steps of the risk warning method for an industrial product as described above according to the instructions in the program code.

[0040] Compared with the prior art, this application has the following beneficial effects:

[0041] The embodiment of the present application provides a risk warning method and related device for industrial products, which extracts features from industrial product data to obtain feature data of the industrial product data, inputs the feature data of the industrial product data into a hierarchical warning model, and obtains the risk level of the industrial product data; and performs risk warning on the industrial product data according to the risk level of the industrial product data. In the embodiment of the present application, the hierarchical warning model is obtained by training based on the least squares support vector machine LSSVM algorithm using feature data sets and label data sets as training data. The LSSVM algorithm can handle classification and regression tasks in high-dimensional input space. The risk level of industrial product data is divided based on the hierarchical warning model, which can improve the accuracy of risk level division. Risk warning of industrial products based on accurate risk levels can effectively improve the accuracy of risk warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0043] Figure 1 A flow chart of a risk warning method for industrial products provided in an embodiment of the present application;

[0044] Figure 2 A feature extraction structure diagram based on a BERT network provided in an embodiment of the present application;

[0045] Figure 3 A least squares support vector machine model diagram provided in an embodiment of the present application;

[0046] Figure 4 An overall architecture diagram of a risk warning method provided in an embodiment of the present application;

[0047] Figure 5A schematic structural diagram of a risk warning device for an industrial product is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0048] As described above, with the development of global industrialization, the safety of industrial products has attracted much attention. Accidents caused by industrial products may cause damage to people's safety and property. In order to avoid industrial product accidents and protect people's life and property safety, there is an urgent need for a solution that can warn of the risks of industrial products before industrial product accidents occur.

[0049] Based on this, an embodiment of the present application provides a risk warning method for industrial products, which obtains feature data of the industrial product data by performing feature extraction on the industrial product data, inputs the feature data of the industrial product data into a graded warning model, and obtains the risk level of the industrial product data; and performs risk warning on the industrial product data according to the risk level of the industrial product data.

[0050] In an embodiment of the present application, the graded warning model is obtained by training based on the least squares support vector machine LSSVM algorithm using a feature data set and a label data set as training data. The LSSVM algorithm is capable of handling classification and regression tasks in a high-dimensional input space. The risk level of industrial product data is divided based on the graded warning model, which can improve the accuracy of the risk level division. Risk warnings for industrial products based on accurate risk levels can effectively improve the accuracy of risk warnings.

[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] See also Figure 1 , which is a flow chart of a risk warning method for industrial products provided in an embodiment of the present application, including S101-S103.

[0053] S101. Extract features from industrial product data to obtain feature data of the industrial product data.

[0054] In a possible implementation, before extracting features from the industrial product data to obtain feature data of the industrial product data, unstructured industrial product data may be acquired, and the unstructured industrial product data may be preprocessed to obtain the preprocessed industrial product data.

[0055] Exemplarily, the industrial product data acquired in the embodiments of the present application is unstructured data. By preprocessing the unstructured industrial product data, preprocessed industrial product data can be obtained to facilitate the subsequent risk grading process.

[0056] Among them, the preprocessing methods include but are not limited to removing abnormal data, removing duplicate data, etc.

[0057] In a possible implementation, feature extraction is performed on industrial product data to obtain risk feature data of the industrial product data; based on a bidirectional encoder representation method (Bidirectional Encoder Representations from Transformers, BERT) network, the risk feature data of the industrial product data is converted into multi-dimensional feature data.

[0058] The number of dimensions corresponds to the number of risk levels.

[0059] Bidirectional Encoder Representations from Transformers (BERT) network has the characteristics of bidirectional context understanding, pre-training and fine-tuning, Transformer architecture, multi-task learning, open source and scalability, which make it have good performance in the field of natural language processing (NLP). Among them, the Transformer architecture is a sequence-to-sequence (seq2seq) based model, which has a wide range of applications in the field of NLP, including but not limited to machine translation, text generation, text classification, sentiment analysis, question-answering systems, etc. Figure 2 As shown, this figure is a feature extraction structure diagram based on the BERT network provided in an embodiment of the present application.

[0060] Based on this, in an embodiment of the present application, feature extraction can be performed on the unstructured industrial product data based on the BERT network, and the risk feature data of the obtained industrial product data can be converted into multi-dimensional feature data.

[0061] For example, in the embodiment of the present application, risk influencing factors of risk events are obtained based on the risk events that have occurred, and the quality and safety characteristics of industrial products are comprehensively considered. The four dimensions of the possibility of risk events, the severity of similar events, the influence of public opinion, and social impact factors are used as the four-dimensional standard for industrial product risk warning.

[0062] Based on the four dimensions of the possibility of risk events, the severity of similar events, the influence of public opinion and social impact factors, the BERT network is used to extract features of the unstructured industrial product data, and the risk feature data of the obtained industrial product data is converted into four-dimensional feature data.

[0063] S102: Input characteristic data of industrial product data into a graded early warning model to obtain the risk level of the industrial product data.

[0064] The hierarchical warning model is obtained by training the feature data set and the label data set based on the least squares support vector machine (LSSVM) algorithm. Figure 3 As shown, this figure is a least squares support vector machine model diagram provided in an embodiment of the present application.

[0065] The feature data set includes the feature data of the industrial product data in the training data set, and the label data set includes the risk level of the feature data of the industrial product data in the training data set.

[0066] In a possible implementation, the training steps of the hierarchical early warning model are as follows:

[0067] Step 21: extract features of the industrial product data in the training data set to obtain a feature data set.

[0068] Illustratively, in the embodiments of the present application, industrial product data related to industrial product quality and safety may be collected, and these industrial product data related to industrial product quality and safety may be preprocessed to obtain a training data set.

[0069] The sources of industrial product data related to industrial product quality and safety may include, but are not limited to, industrial product quality and safety spot check reports, industrial product quality and safety issue notifications, public opinion information on specific events, etc. Since these industrial product data are all unstructured data, in the embodiments of the present application, the collected unstructured industrial product data can be preprocessed and the preprocessed industrial product data can be used as a training data set.

[0070] In the embodiment of the present application, by collecting industrial product data related to industrial product quality and safety from multiple sources, a diverse and large amount of training data can be obtained. A graded early warning model is trained based on the training data, which can improve the accuracy of the output results of the graded early warning model.

[0071] In one example, similar to the process of extracting features from industrial product data in S101 above, the BERT network can be used to extract features from the industrial product data in the training data set to obtain a feature data set containing multiple four-dimensional feature data, which will not be repeated here.

[0072] Step 22: Based on the industrial product quality standard, the risk level of the feature data in the feature data set is marked to obtain a label data set corresponding to the feature data set.

[0073] Exemplarily, in the embodiment of the present application, the risk level of the feature data can be divided into four risk levels: low risk, medium risk, high risk and extremely high risk based on the data range of the four-dimensional feature data.

[0074] In a possible implementation, the risk level of the feature data in the feature data set can be annotated manually based on the industrial product quality standard to obtain a label data set corresponding to the feature data set. Exemplarily, the risk level of each feature data can be annotated in the last column of the corresponding feature data to generate label data corresponding to the feature data. The label data set corresponding to the feature data set includes label data corresponding to each feature data in the feature data set.

[0075] Step 23: Using the feature data set and the label data set as training data, training is performed based on the LSSVM algorithm to obtain the hierarchical warning model.

[0076] In one possible implementation, in order to improve the classification efficiency, a separation hyperplane of the feature data set can be generated based on the LSSVM algorithm; based on the separation hyperplane, the risk level of the feature data in the feature data set is determined, the risk level of the feature data in the feature data set is marked, and a label data set corresponding to the feature data set is obtained.

[0077] The separating hyperplane is defined by the separating hyperplane weight (W) and the bias (b), and the separating hyperplane of the feature data set can be used to distinguish feature data of different risk levels. For example, for the separating hyperplane corresponding to the feature data of low risk, the distance between the feature data of low risk and the separating hyperplane is relatively close, while the distance between the feature data of other risk levels and the separating hyperplane is relatively far.

[0078] As shown in formula (1), it is an expression of the separating hyperplane.

[0079]

[0080] In the embodiment of the present application, the training data can be divided into a training set and a test set. For example, the data in the training set can account for 75% of the total data, and the data in the test set can account for 25% of the total data.

[0081] Based on formula (1), the weight and bias of the separating hyperplane are calculated using the data in the training set. Based on the chaotic matrix shown in formula (2), the accuracy of the risk classification determined based on the separating hyperplane is calculated using the test set data and the separating hyperplane.

[0082]

[0083] In an embodiment of the present application, the feature data in the test set can be divided into risk levels to obtain the risk level of the feature data in the test set, which can be compared with the label data in the test set. Based on the comparison results, the accuracy of the risk grading determined based on the separating hyperplane can be determined.

[0084] In one example, by calculating the distance between the feature data and the separating hyperplane, it can be determined whether the current separating hyperplane is the best separating hyperplane, so that risk classification can be performed based on the best separating hyperplane.

[0085] The method of calculating the distance can be expressed by formula (3) as follows:

[0086]

[0087] Among them, R i Represents geometric distance; r i represents the function interval; y i ∈{1, -1}; x i ∈R n , represents the n-dimensional training sample input vector; W represents the segmentation hyperplane weight; b represents the bias.

[0088] In the embodiment of the present application, the optimal separation hyperplane can be determined by determining the geometric distance of the maximum training sample. The geometric distance of the maximum training sample can be expressed as follows by formula (4):

[0089]

[0090] Since the value of the function interval does not affect the problematic solution, the function interval can be set to 1, and the geometric distance of the maximum training sample is obtained as shown in formula (5).

[0091]

[0092] In the embodiment of the present application, considering the existence of singular points and nonlinear training data, the original training data is mapped to a higher dimensional space, and the Lagrange function is expressed as follows:

[0093]

[0094] Where L(W,B,e,α) represents the Lagrange function; λ represents the regularization parameter; e i represents the error variable; represents the mapping function; α i Represents the Lagrange multiplier.

[0095] By solving equation (6), taking the derivative of each variable and setting the derivative to 0, we can get equation (7). Based on equation (7), we can solve α = [α1, α2, ..., α m ] T and b.

[0096]

[0097] in, is the unit matrix; K is the m×m kernel matrix, and is the kernel parameter. Based on formula (7), the output of the LSSVM network, that is, the separation hyperplane, can be obtained as:

[0098]

[0099] Among them, k(x i , x j ) indicates that the kernel function is a radial basis kernel function. In order to improve the accuracy of risk grading, in the embodiment of the present application, the penalty factor and the kernel function parameters can be optimized based on the Improved Sparrow Search Algorithm (ISSA) algorithm to obtain an optimized penalty factor and an optimized kernel function parameter; based on the optimized penalty factor, the optimized kernel function parameter and the LSSVM algorithm, a separation hyperplane of the feature data set is generated.

[0100] The ISSA algorithm in the embodiment of the present application uses the biological characteristics of the sparrow search algorithm (SSA) algorithm, in which the searchers in the sparrow population conduct a global search for the best foraging location and the followers follow the searchers to forage, and adds a dynamic adjustment strategy for the sparrow population. In the dynamic search process, the number of searchers and followers can be dynamically changed, and the number of searchers can be increased in the later stage of the search to avoid falling into the local optimum. The dynamic change process can be expressed by formula (9) as follows:

[0101] PD1=PD+R1(t / T) (9)

[0102] Among them, PD1 represents the proportion of searchers that are continuously updated; PD represents the proportion of original searchers; t represents the current iteration round; T represents the maximum iteration round; R1 represents the upper limit of the searcher increase ratio.

[0103] From formula (9), we can see that the number of searchers increases with the increase of iteration rounds. Realizing dynamic search can effectively avoid falling into the problem of local optimality.

[0104] In the embodiment of the present application, the position update function of each round of iteration is optimized based on the ISSA algorithm. The position update function can be expressed as follows by formula (10):

[0105] x new =x prey +F×β×I×x prey +F×r1×α×d1×|cos(2πr2)×[1-cos(2πr3)]| (10)

[0106] Among them, F×β×I×x prey is the first product; F×r1×α×d1×|cos(2πr2)×[1-cos(2πr2)]| is the second product; x new Indicates the updated location information; x prey represents the current global optimal position information; F represents the sign information of changing the optimization direction; β represents the foraging ability information; I represents the ability to find food; d i Represents the distance information to the prey; r1, r2, r3 are the first random number, the second random number, and the third random number, respectively, and their value range is [0, 1].

[0107] In the embodiment of the present application, the local search formula of the follower is optimized based on the ISSA algorithm, as shown in formula (11):

[0108]

[0109] Where w is the current disturbance; w max and w min are the maximum disturbance and the minimum disturbance respectively; f i 、f g 、f w They represent the fitness value of the current position, the fitness value of the best position, and the fitness value of the worst position respectively; Levy (β) represents the flight step length; u and v are the fourth and fifth random numbers of the normal distribution respectively; α represents the step scaling factor.

[0110] In the embodiment of the present application, the ISSA algorithm is used to optimize the penalty factor and the kernel function parameters based on the position update function shown in the above formula (10) and the local search formula of the follower shown in formula (11) to obtain the optimized penalty factor and the optimized kernel function parameters.

[0111] In a possible implementation, the separation hyperplane of the feature data set is generated based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, including: based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, generating a number of separation hyperplanes corresponding to the number of risk levels.

[0112] Exemplarily, when the risk levels include low risk, medium risk, high risk, and extremely high risk, four separation hyperplanes can be obtained by combining the ISSA algorithm with the LSSVM algorithm. The four separation hyperplanes include a separation hyperplane corresponding to low risk, a separation hyperplane corresponding to medium risk, a separation hyperplane corresponding to high risk, and a separation hyperplane corresponding to extremely high risk.

[0113] Among them, the separation hyperplane corresponding to low risk can be used to separate particles of characteristic data representing low risk from particles of characteristic data representing other risk levels; the separation hyperplane corresponding to medium risk can be used to separate particles of characteristic data representing medium risk from particles of characteristic data representing other risk levels; the separation hyperplane corresponding to high risk can be used to separate particles of characteristic data representing high risk from particles of characteristic data representing other risk levels; the separation hyperplane corresponding to extremely high risk can be used to separate particles of characteristic data representing extremely high risk from particles of characteristic data representing other risk levels.

[0114] S103. Perform risk warning on the industrial product data according to the risk level of the industrial product data.

[0115] In an embodiment of the present application, after obtaining the risk level of industrial product data, different early warning measures can be taken for industrial product data of different risk levels to prevent industrial product quality and safety incidents before they occur and minimize the harm caused by them.

[0116] To sum up, in the embodiments of the present application, BERT, LSSVM and ISSA algorithms can be combined to divide industrial product data into risk levels, which can improve the accuracy of risk level division, and risk warnings for industrial products based on accurate risk levels can effectively improve the accuracy of risk warnings.

[0117] Combine the following Figure 4 An overall introduction to the method provided in the embodiments of the present application is given.

[0118] In the embodiment of the present application, the collected unstructured industrial product data is input into the BIL industrial product classification warning model, and the risk warning result can be output.

[0119] The BIL industrial product classification warning model is constructed by combining the BERT network, ISSA, and LSSVM algorithms. The BIL industrial product classification warning model includes a BERT network and a classification warning model, wherein the BERT network is used to extract features from unstructured industrial product data to obtain feature data.

[0120] The hierarchical early warning model is based on a large amount of unstructured industrial product data as the training data set and is trained based on the ISSA-LSSVM algorithm.

[0121] In the embodiment of the present application, the process of obtaining the graded warning model based on ISSA-LSSVM algorithm training is as follows:

[0122] Based on the BERT network, feature extraction is performed on the industrial product data in the training data set to obtain a feature data set. Each feature data in the feature data set can be represented by data in four dimensions: S1, S2, S3, and S4.

[0123] Based on the industrial product quality standard, the risk level of the feature data in the feature data set is marked to obtain a label data set corresponding to the feature data set. Figure 4 In , Y1 can represent the label data corresponding to the feature data.

[0124] The feature data set and the label data set are input into the ISSA-LSSVM model training module. The ISSA parameter optimization module in the ISSA-LSSVM model training module can perform population initialization based on the feature data set and the label data set, calculate the fitness value based on the LSSVM algorithm, update the optimal fitness value and the population position, and output the optimized penalty factor and the optimized kernel function parameter. Based on the optimized penalty factor, the optimized kernel function parameter and the LSSVM algorithm, a separation hyperplane corresponding to the number of risk levels is generated. The process of generating a separation hyperplane corresponding to the number of risk levels is the same as the above S102, which will not be repeated here.

[0125] A graded warning model can be constructed based on the separation hyperplanes corresponding to the number of risk levels. The graded warning model can execute the above-mentioned methods S102-S103, which will not be repeated here.

[0126] This application provides a risk warning device for industrial products, see Figure 5 , this figure is a structural schematic diagram of a risk warning device for an industrial product provided in an embodiment of the present application. Its specific implementation method is consistent with the implementation method and the technical effect achieved in the embodiment of the above method, and some contents will not be repeated here.

[0127] A risk warning device 1100 for industrial products, comprising:

[0128] Extraction unit 1101, classification unit 1102 and early warning unit 1103;

[0129] The extraction unit 1101 is used to extract features from the industrial product data to obtain feature data of the industrial product data;

[0130] The grading unit 1102 is used to input the characteristic data of the industrial product data into a grading warning model to obtain the risk level of the industrial product data; the grading warning model is obtained by training based on the least squares support vector machine LSSVM algorithm using the characteristic data set and the label data set as training data, the characteristic data set includes the characteristic data of the industrial product data in the training data set, and the label data set includes the risk level of the characteristic data of the industrial product data in the training data set;

[0131] The early warning unit 1103 is used to issue a risk early warning to the industrial product data according to the risk level of the industrial product data.

[0132] In a possible implementation, the device includes a hierarchical early warning model training unit;

[0133] The hierarchical early warning model training unit is used to:

[0134] Extract features of industrial product data in the training data set to obtain a feature data set;

[0135] Based on the industrial product quality standard, marking the risk level of the feature data in the feature data set, and obtaining a label data set corresponding to the feature data set;

[0136] The feature data set and the label data set are used as training data, and training is performed based on the LSSVM algorithm to obtain the hierarchical warning model.

[0137] In a possible implementation, the hierarchical early warning model training unit is specifically used to:

[0138] Based on the LSSVM algorithm, a separation hyperplane of the feature data set is generated;

[0139] Based on the segmentation hyperplane, the risk level of the feature data in the feature data set is determined, the risk level of the feature data in the feature data set is marked, and a label data set corresponding to the feature data set is obtained.

[0140] In a possible implementation, the hierarchical early warning model training unit is specifically used to:

[0141] Based on the ISSA algorithm, the penalty factor and kernel function parameters are optimized to obtain the optimized penalty factor and kernel function parameters;

[0142] Based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, a separation hyperplane of the feature data set is generated.

[0143] In a possible implementation, the hierarchical early warning model training unit is specifically used to:

[0144] Using the ISSA algorithm, based on the position update function and the follower's local search formula, the penalty factor and kernel function parameters are optimized to obtain the optimized penalty factor and kernel function parameters.

[0145] The position update function represents the relationship between the current global optimal position information, the product of the first product and the second product, and the updated position information; the first product is the product of the flag information for changing the optimal direction, the foraging ability information, the food finding ability information, and the current global optimal position; the second product is the product of the flag information for changing the optimal direction, the first random number, the step scaling factor, the distance information to the prey, the absolute value of the cosine value of the product between the second random number and 2π, and the absolute value of the difference between the cosine value of the product between the third random number and 2π minus 1;

[0146] The first sub-formula in the local search formula of the follower represents the relationship between the current disturbance and the fitness value of the current position, the fitness value of the best position and the fitness value of the worst position, and the second sub-formula in the local search formula of the follower represents the relationship between the flight step size and the step scaling factor, the fourth random number and the fifth random number of the normal distribution.

[0147] In a possible implementation, the hierarchical early warning model training unit is specifically used to:

[0148] Based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, a number of separation hyperplanes corresponding to the number of risk levels is generated.

[0149] In a possible implementation, the extraction unit is specifically configured to:

[0150] Extract features from industrial product data to obtain risk feature data of the industrial product data;

[0151] Based on the BERT network, the risk feature data of the industrial product data is converted into multi-dimensional feature data, and the number of the dimensions corresponds to the number of risk levels.

[0152] In a possible implementation, the device further includes: a preprocessing unit;

[0153] The preprocessing unit is used to: acquire unstructured industrial product data; and preprocess the unstructured industrial product data to obtain the preprocessed industrial product data.

[0154] To sum up, in the embodiments of the present application, BERT, LSSVM and ISSA algorithms can be combined to divide industrial product data into risk levels, which can improve the accuracy of risk level division, and risk warnings for industrial products based on accurate risk levels can effectively improve the accuracy of risk warnings.

[0155] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A risk early warning method for industrial products, characterized in that: include: Extracting features from industrial product data to obtain feature data of the industrial product data; Inputting the characteristic data of the industrial product data into the hierarchical early warning model to obtain the risk level of the industrial product data; the hierarchical early warning model is obtained by training based on the least squares support vector machine LSSVM algorithm using the characteristic data set and the label data set as training data, the characteristic data set includes the characteristic data of the industrial product data in the training data set, and the label data set includes the risk level of the characteristic data of the industrial product data in the training data set; According to the risk level of the industrial product data, a risk warning is issued for the industrial product data.

2. The method according to claim 1, characterized in that The training steps of the hierarchical early warning model include: Extract features of industrial product data in the training data set to obtain a feature data set; Based on the industrial product quality standard, marking the risk level of the feature data in the feature data set, and obtaining a label data set corresponding to the feature data set; The feature data set and the label data set are used as training data, and training is performed based on the LSSVM algorithm to obtain the hierarchical warning model.

3. The method according to claim 2, characterized in that The step of marking the risk level of the feature data in the feature data set based on the industrial product quality standard to obtain a label data set corresponding to the feature data set includes: Based on the LSSVM algorithm, a separation hyperplane of the feature data set is generated; Based on the segmentation hyperplane, the risk level of the feature data in the feature data set is determined, the risk level of the feature data in the feature data set is marked, and a label data set corresponding to the feature data set is obtained.

4. The method according to claim 3, characterized in that: The step of generating a separation hyperplane of the feature data set based on the LSSVM algorithm includes: Based on the improved sparrow search ISSA algorithm, the penalty factor and kernel function parameters are optimized to obtain the optimized penalty factor and kernel function parameters; Based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, a separation hyperplane of the feature data set is generated.

5. The method according to claim 4, characterized in that The improved sparrow search ISSA algorithm is used to optimize the penalty factor and the kernel function parameters to obtain the optimized penalty factor and the optimized kernel function parameters, including: Using the ISSA algorithm, based on the position update function and the follower's local search formula, the penalty factor and kernel function parameters are optimized to obtain the optimized penalty factor and kernel function parameters. The position update function represents the relationship between the current global optimal position information, the product of the first product and the second product, and the updated position information; the first product is the product of the flag information for changing the optimal direction, the foraging ability information, the food finding ability information, and the current global optimal position; the second product is the product of the flag information for changing the optimal direction, the first random number, the step scaling factor, the distance information to the prey, the absolute value of the cosine value of the product between the second random number and 2π, and the absolute value of the difference between the cosine value of the product between the third random number and 2π minus 1; The first sub-formula in the local search formula of the follower represents the relationship between the current disturbance and the fitness value of the current position, the fitness value of the best position and the fitness value of the worst position, and the second sub-formula in the local search formula of the follower represents the relationship between the flight step size and the step scaling factor, the fourth random number and the fifth random number of the normal distribution.

6. The method according to claim 4, characterized in that The step of generating a separation hyperplane of the feature data set based on the optimized penalty factor, the optimized kernel function parameter and the LSSVM algorithm includes: Based on the optimized penalty factor, the optimized kernel function parameters and the LSSVM algorithm, a number of separation hyperplanes corresponding to the number of risk levels is generated.

7. The method according to claim 1, characterized in that The extracting features of the industrial product data to obtain feature data of the industrial product data includes: Extract features from industrial product data to obtain risk feature data of the industrial product data; Based on the BERT network, the risk feature data of the industrial product data is converted into multi-dimensional feature data, and the number of the dimensions corresponds to the number of risk levels.

8. The method according to any one of claims 1 to 7, characterized in that: Before extracting the features of the industrial product data to obtain the feature data of the industrial product data, the method further includes: Obtain unstructured industrial product data; The unstructured industrial product data is preprocessed to obtain the preprocessed industrial product data.

9. A risk warning device for industrial products, characterized in that: include: Extraction unit, classification unit and early warning unit; The extraction unit is used to extract features from the industrial product data to obtain feature data of the industrial product data; The grading unit is used to input the characteristic data of the industrial product data into a grading warning model to obtain the risk level of the industrial product data; the grading warning model is obtained by training based on a least squares support vector machine (LSSVM) algorithm using a characteristic data set and a label data set as training data, wherein the characteristic data set includes the characteristic data of the industrial product data in the training data set, and the label data set includes the risk level of the characteristic data of the industrial product data in the training data set; The early warning unit is used to issue a risk early warning to the industrial product data according to the risk level of the industrial product data.

10. A computing device, characterized in that The computing device includes: a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the steps of the risk warning method for industrial products as described in any one of claims 1 to 8 according to the instructions in the program code.

Images