Carbon emission factor dynamic matching method based on large language model

Abstract

According to the carbon emission factor dynamic matching method based on the large language model, efficient and accurate matching is achieved through multi-step optimization. The method comprises the following steps: firstly, collecting carbon emission factors and activity data from multiple sources, and cleaning and normalizing the data; then, performing deep semantic analysis on the data by utilizing a pre-trained large language model subjected to field fine adjustment to generate a high-quality semantic vector; the system unifies data formats and completes physical quantity conversion through a unit identification and conversion mechanism; in the matching process, the system performs intelligent matching by calculating cosine similarity of semantic vectors, and under the condition that the similarity is close, a weighted average mechanism is applied to optimize a result, so that the matching precision is further improved; meanwhile, the system analyzes the relevance of newly submitted data by utilizing a factor library dynamic updating mechanism, dynamically expands the factor library and continuously optimizes the matching performance; the whole method improves the matching precision and efficiency, reduces manual intervention and errors, and reduces the calculation cost of carbon emission.

Description

Technical Field

[0001] The present invention belongs to the field of carbon emission calculation, and particularly relates to a method for dynamically matching carbon emission factors based on a large language model. Background Art

[0002] Although certain achievements have been made in carbon emission calculation in the prior art, the carbon emission calculation process still faces multiple technical challenges, especially in the link of matching activity data with carbon emission factors. In practical applications, the activity data of different industries has the characteristics of diversity and complexity, and the problems of data standardization and inconsistent terms further increase the difficulty of carbon emission factor matching. Existing methods are insufficient in dealing with complex data and achieving efficient matching, and it is difficult to meet the requirements of accurate and automated carbon emission calculation.

[0003] Currently, the calculation method of carbon emissions mainly relies on the product relationship between activity data and carbon emission factors, and is carried out in accordance with relevant industry specifications such as the "Standard for Calculating Carbon Emissions from Buildings" GB / T 51366-2019. However, in practical applications, there are many difficulties in the matching between activity data and carbon emission factors. The sources of activity data are extensive, including hundreds of different materials, energy types, and thousands of construction and process activities; the carbon emission factor library covers multiple industries, different regions, and a dynamically changing time dimension. In the matching process, problems such as inconsistent data units are often faced, such as the conversion between volume and mass units, and the non-standardization of material descriptions also needs to be processed, such as the same material may have multiple naming methods. In addition, the matching work also involves unit conversion of complex physical quantities, such as the use of density or calorific value. These problems make the current method require a large amount of manual intervention in matching, and complete the matching by analyzing item by item and manually selecting appropriate carbon emission factors. This method is not only inefficient, consuming a large amount of human resources when dealing with a large amount of data, but also easily leads to a decrease in the accuracy and consistency of the calculation due to human errors.

[0004] In the current carbon emission factor matching process, the complexity problems of unit conversion and semantic understanding are still the main technical bottlenecks. Especially when facing diverse materials, complex process descriptions, and dynamically changing new materials, new technologies, and new processes, these problems are more prominent. Therefore, there is an urgent need for a more intelligent and automated solution that can comprehensively handle the parsing and processing of complex data, accurately match carbon emission factors, and provide reliable and scalable technical support for carbon emission calculation. Summary of the Invention

[0005] To solve the above problems, the present invention discloses a dynamic matching method for carbon emission factors based on a large language model. Through the large language model, in-depth semantic analysis is performed on activity data and carbon emission factors, and an intelligent algorithm is combined to construct and maintain a dynamic carbon emission factor library, realizing the precise matching of carbon emission factors and the dynamic update of the factor library, thereby greatly improving the accuracy and efficiency of carbon emission calculation.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A dynamic matching method for carbon emission factors based on a large language model, comprising the following steps:

[0008] Step A1: Collect carbon emission factor data and related activity data from multiple data sources (such as reports released by the government, industry standards, academic research, and enterprise internal records), and perform preprocessing on them. The preprocessing includes data cleaning and normalization to ensure the consistency and integrity of the data format.

[0009] Step A2: The system selects a large language model (such as ChatGLM) as the basic model, establishes language understanding ability through pre-training of general corpus, and combines fine-tuning of multi-industry domain-specific corpus (covering industries such as industry, energy, transportation, and construction) to enhance the model's parsing ability for complex terms and professional descriptions. During the fine-tuning process, the Adam optimization algorithm is used to optimize the model to improve the quality and consistency of semantic vector generation, laying a foundation for subsequent efficient and precise matching.

[0010] Step A3: The system uses the large language model to parse the semantic content of the input data, automatically identifies units related to physical quantities, such as cubic meters and kilograms, and standardizes units in different expression forms. The system converts the input unit into a standard unit through a unified rule, for example, converts kilograms to kilograms, and completes necessary unit conversions based on a predefined unit conversion rule library. For complex conversions involving physical properties, such as density-based unit conversions, the system can dynamically query relevant attribute data and perform calculations to ensure the accuracy of the conversion results and guarantee data consistency and comparability.

[0011] Step A4: The system uses the fine-tuned large language model to perform in-depth semantic analysis on multi-industry activity data (such as energy use, material consumption, process flows, etc.) and carbon emission factors, and converts the original data description into a high-dimensional semantic vector. During the semantic analysis process, the model extracts the deep semantic features of the data and generates semantic vectors to represent the inherent semantic meaning of the data. To ensure the comparability and consistency of data semantic vectors, the system converts the high-dimensional semantic vectors into a fixed-dimensional representation through average pooling technology, providing high-quality input for subsequent precise matching of carbon emission factors.

[0012] Step A5: The system quantifies the semantic similarity between the semantic vectors of project activity data and the factors in the carbon emission factor library by calculating the cosine similarity. The similarity score represents the matching degree between the activity data and the factors. If the similarity score of a certain factor in the factor library is higher than the preset threshold and significantly higher than the scores of other factors, that is, there are no multiple factors with similar similarity scores, the system directly determines that the factor is successfully matched, completes the matching process, and outputs the corresponding carbon emission factor value. At this time, there is no need to enter further optimization steps, and the matching task can end.

[0013] Step A6: When a unique high-similarity matching factor cannot be found in Step A5, that is, the similarity scores of multiple factors are close and all higher than the threshold, the system will perform matching optimization through weighted average calculation. The system assigns weights to each factor according to the similarity scores and uses these weights to perform weighted average on the carbon emission values, so as to obtain a more accurate comprehensive carbon emission factor value. Through the weighted average mechanism, the system can more accurately reflect the actual carbon emissions of project activities, especially in the case of similar similarities, avoiding the deviation that may be caused by the selection of a single factor.

[0014] Step A7: To ensure the timeliness and accuracy of the factor library, the present invention designs a dynamic update mechanism for the factor library. Users can submit new material data or process descriptions through the system interface or API. The system formats the input data and generates corresponding semantic vectors through a large language model. The system compares the generated semantic vectors with the existing data in the factor library. If the similarity exceeds the set threshold, the system adds the new data to the factor library and automatically updates the relevant metadata (such as data source, update time, etc.) to ensure the continuous optimization and maintenance of the factor library.

[0015] To support the implementation of the above method, the present invention further provides a carbon emission factor matching system, which includes the following modules:

[0016] Data input and preprocessing module: used to receive carbon emission factor data and activity data, perform data cleaning and format standardization processing, remove redundant information, correct error data, ensure the consistency of data format and content, and provide high-quality data input for subsequent semantic parsing.

[0017] Semantic parsing module: connected to the data input and preprocessing module, uses a large language model to perform in-depth semantic parsing on the input data, extracts key semantic features and generates high-dimensional semantic vectors. Through the high-precision semantic vector representation, the semantic parsing module provides accurate semantic information support for factor matching, ensuring that the semantic correlation between activity data and carbon emission factors can be accurately calculated.

[0018] Unit Identification and Conversion Module: Connected to the semantic parsing module, it automatically identifies the unit types in the input data and completes unit conversion according to the predefined rule library. This module supports the conversion of complex physical quantities, ensures that all units are standardized, avoids matching errors caused by unit differences, and improves data consistency and comparability at the same time.

[0019] Factor Matching Module: Connected to the semantic parsing module, it quantifies the semantic correlation between the active data semantic vector and the carbon emission factor semantic vector by calculating the cosine similarity. According to the similarity score, it identifies and selects the most relevant carbon emission factor. When the similarity scores of multiple factors are close, the system calls the weighted average module to further optimize the matching result.

[0020] Weighted Average Module: Connected to the factor matching module, it starts the weighted average mechanism when the similarity scores of multiple factors are close. The weighted average module assigns weights according to the similarity scores of the factors and generates a comprehensive carbon emission factor value through weight calculation, thus avoiding the deviation that may be caused by the selection of a single factor and further optimizing the accuracy and stability of the matching result.

[0021] Factor Library Update Module: Connected to the factor matching module, it is used for the dynamic update of the factor library content. When receiving newly submitted data, the module determines its degree of association with the existing factors through similarity analysis. For new data for which no similar factors are found, the module adds it to the factor library and records the data source, update time, and quality identification to ensure the real-time update and optimization of the factor library content to meet the dynamically changing carbon emission calculation requirements.

[0022] The beneficial effects of the present invention are as follows:

[0023] By introducing large language models and intelligent algorithms, the matching accuracy and calculation efficiency of carbon emission factors are significantly improved, and the need for manual intervention is reduced. The automated processing and dynamic update functions of the system make the carbon emission calculation process more efficient, reduce errors caused by human operations, and improve the reliability and timeliness of the overall calculation. Description of the Drawings

[0024] Figure 1 It is a step flow chart of a method for dynamically matching carbon emission factors based on a large language model provided by the present invention.

[0025] Figure 2 It is a matching relationship diagram formed based on a large language model.

[0026] Figure 3 It is a relationship diagram of the carbon emission factor library update mechanism. Detailed Embodiment

[0027] The present invention will be further illustrated below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0028] As Figure 1 shown, a dynamic matching method for carbon emission factors based on a large language model according to the present invention includes the following steps:

[0029] Step A1: Data collection and preprocessing.

[0030] Step A11: Data collection. Carbon emission factor data and related activity data are collected from multiple sources. These data may be sourced from government reports, industry standards, academic research, enterprise internal records, etc. The collected data has a wide range, covering aspects such as materials, process procedures, energy use, etc.

[0031] Step A12: Data preprocessing. The collected data is cleaned and normalized, including removing redundant information, filling in missing values, and unifying data formats (such as date formats, unit formats, etc.). The system standardizes all data to ensure data consistency and comparability. In this process, a large language model can be used to identify and standardize non-standard terms and fuzzy descriptions in the data and convert them into a unified format.

[0032] Step A2: Training and fine-tuning of the large language model.

[0033] Step A21: Model architecture and training. The large language model is used to train and fine-tune the activity data. It includes two steps:

[0034] Step A211: Model architecture. A bidirectional large language model based on the Transformer architecture is adopted, and the self-attention mechanism is used to generate semantic vectors. The calculation formula of the self-attention mechanism is as follows:

[0035]

[0036] where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key. This mechanism enables the model to capture complex context relationships in the input text and generate deep semantic vectors. These semantic vectors have extremely high adaptability and accuracy for complex terms and descriptions unique to the construction field.

[0037] Step A212: Pre-training and fine-tuning. The large language model is first pre-trained on a large-scale general corpus to learn basic language patterns. Subsequently, the system fine-tunes the model using a domain-specific corpus so that the model can better handle terms and descriptions unique to each industry. In the fine-tuning process, the cross-entropy loss function is used to optimize the model parameters:

[0038]

[0039] Among them, y ij is the actual label, is the probability distribution predicted by the model, N is the number of samples, and C is the number of categories.

[0040] Step A22: Model optimization and adaptation. Use the Adam optimization algorithm to optimize the model and evaluate the model performance through the validation set. The optimization formula is as follows:

[0041] m t = β 1 m t-1 +(1 - β 1 )g t

[0042] v t = β 2 v t-1 +(1 - β 2 )g t 2

[0043]

[0044] Among them, β 1 and β 2 represent the exponential decay rates of momentum and second-order momentum estimation, used to control the weights of the first-order and second-order momenta in historical data;

[0045] g t represents the gradient at time t, which is the partial derivative of the loss function with respect to the model parameters; m t represents the estimated value of momentum, indicating the exponentially weighted moving average of the first-order momentum of the gradient;

[0046] v t represents the estimated value of the second-order momentum of the square of the gradient, used to measure the variance of the gradient;

[0047] and are the estimated values after bias correction for m t and v t respectively, eliminating the influence of small samples in the early stage;

[0048] θ t are the model parameters;

[0049] α is the learning rate.

[0050] Through the continuous introduction of the optimization algorithm and domain-specific data, the model has dynamic adaptability and can maintain efficient and accurate performance in the face of changing building data and terms.

[0051] Step A3: Automated Unit Recognition and Conversion

[0052] Step A31: The automated process of unit recognition. This process specifically includes two steps:

[0053] Step A311: Context Analysis and Keyword Detection: The system performs context analysis on the input text, identifies keywords related to physical quantities such as "volume", "weight", etc. through a large language model, and determines the corresponding units such as "m 3 ", "kg", etc.

[0054] Step A312: The system normalizes the recognized units in different forms. For example, it unifies "kilogram", "kilogram" into "kg", and "cubic meter" into "m 3 ". Through natural language understanding, the system can ensure that all input units can be consistently recognized and processed.

[0055] Step A32: Unit Conversion and Verification. After unit recognition is completed, the system automatically performs unit conversion according to a predefined unit conversion rule library. The specific process is as follows:

[0056] Step A321: Query the conversion rule library. The system has a built-in unit conversion rule library that contains conversion rules between common physical quantities. The system automatically queries the corresponding conversion formula from the rule library according to the recognized unit type. For example, the system queries the conversion rule between mass and volume:

[0057] Mass (kg) = Volume (m 3 ) × Density (kg / m 3 )

[0058] In addition, this rule library also contains various material densities that are continuously updated and expanded to facilitate automated conversion. After the conversion is completed, the system automatically verifies the conversion result to ensure that the data meets the expected accuracy requirements, and by comparing the original data with the converted data, ensures the accuracy of unit conversion. During this process, the large language model automatically identifies and corrects possible unit usage errors by parsing the context information of the input text.

[0059] Step A4: Semantic Parsing and Vectorization.

[0060] Step A41: Semantic Parsing. The system uses a fine-tuned adaptive large language model to perform semantic parsing on the standardized construction project activity data. The model captures the semantic information in the input data through the self-attention mechanism and generates semantic vectors. The formula for semantic parsing is:

[0061] v input = Model(xinput )

[0062] Among them, v input represents the semantic vector of the input text, and x input is the original input text. The generated semantic vector contains the deep semantic information of the input data, providing a basis for subsequent similarity calculation and factor matching.

[0063] Step A42: Vectorization processing. The system performs fixed-dimensional processing on the semantic vector through average pooling technology to ensure the consistency of the vector. The specific processing formula is as follows:

[0064]

[0065] Among them, v is the finally generated fixed-dimensional semantic vector, and h i is the output of the hidden layer of the large language model, and n is the length of the input text. The fixed-dimensional semantic vector ensures that different input data has a consistent semantic representation when calculating similarity, thereby improving the accuracy of matching.

[0066] Step A5: Cosine similarity calculation and factor matching.

[0067] Step A51: Cosine similarity calculation. The system performs factor matching by calculating the cosine similarity between the semantic vector of the project activity data and the vector in the carbon emission factor library. The calculation formula is as follows:

[0068]

[0069] Among them, v input is the semantic vector of the project activity data, and v factor is the semantic vector of the carbon emission factor. Through this high-precision semantic parsing and matching mechanism, the system can find the most relevant matching item from the factor library.

[0070] Step A52: Matching threshold judgment. After completing the cosine similarity calculation, the system compares the calculation result with the pre-set threshold τ. If the similarity score is higher than the threshold τ, the system considers that the current carbon emission factor matches the project activity data successfully and proceeds to the next step; otherwise, the system may choose not to match or perform other processing measures.

[0071] Step A6: Matching optimization of the weighted average mechanism. When there are multiple carbon emission factors with similarity scores very close to each other that are greater than the threshold τ, to further improve the accuracy of matching, the system introduces a weighted average mechanism for matching optimization. The specific steps are as follows:

[0072] Step A61: Weight calculation. When the similarity scores of multiple carbon emission factors are close, the system performs a weighted average calculation on these factors. The system first assigns weights to each factor according to the similarity score, and the weight calculation formula is:

[0073]

[0074] where ω i is the weight of the i-th factor, S i is the similarity score of this factor, and l is the total number of all close factors.

[0075] Step A62: Weighted average calculation. The system uses the above weights to perform a weighted average on the carbon emission values of each factor, and finally obtains the comprehensive carbon emission factor value:

[0076]

[0077] where F final is the final comprehensive carbon emission factor value, ω i is the weight of the i-th factor, and F i is the carbon emission value of the i-th factor.

[0078] Step A7: Dynamic factor library update mechanism.

[0079] Step A71: Dynamic update of the factor library. To ensure the timeliness and accuracy of the factor library, the present invention designs a dynamic factor library update mechanism. After the user submits new building material data or process descriptions through the system interface or API, the system will format the input data and generate corresponding semantic vectors through an adaptive large language model.

[0080] Step A72: Matching and storage of new data. The system compares the semantic vectors of the new data with the existing data in the factor library. If the similarity score exceeds the set threshold, the system adds the new data to the factor library and simultaneously updates the metadata related to this entry. The update process ensures data consistency and integrity through the database transaction mechanism. The management of metadata includes the following contents:

[0081] Data source record: Record the source information of the new factor, such as user input, automatic collection, or import from an external database.

[0082] Update timestamp: Record the last update time of the new factor entry to ensure that the system can track the update history of the data.

[0083] Data quality identification: The system evaluates the quality of the new data and stores the quality level in the metadata for subsequent calculations.

[0084] Usage frequency tracking: The system records the usage frequency of each factor to dynamically adjust relevant policies during the matching or optimization process.

[0085] Similarity score storage: The system saves the similarity scores calculated during the matching process for review or further optimization when needed.

[0086] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. A dynamic matching method for carbon emission factors based on a large language model, characterized in that: The following steps are involved: Step 1: Collect carbon emission factor data and related activity data from multiple data sources, and perform data cleaning and normalization to ensure the consistency and integrity of the data format; Step 2: Select a large language model as the base model, establish basic language understanding capabilities through pre-training on general corpora, and fine-tune the model with specialized corpora in specific fields so that it can accurately understand and process complex terms and descriptions; The model is optimized using the Adam optimization algorithm to improve the quality and consistency of semantic vector generation; Step 3: The system uses a large language model to parse the semantic content in the input data, automatically identify the unit type related to the physical quantity, and standardize the identified units; The system performs unit conversion operations based on a predefined unit conversion rule library. For complex unit conversions involving physical properties, the system performs precise calculations by associating semantic information and querying related attribute data, thereby ensuring the accuracy of the conversion results and the consistency of the data. Step 4: The system uses a fine-tuned large language model to perform deep semantic analysis on the activity data, extract the deep semantic features of the data, and convert them into high-dimensional semantic vectors. To ensure the uniformity and comparability of the semantic vectors, the system uses average pooling technology to convert high-dimensional vectors into fixed-dimensional semantic representations. Step 5: The system calculates the cosine similarity between the semantic vector of the project activity data and the semantic vector of each factor in the carbon emission factor library to quantify the semantic correlation between the activity data and the factors; Based on the calculation results, the system sorts all factors in the factor library by similarity score; When the similarity score of a factor is higher than the preset threshold and significantly higher than other factors, the system takes it as the best matching factor, completes the matching and outputs the carbon emission value of the factor as the result; Step 6: During the matching process, if the similarity scores of multiple factors are close and all above the preset threshold, the system will optimize the calculation through the weighted average mechanism; the system assigns weights according to the similarity scores of each factor, and combines these weights to perform a weighted average of the carbon emission values ​​of the candidate factors to generate a comprehensive carbon emission factor value as the final matching result; This optimization mechanism can avoid bias in factor selection, ensuring that the matching results are more accurate and consistent with the carbon emission characteristics of the actual project; Step 7: The system uses the dynamic factor library update mechanism to perform a preliminary matching analysis on the newly submitted material data or process description to determine whether the new data is sufficiently similar or complementary to the data in the existing factor library. Only when the new data cannot be effectively matched with the factors in the current factor library will the system identify it as new data and add it to the factor library. In addition, the system automatically records relevant metadata to ensure that the factor library always remains real-time and accurate.

2. The dynamic carbon emission factor matching method based on a large language model according to claim 1 is characterized in that: The data cleaning and normalization steps described in step 1 include: removing redundant information to improve data processing efficiency and reduce noise; correcting erroneous data, including detecting and correcting outliers or format problems to ensure data accuracy and completeness; and unifying the format and units of all data to standardize data from different sources.

3. The dynamic carbon emission factor matching method based on a large language model according to claim 1 is characterized in that: The model optimization step described in step 2 includes: pre-training the large language model with general corpus to form the model's basic language understanding ability and generate an initial semantic vector; then, fine-tuning the model with special corpus in the fields of construction, industrial manufacturing, and transportation so that it can accurately parse and understand the complex terms and semantic structures of multiple industries; using the Adam optimization algorithm to optimize the model parameters to improve the accuracy and consistency of the generated semantic vectors, thereby achieving efficient matching of carbon emission factors in multiple fields.

4. The dynamic carbon emission factor matching method based on a large language model according to claim 1 is characterized in that: The unit recognition and conversion steps described in step 3 include: the system automatically recognizes the unit type in the text through a large language model; unifies the recognized units into system standard units according to a predefined unit conversion rule library; and performs complex unit conversions in combination with physical properties to ensure the accuracy of the conversion results.

5. The dynamic carbon emission factor matching method based on a large language model according to claim 1 is characterized in that: The semantic parsing step described in step 4 extracts deep semantic information of activity data through a large language model, covering the fields of construction, industrial manufacturing and transportation; and uses average pooling technology to convert high-dimensional semantic vectors into fixed dimensions to ensure the uniformity and comparability of semantic representation to support subsequent similarity calculation and efficient factor matching.

6. A carbon emission factor matching system, characterized in that: Contains the following modules: Data input and preprocessing module, used to receive carbon emission factor data and activity data, perform data cleaning and format normalization, remove redundant information, correct erroneous data, and ensure consistency of data format and content; The semantic parsing module is connected to the data input and preprocessing module. It uses a large language model to perform deep semantic parsing on the input data, extract key semantic features and generate high-dimensional semantic vectors to provide high-quality semantic representation for subsequent matching. The unit recognition and conversion module is connected to the semantic parsing module to automatically identify the unit type in the data and complete the unit conversion according to the predefined rule base; this module supports the conversion of complex physical quantities, ensures that all units are standardized, and improves the consistency and comparability of data; The factor matching module, connected with the semantic parsing module, quantifies the semantic relevance by calculating the cosine similarity between the semantic vectors of activity data and carbon emission factors; According to the similarity score, identify and select the most matching carbon emission factor to complete the efficient matching task; The weighted average module is connected to the factor matching module. When the similarity scores of multiple carbon emission factors are close, the weighted average mechanism is activated. This module assigns weights to factors according to the similarity scores and generates a comprehensive carbon emission factor value through weight calculation to optimize the accuracy and stability of the matching results. The factor library update module is connected to the factor matching module and is used to dynamically update the content of the factor library. When the system receives newly submitted data, it determines whether it is significantly different from the existing factors through similarity analysis; for new data that does not find similar factors, the module adds it to the factor library and records the data source, update time and quality information to ensure real-time update and optimization of the factor library content.

7. The carbon emission factor matching system according to claim 6, characterized in that: The semantic parsing module uses a large language model to perform deep semantic parsing on the input data and converts text information into a high-dimensional semantic vector. The module further quantifies the semantic similarity between the semantic vector of the input data and the semantic vector of each factor in the carbon emission factor library by calculating the cosine similarity between the two. The similarity score is determined based on the calculation result and is used to sort the carbon emission factors to ensure that the factor with the highest similarity score is selected as the best matching result. Through this high-precision semantic analysis and similarity measurement, the system achieves efficient matching of carbon emission factors, supports deep semantic association analysis of multi-domain data, and improves the accuracy and stability of the matching process.

8. The carbon emission factor matching system according to claim 6, characterized in that: The weighted average module calculates the comprehensive carbon emission factor value by analyzing and weighting the similarity scores of multiple carbon emission factors; The weight assignment process is based on the normalization of the factor similarity scores, ensuring that factors with higher similarity have greater weight in the calculation; By multiplying and accumulating each carbon emission factor value with its weight, the module generates a comprehensive factor value as the final matching result when multiple factors are close in similarity, thereby improving the matching accuracy and stability and reducing the deviation that may be caused by the selection of a single factor; As a part of the factor matching module, the weighted average module is used to optimize the matching results, and is particularly suitable for processing factor sets with similar similarity scores.

Images