Dangerous chemicals identification method and device

Abstract

The embodiment of the present application discloses a method for identifying hazardous chemicals, which includes: obtaining chemical information data, and constructing a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm of deep learning; defining text variables and numerical variables based on the naming rules and properties of chemicals, extracting hazardous chemical determination conditions from the chemical information data, and establishing a hazardous chemical rule library based on text variables, numerical variables and hazardous chemical determination conditions; establishing a hazardous chemical data model based on graph neural network technology based on the chemical information data; constructing a hazardous chemical identification system using the hazardous chemical semantic library, hazardous chemical rule library and hazardous chemical data model; obtaining information on chemicals to be classified, and using the hazardous chemical identification system to classify the chemicals to be classified in the information on chemicals to be classified, and obtaining hazardous chemical determination results. The present application realizes more intelligent and accurate identification of hazardous chemicals.

Description

Technical Field

[0001] The present application relates to the field of data analysis technology, and in particular to a method and device for identifying hazardous chemicals. Background Art

[0002] Dangerous chemicals have strict safety requirements during production, storage, transportation and use. In the process of identifying hazardous chemicals, due to the non-standard naming of some chemicals and the limitations of the catalog, direct comparison with the catalog often leads to misjudgment or even missed judgment. For example, a large number of chemicals are not included in the catalog but still have the characteristics of hazardous chemicals.

[0003] Therefore, how to identify hazardous chemicals efficiently and accurately has become an urgent problem to be solved. Summary of the invention

[0004] The present application provides a method and device for identifying hazardous chemicals to achieve more intelligent and accurate hazardous chemical identification.

[0005] This application provides the following solutions:

[0006] According to a first aspect, a method for identifying hazardous chemicals is provided, the method comprising: acquiring chemical information data, and constructing a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm of deep learning; defining text variables and numerical variables based on the naming rules and properties of chemicals, extracting hazardous chemical determination conditions from the chemical information data, and establishing a hazardous chemical rule library based on the text variables, the numerical variables and the hazardous chemical determination conditions; establishing a hazardous chemical data model based on the graph neural network technology based on the chemical information data; constructing a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model; acquiring information on chemicals to be classified, and classifying the chemicals to be classified in the information on chemicals to be classified using the hazardous chemical identification system to obtain hazardous chemical determination results.

[0007] According to an achievable method in an embodiment of the present application, the construction of a hazardous chemicals semantic library based on the chemical information data using a deep learning multimodal data fusion algorithm includes: using a recurrent neural network to process text information in the chemical information data to obtain data features of different modalities; integrating the data features of different modalities in the fusion layer of the recurrent neural network, and inputting the integration results into a fully connected layer for data classification to obtain a hazardous chemicals semantic library; wherein, an attention mechanism is introduced in the data classification process to automatically select and strengthen relevant data feature information.

[0008] According to an achievable method in an embodiment of the present application, establishing a hazardous chemicals data model based on the chemical information data and graph neural network technology includes: obtaining chemical names, chemical properties and the relationship between chemicals from the chemical information data; representing the chemical names and the chemical properties as nodes, and the relationship between chemicals as edges; using a graph convolutional network to extract the relationship features between nodes and edges, using a graph reasoning algorithm based on the relationship features to derive the features of chemicals that are not clearly identified, and constructing a knowledge graph in combination with the chemical information data; and constructing a hazardous chemicals data model based on the knowledge graph.

[0009] According to an achievable method in an embodiment of the present application, the use of the hazardous chemicals identification system to classify the chemicals to be classified in the chemical information to be classified to obtain a hazardous chemical determination result includes: preprocessing the chemical information to be classified to obtain input information, the preprocessing including: invalid text fragment filtering and / or synonym processing; inputting the input information into the hazardous chemicals semantic library, the hazardous chemicals semantic library performing semantic matching on the input information to obtain a semantic matching result; combining the semantic matching result and the hazardous chemicals determination condition generated by the hazardous chemicals rule library, and using the hazardous chemicals data model to classify the chemicals to be classified to obtain a hazardous chemical determination result.

[0010] According to an achievable method in an embodiment of the present application, the hazardous goods judgment condition generated by combining the semantic matching result and the hazardous goods rule library is used to classify the chemicals to be classified using the hazardous chemicals data model to obtain the hazardous chemicals judgment result, including: if the semantic matching result is a complete mismatch, the chemicals to be classified are marked as non-hazardous chemicals; if the semantic matching result is a match or a suspected match, the matching keywords of the hazardous chemicals semantic library are recorded; the information of the chemicals to be classified for which the semantic matching result is a match or a suspected match, the matching keywords and the hazardous goods judgment condition are input into the hazardous chemicals data model, and the hazardous chemicals judgment result is obtained according to the output of the hazardous chemicals data model.

[0011] According to an achievable method in an embodiment of the present application, the construction of a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model includes: constructing a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model, and training the hazardous chemical identification system, the training including: obtaining training data including multiple training samples, the training samples including chemical information samples and their corresponding hazardous chemical determination result samples; the hazardous chemical semantic library outputs a semantic matching result based on the chemical information sample; the hazardous chemical rule library outputs a hazardous chemical determination condition based on the chemical information sample; different weights are assigned to the semantic matching result, the hazardous chemical determination condition and the chemical information sample, the hazardous chemical data model performs model inference after weighted fusion of the semantic matching result, the hazardous chemical determination condition and the chemical information sample according to the weight, and outputs a determination result; wherein the training objectives include: minimizing the difference between the generated determination result and the hazardous chemical determination result sample; and adjusting the weight value according to the training objective.

[0012] According to an achievable method in an embodiment of the present application, the training also includes: optimizing the recognition strategy using reinforcement learning, and dynamically adjusting the parameters of the hazardous chemicals data model according to the training objectives; wherein the reinforcement learning strategy optimization includes one or more of the following: converting the chemical identification task into a decision-making process in reinforcement learning; designing a multi-level reward mechanism to guide the hazardous chemicals data model to optimize the recognition strategy; dynamically adjusting the recognition path, and giving priority to high-weight reasoning paths when the input data is incomplete or ambiguous.

[0013] According to a second aspect, a hazardous chemical identification device is provided, the device comprising: a semantic library construction unit, configured to acquire chemical information data, and construct a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm of deep learning; a rule library construction unit, configured to define text variables and numerical variables based on the naming rules and attributes of chemicals, extract hazardous chemical determination conditions from the chemical information data, and establish a hazardous chemical rule library based on the text variables, the numerical variables and the hazardous chemical determination conditions; a data model construction unit, configured to establish a hazardous chemical data model based on the graph neural network technology based on the chemical information data; a determination result acquisition unit, configured to construct a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model; acquire information on chemicals to be classified, and classify the chemicals to be classified in the information on chemicals to be classified using the hazardous chemical identification system to obtain hazardous chemical determination results.

[0014] According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

[0015] According to a fourth aspect, an electronic device is provided, comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, and the program instructions, when read and executed by the one or more processors, execute the steps of any one of the methods described in the first aspect.

[0016] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0017] (1) This application acquires chemical information data and uses a multimodal data fusion algorithm of deep learning to construct a semantic library of hazardous chemicals, which can achieve accurate semantic analysis of chemical information; a hazardous chemical rule library is constructed by combining naming rules and attributes to ensure the logic and comprehensiveness of hazardous chemical determination; the hazardous chemical data model established based on graph neural networks further enhances the ability to understand the complex relationships between chemicals. Ultimately, the recognition system constructed using the semantic library, rule library, and data model can efficiently and accurately classify the information of the chemicals to be tested, providing technical support for the safe management of hazardous chemicals.

[0018] (2) This application uses a recurrent neural network to extract text features from chemical information data, and a multimodal fusion algorithm to conduct a comprehensive analysis of features of different data types, which can effectively improve the efficiency and accuracy of semantic library construction. Introducing an attention mechanism in the classification process can automatically select and strengthen feature information that is critical to the identification of hazardous chemicals, thereby greatly improving the adaptability of the semantic library to complex and diverse chemical data.

[0019] (3) The hazardous chemicals data model based on graph neural network in this application can deeply mine the implicit correlation between chemicals, thereby deriving the characteristics of chemicals that are not clearly identified. Through knowledge graph reasoning technology, the reasoning ability and dynamic adaptability of the model are further enhanced, making the hazardous chemicals data model perform better when dealing with complex chemical environments. The knowledge graph constructed in combination with chemical information data not only enriches the knowledge structure of the model, but also significantly improves the accuracy of hazardous chemical identification.

[0020] (4) This application improves the standardization and effectiveness of input data and reduces error interference by preprocessing the information of chemicals to be classified; uses a semantic library for semantic matching and generates judgment conditions in combination with a rule library to ensure the reliability of hazardous chemical classification; and further uses a hazardous chemical data model to conduct in-depth analysis of suspected hazardous chemicals, significantly improving the overall accuracy of the system's classification judgment.

[0021] (5) This application combines the semantic matching results with the judgment conditions generated by the rule base. The hazardous chemicals data model can analyze the properties of suspected hazardous chemicals more finely and dynamically adjust the classification strategy. In the case of complete mismatch, non-hazardous chemicals can be quickly excluded to reduce the system burden; for matched and suspected matched chemicals, in-depth analysis is performed by recording matching keywords and judgment conditions to ensure the accuracy and logic of the classification results. Overall, the classification efficiency and accuracy of the system have been significantly improved.

[0022] (6) This application constructs a recognition system through the comprehensive application of semantic libraries, rule libraries and data models, which can make full use of multimodal data and chemical knowledge graphs for efficient classification. During the training process, weights are assigned to different data sources, and the model performance is dynamically optimized with the goal of minimizing the difference between the judgment results and the actual sample results, thereby significantly improving the system's learning ability and judgment accuracy.

[0023] (7) This application uses a reinforcement learning strategy to dynamically optimize the recognition path. When the input data is incomplete or ambiguous, the reasoning path with a higher weight can be given priority, thereby ensuring the stability of the system and the accuracy of classification. The design of the multi-level reward mechanism enables the system to continuously adjust the model parameters and weight values, optimize the recognition strategy, and further improve the adaptability and judgment effect to complex chemical environments.

[0024] Of course, any product implementing the present application does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] 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 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 work.

[0026] Figure 1 is a flow chart of a method for identifying hazardous chemicals provided in an embodiment of the present application;

[0027] Figure 2 A structural block diagram of a hazardous chemical identification device provided in an embodiment of the present application;

[0028] Figure 3 A schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0030] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0031] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0032] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0033] In practice, current hazardous chemical identification methods mainly rely on manual review and can only use a single data type, resulting in many deficiencies. The manual review process usually takes a lot of time and human resources, especially when the amount of chemical information is huge and the naming is complex, omissions and errors are prone to occur. At the same time, a single data type (such as text descriptions or a single attribute) cannot fully capture the diversity and complexity of chemicals, and it is difficult to effectively respond to the needs of ambiguous information, incomplete data, and multimodal data fusion in practical applications. These limitations make it difficult for existing methods to meet actual needs in terms of recognition accuracy and efficiency, and more intelligent and automated solutions are urgently needed.

[0034] In view of this, the present application provides a new idea. Figure 1 A flowchart of a method for identifying hazardous chemicals based on an artificial intelligence model provided in an embodiment of the present application is shown in FIG. Figure 1 As shown in , the method may include the following steps:

[0035] Step 101: Acquire chemical information data, and construct a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm based on deep learning;

[0036] Step 102: defining text variables and numerical variables according to the naming rules and properties of chemicals, extracting hazardous chemical determination conditions from the chemical information data, and establishing a hazardous chemical rule base according to the text variables, the numerical variables and the hazardous chemical determination conditions;

[0037] Step 103: Establishing a hazardous chemicals data model based on the chemical information data and graph neural network technology;

[0038] Step 104: construct a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model; obtain information on chemicals to be classified, and use the hazardous chemical identification system to classify the chemicals to be classified in the information on chemicals to be classified to obtain a hazardous chemical determination result.

[0039] It can be seen from the above process that this application can achieve accurate semantic analysis of chemical information by acquiring chemical information data and using a multimodal data fusion algorithm of deep learning to build a hazardous chemical semantic library; combining naming rules and attributes to build a hazardous chemical rule library to ensure the logic and comprehensiveness of hazardous chemical judgments; the hazardous chemical data model based on graph neural networks further enhances the ability to understand the complex relationships between chemicals. Ultimately, the recognition system built using the semantic library, rule library, and data model can efficiently and accurately classify the information of the chemicals to be tested, providing technical support for the safe management of hazardous chemicals.

[0040] First, the above step 101, namely "obtaining chemical information data, and constructing a hazardous chemical semantic library based on the chemical information data using a deep learning multimodal data fusion algorithm" is described in detail in conjunction with an embodiment.

[0041] This application first establishes a semantic library of hazardous chemicals through chemical information data. Hazardous chemicals are dangerous chemicals, which is the abbreviation for hazardous chemicals in this field. The semantic library of hazardous chemicals is used to comprehensively store and organize semantic information related to hazardous chemicals. The semantic library integrates various attribute information of chemicals (such as text descriptions, numerical features, and image data) to provide a unified semantic model support for the intelligent identification of hazardous chemicals.

[0042] In the identification of hazardous chemicals, chemical information data is the core data source, which includes multi-dimensional information such as the name of the chemical, CAS number, ingredient information, physical and chemical properties, hazard classification, etc. Chemical information data also includes external knowledge bases and regulatory standards (such as the classification of hazardous chemicals in different countries, transportation rules, etc.).

[0043] However, this information often exists in multimodal form, such as text descriptions, numerical parameters, and image data. Therefore, it is necessary to uniformly model it through a multimodal data fusion algorithm based on deep learning to build a semantic library for hazardous chemicals.

[0044] The multimodal data fusion algorithm of deep learning refers to an algorithm that extracts, associates and integrates data from multiple different modalities (such as text, numerical values, images, etc.) through deep learning technology. This method aims to fully explore the complementarity of information from different modalities to improve the efficiency and accuracy of data processing and analysis. The multimodal data fusion algorithm can effectively process different types of data features. Specifically, first, the text information of chemicals (such as names, ingredient descriptions) can be used to extract sequence features using recurrent neural networks (RNN) or their variants (such as LSTM, GRU) to learn the naming rules and associations of chemicals. Secondly, for the numerical data of chemicals (such as boiling point, flash point, toxicity index), a fully connected network is used for processing to extract its potential feature distribution. For image data (such as the structural formula of chemicals), its visual features can be extracted through convolutional neural networks (CNN).

[0045] For example, when building a semantic database for hazardous chemicals, the basic information of a chemical may include:

[0046] Name: Methanol

[0047] CAS No.: 67-56-1

[0048] Boiling point: 64.7°C

[0049] Flash point: 11°C

[0050] Structural formula: molecular formula CH 3 2D image of -OH

[0051] The algorithm first extracts text features from the name "methanol" and uses the model to identify its high-frequency semantic features related to "alcohol chemicals". At the same time, the numerical processing module identifies its flash point below 30°C and its numerical features that meet the flammability standards. Combined with the structural image, the visual processing module extracts the hydroxyl (-OH) characteristic in its molecules. These features are integrated into a unified feature vector in the multimodal data fusion layer and eventually included in the hazardous chemicals semantic library.

[0052] As an implementable method, when constructing a semantic library of hazardous chemicals, the present application uses a recurrent neural network to process the text information in the chemical information data to obtain data features of different modalities; the data features of different modalities are integrated in the fusion layer of the recurrent neural network, and the integration results are input into the fully connected layer for data classification to obtain a semantic library of hazardous chemicals; wherein, an attention mechanism is introduced in the data classification process to automatically select and strengthen relevant data feature information.

[0053] Specifically, we first collect the names and descriptions of chemicals, and use a recurrent neural network (RNN) to process this text information. Through RNN, we can extract the lexical features of each chemical and capture the correlation between sequences, such as the composition structure, chemical name, and properties of the chemical. Next, in the fusion layer of the recurrent neural network, we integrate text features with other possible modal data, such as the physical properties or image information of the chemical. These different modal data are unified into a comprehensive feature space. Finally, the fused features are input into the fully connected layer for classification. In this process, the attention mechanism automatically selects important features related to the classification task. For example, when the classification involves the potential hazardous properties of certain chemicals, the model automatically enhances the features related to the hazard and ignores irrelevant background information, thereby improving the accuracy of the classification results.

[0054] The above step 102, namely "defining text variables and numerical variables according to the naming rules and properties of chemicals, extracting hazardous chemical determination conditions from the chemical information data, and establishing a hazardous chemical rule library according to the text variables, numerical variables and hazardous chemical determination conditions" is described in detail below in conjunction with an embodiment.

[0055] This application analyzes the naming rules and physical and chemical properties of chemicals, extracts the conditions for hazard determination, and organizes these conditions in the form of rules to form a hazardous chemical rule library, providing a standardized basis for subsequent hazardous chemical identification.

[0056] First, define text variables and numerical variables. Text variables mainly refer to information such as the name, category, and specific key descriptors of chemicals; numerical variables refer to the physical and chemical parameters of chemicals, such as flash point, boiling point, toxicity threshold, etc.; then, extract the judgment conditions for hazardous chemicals, and extract the rule conditions related to hazard judgment based on the relevant laws and regulations on hazardous chemicals in the chemical information data. For example, "liquids with a flash point below 60°C can be classified as flammable liquids." Finally, organize and structure the extracted judgment conditions to form a hierarchical and extensible rule base. The rule base not only contains fixed judgment conditions, but also supports dynamic expansion of new rules according to industry needs.

[0057] Take the hazard determination of "ethanol" as an example:

[0058] Defining text variables and numeric variables

[0059] Text variables include: chemical name "ethanol", hazard category description "flammable liquid", "toxic", etc.

[0060] Numerical variables include: flash point (13°C), boiling point (78°C), toxicity threshold (LD50=7060mg / kg), etc.

[0061] Extraction of hazardous chemicals

[0062] According to the Catalogue of Hazardous Chemicals and international standards, the following criteria are extracted:

[0063] Condition 1: If the "flash point" is < 60°C, it is classified as a "flammable liquid".

[0064] Condition 2: If "LD50" ≤ 5000 mg / kg, it is classified as "toxic".

[0065] Establish a hazardous chemicals rules database

[0066] Based on the above extracted conditions, the following rule entries are generated and stored in the rule base:

[0067] Rule 1: IF flash point < 60°C THEN flammable liquid.

[0068] Rule 2: IF LD50 ≤ 5000mg / kg THEN it is a toxic chemical.

[0069] Rule 3: IF name contains "alcohol" AND category description is "flammable" THEN flammable liquid.

[0070] Through the establishment of a rule base, the hazardous chemicals identification system can quickly make a comprehensive judgment on the hazard of ethanol based on its physical and chemical parameters and text descriptions when processing ethanol-related information.

[0071] The above step 103, i.e., "establishing a hazardous chemicals data model based on graph neural network technology according to the chemical information data," is described in detail below in conjunction with an embodiment.

[0072] The establishment of a hazardous chemicals data model based on Graph Neural Network (GNN) technology is to capture and analyze correlation and structural information by representing complex data as a graph structure when processing chemical information. Graph neural network is a deep learning method for processing graph data, which processes complex data structures through the representation of nodes and edges. Chemical data often involves the structure between molecules, chemical bonds, and the relationship between substances, which can all be represented by graphs. In a graph neural network, each node represents a component of a chemical (such as an atom or functional group), and the edge represents the chemical relationship between different chemicals, such as a covalent bond or similarity relationship.

[0073] In chemical data processing, graph structures can capture chemical structures, properties, and associations with other chemicals. For example, chemical A may be connected to chemicals B and C through molecular structure, and the chemical connection between the two is represented by edges. Through graph neural networks, these relationships can be extracted through node embedding and edge embedding, thereby analyzing the risks of chemicals.

[0074] Through graph neural networks, chemical data models can learn key safety features in complex data relationships, such as toxicity, reactivity, stability, etc. This model can more intuitively represent the relationship between chemicals, thereby helping with hazard assessment and classification.

[0075] This application is based on graph neural network technology. Building a hazardous chemical data model can be achieved through a variety of methods. As an implementable method, a chemical may consist of multiple molecular structures, and each structure can be regarded as a node in the graph. The edges in the graph represent the chemical connections between molecules, such as chemical reactions between functional groups, or changes in chemicals during the reaction process. Through graph neural networks, the model can learn the relationship between these nodes and edges and extract the data features of hazardous materials.

[0076] As another feasible method, the chemical name, chemical properties and the relationship between chemicals can be obtained from the chemical information data; the chemical name and the chemical properties are represented as nodes, and the relationship between chemicals is represented as edges; a graph convolutional network is used to extract the relationship features between nodes and edges, and a graph reasoning algorithm is used based on the relationship features to derive the characteristics of chemicals that are not clearly identified, and a knowledge graph is constructed in combination with the chemical information data; and a hazardous chemical data model is constructed based on the knowledge graph.

[0077] Specifically, the basic elements of chemicals are extracted from chemical information data, including chemical names, properties (such as molecular structure, physical properties, etc.) and the associations between different chemicals. For example, chemical A may have a chemical connection or reaction process with chemical B, C, and D, forming a complex relationship network; in the graph neural network, the chemical name and chemical properties are represented as nodes, and the associations between chemicals are represented as edges. For example, node A may represent the name of chemical A, while node B may represent its molecular structure properties. Edges represent reactions, similarities, or other associations between chemicals. In the graph neural network, a graph convolutional network is used to extract the relationship features between nodes and edges. Through convolution operations, the model can extract complex features between chemicals from the graph structure, such as structural similarity, reaction paths, etc. In order to derive the features of chemicals that are not clearly identified, a graph reasoning algorithm can be used to infer unknown information from existing data. For example, through graph reasoning, the potential reactivity or toxicity of a chemical can be predicted and this information can be integrated into the knowledge graph. Finally, through the above process, a comprehensive knowledge graph of hazardous chemicals is constructed by combining chemical information data and derived features. This map contains relevant characteristics, risk information, structural information, etc. of different chemicals, supporting the comprehensive management and analysis of hazardous chemicals.

[0078] For example, for chemicals A, B, and C, they have molecular structures, physical properties, and toxicity data, respectively. Chemical A is taken as a node, and the attributes of the node include name and chemical structure; the edge between chemicals A and B represents the chemical reaction relationship. In the graph convolutional network, the chemical relationship characteristics between nodes A and B are extracted through the edge between them. Then, through the graph reasoning algorithm, the potential toxicity characteristics of chemical C can be predicted and added to the knowledge graph to form a more comprehensive hazardous chemical data model. This model can help to carry out more accurate classification and risk assessment of hazardous materials.

[0079] The hazardous chemicals data model is obtained through pre-training. Specifically, PyTorch is an open source deep learning framework that can establish a residual network 108-layer semantic model framework based on PyTorch. The chemicals to be classified are converted into arrays through word vectors, and data preprocessing is performed through a 7×7 convolution layer and a maximum pooling layer; 108 residual blocks are processed by 3×3 convolution gradients; the final output data is mapped to hazardous chemicals, unknown, and non-hazardous chemicals through an average pooling layer and a fully connected layer. The self-developed residual network 108-layer semantic model has a high rule model matching value for hazardous chemicals, otherwise it is treated as unknown.

[0080] The following is a detailed description of step 104, namely, "using the hazardous chemicals semantic library, the hazardous chemicals rule library and the hazardous chemicals data model to build a hazardous chemicals identification system; obtaining information on chemicals to be classified, and using the hazardous chemicals identification system to classify the chemicals to be classified in the information on chemicals to be classified, and obtaining hazardous chemicals determination results" in conjunction with an embodiment.

[0081] This application uses a hazardous chemical semantic library, a hazardous chemical rule library and a hazardous chemical data model to build a hazardous chemical identification system. The system can perform risk assessment and classification of chemicals through semantic matching, rule judgment and data modeling, thereby improving the level of safety management.

[0082] As an implementable method, in the hazardous chemicals identification system of the present application, the input is the information of chemicals to be classified, which is information describing the chemicals to be classified, and may include: the name of the chemicals to be classified, CAS number, ingredient information, physical and chemical properties, hazard classification and other multi-dimensional information.

[0083] As an implementable method, the present application can pre-process the chemical information to be classified before inputting the chemical information to the hazardous chemical identification system to obtain input information, and the pre-processing includes: invalid text fragment filtering and / or synonym processing. In the chemical information, there may be some useless or meaningless text fragments, such as incomplete and redundant descriptions. These fragments will interfere with the learning process of the model and reduce the classification accuracy. Therefore, by filtering out these invalid fragments, the chemical information can be simplified and meaningful content can be retained; the chemical name or description may contain multiple similar but different expressions, such as "toluene" and "benzyl methane", which are synonyms or near-synonyms. Through synonym processing, these expressions can be unified to ensure that the model can more accurately identify the chemical category.

[0084] After obtaining the input information, the input information is input into the hazardous goods semantic library, and the hazardous goods semantic library performs semantic matching on the input information to obtain a semantic matching result; combining the semantic matching result and the hazardous goods determination condition generated by the hazardous goods rule library, the hazardous chemicals data model is used to classify the chemicals to be classified to obtain a hazardous chemicals determination result.

[0085] Specifically, the chemicals to be classified are first input into the hazardous chemicals semantic library and the hazardous chemicals rule library respectively, and the hazardous chemicals semantic library performs semantic matching to generate matching results; at the same time, the hazardous chemicals rule library generates corresponding rules for the chemicals to be classified; the information of the chemicals to be classified, the matching results and the corresponding rules are input into the hazardous chemicals data model, and the hazardous chemicals data model performs reasoning and outputs the reasoning results, and obtains the judgment results based on the reasoning results. Among them, the hazardous chemicals data model can directly output the judgment results, or it can output the probability that the chemicals to be classified are hazardous chemicals. The judgment results of hazardous chemicals are hazardous chemicals, unknown, and non-hazardous chemicals.

[0086] When the hazardous chemicals data model outputs the probability that the chemical to be classified is a hazardous chemical, the hazardous chemical judgment threshold M1 and the non-hazardous chemical threshold M2 can be set. After the hazardous chemicals data model performs reasoning, it outputs the probability that the chemical to be identified is a hazardous chemical. If the probability is less than M2, the chemical is judged to be a non-hazardous chemical; if the probability is greater than M1, the chemical is judged to be a hazardous chemical. The specific values ​​of M1 and M2 can be pre-set or modified during the model training process.

[0087] Preferably, the present application can perform pre-classification according to the semantic matching results of the hazardous chemicals semantic library. Specifically: when combining the semantic matching results and the hazardous goods judgment conditions generated by the hazardous goods rule library, using the hazardous chemicals data model to generate the hazardous chemicals judgment results, it includes: if the semantic matching result is a complete mismatch, marking the chemical information as non-hazardous chemical; if the semantic matching result is a match or a suspected match, recording the matching keywords of the hazardous chemicals semantic library; inputting the to-be-classified chemical information, matching keywords and hazardous goods judgment conditions whose semantic matching results are a match or a suspected match into the hazardous chemicals data model, and obtaining the hazardous chemicals judgment results generated by the hazardous chemicals data model.

[0088] Among them, matching keywords are generated by the hazardous chemicals semantic library, which can be the full name or abbreviation of the chemical, the specific components or ingredients of the chemical, the use or industry of the chemical, words related to hazardous properties, such as "flammable", "corrosive", "toxic", etc., and descriptions of the state, concentration, and ratio of the chemical, such as "concentration higher than 5%", "mixture", "solution", etc.

[0089] When constructing a hazardous chemical identification system, this application requires training of the system. The training includes: obtaining training data including multiple training samples, wherein the training samples include chemical information samples and their corresponding hazardous chemical determination result samples; specifically, the training samples are multiple groups of chemical information samples and their corresponding hazardous chemical determination result samples, and the training samples can be obtained through manual annotation, or from historical records or certain professional databases.

[0090] The chemical information sample is input into the hazardous chemical identification system, and the hazardous chemical semantic library outputs the semantic matching result according to the chemical information sample; the hazardous chemical rule library outputs the hazardous chemical judgment condition according to the chemical information sample; the matching result, hazardous chemical judgment condition and hazardous chemical information sample are input into the hazardous chemical data model, and the hazardous chemical data model outputs the judgment result. The training objectives include: minimizing the difference between the generated judgment result and the hazardous chemical judgment result sample.

[0091] Preferably, different weights are assigned to the semantic matching results, the hazardous materials determination conditions and the chemical information samples, and the hazardous chemicals data model performs weighted fusion on the inputs according to the weights and performs inference output to obtain output results; and the weight values ​​are adjusted according to the training objectives.

[0092] In order to minimize the difference between the generated hazardous chemicals determination results and the sample data, a loss function can be defined. The loss function is used to measure the error between the model prediction value and the true value and guide the optimization of model parameters. Select a suitable optimization algorithm, calculate the gradient through back propagation, and update the model parameters to minimize the loss function value. Dynamically adjust the weights of different modules in the model according to the model performance during training (such as the accuracy on the validation set).

[0093] Each time a chemical is processed, the system extracts semantic matching results from the hazardous materials semantic library based on the chemical information. These results may be complete matches, partial matches, or mismatches. On this basis, different weights are assigned to these semantic matching results, judgment conditions, and samples to balance the accuracy of the model output. When processing inputs, the system combines the weight of each input with the corresponding semantic features through weighted fusion, performs reasoning operations, and thus obtains the judgment results of hazardous chemicals. Based on the difference between the output results and the actual judgment result samples, the system adjusts the values ​​of different weights to optimize the model's predictive ability and reduce errors.

[0094] For example, for "sulfuric acid", due to high-weight input features such as "strong acidity", the system tends to output "hazardous chemicals". If the system outputs a non-hazardous chemical when processing a "sulfuric acid" sample, and there is a difference with the actual judgment result, the system will adjust the weight related to "strong acidity" in the model to increase its influence, thereby optimizing the system's prediction ability.

[0095] Furthermore, the training also includes: using reinforcement learning to optimize the recognition strategy and dynamically adjusting the parameters of the hazardous chemicals data model according to the training objectives.

[0096] The reinforcement learning strategy optimization includes one or more of the following:

[0097] The chemical identification task is transformed into a decision-making process in reinforcement learning. In each decision-making process, the system generates various possible identification paths based on the input chemical information and selects the best path. Through iterative optimization, the system can more accurately identify hazardous chemicals. For example, for the chemical "sulfuric acid", the system may consider different paths, such as analyzing its degree of hazard based on chemical properties, toxicity or reactivity. These paths are continuously adjusted through reinforcement learning, and finally the path that best matches the actual hazard is selected for judgment.

[0098] Design a multi-level reward mechanism to guide the hazardous chemicals data model to optimize the recognition strategy. The multi-level reward mechanism helps the system to perform reinforcement learning at different stages to guide the model to optimize the recognition strategy. For example, the initial reward may be based on basic features, such as chemical reactivity. As the accuracy of recognition increases, the reward mechanism will give higher rewards to optimize the deep learning ability of the model. For example, when the system processes "sulfuric acid", the initial stage may only give low rewards based on basic features, such as acid strength; as the inference accuracy increases, the system's rewards gradually increase to encourage the model to optimize more complex features, such as toxic reactions, environmental impacts, etc.

[0099] Among them, the reward strategy can be set according to the evaluation indicators. The evaluation indicators for the hazardous chemicals identification system can be divided into two categories:

[0100] Recognition rate = number of recognitions / number of inputs

[0101] Accuracy = correct number of samples / total number of samples

[0102] The higher the recognition rate, the less data is left for manual judgment, and the more efficient the system processing is. The accuracy rate is the reliability of the system. The higher the accuracy rate, the higher the reliability of the system. The improvement of recognition rate depends on the number of scenarios of model training sample data, and the improvement of accuracy rate depends on the accuracy of sample data and low conflict rate. The accuracy rate can also be improved by adjusting the hazardous chemical judgment threshold M1 and the non-hazardous chemical threshold M2, but as the accuracy rate increases, the recognition rate will also decrease.

[0103] Dynamically adjust the identification path, and give priority to high-weighted reasoning paths when the input data is incomplete or ambiguous. When processing incomplete or ambiguous input data, the system dynamically adjusts the identification path and gives priority to high-weighted reasoning paths, thereby improving the accuracy of the prediction results. This allows for more robust determination of hazardous chemicals under uncertain circumstances. For example, for the input "a certain unknown acidic liquid", due to incomplete data, the system will give priority to high-weighted paths of "acid strength" and "corrosive reaction" to improve the reliability of hazardous chemical identification and reduce erroneous determinations.

[0104] The above method provided in the embodiment of the present application can be applied to a variety of application scenarios, including but not limited to: in the chemical, pharmaceutical and other industrial fields, the management of hazardous chemicals is crucial. The present application can help enterprises automatically identify and classify chemicals to ensure that the storage, transportation and use of hazardous chemicals meet safety standards. For example, in a production workshop, the system can identify in real time whether there are potential hazardous chemicals and provide timely safety warnings; after a disaster or accident, it is crucial to quickly identify hazardous chemicals. The present application can quickly process on-site chemical information, provide accurate hazardous chemical determination, and help emergency management departments formulate safety measures. In addition, the methods and devices provided in the present application can also be applied to multiple fields such as environmental monitoring and risk assessment, education and training, and chemical supply chain management.

[0105] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0106] According to an embodiment of another aspect, a hazardous chemical identification device is provided. Figure 2 A schematic block diagram of a hazardous chemical identification device according to an embodiment is shown, Figure 2 As shown, the device 200 includes:

[0107] The semantic library construction unit 201 is configured to obtain chemical information data, and construct a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm of deep learning;

[0108] A rule base building unit 202 is configured to define text variables and numerical variables according to the naming rules and properties of chemicals, extract hazardous chemical determination conditions from the chemical information data, and establish a hazardous chemical rule base according to the text variables, the numerical variables and the hazardous chemical determination conditions;

[0109] The data model building unit 203 is configured to build a hazardous chemical data model based on the chemical information data and graph neural network technology;

[0110] The determination result acquisition unit 204 is configured to construct a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model; obtain information on chemicals to be classified, and classify the chemicals to be classified in the information on chemicals to be classified using the hazardous chemical identification system to obtain a hazardous chemical determination result.

[0111] As an implementable manner, the semantic library construction unit 201 can be configured to construct a hazardous chemicals semantic library based on the chemical information data using a multimodal data fusion algorithm of deep learning as follows: using a recurrent neural network to process text information in the chemical information data to obtain data features of different modalities; integrating the data features of different modalities in the fusion layer of the recurrent neural network, and inputting the integration results into the fully connected layer for data classification to obtain a hazardous chemicals semantic library; wherein, an attention mechanism is introduced in the data classification process to automatically select and strengthen relevant data feature information.

[0112] As an implementable manner, the data model building unit 203 can be configured to establish a hazardous chemical data model based on the chemical information data and based on the graph neural network technology as follows: obtain the chemical name, chemical properties and the relationship between chemicals from the chemical information data; represent the chemical name and the chemical properties as nodes, and the relationship between chemicals as edges; use a graph convolutional network to extract the relationship features between nodes and edges, use a graph reasoning algorithm based on the relationship features to derive the characteristics of chemicals that are not clearly identified, and build a knowledge graph in combination with the chemical information data; and build a hazardous chemical data model based on the knowledge graph.

[0113] As an implementable manner, the determination result acquisition unit 204 can be configured to: pre-process the chemical information to be classified to obtain input information when using the hazardous chemical identification system to classify the chemicals to be classified in the chemical information to be classified to obtain the hazardous chemical determination result, wherein the pre-processing includes: invalid text fragment filtering and / or synonym processing; inputting the input information into the hazardous chemical semantic library, and the hazardous chemical semantic library performs semantic matching on the input information to obtain a semantic matching result; combining the semantic matching result and the hazardous chemical determination condition generated by the hazardous chemical rule library, and using the hazardous chemical data model to classify the chemicals to be classified to obtain a hazardous chemical determination result.

[0114] As an implementable manner, the determination result acquisition unit 204 classifies the chemicals to be classified using the hazardous chemicals data model in combination with the semantic matching result and the hazardous goods determination condition generated by the hazardous goods rule library, and when obtaining the hazardous chemicals determination result, it can be configured as follows: if the semantic matching result is a complete mismatch, the chemicals to be classified are marked as non-hazardous chemicals; if the semantic matching result is a match or a suspected match, the matching keywords of the hazardous chemicals semantic library are recorded; the information of the chemicals to be classified for which the semantic matching result is a match or a suspected match, the matching keywords and the hazardous goods determination condition are input into the hazardous chemicals data model, and the hazardous chemicals determination result is obtained according to the output of the hazardous chemicals data model.

[0115] As an implementable manner, the judgment result acquisition unit 204 can be configured to: construct a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model, and train the hazardous chemical identification system, wherein the training includes: obtaining training data including multiple training samples, wherein the training samples include chemical information samples and their corresponding hazardous chemical judgment result samples; the hazardous chemical semantic library outputs a semantic matching result according to the chemical information sample; the hazardous chemical rule library outputs a hazardous chemical judgment condition according to the chemical information sample; different weights are assigned to the semantic matching result, the hazardous chemical judgment condition and the chemical information sample, and the hazardous chemical data model performs model inference after weighted fusion of the semantic matching result, the hazardous chemical judgment condition and the chemical information sample according to the weight, and outputs a judgment result; wherein the training objectives include: minimizing the difference between the generated judgment result and the hazardous chemical judgment result sample; and adjusting the weight value according to the training objective.

[0116] As an implementable method, the determination result acquisition unit 204 can be configured during training to: optimize the recognition strategy using reinforcement learning, and dynamically adjust the parameters of the hazardous chemicals data model according to the training objectives; wherein the reinforcement learning strategy optimization includes one or more of the following: converting the chemical identification task into a decision-making process in reinforcement learning; designing a multi-level reward mechanism to guide the hazardous chemicals data model to optimize the recognition strategy; dynamically adjust the recognition path, and give priority to the high-weight reasoning path when the input data is incomplete or ambiguous.

[0117] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0118] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0119] In addition, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of any one of the methods in the aforementioned method embodiments are implemented.

[0120] And an electronic device, comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, and the program instructions, when read and executed by the one or more processors, execute the steps of any of the methods described in the aforementioned method embodiments.

[0121] The present application also provides a computer program product, including a computer program, which implements the steps of any one of the methods in the aforementioned method embodiments when executed by a processor.

[0122] in, Figure 3 The architecture of the electronic device is shown as an example, which may include a processor 310, a video display adapter 311, a disk drive 312, an input / output interface 313, a network interface 314, and a memory 320. The processor 310, the video display adapter 311, the disk drive 312, the input / output interface 313, the network interface 314, and the memory 320 may be communicatively connected via a communication bus 330.

[0123] The processor 310 may be implemented by a general-purpose CPU, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in this application.

[0124] The memory 320 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 320 can store an operating system 321 for controlling the operation of the electronic device 300, and a basic input and output system (BIOS) 322 for controlling the low-level operation of the electronic device 300. In addition, a web browser 323, a data storage management system 324, and a hazardous chemical identification device 325, etc. can also be stored. The above-mentioned hazardous chemical identification device 325 can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present application. In short, when the technical solution provided in the present application is implemented by software or firmware, the relevant program code is stored in the memory 320 and is called and executed by the processor 310.

[0125] The input / output interface 313 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0126] The network interface 314 is used to connect to a communication module (not shown) to achieve communication interaction between the device and other devices. The communication module can achieve communication through a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WIFI, Bluetooth, etc.).

[0127] The bus 330 comprises a pathway for transmitting information between the various components of the device (eg, the processor 310 , the video display adapter 311 , the disk drive 312 , the input / output interface 313 , the network interface 314 , and the memory 320 ).

[0128] It should be noted that, although the above device only shows a processor 310, a video display adapter 311, a disk drive 312, an input / output interface 313, a network interface 314, a memory 320, a bus 330, etc., in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include components necessary for implementing the solution of the present application, and does not necessarily include all the components shown in the figure.

[0129] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a computer program product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0130] The technical solution provided by the present application is described in detail above. The principle and implementation method of the present application are described in detail using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A method for identifying hazardous chemicals, characterized in that: The method includes: Acquire chemical information data, and build a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm based on deep learning; According to the naming rules and properties of chemicals, text variables and numerical variables are defined, hazardous chemical determination conditions are extracted from the chemical information data, and a hazardous chemical rule library is established according to the text variables, the numerical variables and the hazardous chemical determination conditions; According to the chemical information data, a hazardous chemical data model is established based on graph neural network technology; A hazardous chemical identification system is constructed by using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model, and the hazardous chemical identification system is trained, wherein the training comprises: obtaining training data including a plurality of training samples, wherein the training samples include chemical information samples and their corresponding hazardous chemical determination result samples; The hazardous chemicals semantic library outputs a semantic matching result according to the chemical information sample; the hazardous chemicals rule library outputs a hazardous chemicals determination condition according to the chemical information sample; Different weights are assigned to the semantic matching results, the hazardous goods determination conditions and the chemical information samples. The hazardous chemicals data model performs model reasoning after weighted fusion of the semantic matching results, the hazardous goods determination conditions and the chemical information samples according to the weights, and outputs the determination results; wherein the training objectives include: minimizing the difference between the generated determination results and the hazardous chemicals determination result samples; adjusting the weights according to the training objectives; Acquire information of chemicals to be classified, preprocess the information of chemicals to be classified, and obtain input information, wherein the preprocessing includes: filtering invalid text fragments and / or synonym processing; inputting the input information into the hazardous chemicals semantic library, wherein the hazardous chemicals semantic library performs semantic matching on the input information to obtain a semantic matching result; combining the semantic matching result with the hazardous chemicals judgment condition generated by the hazardous chemicals rule library, classifying the chemicals to be classified using the hazardous chemicals data model, and obtaining a hazardous chemicals judgment result.

2. The method according to claim 1, characterized in that The step of constructing a hazardous chemicals semantic library based on the chemical information data and utilizing a multimodal data fusion algorithm of deep learning includes: Using a recurrent neural network to process text information in the chemical information data to obtain data features of different modalities; The different modal data features are integrated in the fusion layer of the recurrent neural network, and the integration results are input into the fully connected layer for data classification to obtain a hazardous chemicals semantic library; wherein, an attention mechanism is introduced in the data classification process to automatically select and strengthen relevant data feature information.

3. According to the method of claim 1, establishing a hazardous chemicals data model based on the chemical information data and graph neural network technology comprises: Obtaining chemical names, chemical properties, and associations between chemicals from the chemical information data; The chemical name and the chemical attribute are represented as nodes, and the association relationship between chemicals is represented as edges; A graph convolutional network is used to extract the relationship features between nodes and edges, and a graph inference algorithm is used based on the relationship features to deduce the features of chemicals that are not clearly identified, and a knowledge graph is constructed in combination with the chemical information data; A hazardous chemicals data model is constructed based on the knowledge graph.

4. According to the method of claim 1, the combining the semantic matching result and the hazardous chemical determination condition generated by the hazardous chemical rule library, and using the hazardous chemical data model to classify the chemicals to be classified to obtain the hazardous chemical determination result comprises: If the semantic matching result is a complete mismatch, the chemical to be classified is marked as a non-hazardous chemical; If the semantic matching result is a match or a suspected match, the matching keywords of the hazardous chemicals semantic library are recorded; The to-be-classified chemical information, the matching keywords and the hazardous chemical determination conditions for which the semantic matching results are matched or suspected matched, are input into the hazardous chemical data model, and the hazardous chemical determination result is obtained according to the output of the hazardous chemical data model.

5. The method according to claim 1, wherein the training further comprises: Utilize reinforcement learning to optimize the identification strategy, and dynamically adjust the parameters of the hazardous chemicals data model according to the training objectives; wherein the reinforcement learning to optimize the identification strategy includes one or more of the following: Convert the chemical identification task into a decision-making process in reinforcement learning; Design a multi-level reward mechanism to guide the hazardous chemicals data model to optimize the identification strategy; Dynamically adjust the recognition path and give priority to the reasoning path with high weight when the input data is incomplete or ambiguous.

6. A hazardous chemical identification device, characterized in that: The device comprises: A semantic library construction unit is configured to obtain chemical information data, and construct a hazardous chemical semantic library based on the chemical information data using a multimodal data fusion algorithm of deep learning; A rule base building unit is configured to define text variables and numerical variables according to the naming rules and properties of chemicals, extract hazardous chemical determination conditions from the chemical information data, and establish a hazardous chemical rule base according to the text variables, the numerical variables and the hazardous chemical determination conditions; A data model building unit is configured to build a hazardous chemical data model based on the chemical information data and graph neural network technology; The determination result acquisition unit constructs a hazardous chemical identification system using the hazardous chemical semantic library, the hazardous chemical rule library and the hazardous chemical data model, and trains the hazardous chemical identification system, wherein the training comprises: acquiring training data comprising a plurality of training samples, wherein the training samples comprise chemical information samples and their corresponding hazardous chemical determination result samples; the hazardous chemical semantic library outputs a semantic matching result according to the chemical information sample; the hazardous chemical rule library outputs a hazardous chemical determination condition according to the chemical information sample; different weights are assigned to the semantic matching result, the hazardous chemical determination condition and the chemical information sample, and the hazardous chemical data model assigns different weights to the semantic matching result, the hazardous chemical determination condition and the chemical information sample according to the weights. The method performs model reasoning after weighted fusion and outputs a judgment result; wherein the training objectives include: minimizing the difference between the generated judgment result and the hazardous chemical judgment result sample; adjusting the weight according to the training objectives; obtaining information of chemicals to be classified, preprocessing the information of chemicals to be classified, and obtaining input information, wherein the preprocessing includes: invalid text fragment filtering and / or synonym processing; inputting the input information into the hazardous chemical semantic library, and the hazardous chemical semantic library performs semantic matching on the input information to obtain a semantic matching result; combining the semantic matching result with the hazardous chemical judgment condition generated by the hazardous chemical rule library, and using the hazardous chemical data model to classify the chemicals to be classified to obtain a hazardous chemical judgment result.

7. An electronic device, characterized in that: include: one or more processors; And a memory associated with the one or more processors, the memory is used to store program instructions, and when the program instructions are read and executed by the one or more processors, the steps of the method described in any one of claims 1 to 5 are executed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images