Emission factor matching method and device, intelligent agent and storage medium

Abstract

The application discloses an emission factor matching method and device, an intelligent agent and a storage medium, and belongs to the technical field of data processing. According to the application, the description information of a target product input by a user is directly used to generate a target material name corresponding to the target product, a target material is searched in a factor material library according to the target material name, and a target emission factor corresponding to the target material is mapped in the factor material library according to a mapping relationship between emission factors and materials. Thus, the complex emission factor matching problem is converted into a material semantic matching problem by first matching the target material according to the factor material library and the target material name of the target product, and then returning to the factor material library according to the target material and the mapping relationship to map the target emission factor. The emission factor matching accuracy can be improved without going through multiple uncertain processes such as translation and retrieval, the semantic deviation caused by language difference and the influence on the matching result can be reduced, and the emission factor matching accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to emission factor matching methods, apparatus, intelligent agents, and storage media. Background Technology

[0002] In carbon footprint calculation systems, the selection of emission factors typically relies on text matching between the emission source name and the factor name in the emission factor database. However, because the emission factor database is primarily stored in foreign languages ​​and the factor names are highly specialized and standardized, issues such as translation bias, difficulties in cross-language matching, and inconsistencies in material semantics arise, leading to inaccurate factor matching. Summary of the Invention

[0003] This application provides an emission factor matching method, apparatus, intelligent agent, and storage medium to improve the accuracy of emission factor location matching.

[0004] Firstly, an emission factor matching method is provided, the method comprising: Generate the target material name corresponding to the target product based on the description information of the target product entered by the user; Search for target materials whose names match the target material name in a pre-built factor material library; the factor material library includes the mapping relationship between emission factors and their respective materials; Based on the mapping relationship, the emission factor corresponding to the target material is determined in the factor material library and used as the target emission factor for matching the target product.

[0005] In some embodiments, the method for constructing a factor material library includes: Extract the material corresponding to each emission factor from the attribute information of the emission factors contained in the preset factor library to form a primary mapping relationship between emission factors and materials; Based on the initial mapping relationship, emission factors belonging to the same material are mapped to the same material to generate an updated mapping relationship; Based on the updated mapping relationship, a factor material library is formed.

[0006] In some embodiments, the target material name includes at least one of a surface semantic name, a standard substance name, and a deep standard name; the preset priority of the surface semantic name, the standard substance name, and the deep standard name decreases in descending order.

[0007] In some embodiments, searching for a target material whose name matches the target material name in a pre-built factor material library includes: Using an embedding model, the surface semantic name is matched in the factor material library to obtain the first matching result; in response to the first matching result satisfying the preset matching conditions, the target material is determined based on the first matching result. In response to the first matching result not meeting the preset matching conditions, the embedding model is used to match the standard substance name in the factor material library to obtain the second matching result; In response to the second matching result satisfying the preset matching conditions, the target material is determined based on the second matching result; In response to the second matching result not meeting the preset matching conditions, the deep standard name is matched in the factor material library using the embedding model to obtain the third matching result; The target material is determined based on the first matching result, the second matching result, and the third matching result.

[0008] In some embodiments, determining the target material based on a first matching result, a second matching result, and a third matching result includes: Using a reordering model, the first matching result, the second matching result, and the third matching result are reordered respectively. Select the first preset number of matching results from the first reordered matching results as the first candidate set; Select a second preset number of matching results from the reordered second matching results as the second candidate set; Select the third preset number of matching results from the reordered third matching results as the third candidate set; Using a large language model, the target material is determined based on the first candidate set, the second candidate set, and the third candidate set.

[0009] In some embodiments, the first preset quantity is less than the second preset quantity, and the second preset quantity is less than the third preset quantity.

[0010] Secondly, an emission factor matching device is also provided, the device comprising: The generation module is used to generate the target material name corresponding to the target product based on the description information of the target product input by the user. The search module is used to search for target materials that match the target material name in a pre-built factor material library; wherein, the factor material library includes the mapping relationship between emission factors and their respective materials; The determination module is used to determine the emission factor corresponding to the target material in the factor material library based on the mapping relationship, and use it as the target emission factor for matching the target product.

[0011] Thirdly, an emission factor matching agent is also provided for performing the method described in the first aspect.

[0012] Fourthly, a computer-readable storage medium is also provided, on which a computer program is stored, the computer program being loaded by a processor to perform the steps of the method described in the first aspect.

[0013] Beneficial Effects: This application provides an emission factor matching method, apparatus, intelligent agent, and storage medium. The emission factor matching method provided in this application directly generates the target material name corresponding to the target product by using the description information of the target product input by the user. Based on the target material name, it searches for a matching target material in the factor material library. Then, based on the mapping relationship between emission factors and materials, it maps the target emission factor corresponding to the target material in the factor material library. Thus, by first matching the target material based on the factor material library and the target material name of the target product, and then returning to the factor material library to map the target emission factor based on the target material and the mapping relationship, the complex emission factor matching problem is transformed into a material semantic matching problem. This eliminates the need for multiple uncertain processes such as translation and retrieval, reduces semantic deviation caused by language differences and their impact on the matching results, and helps improve the accuracy of emission factor matching. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of an emission factor matching method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a factor material library provided in an embodiment of this application; Figure 3 This is a schematic diagram of another factor material library provided in an embodiment of this application; Figure 4 This is the overall process architecture of the emission factor matching method provided in the embodiments of this application; Figure 5 This is a schematic diagram of the principle structure of an emission factor matching device provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0018] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.

[0019] The use of "applies to" or "configured to" in this application implies open and inclusive language, which does not exclude the applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more of the stated conditions or values ​​may in practice be based on additional conditions or values ​​beyond those stated.

[0020] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0021] The applicant's research revealed that existing carbon footprint calculation systems typically rely on textual matching between emission source names and factor names in emission factor databases when selecting emission factors. However, emission factor databases (e.g., Ecoinvent) are primarily stored in English, and their factor names are highly specialized and standardized, leading to the following shortcomings in practical applications: First, there is a problem with the complexity of cross-language matching, which relies on human experience. For example, when emission sources are entered in Chinese, it is usually necessary to first translate the emission source name into English, then manually or semi-automatically analyze its composition, and finally perform a fuzzy search in the emission factor database based on the translation results. This process relies on human understanding and multiple attempts, making it complex and inefficient.

[0022] Second, there are semantic discrepancies between the translated results and the standard descriptions in the factor database. For example, because the material names in the emission factor database often use technical terms, synonyms, or industry-specific naming conventions, the translated English descriptions of emission sources often cannot directly correspond to the standard material names in the factor database. This can easily lead to situations where the semantics are similar but the text is inconsistent, resulting in search failures or the need to repeatedly adjust keywords for searching.

[0023] Third, inconsistencies between material abbreviations and full names lead to matching difficulties. For example, in actual operations, some emission sources are only represented by abbreviations or acronyms (e.g., polyvinyl chloride (PVC)). Further analysis of their full names or standard material names is required before factor matching can be performed. However, relevant technologies typically lack the ability to automatically identify the relationship between abbreviations and standard material names, increasing the cost of manual intervention.

[0024] Fourth, the retrieval process is highly repetitive and inefficient overall. For example, due to translation uncertainties, differences in material descriptions, and abbreviation issues, related technologies often require multiple translations, retrievals, and manual comparisons, resulting in a time-consuming and unstable process for selecting emission factors, making it difficult to support large-scale or automated carbon footprint modeling needs.

[0025] In view of this, embodiments of this application provide an emission factor matching method, apparatus, intelligent agent, and storage medium. This method directly generates the target material name corresponding to the target product from the user-input description information. It then searches for a matching target material in a factor material library based on the target material name, and finally maps the target emission factor corresponding to the target material in the factor material library based on the mapping relationship between emission factors and materials. Thus, by first matching the target material based on the factor material library and the target product's target material name, and then mapping the target material back to the factor material library based on the mapping relationship, the complex emission factor matching problem is transformed into a material semantic matching problem. This eliminates the need for multiple uncertain processes such as translation followed by retrieval, reducing semantic bias caused by language differences and their impact on the matching results, thereby improving the accuracy of emission factor matching.

[0026] Figure 1 This is a flowchart of an emission factor matching method provided in an embodiment of this application. (See attached document.) Figure 1The method includes the following steps S100 to S300.

[0027] Step S100: Generate the target material name corresponding to the target product based on the description information of the target product entered by the user.

[0028] The target product refers to the physical object, material, activity process, intermediate, etc. that needs to be matched with the emission factor, such as basic chemicals (e.g., nylon, polyethylene), materials containing abbreviations (e.g., PVC), specific process materials, emission activities, waste, etc.

[0029] The description information of the target product is semantic description information. For example, the product name, material, emission source, and process stage entered by the user are semantic description information. The target material name corresponding to the target product is also semantic description information, specifically information describing the material semantically. Therefore, by directly generating the target material name corresponding to the target product from the user-input description information, it is possible to directly generate the target material name that describes the material semantically from the product's semantic description information. This facilitates subsequent material matching based on semantic information, thus eliminating the need for multiple uncertain processes such as translation before retrieval or repeated manual translation and keyword adjustments. It can reduce semantic bias caused by language differences and their impact on matching results, thereby improving the accuracy of subsequent location matching.

[0030] Step S200: Search for a target material whose name matches the target material name in the pre-built factor material library.

[0031] The factor material library includes the mapping relationship between emission factors and their corresponding materials. The factor material library contains multiple materials and multiple emission factors, each emission factor having a corresponding material, thus establishing a mapping relationship between emission factors and their respective materials.

[0032] By generating the target material name of the target product and searching for the target material name that matches the target material name in the factor material library, the matching target material is found by using the target material name with the semantic description of the material and the factor material library. This makes the matching process more focused on the essential material properties of the emission factors, which helps to improve the consistency and accuracy of the matching. Moreover, the material matching process does not need to go through multiple uncertain processes such as translation before retrieval or repeated manual translation and keyword adjustment. This can reduce the semantic deviation caused by language differences and its impact on the matching results, which is conducive to improving the accuracy of subsequent location matching.

[0033] Step S300: Based on the mapping relationship, determine the emission factor corresponding to the target material in the factor material library as the target emission factor for matching the target product.

[0034] Since there is a certain mapping relationship between emission factors and their corresponding materials, after matching the target material with the target material name based on the factor material library, the emission factor corresponding to the target material can be mapped back to the factor material library based on the target material and the mapping relationship. This can be used as the target emission factor to match the target product, thus transforming the complex emission factor matching problem into material semantic matching, thereby matching accurate emission factors and improving the accuracy of emission factor matching.

[0035] In the technical solution of this application embodiment, the implementation process of the emission factor matching method is as follows: (Refer to...) Figure 1 First, the system generates the target material name corresponding to the target product based on the description information input by the user. Then, it searches a pre-built factor material library for a target material that matches the target material name; this library includes a mapping relationship between emission factors and their corresponding materials. Finally, based on the mapping relationship, the emission factor corresponding to the target material is determined in the factor material library as the target emission factor for matching the target product. Thus, by first matching the target material based on the factor material library and the target product's target material name, and then mapping the target material back to the factor material library to find the target emission factor, the complex emission factor matching problem is transformed into a material semantic matching problem. This eliminates the need for multiple uncertain processes such as translation followed by retrieval, reducing semantic bias caused by language differences and their impact on matching results, thereby improving the accuracy of emission factor matching.

[0036] In some embodiments, step S100 above may include the following steps: Based on the description information, the name of the target material is extracted using a large language model.

[0037] The descriptive information includes at least one of the following: the name of the target product, the emission source, the material, and the process stage. For example, if the target product is a chemical, its name is nylon, its emission source is nylon 6-6, its material is polyamide, and its process stages include polymerization, injection molding, extrusion molding, and glass fiber reinforcement.

[0038] An emission source refers to a specific activity, material, or process that generates greenhouse gas emissions throughout its life cycle, including but not limited to energy consumption, raw material use, transportation activities, chemical reactions, and waste disposal. In this application's embodiments, emission sources are analyzed, classified, and intelligently matched with emission factors.

[0039] Large Language Models (LLMs) refer to artificial intelligence models with natural language understanding and reasoning capabilities, including but not limited to DeepSeek and Qwen. In the technical solutions of this application, by utilizing a large language model, tasks such as semantic understanding of emission sources, process inference, factor matching, and life cycle stage identification of the target product can be achieved based on the description information of the target product, and the target material name of the target product can be extracted from it. For example, assuming the user inputs the description information of the target product including the product name "nylon," the emission source being nylon 6-6, and the material being polyamide, the large language model can use this information to perform semantic understanding of nylon emission sources, process inference, factor matching, and life cycle stage identification to extract the target material name of nylon, such as Nylon, Nylon 6-6, or Polyamide. Furthermore, if the user only inputs the product name, such as "nylon," the large language model can also search for nylon-related information based solely on the product name and analyze its product composition, emission source semantic understanding, process inference, factor matching, and life cycle stage identification information, and extract the target material name of nylon from it. Therefore, by utilizing a large language model, the target material name of the target product can be extracted based on the description information of the target product input by the user. This eliminates the need for manual translation and reliance on experience-based judgment, reduces the impact of language differences on subsequent matching results, and facilitates accurate matching of the target material based on the target material name, thereby improving the accuracy of emission factor matching.

[0040] In some embodiments, the method for constructing a factor material library may include the following steps S410 to S430.

[0041] Step S410: Extract the material corresponding to each emission factor from the attribute information of the emission factors contained in the preset factor library to form a primary mapping relationship between emission factors and materials. The preset factor library is sourced from the Ecoinvent database. Ecoinvent is a globally widely used lifecycle emission factor database, containing environmental emission data across various categories, including raw material production, energy use, transportation activities, and waste disposal. In this embodiment, the Ecoinvent database is used as the emission factor source (i.e., as the preset factor library), and the most suitable emission factors are automatically selected through intelligent factor matching.

[0042] The preset factor library includes multiple emission factors and their corresponding attribute information, including factor name, factor entry, process name, country, and affiliated unit.

[0043] Each emission factor has a corresponding material, thus a primary mapping relationship can be established between them. This primary mapping relationship includes the material and attribute information of the emission factor. By establishing this primary mapping relationship, it becomes easier to accurately map the emission factor to the target material, thereby improving the accuracy of emission factor matching.

[0044] The core material of a corresponding emission factor can be extracted by its attribute information. For example, the core material can be extracted by the factor name of the emission factor.

[0045] For example, assuming the factor name is "printed wiring board production, surface mounted, unspecified, Pb free", the corresponding core material can be extracted as "printed wiring board". Similarly, assuming the factor name is "polyurethane production, flexible foam, MDI-based", the corresponding core material can be extracted as "polyurethane".

[0046] Step S420: Based on the primary mapping relationship, the emission factors belonging to the same material are mapped to the same material to generate an updated mapping relationship.

[0047] Since each emission factor has its corresponding attribute information, a corresponding material can be extracted for each emission factor. Among multiple emission factors, there may be emission factors with the same material. Therefore, by deduplicating and merging emission factors from different countries and with different processes but the same material, the core material, country, and unit of the emission factor are retained as structured vectors and stored in the factor material library (if the same material has multiple countries, they are separated by commas). This maps emission factors belonging to the same material to the same material. Finally, multiple (e.g., about 2000+) core materials (there are a large number of emission factors with the same material but different processes / regions) are extracted from tens of thousands of emission factors and stored in the factor material library. By deduplicating and merging emission factors belonging to the same material based on the primary mapping relationship, the efficiency and accuracy of matching target materials and mapping target emission factors in the factor material library are improved, and the library construction cost is reduced.

[0048] Step S430: Based on the updated mapping relationship, form a factor material library.

[0049] The factor material library refers to a collection of materials constructed from an emission factor database to characterize the core material properties of emission factors. This library identifies and retains representative core material information by parsing and extracting emission factor names and descriptions, supporting semantic matching at the material level. In this embodiment, the factor material library originates from the Ecoinvent emission factor database and is formed by extracting and standardizing core materials from factor names, used for semantic matching and location with user-input material descriptions.

[0050] Figure 2 This is a schematic diagram of the structure of a factor material library provided in an embodiment of this application. Figure 3 This is a schematic diagram of another factor material library structure provided in an embodiment of this application. Exemplarily, the structure of the factor material library includes a core material name (material_name), aliases / synonyms (unit), country (country), whether it is waste (sectors), vector (embedding), identifier (id), and field type (Ic_type), etc. For example, Figure 2 The structure of a factor material library is listed, including id, material, unit, country, and sectors. Figure 3 Another factor material library structure is listed, including country, sectors, Ic_type, and embedding. The core material names include the English main name, Chinese main name, translation, abbreviation, and commonly used industry names.

[0051] In the technical solution of this application embodiment, the specific implementation process of constructing the factor material library is as follows: First, extract the material corresponding to each emission factor from the attribute information of the emission factors contained in the preset factor library to form a primary mapping relationship between emission factors and materials. Then, based on the primary mapping relationship, map emission factors belonging to the same material to the same material to generate an updated mapping relationship. Finally, based on the updated mapping relationship, form the factor material library. Thus, by parsing the emission factor names from the emission factor database, extracting and standardizing the core material information that can characterize the material properties of the emission factors from the emission factor names, and constructing the factor material library, the matching object is transformed from a complete factor entry into a material semantic unit, making the matching process more focused on the essential material properties of the emission factors, thereby improving the accuracy and consistency of the matching. Furthermore, during the construction of the factor material library, the core material corresponding to the emission factor is extracted through the attribute information of the emission factor to establish a primary mapping relationship between the emission factor and the corresponding material. The emission factors belonging to the same material are deduplicated and merged to obtain the updated mapping relationship. This reduces the cost of building the library and facilitates the rapid and accurate matching of the target material based on the factor material library, as well as the rapid and accurate mapping of the target emission factor corresponding to the target material based on the mapping relationship, thereby improving the accuracy of emission factor matching.

[0052] In some embodiments, the target material name includes at least one of a surface semantic name, a standard substance name, and a deep standard name; the preset priority of the surface semantic name, the standard substance name, and the deep standard name decreases in descending order.

[0053] The surface semantic name refers to the direct translation of the target product's name (emEnglishName). For example, if the target product is nylon, its direct translation is nylon. The standard material name refers to the direct material name of the target product (materialDirectEnglishName). For example, if the target product is nylon, its direct material name is nylon6-6. The deep standard name refers to the actual material name of the target product (materialRealEnglishName). For example, if the target product is nylon, its actual material name is polyamide.

[0054] For example, the specific implementation of extracting the target material name using a large language model based on the description information can be output using the following command: "You are now a smart recommendation expert who is proficient in carbon emission factors and the Ecoinvent database."

[0055] You are familiar with the structure and classification system of all material and process datasets in various versions of Ecoinvent (V3.9.1, V3.10, V3.11), and understand their lifecycle phase logic (raw materials, manufacturing, transportation, use, disposal, etc.). Your task consists of two parts: First, based on the user input, determine its material and output the following six attributes: emEnglishName (English name of the emission source, which must be accurate and concise); materialDirectEnglishName (Direct translation of the English name of the material; if the original input is a product name, abbreviation, etc., the full name can be translated directly; if unknown, it can be left blank)); materialRealEnglishName (deep-level material name in English, must be a standard, universally accepted substance or chemical name / classification from the Ecoinvent database or an internationally recognized database; simply copying or translating the input is not acceptable. If the input is a standard name, it can be the same as materialDirectEnglishName)".

[0056] Among them, surface semantic names, standard substance names, and deep standard names have search priorities (i.e., preset priorities). When searching for a target material in the factor material library based on the target material name, according to the preset priorities, the search is first conducted based on the surface semantic name. If a target material is found, subsequent priority searches and matches are not performed (i.e., searches and matches based on standard substance names and deep standard names are not performed). If no target material matching the surface semantic name is found in the factor material library, the search is then conducted based on the standard substance name. If a target material is found, subsequent priority searches and matches are not performed (i.e., searches and matches based on deep standard names are not performed). If no target material matching the standard substance name is found in the factor material library, the search is then conducted based on the deep standard name.

[0057] For example, if the user inputs the description "nylon," the large language model directly generates three target material names: Nylon (surface semantic name), Nylon6-6 (standard material name), and Polyamide (deep standard name). Based on a preset priority, the surface semantic name level takes precedence. The search for Nylon is performed in the factor material library. If a matching target material is found, there's no need to search for Nylon6-6 and Polyamide; that is, no further drilling down to Nylon6-6 and Polyamide is required. If no matching target material is found for Nylon, the search for Nylon6-6 is performed in the factor material library. If a matching target material is found for Nylon6-6, there's no need to search for Polyamide; that is, no further drilling down to Polyamide is required. If no matching target material is found for Nylon6-6, the search for Polyamide is performed in the factor material library.

[0058] Therefore, by using a large language model to directly generate surface semantic names, standard material names, and deep standard names at different semantic granularities based on the user-input description information of the target product, comprehensive coverage of semantic material names in different expressions can be achieved. This improves the accuracy and effectiveness of subsequent searches and matching, and eliminates the need for multiple uncertain processes such as translation before retrieval, reducing semantic bias caused by language differences and their impact on matching results. Furthermore, by setting preset priorities for the three types of names with different semantic granularities, matching is performed according to the preset priority order when searching for matching target materials in the factor material library, and matching stops upon successful matching. This achieves effective matching while also considering matching quality, reducing computational resource consumption and lowering computational costs.

[0059] In some embodiments, step S200 may include steps S210 to S250 as follows: Step S210: Using the embedding model, the surface semantic name is matched in the factor material library to obtain the first matching result; in response to the first matching result satisfying the preset matching conditions, the target material is determined according to the first matching result.

[0060] An embedding model is a semantic representation model used to map text, phrases, or names into fixed-dimensional vector representations. This model semantically encodes the input text, ensuring that semantically similar texts have high similarity in the vector space, thus supporting semantic retrieval and matching based on vector similarity. In this embodiment, the embedding model is used to convert user-input material description information and material names from the factor material library into semantic vectors, providing a foundation for similarity calculation and candidate selection in target material matching.

[0061] For example, the specific implementation of matching the surface semantic name in the factor material library can be as follows: perform a similarity search on the surface semantic name in the vector index of the factor material library, and calculate the corresponding first similarity result (i.e., the first matching result). If the first similarity result meets the preset matching conditions, it means that a target material matching the surface semantic name has been found in the factor material library, and the material corresponding to the first similarity result is taken as the target material.

[0062] The preset matching condition is: similarity greater than or equal to preset similarity. For example, the preset similarity is 100%.

[0063] Step S220: In response to the first matching result not meeting the preset matching conditions, the standard substance name is matched in the factor material library using the embedding model to obtain the second matching result.

[0064] For example, the specific implementation of matching standard substance names in the factor material library can be as follows: perform similarity retrieval on the standard substance names in the vector index of the factor material library, and calculate the corresponding second similarity result (i.e., the second matching result).

[0065] Step S230: In response to the second matching result satisfying the preset matching conditions, determine the target material based on the second matching result.

[0066] If the second similarity result meets the preset matching conditions, it means that a target material matching the standard substance name has been found in the factor material library. In this case, the material corresponding to the second similarity result is taken as the target material.

[0067] Step S240: In response to the second matching result not meeting the preset matching conditions, the deep standard name is matched in the factor material library using the embedded model to obtain the third matching result.

[0068] For example, the specific implementation of matching deep standard names in the factor material library can be: performing a similarity search on deep standard names in the vector index of the factor material library, and calculating the corresponding third similarity result (i.e., the third matching result).

[0069] Step S250: Determine the target material based on the first matching result, the second matching result, and the third matching result.

[0070] Regardless of whether the third similarity result meets the preset matching conditions, the target material will be further determined based on the first, second, and third similarity results, which can reduce misjudgments and improve matching accuracy.

[0071] In the technical solution of this application embodiment, the specific implementation process of searching for a target material that matches the target material name in the pre-built factor material library can be as follows: First, an embedding model is used to generate corresponding material semantic vectors for the three types of names: surface semantic name, standard material name, and deep standard name, and these vector vectors are stored in the vector library. Then, a similarity search is performed in the vector index of the factor material library to obtain the corresponding matching results (e.g., similarity results).

[0072] For example, using an embedding model, the semantic vector of the first material corresponding to the surface semantic name is searched for similarity in the vector index of the factor material library to obtain the first similarity result. If the first similarity result is 100%, it means that a target material matching the surface semantic name has been found in the factor material library, and no further priority search and matching is needed. If the first similarity result is less than 100%, it means that no target material matching the surface semantic name has been found in the factor material library. Then, according to the preset priority, the embedding model is used to search for the semantic vector of the second material corresponding to the standard material name for similarity in the vector index of the factor material library to obtain the second similarity result. If the second similarity result is 100%, it means that a target material matching the standard material name has been found in the factor material library, and no further priority search and matching is needed. If the second similarity result is less than 100%, it means that no target material matching the standard material name was found in the factor material library. Then, according to the preset priority, the embedding model is used to perform similarity retrieval on the third material semantic vector corresponding to the deep standard name in the vector index of the factor material library to obtain the third similarity result (i.e. the third matching result). For the deep standard name level, regardless of whether a target material matching the deep standard name is found, the target material is further determined based on the first matching result, the second matching result, and the third matching result.

[0073] Therefore, by embedding a model to unify the semantic vectorization representation of Chinese emission source material descriptions and English material names in the factor material library, multiple uncertain processes such as translation followed by retrieval are eliminated, overcoming the limitations of language differences and inconsistent textual expressions on matching results. Furthermore, by setting a priority-based, layer-by-layer filtering matching strategy (searching each priority level and stopping when a match is successful), the target material can be matched in most scenarios with only one or two searches, improving matching efficiency and reducing computational resource consumption. The progressive matching chain, from surface semantic names to deep standard names, forms a progressive matching path from coarse to fine matching, achieving a balance between matching efficiency and accuracy. In addition, by setting preset matching conditions, low-quality or invalid matching results can be effectively filtered out, reducing the possibility of misidentifying semantically irrelevant or weakly related materials as successful matches, thus improving the accuracy and effectiveness of matching.

[0074] For example, the vector library can be the Milvus vector database. Milvus is an open-source vector database system primarily used for storing, indexing, and retrieving large-scale vector data based on similarity. This database supports various vector index structures and similarity calculation methods, enabling it to adapt to vector retrieval scenarios of different scales and accuracy requirements while ensuring retrieval efficiency. In this embodiment, Milvus is used as the vector library to store material semantic vectors generated by the embedding model and to support the similarity-based candidate factor material retrieval process.

[0075] In some embodiments, step S250 may include steps S251 to S255.

[0076] Step S251: Using the reordering model, reorder the first matching result, the second matching result, and the third matching result respectively.

[0077] A reranking model is a model that performs refined semantic judgment and reorders candidate results based on the initial retrieval results. This model typically relies on more sophisticated semantic understanding capabilities to evaluate input queries and candidate objects in pairs, thereby improving the accuracy of the final matching results. In this embodiment, the reranking model is used to further semantically discriminate the candidate materials (i.e., the first matching result, the second matching result, and the third matching result) obtained from the embedded model. By comprehensively considering material semantics, application scenarios, and contextual information, the model reorders the candidate factor materials to determine the most matching target material.

[0078] When reordering the matching results using the reordering model, the sorting order can be from largest to smallest according to the matching degree (e.g., similarity) or from smallest to largest. The specific sorting algorithm can be set according to the actual situation, and no specific limitation is made here.

[0079] Step S252: Select the first preset number of matching results from the first reordered matching results as the first candidate set.

[0080] For example, taking the re-sorting based on matching degree from largest to smallest as an example, the first preset number of matching results in the first re-sorted matching results are selected as the first candidate set. For example, the first preset number is 3, but other values ​​are also possible. The specific value can be set according to the actual situation, and no specific limitation is made here.

[0081] Step S253: Select the second preset number of matching results from the reordered second matching results as the second candidate set.

[0082] For example, taking the re-sorting based on matching degree from largest to smallest as an example, the second preset number of matching results in the re-sorted second matching results are selected as the second candidate set. The second preset number is 5, but other values ​​are also possible. The specific value can be set according to the actual situation, and no specific limitation is made here.

[0083] Step S254: Select the third preset number of matching results from the reordered third matching results as the third candidate set.

[0084] For example, taking the re-sorting based on matching degree from largest to smallest as an example, the third preset number of matching results in the re-sorted third matching results are selected as the third candidate set. The third preset number is 10, but other values ​​can also be used. The specific value can be set according to the actual situation, and no specific limitation is made here.

[0085] Step S255: Using a large language model, determine the target material based on the first candidate set, the second candidate set, and the third candidate set.

[0086] Specifically, when the second matching result does not meet the preset matching conditions, an embedding model is used to match the deep standard name in the factor material library to obtain a third matching result. Then, a re-ranking model is used to re-rank the first, second, and third matching results respectively. For example, taking re-ranking by matching degree from largest to smallest, the first preset number of matching results (e.g., the top 3) from the re-ranked first matching results are selected as the first candidate set; the second preset number of matching results (e.g., the top 5) from the re-ranked second matching results are selected as the second candidate set; and the third preset number of matching results (e.g., the top 10) from the re-ranked third matching results are selected as the third candidate set. Then, the first, second, and third candidate sets, along with the user-input description of the target product, are input into a large language model, which selects the most suitable material as the target material. Thus, by combining semantic vector similarity retrieval and a re-ranking model, candidate materials are filtered and refined layer by layer, reducing mismatches caused by relying solely on single similarity calculations and improving the reliability and accuracy of the final material location result. Furthermore, after fine ranking, candidate factor materials (i.e., the first candidate set, the second candidate set, and the third candidate set) are selected. By leveraging the understanding capabilities of the large language model, the most suitable material can be selected from the limited candidate materials, thereby improving the accuracy of the final matching results.

[0087] In some embodiments, the first preset quantity is less than the second preset quantity, and the second preset quantity is less than the third preset quantity.

[0088] For example, the first preset quantity is 3, the second preset quantity is 5, and the third preset quantity is 10. Other values ​​can also be set, and the specific settings can be made according to the actual situation. No specific limitations are made here.

[0089] By setting an increasing relationship where the first preset quantity is less than the second preset quantity, and the second preset quantity is less than the third preset quantity, differentiated recall for names with different semantic granularities can be achieved. This ensures that the candidate set provided to the large language model includes both high-precision candidate materials (i.e., the fewest but most accurate candidates returned from the first matching results corresponding to the surface semantic names) and wide-coverage candidate materials (i.e., the most returned but with the widest coverage from the third matching results corresponding to the deep standard names). This helps the large language model select the most suitable material from the limited candidate materials, thereby improving the accuracy of emission factor matching.

[0090] Figure 4 This is the overall technical architecture of the emission factor matching method provided in the embodiments of this application. For example, see [link to relevant documentation]. Figure 4The overall technical architecture of this emission factor matching method is as follows: First, the data in the preset factor library comes from the Ecoinvent factor library (i.e., the Ecoinvent database). The attribute information of emission factors includes process names and factor entries. Data extraction, transformation, loading, and data ETL are performed from the Ecoinvent database. The extracted data is collected, cleaned, and standardized. Then, a core material library (i.e., a factor material library) is constructed. The core material extractor extracts core materials from the attribute information of emission factors, and emission factors belonging to the same material are deduplicated and merged, retaining identifiers such as units and countries for integration to form the factor material library. Second, the target material name is embedded in a multilingual phrase using an embedding service to generate a vector (i.e., surface semantic name, standard material name, and deep standard name are embedded to generate corresponding material semantic vectors). The generated vector is then input into a vector database storage based on the Milvus system (VectorDatabase Store (based on Milvus), VectorStore (Milvus)) for vector indexing and similarity retrieval. A similarity search and rerank model (Similarity & Rerank) is used for initial screening and cross-encoder reranking to obtain candidate factor materials (i.e., the first, second, and third candidate sets). Finally, the candidate factor materials, along with user-input descriptive information such as product, emission source, material, and context stage, are input into a large language model selector (LLM selector). The LLM selector fuses the candidate and contextual selections to filter out the most suitable target material, mapping it to the corresponding target emission factor. Finally, through an application programming interface (API / Command-Line Interface), the queryable factor results (i.e., the target emission factor) are returned.

[0091] In summary, the emission factor matching method provided in this application directly generates the target material name corresponding to the target product from the description information of the target product input by the user. It then searches for a matching target material in the factor material library based on the target material name, and finally maps the target emission factor corresponding to the target material in the factor material library based on the mapping relationship between emission factors and materials. This transforms the complex emission factor matching problem into a material semantic matching problem, eliminating the need for multiple uncertain processes such as translation and retrieval. It reduces semantic bias caused by language differences and their impact on matching results, achieving automatic identification and accurate positioning of the core material of emission factors. This improves the accuracy of emission factor matching and the efficiency of carbon footprint modeling. It also solves problems in related technologies where emission factor selection relies on manual translation, keyword retrieval, and experience-based judgment, leading to difficulties in cross-language matching, inconsistent material semantics, difficulty in corresponding abbreviations with standard names, and low retrieval efficiency. Furthermore, through a fixed factor material library, vectorized representation, and reordering process, the matching results have high consistency and reproducibility, reducing subjective fluctuations caused by manual operation. It can meet the requirements of engineering and large-scale applications, and can automatically locate emission source materials to factor materials in a short time. It is suitable for multi-emission source and large-scale carbon footprint modeling scenarios, effectively improving the overall modeling efficiency.

[0092] Figure 5 This is a schematic block diagram of an emission factor matching device provided in an embodiment of this application. An embodiment of this application also provides an emission factor matching device, see below. Figure 5 The emission factor matching device 10 includes: a generation module 11, used to generate a target material name corresponding to the target product based on the description information of the target product input by the user; a search module 12, used to search for a target material that matches the target material name in a pre-built factor material library; wherein the factor material library includes a mapping relationship between emission factors and their respective materials; and a determination module 13, used to determine the emission factor corresponding to the target material in the factor material library as the target emission factor for matching the target product based on the mapping relationship.

[0093] The technical solution of this application provides an emission factor matching device. It directly generates the target material name corresponding to the target product by using the description information of the target product input by the user. Based on the target material name, it searches for a matching target material in a factor material library. Then, based on the mapping relationship between emission factors and materials, it maps the target emission factor corresponding to the target material in the factor material library. Thus, by first matching the target material based on the factor material library and the target product's target material name, and then returning to the factor material library based on the target material and the mapping relationship to map the target emission factor, the complex emission factor matching problem is transformed into a material semantic matching problem. This eliminates the need for multiple uncertain processes such as translation followed by retrieval, reducing semantic bias caused by language differences and their impact on the matching results, thereby improving the accuracy of emission factor matching.

[0094] In some embodiments, the emission factor matching device 10 further includes a construction module for: Extract the material corresponding to each emission factor from the attribute information of the emission factors contained in the preset factor library to form a primary mapping relationship between emission factors and materials; Based on the initial mapping relationship, emission factors belonging to the same material are mapped to the same material to generate an updated mapping relationship; Based on the updated mapping relationship, a factor material library is formed.

[0095] In some embodiments, the target material name includes at least one of a surface semantic name, a standard substance name, and a deep standard name; the preset priority of the surface semantic name, the standard substance name, and the deep standard name decreases in descending order.

[0096] In some embodiments, the search module 12 is further configured to: Using an embedding model, the surface semantic name is matched in the factor material library to obtain the first matching result; in response to the first matching result satisfying the preset matching conditions, the target material is determined based on the first matching result. In response to the first matching result not meeting the preset matching conditions, the embedding model is used to match the standard substance name in the factor material library to obtain the second matching result; In response to the second matching result satisfying the preset matching conditions, the target material is determined based on the second matching result; In response to the second matching result not meeting the preset matching conditions, the deep standard name is matched in the factor material library using the embedding model to obtain the third matching result; The target material is determined based on the first matching result, the second matching result, and the third matching result.

[0097] In some embodiments, the search module 12 is further configured to: Using a reordering model, the first matching result, the second matching result, and the third matching result are reordered respectively. Select the first preset number of matching results from the first reordered matching results as the first candidate set; Select a second preset number of matching results from the reordered second matching results as the second candidate set; Select the third preset number of matching results from the reordered third matching results as the third candidate set; Using a large language model, the target material is determined based on the first candidate set, the second candidate set, and the third candidate set.

[0098] In some embodiments, the first preset quantity is less than the second preset quantity, and the second preset quantity is less than the third preset quantity.

[0099] In some embodiments, the generation module 11 is further configured to: Based on the description information, the target material name is extracted using a large language model; the description information includes at least one of the target product name, emission source, material, and process stage.

[0100] This application also provides an emission factor matching agent for executing the emission factor matching method described in any embodiment of this application.

[0101] This application also provides a computer-readable storage medium storing a computer program thereon, which is loaded by a processor to execute the emission factor matching method described in any embodiment of this application.

[0102] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0103] The above provides a detailed description of an emission factor matching method, apparatus, intelligent agent, and storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An emission factor matching method, characterized in that, include: Generate the target material name corresponding to the target product based on the description information of the target product input by the user; Search for a target material whose name matches the target material name in a pre-built factor material library; wherein, the factor material library includes the mapping relationship between emission factors and their respective materials; Based on the mapping relationship, the emission factor corresponding to the target material is determined in the factor material library as the target emission factor for matching the target product.

2. The method according to claim 1, characterized in that, The method for constructing the factor material library includes: Extract the material corresponding to each emission factor from the attribute information of the emission factors contained in the preset factor library to form a primary mapping relationship between emission factors and materials; Based on the primary mapping relationship, emission factors belonging to the same material are mapped to the same material to generate an updated mapping relationship. Based on the updated mapping relationship, the factor material library is formed.

3. The method according to claim 1, characterized in that, The target material name includes at least one of the surface semantic name, the standard substance name, and the deep standard name; the preset priority of the surface semantic name, the standard substance name, and the deep standard name decreases in descending order.

4. The method according to claim 3, characterized in that, The step of searching for a target material that matches the target material name in a pre-built factor material library includes: Using an embedding model, the surface semantic name is matched against the factor material library to obtain a first matching result; in response to the first matching result satisfying a preset matching condition, the target material is determined based on the first matching result. In response to the first matching result not meeting the preset matching conditions, the embedding model is used to match the standard substance name in the factor material library to obtain a second matching result; In response to the second matching result satisfying the preset matching condition, the target material is determined based on the second matching result; In response to the second matching result not meeting the preset matching conditions, the deep standard name is matched in the factor material library using the embedding model to obtain a third matching result; The target material is determined based on the first matching result, the second matching result, and the third matching result.

5. The method according to claim 4, characterized in that, Determining the target material based on the first matching result, the second matching result, and the third matching result includes: Using a reordering model, the first matching result, the second matching result, and the third matching result are reordered respectively. Select a first preset number of matching results from the reordered first matching results as the first candidate set; Select a second preset number of matching results from the reordered second matching results as the second candidate set; A third preset number of matching results from the reordered third matching results are selected as the third candidate set; The target material is determined using a large language model based on the first candidate set, the second candidate set, and the third candidate set.

6. The method according to claim 5, characterized in that, The first preset quantity is less than the second preset quantity, and the second preset quantity is less than the third preset quantity.

7. The method according to claim 1, characterized in that, The step of generating the target material name corresponding to the target product based on the description information of the target product input by the user includes: Based on the description information, the name of the target material is extracted using a large language model; the description information includes at least one of the target product's name, emission source, material, and process stage.

8. An emission factor matching device, characterized in that, The device includes: The generation module is used to generate the target material name corresponding to the target product based on the description information of the target product input by the user; The search module is used to search for target materials that match the name of the target material in a pre-built factor material library; wherein, the factor material library includes the mapping relationship between emission factors and their respective materials; The determination module is used to determine the emission factor corresponding to the target material in the factor material library as the target emission factor for matching the target product, based on the mapping relationship.

9. An emission factor matching intelligent agent, characterized in that, Used to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to perform the steps of the method as described in any one of claims 1 to 7.

Images