Industrial product risk detection method and related device

Abstract

The embodiment of the invention provides an industrial product risk detection method and a related device, and the method comprises the steps: carrying out the feature extraction of a plurality of pieces of graded industrial product data through a risk detection model, and obtaining the feature attributes of the plurality of pieces of industrial product data; based on the association relationship between the feature attributes of the plurality of industrial product data, determining the association degree between the feature attributes; and performing risk detection on the plurality of industrial product data according to the correlation degree to obtain a detection result. In the embodiment of the invention, the risk detection model comprises a bidirectional encoder characterization method model, an improved text convolutional neural network and a bidirectional gating cycle unit network, and the quantities of output feature vectors of the three are the same. A large number of features can be extracted based on the risk detection model, risk detection is performed on multiple pieces of industrial product data based on a large number of feature attributes extracted based on the risk detection model, and the efficiency of risk detection can be effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of risk detection, and in particular to a risk detection 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. Testing the risks of industrial products is an effective means to ensure the safety of industrial products.

[0003] Currently, the risk of industrial products can be predicted by random inspection of industrial products. However, due to the wide variety and large quantity of industrial products, the method of detecting the risk of industrial products by random inspection has the problem of low detection efficiency. Summary of the invention

[0004] Based on the above problems, the present application provides a risk detection 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 method for detecting risks of industrial products, including:

[0007] Based on the risk types of the multiple industrial product data, label the multiple industrial product data to obtain multiple labeled industrial product data;

[0008] Based on the risk index, the plurality of labeled industrial product data are graded to obtain a plurality of graded industrial product data;

[0009] Based on the risk detection model, feature extraction is performed on the industrial product data of the plurality of divided levels to obtain feature attributes of the plurality of industrial product data; the risk detection model includes a bidirectional encoder representation method Bert model, an improved text convolution neural network Text-CNN network and a bidirectional gated recurrent unit Bi-Gru network, and the Bert model, the improved Text-CNN network and the Bi-Gru network have the same number of output feature vectors;

[0010] Determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data;

[0011] According to the correlation degree, risk detection is performed on the plurality of industrial product data to obtain detection results.

[0012] In a possible implementation, the step of classifying the plurality of labeled industrial product data based on the risk index to obtain a plurality of classified industrial product data includes:

[0013] For each of the plurality of graded industrial product data, extract a feature of a first indicator parameter of the graded industrial product data based on a Bert model to obtain a feature vector corresponding to the first indicator parameter;

[0014] Based on the Bert model, the bidirectional gated recurrent unit Bi-Gru network and the improved text convolution neural network Text-CNN network, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters;

[0015] The feature vector corresponding to the first indicator parameter is fused with the feature vectors corresponding to the other indicator parameters to obtain feature attributes of multiple industrial product data.

[0016] In a possible implementation, the Bert model, the bidirectional gated recurrent unit Bi-Gru network, and the improved text convolution neural network Text-CNN network are used to extract features of other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters, including:

[0017] Based on the Bert model, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain a feature vector corresponding to the Bert model;

[0018] Based on the Bi-Gru network, feature extraction is performed on the feature vector corresponding to the Bert model to obtain the feature vector corresponding to the Bi-Gru network; the feature vector corresponding to the Bi-Gru network obtained at the first moment is a linear interpolation between the feature vector corresponding to the Bi-Gru network obtained at the second moment and the candidate feature vector corresponding to the Bi-Gru network; the second moment is the moment before the first moment;

[0019] Based on the Text-CNN network, feature extraction is performed on the feature vector corresponding to the Bi-Gru network to obtain feature vectors corresponding to the other indicator parameters.

[0020] In a possible implementation, determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data includes:

[0021] Performing weight calculation on the characteristic attributes of the plurality of industrial product data to obtain the weight of each characteristic attribute;

[0022] Based on the association relationship between the feature attributes whose weights are greater than or equal to the weight threshold among the feature attributes of the plurality of industrial product data, the association degree between the feature attributes is determined.

[0023] In a possible implementation, determining the degree of association between the feature attributes based on the association relationship between the feature attributes whose weights are greater than or equal to the weight threshold among the feature attributes of the plurality of industrial product data includes:

[0024] Among the characteristic attributes of the plurality of industrial product data, characteristic attributes whose weights are greater than or equal to the weight threshold are taken as preferred characteristic attributes;

[0025] Based on the minimum support and confidence, the correlation degree is calculated for the preferred feature attributes to determine the correlation degree between the feature attributes.

[0026] In a possible implementation, determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data includes:

[0027] Based on the characteristic attributes of the plurality of industrial product data, calculating soft labels of the characteristic attributes of the plurality of industrial product data;

[0028] Based on soft labels of characteristic attributes of a plurality of industrial product data, the correlation between the characteristic attributes is determined.

[0029] In a possible implementation, the labeling of the plurality of industrial product data based on the risk types of the plurality of industrial product data to obtain the plurality of labeled industrial product data includes:

[0030] Preprocessing a plurality of industrial product data to obtain a plurality of preprocessed industrial product data;

[0031] Based on the risk types of the plurality of industrial product data, label processing is performed on the plurality of preprocessed industrial product data to obtain a plurality of labeled industrial product data.

[0032] In a possible implementation, the risk indicator includes a first indicator parameter, a second indicator parameter, a third indicator parameter and a fourth indicator parameter, the first indicator parameter indicates the possibility of risk occurrence, the second indicator parameter indicates the severity of risk occurrence, the third indicator parameter indicates the degree of risk diffusion, and the fourth indicator parameter indicates the social impact factor of risk, and the step of classifying the plurality of labeled industrial product data based on the risk indicator to obtain a plurality of classified industrial product data includes:

[0033] For each labeled industrial product data among the multiple labeled industrial product data, calculating the risk value of the labeled industrial product data according to a first weight corresponding to a first indicator parameter of the labeled industrial product data, a second weight corresponding to a second indicator parameter, a third weight corresponding to a third indicator parameter, and a fourth weight corresponding to a fourth indicator parameter;

[0034] According to the risk values ​​of a plurality of labeled industrial product data, the labeled industrial product data are graded to obtain a plurality of graded industrial product data.

[0035] Second aspect: The embodiment of the present application provides a risk detection device for industrial products, including:

[0036] A label processing unit, a grading unit, a feature extraction unit, a determination unit, and a detection unit;

[0037] The label processing unit is used to perform label processing on the multiple industrial product data based on the risk types of the multiple industrial product data to obtain multiple labeled industrial product data;

[0038] The grading unit is used to grade the plurality of labeled industrial product data based on the risk index to obtain a plurality of graded industrial product data;

[0039] The feature extraction unit is used to extract features from the multiple graded industrial product data based on the risk detection model to obtain feature attributes of the multiple industrial product data; the risk detection model includes a bidirectional encoder representation method Bert model, an improved text convolution neural network Text-CNN network and a bidirectional gated recurrent unit Bi-Gru network, and the Bert model, the improved Text-CNN network and the Bi-Gru network have the same number of output feature vectors;

[0040] The determining unit is used to determine the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data;

[0041] The detection unit is used to perform risk detection on the multiple industrial product data according to the correlation degree to obtain a detection result.

[0042] A third aspect: An embodiment of the present application provides a computer device, the computer device comprising: a processor and a memory;

[0043] The memory is used to store program code and transmit the program code to the processor;

[0044] The processor is used to execute the steps of the risk detection method for an industrial product as described above according to the instructions in the program code.

[0045] Compared with the prior art, this application has the following beneficial effects:

[0046] The embodiment of the present application provides a risk detection method for industrial products. Through the risk detection model, feature extraction can be performed on multiple graded industrial product data to obtain feature attributes of multiple industrial product data; based on the correlation relationship between the feature attributes of multiple industrial product data, the correlation between the feature attributes is determined; according to the correlation, risk detection is performed on multiple industrial product data to obtain detection results. In the embodiment of the present application, the risk detection model includes a bidirectional encoder representation model, an improved text convolutional neural network, and a bidirectional gated recurrent unit network. The number of feature vectors output by the bidirectional encoder representation model, the improved text convolutional neural network, and the bidirectional gated recurrent unit network is the same. Based on the risk detection model, a large number of features can be extracted. Based on the large number of feature attributes extracted by the risk detection model, risk detection is performed on multiple industrial product data, which can effectively improve the efficiency of risk detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] 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.

[0048] Figure 1 A flow chart of a method for risk detection of industrial products is provided for an embodiment of the present application;

[0049] Figure 2 A feature fusion schematic diagram provided for an embodiment of the present application;

[0050] Figure 3 A framework diagram of an improved Text-CNN network provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of a risk detection method for an industrial product provided in an embodiment of the present application;

[0052] Figure 5 A schematic structural diagram of a risk detection device for an industrial product is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0053] Currently, the risk of industrial products can be predicted by random inspection of industrial products. However, due to the wide variety and large quantity of industrial products, the method of detecting the risk of industrial products by random inspection has the problem of low detection efficiency.

[0054] Based on this, the embodiment of the present application provides a risk detection method and related devices for industrial products. Through the risk detection model, feature extraction can be performed on multiple graded industrial product data to obtain feature attributes of multiple industrial product data; based on the correlation relationship between the feature attributes of multiple industrial product data, the correlation between the feature attributes is determined; according to the correlation, risk detection is performed on multiple industrial product data to obtain detection results. In the embodiment of the present application, the risk detection model includes a bidirectional encoder representation model, an improved text convolutional neural network, and a bidirectional gated recurrent unit network. The bidirectional encoder representation model, the improved text convolutional neural network, and the bidirectional gated recurrent unit network have the same number of output feature vectors. Based on the risk detection model, a large number of features can be extracted. Based on the large number of feature attributes extracted by the risk detection model, risk detection is performed on multiple industrial product data, which can effectively improve the efficiency of risk detection.

[0055] 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.

[0056] See also Figure 1 , which is a flow chart of a risk detection method for an industrial product provided in an embodiment of the present application, including S101-S105.

[0057] S101. Based on the risk types of a plurality of industrial product data, label the plurality of industrial product data to obtain a plurality of labeled industrial product data.

[0058] In the embodiment of the present application, multiple industrial product data can be selected from the industrial product safety data set sampled by the State Administration for Market Regulation. Since the selected multiple industrial product data may contain some data noise, missing values, and outliers, it will affect the accuracy of risk prediction. To this end, in the embodiment of the present application, the selected multiple industrial product data can be preprocessed to fill in blank values ​​and process unbalanced data to obtain multiple preprocessed industrial product data.

[0059] Exemplarily, the process of preprocessing the selected multiple industrial product data includes but is not limited to data preprocessing methods such as missing value filling and singular value regression to obtain multiple industrial product data with low noise and high availability.

[0060] For multiple industrial product data, the preprocessed multiple industrial product data can be labeled based on their corresponding risk types to obtain multiple labeled industrial product data. The label information corresponding to each labeled industrial product data can represent the risk type of the industrial product data.

[0061] S102. Based on the risk index, the plurality of labeled industrial product data are classified into grades to obtain a plurality of graded industrial product data.

[0062] Exemplarily, the risk index may include a first index parameter, a second index parameter, a third index parameter and a fourth index parameter, wherein the first index parameter indicates the possibility of risk, the second index parameter indicates the severity of risk, the third index parameter indicates the degree of risk diffusion, and the fourth index parameter indicates the social impact factor of risk.

[0063] The first indicator parameter helps determine which risks are of greatest concern and what measures need to be taken to reduce the probability of these risks. The higher the value of the first indicator parameter, the higher the probability of the risk occurring, and the higher the probability, the greater the impact. The second indicator parameter helps determine which risks may have a significant impact on the enterprise and require appropriate emergency response measures or long-term risk management plans. The third indicator parameter helps determine which risks require more extensive attention and management, and what response mechanisms need to be established to reduce the impact of the spread. The fourth indicator parameter helps management departments to have a more comprehensive understanding of the possible consequences of risk events and take appropriate measures to reduce these impacts.

[0064] In an embodiment of the present application, for each labeled industrial product data among the multiple labeled industrial product data, the risk value of the labeled industrial product data can be calculated based on a first weight corresponding to a first indicator parameter of the labeled industrial product data, a second weight corresponding to a second indicator parameter, a third weight corresponding to a third indicator parameter, and a fourth weight corresponding to a fourth indicator parameter; based on the risk values ​​of the multiple labeled industrial product data, the labeled industrial product data are graded to obtain industrial product data of multiple graded levels.

[0065] In the embodiment of the present application, the labeled industrial product data can be graded based on four dimensions: the possibility of risk occurrence, the severity of risk occurrence, the degree of risk spread, and the social impact factor of risk.

[0066] For example, for the labeled industrial product data, the values ​​corresponding to the first indicator parameter, the second indicator parameter, the third indicator parameter and the fourth indicator parameter are A, B, C and D respectively; the first weight corresponding to the first indicator parameter, the second weight corresponding to the second indicator parameter, the third weight corresponding to the third indicator parameter and the fourth weight corresponding to the fourth indicator parameter are a, b, c and d respectively, then the risk value of the labeled industrial product data can be expressed as (A×a+B×b+C×c+D×d). Based on the size of the risk value, the labeled industrial product data can be graded to obtain industrial product data of multiple grades.

[0067] S103: Based on the risk detection model, extract features of the multiple graded industrial product data to obtain feature attributes of the multiple industrial product data.

[0068] Among them, the risk detection model includes a bidirectional encoder representation method (Bidirectional Encoder Representations from Transformers, Bert) model, an improved text convolutional neural network (Text Convolutional Neural Networks, Text-CNN) network and a bidirectional gated recurrent unit (Bidirectional Gated Recurrent Unit, Bi-Gru) network.

[0069] In the embodiment of the present application, the number of feature vectors output by the improved Text-CNN network is the same as the number of feature vectors output by the Bert model and the Bi-Gru network. For example, the number of feature vectors output by the improved Text-CNN network can be adjusted from 1 to 32.

[0070] In a possible implementation, for each of the multiple classified industrial product data, feature extraction is performed on the first indicator parameter of the classified industrial product data based on the Bert model to obtain a feature vector corresponding to the first indicator parameter. Based on the Bert model, the bidirectional gated recurrent unit Bi-Gru network and the improved text convolution neural network Text-CNN network, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters. Figure 2 As shown, this figure is a feature fusion schematic diagram provided in an embodiment of the present application.

[0071] In the embodiment of the present application, a Bert model with enhanced semantic information extraction capability is adopted, and an encoder (Transformer Encoder) of a Transformer model is used as a feature extractor to perform feature extraction on the first indicator parameter, thereby improving the accuracy of feature extraction.

[0072] Based on the Bert model, the bidirectional gated recurrent unit Bi-Gru network and the improved text convolution neural network Text-CNN network, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters.

[0073] Exemplarily, when the first indicator parameter represents the possibility of a risk occurring, other indicator parameters may include a second indicator parameter representing the severity of the risk occurring, a third indicator parameter representing the degree of risk diffusion, and a fourth indicator parameter representing a social impact factor of the risk.

[0074] In a possible implementation, the Bert model, the bidirectional gated recurrent unit Bi-Gru network, and the improved text convolution neural network Text-CNN network are used to extract features of other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters, including:

[0075] Based on the Bert model, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain a feature vector corresponding to the Bert model;

[0076] Based on the Bi-Gru network, feature extraction is performed on the feature vector corresponding to the Bert model to obtain the feature vector corresponding to the Bi-Gru network; the feature vector corresponding to the Bi-Gru network obtained at the first moment is a linear interpolation between the feature vector corresponding to the Bi-Gru network obtained at the second moment and the candidate feature vector corresponding to the Bi-Gru network; the second moment is the moment before the first moment.

[0077] Exemplarily, the feature vector corresponding to the Bi-Gru network obtained at the first moment is shown in formula (1):

[0078]

[0079] Where t represents the first moment; h t is the output of the Bi-Gru network at the first moment; h t-1 is the output of the Bi-Gru network at the second moment; represents the candidate output corresponding to the Bi-Gru network; z t It is the update gate, which indicates the degree of content update.

[0080] zt It can be expressed by formula (2) as follows:

[0081] z t =σ(W z x t +U z h t-1 +b z ) (2)

[0082] It can be expressed as follows by formula (3):

[0083]

[0084] Among them, r t is a set of reset gates, which can be expressed by formula (4) as follows:

[0085] r t =σ(W r x t +U r h t-1 +b r ) (4)

[0086] Among them, W z , W h , W r are the weight matrices of the update gate, candidate output, and reset gate of the neuron at the first moment; U z , U h , U r are the weight matrices of the update gate, candidate output, and reset gate loop input at the first moment; b z 、b h 、b r They are the offset vectors of the update gate, candidate output, and reset gate at the first moment respectively.

[0087] After obtaining the feature vector corresponding to the Bi-Gru network, feature extraction is performed on the feature vector corresponding to the Bi-Gru network based on the improved Text-CNN network to obtain the feature vector corresponding to the other indicator parameters. Figure 3 As shown in FIG. 1 , the figure is a framework diagram of an improved Text-CNN network provided in an embodiment of the present application. The loss function of the improved Text-CNN network can be expressed as follows by formula (5):

[0088] L1(θ, θ1, θ2)=L1(θ, θ1)+L1(θ, θ2) (5)

[0089] By performing feature fusion on the feature vector corresponding to the first indicator parameter and the feature vector corresponding to the other indicator parameters, feature attributes of multiple industrial product data can be obtained.

[0090] S104: Determine the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data.

[0091] In a possible implementation, weight calculation may be performed on the characteristic attributes of the plurality of industrial product data to obtain the weight of each characteristic attribute. Based on the correlation relationship between the characteristic attributes whose weights are greater than or equal to the weight threshold among the characteristic attributes of the plurality of industrial product data, the correlation between the characteristic attributes is determined.

[0092] In an embodiment of the present application, among the characteristic attributes of the multiple industrial product data, the characteristic attributes whose weights are greater than or equal to the weight threshold can be used as preferred characteristic attributes; based on the minimum support and confidence, the correlation degree of the preferred characteristic attributes is calculated to determine the correlation degree between the characteristic attributes.

[0093] Different indicator parameters have different impacts on the industrial product risk detection process. In one possible implementation, based on the characteristic attributes of the multiple industrial product data, soft labels of the characteristic attributes of the multiple industrial product data are calculated; based on the soft labels of the characteristic attributes of the multiple industrial product data, the correlation between the characteristic attributes is determined.

[0094] Among them, the soft labels based on the characteristic attributes of multiple industrial product data can determine the frequent item set association rules to determine the association degree between the characteristic attributes.

[0095] In the embodiment of the present application, risk analysis can be performed based on frequent item set association rules, and the data values ​​corresponding to the characteristic attributes can be sorted to obtain the fluctuation range of the data values ​​of multiple characteristic attributes. The fluctuation range can be expressed as [V min , V max ], where V min and V max They are the minimum and maximum values ​​of the data values ​​of multiple feature attributes.

[0096] S105. Perform risk detection on the plurality of industrial product data according to the correlation degree to obtain a detection result.

[0097] In an embodiment of the present application, based on a risk detection model constructed by a Bert model, a Bi-Gru network, and an improved Text-CNN network, risk detection can be performed on industrial product data from multiple dimensions, thereby improving the accuracy of risk prediction and downgrading or risk isolation processing based on the detection results.

[0098] For ease of understanding, the following Figure 4 A risk detection method for industrial products provided in an embodiment of the present application is generally introduced.

[0099] In an embodiment of the present application, the industrial product data set sampled by the State Administration for Market Regulation can be selected to obtain multiple industrial product data. Based on the data preprocessing module, the attributes of multiple industrial product data are determined, and the multiple industrial product data are preprocessed to obtain multiple preprocessed industrial product data. The preprocessed multiple industrial product data are subjected to service object and security type-oriented label processing to obtain multiple labeled industrial product data.

[0100] The data preprocessing module sends multiple labeled industrial product data to the multi-dimensional risk data grading module. The multi-dimensional risk data grading module can perform risk grading on the multiple labeled industrial product data based on multi-dimensional risk indicators to obtain multiple graded industrial product data.

[0101] The multi-dimensional risk data classification module sends a plurality of classified industrial product data to the industrial product risk prediction module. The industrial product risk prediction module includes a feature extraction module, a soft-label-based feature fusion module, and a risk prediction module.

[0102] The feature extraction module can realize feature extraction of multiple graded industrial product data based on the Bert model, Bi-Gru network and improved Text-CNN network. The feature fusion module based on soft labels performs feature fusion on the extracted feature vectors to obtain the fused feature vectors, and obtains the frequent item set association rules based on the fused feature vectors. The association degree can be determined based on the frequent item set association rules.

[0103] The risk prediction module performs risk detection on the plurality of industrial product data according to the correlation degree to obtain a detection result.

[0104] In summary, in the embodiment of the present application, the risk detection model includes a bidirectional encoder representation model, an improved text convolutional neural network and a bidirectional gated recurrent unit network, and the number of output feature vectors of the bidirectional encoder representation model, the improved text convolutional neural network and the bidirectional gated recurrent unit network is the same. Based on the risk detection model, a large number of features can be extracted. Based on the large number of feature attributes extracted by the risk detection model, risk detection can be performed on multiple industrial product data, which can effectively improve the efficiency of risk detection.

[0105] This application provides a risk detection device for industrial products, see Figure 5 , this figure is a structural schematic diagram of a risk detection 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.

[0106] A risk detection device 1100 for industrial products, comprising:

[0107] A label processing unit 1101, a level classification unit 1102, a feature extraction unit 1103, a determination unit 1104, and a detection unit 1105;

[0108] The label processing unit 1101 is used to perform label processing on the multiple industrial product data based on the risk types of the multiple industrial product data to obtain multiple labeled industrial product data;

[0109] The grading unit 1102 is used to grade the plurality of labeled industrial product data based on the risk index to obtain a plurality of graded industrial product data;

[0110] The feature extraction unit 1103 is used to extract features from the multiple graded industrial product data based on the risk detection model to obtain feature attributes of the multiple industrial product data; the risk detection model includes a bidirectional encoder representation method Bert model, an improved text convolution neural network Text-CNN network and a bidirectional gated recurrent unit Bi-Gru network, and the Bert model, the improved Text-CNN network and the Bi-Gru network have the same number of output feature vectors;

[0111] The determining unit 1104 is used to determine the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data;

[0112] The detection unit 1105 is used to perform risk detection on the multiple industrial product data according to the correlation degree to obtain a detection result.

[0113] In a possible implementation, the level division unit is specifically configured to:

[0114] For each of the plurality of graded industrial product data, extract a feature of a first indicator parameter of the graded industrial product data based on a Bert model to obtain a feature vector corresponding to the first indicator parameter;

[0115] Based on the Bert model, the bidirectional gated recurrent unit Bi-Gru network and the improved text convolution neural network Text-CNN network, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters;

[0116] The feature vector corresponding to the first indicator parameter is fused with the feature vectors corresponding to the other indicator parameters to obtain feature attributes of multiple industrial product data.

[0117] In a possible implementation, the level division unit is specifically configured to:

[0118] Based on the Bert model, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain a feature vector corresponding to the Bert model;

[0119] Based on the Bi-Gru network, feature extraction is performed on the feature vector corresponding to the Bert model to obtain the feature vector corresponding to the Bi-Gru network; the feature vector corresponding to the Bi-Gru network obtained at the first moment is a linear interpolation between the feature vector corresponding to the Bi-Gru network obtained at the second moment and the candidate feature vector corresponding to the Bi-Gru network; the second moment is the moment before the first moment;

[0120] Based on the Text-CNN network, feature extraction is performed on the feature vector corresponding to the Bi-Gru network to obtain feature vectors corresponding to the other indicator parameters.

[0121] In a possible implementation manner, the determining unit is specifically configured to:

[0122] Performing weight calculation on the characteristic attributes of the plurality of industrial product data to obtain the weight of each characteristic attribute;

[0123] Based on the association relationship between the feature attributes whose weights are greater than or equal to the weight threshold among the feature attributes of the plurality of industrial product data, the association degree between the feature attributes is determined.

[0124] In a possible implementation, the determining unit is specifically configured to: select, among the characteristic attributes of the plurality of industrial product data, characteristic attributes whose weights are greater than or equal to a weight threshold as preferred characteristic attributes;

[0125] Based on the minimum support and confidence, the correlation degree is calculated for the preferred feature attributes to determine the correlation degree between the feature attributes.

[0126] In a possible implementation manner, the determining unit is specifically configured to:

[0127] Based on the characteristic attributes of the plurality of industrial product data, calculating soft labels of the characteristic attributes of the plurality of industrial product data;

[0128] Based on soft labels of characteristic attributes of a plurality of industrial product data, the correlation between the characteristic attributes is determined.

[0129] In a possible implementation, the label processing unit is specifically configured to:

[0130] Preprocessing a plurality of industrial product data to obtain a plurality of preprocessed industrial product data;

[0131] Based on the risk types of the plurality of industrial product data, label processing is performed on the plurality of preprocessed industrial product data to obtain a plurality of labeled industrial product data.

[0132] In a possible implementation, the risk index includes a first indicator parameter, a second indicator parameter, a third indicator parameter and a fourth indicator parameter, the first indicator parameter indicates the possibility of risk occurrence, the second indicator parameter indicates the severity of risk occurrence, the third indicator parameter indicates the degree of risk diffusion, the fourth indicator parameter indicates the social impact factor of risk, and the level classification unit is specifically used to:

[0133] For each labeled industrial product data among the multiple labeled industrial product data, calculating the risk value of the labeled industrial product data according to a first weight corresponding to a first indicator parameter of the labeled industrial product data, a second weight corresponding to a second indicator parameter, a third weight corresponding to a third indicator parameter, and a fourth weight corresponding to a fourth indicator parameter;

[0134] According to the risk values ​​of a plurality of labeled industrial product data, the labeled industrial product data are graded to obtain a plurality of graded industrial product data.

[0135] In summary, in the embodiment of the present application, the risk detection model includes a bidirectional encoder representation model, an improved text convolutional neural network and a bidirectional gated recurrent unit network, and the number of output feature vectors of the bidirectional encoder representation model, the improved text convolutional neural network and the bidirectional gated recurrent unit network is the same. Based on the risk detection model, a large number of features can be extracted. Based on the large number of feature attributes extracted by the risk detection model, risk detection can be performed on multiple industrial product data, which can effectively improve the efficiency of risk detection.

[0136] An embodiment of the present application provides a computer device, the computer device comprising: a processor and a memory;

[0137] The memory is used to store program code and transmit the program code to the processor;

[0138] The processor is used to execute the steps of the risk detection method for an industrial product as described above according to the instructions in the program code.

[0139] 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 detection method for industrial products, characterized in that: include: Based on the risk types of the multiple industrial product data, label the multiple industrial product data to obtain multiple labeled industrial product data; Based on the risk index, the plurality of labeled industrial product data are graded to obtain a plurality of graded industrial product data; Based on the risk detection model, feature extraction is performed on the industrial product data of the plurality of divided levels to obtain feature attributes of the plurality of industrial product data; the risk detection model includes a bidirectional encoder representation method Bert model, an improved text convolution neural network Text-CNN network and a bidirectional gated recurrent unit Bi-Gru network, and the Bert model, the improved Text-CNN network and the Bi-Gru network have the same number of output feature vectors; Determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data; According to the correlation degree, risk detection is performed on the plurality of industrial product data to obtain detection results.

2. The method according to claim 1, characterized in that The step of classifying the plurality of labeled industrial product data based on the risk index to obtain a plurality of classified industrial product data includes: For each of the plurality of graded industrial product data, extract a feature of a first indicator parameter of the graded industrial product data based on a Bert model to obtain a feature vector corresponding to the first indicator parameter; Based on the Bert model, the bidirectional gated recurrent unit Bi-Gru network and the improved text convolution neural network Text-CNN network, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters; The feature vector corresponding to the first indicator parameter is fused with the feature vectors corresponding to the other indicator parameters to obtain feature attributes of multiple industrial product data.

3. The method according to claim 2, characterized in that The method based on the Bert model, the bidirectional gated recurrent unit Bi-Gru network and the improved text convolution neural network Text-CNN network performs feature extraction on other indicator parameters other than the first indicator parameter to obtain feature vectors corresponding to the other indicator parameters, including: Based on the Bert model, feature extraction is performed on other indicator parameters other than the first indicator parameter to obtain a feature vector corresponding to the Bert model; Based on the Bi-Gru network, feature extraction is performed on the feature vector corresponding to the Bert model to obtain the feature vector corresponding to the Bi-Gru network; the feature vector corresponding to the Bi-Gru network obtained at the first moment is a linear interpolation between the feature vector corresponding to the Bi-Gru network obtained at the second moment and the candidate feature vector corresponding to the Bi-Gru network; the second moment is the moment before the first moment; Based on the Text-CNN network, feature extraction is performed on the feature vector corresponding to the Bi-Gru network to obtain feature vectors corresponding to the other indicator parameters.

4. The method according to claim 1, characterized in that: The determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data includes: Performing weight calculation on the characteristic attributes of the plurality of industrial product data to obtain the weight of each characteristic attribute; Based on the association relationship between the feature attributes whose weights are greater than or equal to the weight threshold among the feature attributes of the plurality of industrial product data, the association degree between the feature attributes is determined.

5. The method according to claim 4, characterized in that The determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes whose weights are greater than or equal to the weight threshold among the characteristic attributes of the plurality of industrial product data comprises: Among the characteristic attributes of the plurality of industrial product data, characteristic attributes whose weights are greater than or equal to the weight threshold are taken as preferred characteristic attributes; Based on the minimum support and confidence, the correlation degree is calculated for the preferred feature attributes to determine the correlation degree between the feature attributes.

6. The method according to claim 1, characterized in that The determining the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data includes: Based on the characteristic attributes of the plurality of industrial product data, calculating soft labels of the characteristic attributes of the plurality of industrial product data; Based on soft labels of characteristic attributes of a plurality of industrial product data, the correlation between the characteristic attributes is determined.

7. The method according to claim 1, characterized in that The step of labeling the plurality of industrial product data based on the risk types of the plurality of industrial product data to obtain a plurality of labeled industrial product data includes: Preprocessing a plurality of industrial product data to obtain a plurality of preprocessed industrial product data; Based on the risk types of the plurality of industrial product data, label processing is performed on the plurality of preprocessed industrial product data to obtain a plurality of labeled industrial product data.

8. The method according to claim 1, characterized in that The risk index includes a first index parameter, a second index parameter, a third index parameter and a fourth index parameter, wherein the first index parameter indicates the possibility of risk occurrence, the second index parameter indicates the severity of risk occurrence, the third index parameter indicates the degree of risk diffusion, and the fourth index parameter indicates the social impact factor of risk. Based on the risk index, the plurality of labeled industrial product data are graded to obtain a plurality of graded industrial product data, including: For each labeled industrial product data among the multiple labeled industrial product data, calculating the risk value of the labeled industrial product data according to a first weight corresponding to a first indicator parameter of the labeled industrial product data, a second weight corresponding to a second indicator parameter, a third weight corresponding to a third indicator parameter, and a fourth weight corresponding to a fourth indicator parameter; According to the risk values ​​of a plurality of labeled industrial product data, the labeled industrial product data are graded to obtain a plurality of graded industrial product data.

9. A risk detection device for industrial products, characterized in that: include: A label processing unit, a grading unit, a feature extraction unit, a determination unit, and a detection unit; The label processing unit is used to perform label processing on the multiple industrial product data based on the risk types of the multiple industrial product data to obtain multiple labeled industrial product data; The grading unit is used to grade the plurality of labeled industrial product data based on the risk index to obtain a plurality of graded industrial product data; The feature extraction unit is used to extract features from the multiple graded industrial product data based on the risk detection model to obtain feature attributes of the multiple industrial product data; the risk detection model includes a bidirectional encoder representation method Bert model, an improved text convolution neural network Text-CNN network and a bidirectional gated recurrent unit Bi-Gru network, and the Bert model, the improved Text-CNN network and the Bi-Gru network have the same number of output feature vectors; The determining unit is used to determine the degree of association between the characteristic attributes based on the association relationship between the characteristic attributes of the plurality of industrial product data; The detection unit is used to perform risk detection on the multiple industrial product data according to the correlation degree to obtain a detection result.

10. A computer device, characterized in that: The computer device comprises: 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 detection method for industrial products as described in any one of claims 1 to 8 according to the instructions in the program code.

Images