Industrial enterprise carbon pollution emission data list generation method based on artificial intelligence

Abstract

The invention provides an industrial enterprise carbon pollution emission data list generation method based on artificial intelligence, and relates to the technical field of industrial source carbon pollution emission list accounting, and the method comprises the steps: obtaining raw material information, product information, process information, process information, scale information and treatment facility information submitted by an enterprise; determining an accuracy identification result of the various kinds of information; screening industry pollution discharge information in the database; obtaining corrected raw material information, product information, process information, process information, scale information and treatment facility information; determining emission data of the various pollutants; and generating an industrial enterprise carbon pollution emission data list. According to the invention, by means of a production and pollution discharge coefficient system, a pollutant comparison table can be combed and established, process coefficients are combined, production and pollution discharge links during filling and reporting of enterprises and omission of necessary pollutants are reduced, activity level data thresholds of products, energy consumption and the like are determined, abnormal data in production activity levels are reduced, and the rationality of basic data is improved.

Description

Technical Field

[0001] This invention relates to the field of industrial carbon pollution emission inventory accounting technology, and in particular to an artificial intelligence-based method for generating a data inventory of carbon pollution emissions from industrial enterprises. Background Technology

[0002] Air pollutant / greenhouse gas emission inventories comprehensively identify various air pollutants and greenhouse gas emission sources, serving as a crucial foundation for formulating and implementing pollution reduction and carbon reduction policies and providing fundamental support for management decisions. Currently, existing pollutant / greenhouse gas inventory data products include the China Multiscale Emission Inventory Model (MEIC), the China High Spatial Resolution Emission Grid Data (CHRED), and the China Carbon Accounting Database (CEADs), all covering periods prior to 2020 and experiencing slow data updates. Traditional inventories primarily rely on macro-level data for calculation, resulting in low accuracy, unclear calculation methods and data sources, and significant challenges in data application and quality control. Furthermore, they incompletely cover gas types and emission sources, and their indicator coverage is insufficient. More importantly, traditional inventories are time-consuming and labor-intensive to compile, failing to meet the demands of refined environmental management in terms of spatiotemporal accuracy.

[0003] Currently, emission source statistical surveys are a routine task. Survey subjects fill out forms according to the "Emission Source Statistical Survey System," and the final data results are generated after review at each level—county, city, provincial, and national—by environmental protection departments. This has established a multi-dimensional data chain encompassing national, provincial, municipal, county, industry, enterprise, production line, and equipment levels, providing annual, quarterly, and monthly data in terms of temporal accuracy. Regarding statistical indicators, they cover production activity levels, energy consumption, and the operation of pollution control facilities, basically meeting the basic data requirements for compiling inventories with higher spatiotemporal precision. However, there are still shortcomings in the indicator settings, such as the lack of ammonia, carbon dioxide, and PM2.5. 10 PM 2.5 Important indicators such as BC and OC are lacking; however, industrial source emission accounting involves the industrial source reporting entities adding their own accounting modules. The varying skill levels of reporting personnel make this process prone to omissions, duplicate accounting, and unreasonable or even abnormal data entry regarding production activity levels. Data quality relies heavily on manual verification and platform review. The quality of emission source statistics depends primarily on manual review and platform-defined rules for verification. For situations where pollutant accounting is difficult, especially with numerous equipment, long production processes, multiple chains, and complex production technologies, the detection and correction rates for hidden problems such as omissions and duplicate accounting are low. Therefore, simple emission source statistics cannot meet the current needs of refined environmental management.

[0004] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides an artificial intelligence-based method for generating a carbon emission data inventory of industrial enterprises. It can solve the problems of single technical accounting indicators, low accounting accuracy, low spatiotemporal resolution, time-consuming and labor-intensive, and unable to be carried out on a regular basis. It can also solve the problems of incomplete statistical indicators and low accuracy of emission source statistical data, so as to meet the current needs of refined environmental management.

[0006] According to a first aspect of the present invention, a method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence is provided, comprising: Obtain information on raw materials, products, processes, technologies, scale, and treatment facilities submitted by enterprises; The information self-verification model is used to verify the information of raw materials, products, processes, technology, scale, and treatment facilities to determine the accuracy of various types of information. If there is information indicating an abnormality in the accuracy of the identification results, then industry pollution discharge information will be screened from the database based on at least one of the multiple pieces of information submitted by the enterprise. By using industry pollution discharge information, various information reported by enterprises, and pollution discharge data correlation models, we can obtain corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. Based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, the emission data of various pollutants are determined; Based on emission data of various pollutants, a data list of carbon emissions from industrial enterprises is generated.

[0007] According to the present invention, the accuracy identification results of various types of information are determined, including: By using the text recognition sub-model of the information self-verification model, features are extracted from multiple category names in the raw material information to obtain category name feature information; The data corresponding to the category name is concatenated with the category name feature information to obtain the category verification feature information; The graph attention sub-model of the information self-verification model is used to process the verification feature information of each category to obtain the verification feature information of each category; By splicing together the verification feature information of each category, the raw material information verification feature information is obtained; The verification feature information of each category is processed by the first multi-layer perception layer of the information self-verification model to obtain the first probability information that the data corresponding to each category name is abnormal data. The raw material information verification feature information is processed by the second multi-layer perception layer of the information self-verification model to obtain the second probability information of raw material information anomalies; Based on the first probability information and the second probability information, the accuracy of the raw material information identification result is determined.

[0008] According to the present invention, the training steps of the information self-verification model include: Obtain raw material information from multiple samples; The data of at least one category in some sample raw material information from multiple sample raw material information are modified according to a preset error type to obtain negative raw material information samples; The information self-verification model is used to process the sample raw material information or negative raw material information to obtain the first sample probability information that the data corresponding to each category name is abnormal data, and the second sample probability information that the sample raw material information or negative raw material information is abnormal. Based on the probability information of the first sample, the probability information of the second sample, and the data corresponding to the modified category, the loss function of the information self-verification model is obtained. The information self-calibration model is trained based on the loss function of the information self-calibration model to obtain the trained information self-calibration model.

[0009] According to the present invention, obtaining the loss function of the information self-verification model includes: The graph attention sub-model of the information self-verification model determines the association weights of each category; Determine the first weight of the category of the modified data in the negative sample of raw material information based on the preset error type; The calculated weights for each category are determined based on the first weight and the associated weights. For negative samples of raw material information, the loss value of the negative samples of raw material information is determined based on the calculated weights, the first sample probability information and the second sample probability information of each category. For sample raw material information, the loss value of the sample raw material information is determined based on the first sample probability information and the second sample probability information of each category; Based on the loss value of the negative sample of raw material information and the loss value of the sample raw material information, the loss function of the information self-verification model is determined.

[0010] According to the present invention, obtaining corrected raw material information, product information, process information, technology information, scale information, and treatment facility information includes: Based on the accuracy identification results, the category verification feature information is labeled to obtain the category proofreading feature information; By using the first self-attention mechanism of the sewage discharge data association model, the verification feature information of each category of each piece of information is processed to obtain the first-level output verification feature information corresponding to each category of each piece of information. By using the feature extraction hierarchy of the pollution discharge data association model, the industry pollution discharge information is processed to obtain industry pollution discharge characteristic information; By using the cross-attention mechanism of the pollution discharge data association model, the industry pollution discharge characteristic information and the primary output verification characteristic information are processed to obtain the secondary output verification characteristic information corresponding to each category. By using the second self-attention mechanism of the sewage discharge data association model, the secondary output verification feature information corresponding to each category is processed to obtain the tertiary output verification feature information of each category. Input the three-level output correction feature information corresponding to the categories marked as abnormal into the third multi-level perception layer of the sewage data association model to obtain the corrected data; Based on the corrected data, corrected raw material information, product information, process information, technology information, scale information, and treatment facility information are obtained.

[0011] According to the present invention, the training steps of the sewage discharge data association model include: Obtain information from multiple samples; Data from at least one category in a subset of multiple sample data is modified according to a preset error type to obtain negative information samples. By processing sample information or negative samples through a sewage discharge data association model, we can obtain the first-level sample output verification feature information, second-level sample output verification feature information, and third-level sample output verification feature information corresponding to each category. The first anomaly identification probability information of each category is obtained by processing the first-level sample output verification feature information of each category through the fourth multi-layer perception layer. The fifth multi-layer perception layer processes the second-level sample output verification feature information corresponding to each category to obtain the second anomaly recognition probability information of each category. The sixth multi-layer perception layer processes the proofreading feature information of the third-level sample output corresponding to each category to obtain the third anomaly recognition probability information of each category. Through the third multi-layer perception layer, the calibration feature information of the three-level sample output corresponding to each category is processed to obtain the corrected data corresponding to each category. Based on the corrected data for each category, the data for the unmodified categories, the data for the modified categories before modification, the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information, the loss function of the sewage discharge data association model is obtained. Based on the loss function of the sewage discharge data association model, the sewage discharge data association model is trained to obtain the trained sewage discharge data association model.

[0012] According to the present invention, obtaining the loss function of the sewage discharge data association model includes: Based on the corrected data for each category and the data for the unmodified categories, obtain the data consistency loss function; Based on the corrected data for each category and the uncorrected data for the modified category, the data correction loss function is obtained. Based on the first, second, and third anomaly identification probability information for each category, the anomaly identification loss function is obtained. Based on the data consistency loss function, data correction loss function, and anomaly identification loss function, the loss function of the sewage discharge data association model is determined.

[0013] According to a second aspect of the present invention, an artificial intelligence-based system for generating an inventory of carbon emissions from industrial enterprises is provided, comprising: The acquisition module is used to acquire information on raw materials, products, processes, technologies, scale, and treatment facilities reported by enterprises. The self-verification module is used to verify the information of raw materials, products, processes, technology, scale, and treatment facilities through the information self-verification model, and determine the accuracy of the identification results of various types of information. The filtering module is used to filter industry pollution discharge information from the database based on at least one of the various information submitted by the enterprise if there is information with abnormal accuracy identification results. The calibration module is used to obtain calibrated raw material information, product information, process information, technology information, scale information, and treatment facility information by using industry pollution discharge information, various information reported by enterprises, and pollution discharge data association models. The data module is used to determine the emission data of various pollutants based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. The inventory module is used to generate an inventory of carbon emissions from industrial enterprises based on emission data of various pollutants.

[0014] According to a third aspect of the present invention, an artificial intelligence-based industrial enterprise carbon emission inventory generation device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the artificial intelligence-based industrial enterprise carbon emission inventory generation method.

[0015] According to a fourth aspect of the present invention, a computer-readable storage medium is characterized in that it stores computer program instructions thereon, which, when executed by a processor, implement the artificial intelligence-based method for generating an industrial enterprise carbon emissions inventory.

[0016] By adopting the above technical solution, the present invention can achieve the following technical effects: According to this invention, based on the raw material information, product information, process information, technological information, scale information, and treatment facility information submitted by enterprises, the accuracy identification results of various information can be determined. Industry pollution discharge information is then filtered from the database to obtain corrected raw material information, product information, process information, technological information, scale information, and treatment facility information, thereby determining the emission data of various pollutants and generating a carbon pollution emission data list for industrial enterprises. By utilizing the production and pollution discharge coefficient system, a pollutant comparison table and combined process coefficients can be established to reduce omissions of production and pollution discharge links and mandatory pollutants during enterprise reporting. Thresholds for activity levels such as product and energy consumption can be determined. The advantages of emission source statistical data can be fully utilized to establish a carbon pollution emission accounting system adapted to it and conduct automatic accounting. Furthermore, the integration of other data further improves the accuracy of the calculation results, reduces abnormal data in production activity levels, and improves the rationality of basic data. The raw material information, product information, process information, technological information, scale information, and treatment facility information submitted by enterprises can also be obtained, providing basic data for determining the accuracy identification results of various information. The existing coefficient database can be supplemented and improved. Based on the statistical coefficient system, a standardized and unified six-level classification source inventory emission coefficient database, refined to the process level, has been established, expanding the pollutant indicators of the existing emission source statistical coefficient system. Furthermore, based on the existing "four same" combination, coefficients for NH3, PM10, PM2.5, BC, and OC, along with corresponding efficiencies of different treatment facilities, have been added. Compared with traditional source inventory coefficients, this significantly expands the combinations of traditional source inventory coefficients. When training the information self-verification model, negative samples of raw material information can be manually set, and the loss values ​​of the negative samples and the sample raw material information can be obtained. This allows the determination of the loss function of the information self-verification model, which is then trained to obtain the trained information self-verification model. Considering the impact of incorrect category data on related category data, association weights have been determined. Furthermore, considering the possibility that incorrect category data is intentionally modified, a first weight has been determined, leading to the determination of the calculation weights. This ensures that the loss value of the negative raw material information samples accurately describes the error when the information self-verification model identifies incorrect raw material information. This approach improves the accuracy and relevance of training, enhancing the performance of the information self-calibration model. When training the sewage discharge data association model, data consistency loss function, data correction loss function, and anomaly identification loss function can be determined, thereby establishing the overall loss function for the sewage discharge data association model. Considering the scenarios of incorrectly correcting data in categories that do not require correction and failing to correctly correct data in categories that do require correction, data consistency loss function and data correction loss function are set. This improves the accuracy and relevance of training, enhancing the performance of the information self-calibration model.Furthermore, based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, the emission data of various pollutants can be determined, thereby generating a carbon pollution emission data list for industrial enterprises. A supplementary identification technology for the entire process of enterprise pollution control has been established, particularly capable of comprehensively identifying combined processes in the production process, achieving integrated accounting for combined processes. A verification technology for enterprise production activity levels has also been established, guiding enterprises to supplement their reporting based on the accuracy of the identification results, ensuring the completeness of major production and pollution control accounting links to the greatest extent possible. A rapid carbon pollution inventory accounting system has also been established, supplementing environmental statistics data based on the production and pollution factor database and combined processes, identifying problems in enterprise environmental statistics accounting, and quickly generating an inventory. This reduces abnormal data at the production activity level and improves the rationality of basic data.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort. Figure 1 An exemplary flowchart illustrates a method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence, according to an embodiment of the present invention. Figure 2 An exemplary flowchart illustrating the determination of the accuracy identification result of various types of information according to an embodiment of the present invention is shown; Figure 3 An exemplary flowchart illustrating the process of obtaining corrected raw material information, product information, process information, technology information, scale information, and treatment facility information according to an embodiment of the present invention is shown. Figure 4 An exemplary schematic diagram of an artificial intelligence-based industrial enterprise carbon emission data inventory generation system according to an embodiment of the present invention is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 An exemplary flowchart illustrates a method for generating an industrial enterprise carbon pollution emission data inventory based on artificial intelligence according to an embodiment of the present invention, the method comprising: Step S1: Obtain the raw material information, product information, process information, technology information, scale information, and treatment facility information submitted by the enterprise; Step S2: Using the information self-verification model, the information on raw materials, products, processes, technology, scale, and treatment facilities is verified to determine the accuracy of the identification results for various types of information. Step S3: If there is information indicating an abnormal accuracy identification result, then filter industry pollution discharge information in the database based on at least one of the multiple pieces of information submitted by the enterprise. Step S4: Obtain corrected raw material information, product information, process information, technology information, scale information, and treatment facility information by using industry pollution discharge information, various information reported by enterprises, and pollution discharge data association models; Step S5: Based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, determine the emission data of various pollutants; Step S6: Generate a list of carbon emissions data for industrial enterprises based on the emission data of various pollutants.

[0022] The AI-based method for generating an industrial enterprise carbon emission data list according to an embodiment of the present invention can determine the accuracy of various information based on the raw material information, product information, process information, technology information, scale information, and treatment facility information submitted by the enterprise. It then filters industry pollution discharge information from a database to obtain corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, thereby determining the emission data of various pollutants and generating an industrial enterprise carbon emission data list. By leveraging the production and emission coefficient system, a pollutant comparison table and combined process coefficients can be established, reducing omissions of production and emission links and mandatory pollutants during enterprise reporting. Thresholds for activity levels such as product and energy consumption can be determined, fully utilizing the advantages of emission source statistical data to establish a carbon emission accounting system adapted to it and conduct automatic accounting. Furthermore, integrating other data further improves the accuracy of the calculation results, reduces abnormal data in production activity levels, and enhances the rationality of basic data.

[0023] Example 1: According to embodiments of the present invention, addressing the technical problems of relatively low temporal accuracy and spatial resolution, and the difficulty in data application and quality control of traditional inventories, the present invention proposes a high-precision pollutant and greenhouse gas accounting method and a data inventory generation method based on current emission source statistical survey data. This method can economically, quickly, and systematically achieve the routine compilation of carbon pollution inventories, providing a new technical path for improving the scientificity and practicality of emission inventories.

[0024] According to embodiments of the present invention, existing coefficient databases can be supplemented and improved. Based on a statistical coefficient system, a standardized and unified six-level classification source inventory pollution emission coefficient database, refined to the process level, is established, expanding the pollutant indicators of the existing emission source statistical coefficient system. Regarding resolution, traditional source inventories calculate industrial carbon emissions according to pollution sources, categorizing them into fossil fuel combustion, stationary combustion sources, process sources, solvent use sources, biomass combustion, etc., resulting in low resolution. In contrast, the carbon pollution inventory proposed in this invention calculates emissions for individual survey objects; for industries such as steel and cement, it can even be refined to the process level, achieving higher spatial accuracy. Regarding the types of pollutants calculated, traditional inventories mainly involve SO2, NOx, TSP, and PM2.5. 10 PM 2.5 Some pollutants in BC, OC, and NH3 are relatively few in number, while the carbon pollution fusion inventory includes air pollutants SO2, NOx, TSP, and PM2.5. 10 PM 2.5The coverage is more comprehensive, encompassing carbon dioxide, carbon monoxide, organic carbon (OC), NH3, and greenhouse gases such as CO2, CH4, N2O, and HFCs. Regarding accounting methods and activity levels, traditional source inventories use emission coefficients, primarily based on industry-product coefficients, with some industries considering processes. The classification is relatively coarse, resulting in only about 300 production combinations. Activity levels are based on industry product output, and the number is limited. This invention's integrated carbon pollution inventory primarily uses a coefficient method, while also incorporating material balance and monitoring data methods. The calculation is based on coefficients for production stages, equipment, or processes, while also considering products, processes, and scale. The classification standards are more detailed, resulting in approximately 20,000 combinations. Enterprise data can be accurate to the process / equipment level. VOCs from organic solvent usage are also included. S The material balance algorithm is adopted, and the source-term method is used for VOCs from common sources, which is more accurate than the previous method of calculating VOCs using a uniform coefficient across the entire plant. Furthermore, accurate emission data can be calculated based on the emission coefficient database, providing training data for model training.

[0025] According to an embodiment of the present invention, in step S1, the raw material information, product information, process information, technology information, scale information, and treatment facility information reported by the enterprise are obtained. The enterprise's products, raw materials, processes, technology, scale, and treatment facilities in the emission source statistical basic data reported by the enterprise are the raw material information, product information, process information, technology information, scale information, and treatment facility information.

[0026] In the example, when obtaining raw material information, the raw material combination of the enterprise's products in the accounting process can be identified; based on the identified products, combined with the correspondence between major products and raw materials in the production and pollution discharge coefficient system, the raw materials filled in by the enterprise in the basic information table can be identified to supplement and determine the raw material combination of the products; for the raw materials corresponding to the newly identified products, combined with the correspondence between major products and raw materials in the production and pollution discharge coefficient system and the raw materials already filled in by the enterprise in the basic information table, the raw material combination of the products can be supplemented and determined; if the newly added products do not match the coefficient combination of the already filled raw materials, the raw materials with the highest filling frequency in the industry of the enterprise can be matched to supplement and determine the raw material combination of the products.

[0027] In the example, when obtaining product information, the system can identify the products that the company has already filled in in the basic information table, as well as the products involved in the accounting process; based on the correspondence between the main products and raw materials in the production and pollution discharge coefficient system and the raw materials that the company has already filled in in the basic information table, new products can be identified; for companies that have wastewater discharge accounting but no corresponding exhaust gas discharge accounting, new products can be identified in the process of accounting for the company's wastewater discharge.

[0028] In the example, when acquiring process information, the system can identify the processes for raw materials in the production process already reported by the enterprise, verify them according to the "Production and Discharge Coefficient," and supplement them if any are missing. For example, under the same combination of raw materials, the process with the highest reporting frequency nationwide is selected for supplementation. For example, for a newly added combination of raw materials, if the coefficient has only one process, that process is selected; if there are multiple processes, the process with the highest frequency is automatically matched based on big data. For enterprises whose production processes include continuous processes, that is, the production process of a certain product includes at least two production processes, and these processes need to be implemented in a specific order, the continuous processes are identified and verified. If any are missing, the missing processes are supplemented. For example, under the same combination of raw materials, the process with the highest reporting frequency nationwide is selected. Supplement the processes with the highest reporting frequency; identify whether necessary general processes should be included, such as boilers, kilns, casting, and painting. For industries and enterprises lacking necessary general processes, supplement them with general processes. For example, under the same fuel, product, or material, select the relevant parameters of the general processes with the highest reporting frequency nationwide and supplement them; identify whether necessary general source items should be included, such as particulate matter emissions from stockpiles, storage tanks, use of industrial anti-corrosion coatings, storage and loading of volatile organic liquids, dynamic and static sealing points of equipment, circulating water cooling towers, flares, and volatile organic compounds from solid material stockpiles. For industries and enterprises lacking necessary general source items, supplement them with general source items. For example, under the same material, select the relevant parameters of the general processes with the highest reporting frequency nationwide and supplement them.

[0029] In the example, when obtaining process information, scale information, and treatment facility information, the system can identify the product information, raw material information, process information, technology information, scale information, and treatment facility information already submitted by the enterprise, and supplement any missing process information, scale information, and treatment facility information. For missing process information, scale information, and treatment facility information, the system selects the corresponding coefficients based on the most frequently submitted scale, process, and treatment facility efficiency of enterprises nationwide under the same product raw material process. For missing only scale information and treatment facility information, the system selects the corresponding coefficients based on the most frequently submitted scale and treatment facility efficiency of enterprises nationwide under the same product raw material process. For missing only treatment facility information, the system selects the corresponding coefficient based on the most frequently submitted treatment facility efficiency of enterprises nationwide under the same product raw material process production scale.

[0030] According to embodiments of the present invention, based on the basic statistical data of emission sources submitted by enterprises, and after the above preliminary verification, information on raw materials, products, processes, technologies, scale, and treatment facilities can be obtained. Furthermore, after integration, this information can be used to calculate the discharge volume of various pollutants.

[0031] This method allows for the acquisition of information submitted by enterprises, including raw material information, product information, process information, technological information, scale information, and treatment facility information. It provides fundamental data for determining the accuracy of identification results based on this diverse range of information.

[0032] Example 2: Figure 2 A flowchart illustrating the accuracy identification results of various types of information according to an embodiment of the present invention is shown as an example.

[0033] According to an embodiment of the present invention, in step S2, the information self-verification model is used to verify the raw material information, product information, process information, technology information, scale information, and treatment facility information respectively, and to determine the accuracy identification results of various information, including: using the text recognition sub-model of the information self-verification model to extract features from multiple category names in the raw material information to obtain category name feature information; concatenating the data corresponding to the category name with the category name feature information to obtain category verification feature information; processing the verification feature information of each category using the graph attention sub-model of the information self-verification model to obtain verification feature information of each category; concatenating the verification feature information of each category to obtain raw material information verification feature information; processing the verification feature information of each category using the first multi-layer perception layer of the information self-verification model to obtain first probability information that the data corresponding to each category name is abnormal data; processing the verification feature information of the raw material information using the second multi-layer perception layer of the information self-verification model to obtain second probability information that the raw material information is abnormal; and determining the accuracy identification result of the raw material information based on the first probability information and the second probability information.

[0034] According to an embodiment of the present invention, by using a text recognition sub-model of an information self-verification model (e.g., BERT, word embedding model, etc.), features are extracted from multiple category names (e.g., raw material name, consumption quantity, etc.) in the raw material information, resulting in a vector, which is the category name feature information, describing the features of multiple category names in the raw material information. Further, by concatenating the data corresponding to the category name with the category name feature information, new vectors corresponding to each category name are obtained, which are the category verification feature information, describing the features of each category name and data.

[0035] According to an embodiment of the present invention, the verification feature information of each category is processed by a graph attention sub-model of the information self-verification model to obtain the verification feature information of each category. When processing the verification feature information of each category using the graph attention sub-model of the information self-verification model, each category verification feature information can be treated as a node in a graph structure, and the connection weights between each node can be obtained, which can describe the tightness of the relationship between the verification feature information of each category. When determining the connection weights, the category verification feature information corresponding to two nodes with a connection relationship is concatenated, and the corresponding value is obtained through the fully connected layer of the graph attention sub-model, which can be used as the connection weight. The parameters of the fully connected layer can be determined through training. Each category's verification feature information, along with the category verification feature information connected to a certain category's verification feature information, is weighted and summed using corresponding connection weights. This yields an aggregated vector of the category verification feature information, which is the processed category verification feature information (i.e., the node's output vector), and thus the verification feature information. This aggregates the features of the category itself, as well as features from other categories, providing a more comprehensive description of the category's characteristics. By fusing the features of various categories, it can more comprehensively describe the content, meaning, and other features of the category. The above processing can be performed on each category's verification feature information to obtain the verification feature information for each category. Furthermore, the verification feature information of each category is concatenated to obtain the raw material information verification feature information. The characteristics of the raw material information can be described from its various categories.

[0036] According to an embodiment of the present invention, the verification feature information of each category is processed by the first multi-layer perception layer of the information self-verification model (e.g., including a fully connected layer and an activation layer, wherein the activation layer is a network layer processed using the sigmoid activation function). This allows the probability information that the data corresponding to each category name is abnormal data to be obtained, which is the first probability information. For example, the probability of an abnormal raw material name is 0.2, and the probability of an abnormal raw material consumption is 0.6. Similar to obtaining the first probability information, the verification feature information of the raw material information can be processed by the second multi-layer perception layer of the information self-verification model to obtain the probability of abnormal raw material information, which is the second probability information.

[0037] According to an embodiment of the present invention, the accuracy identification result of raw material information is determined based on first probability information and second probability information. The accuracy identification result of raw material information may include the first probability information, the second probability information, and whether the raw material information is abnormal, and whether the data corresponding to each category name in the raw material information is abnormal. For example, when the first probability information is greater than 0.5, the data corresponding to the category name of the raw material information can be considered abnormal data; when the second probability information is greater than 0.5, the raw material information as a whole can be considered abnormal. This method can determine whether there are any abnormalities in the reporting of raw material information, and whether there are any abnormalities in the reporting of raw material information category names and data. By judging whether there are any abnormalities in the reporting of raw material information from the whole to the part and performing mutual verification, the accuracy of the accuracy identification result is improved. Based on the same processing method, the accuracy identification results of product information, process information, technology information, scale information, and treatment facility information can be obtained.

[0038] Example 3: According to an embodiment of the present invention, the training steps of the information self-verification model include: acquiring multiple sets of sample raw material information; modifying the data of at least one category in some sample raw material information according to a preset error type to obtain negative samples of raw material information; processing the sample raw material information or negative samples of raw material information through the information self-verification model to obtain first sample probability information that the data corresponding to each category name is abnormal data, and second sample probability information that the sample raw material information or negative samples of raw material information are abnormal; obtaining the loss function of the information self-verification model based on the first sample probability information, the second sample probability information, and the data corresponding to the modified categories; and training the information self-verification model based on the loss function of the information self-verification model to obtain the trained information self-verification model.

[0039] According to an embodiment of the present invention, multiple sets of previously verified correct raw material information are obtained, which constitute the sample raw material information. Data for at least one category in a portion of the sample raw material information is modified according to a preset error type (e.g., category name error, consumption error, process omission, data unit error, etc., and the preset error type includes error situations for various categories of data), artificially setting erroneous sample raw material information, which constitutes the negative sample of raw material information. Similar to obtaining the first probability information and the second probability information, the sample raw material information or the negative sample of raw material information is processed through an information self-verification model to obtain first sample probability information that the data corresponding to each category name is abnormal data, and second sample probability information that the sample raw material information or the negative sample of raw material information is abnormal.

[0040] Example 4: According to an embodiment of the present invention, the loss function of the information self-verification model is obtained based on the first sample probability information, the second sample probability information, and the data corresponding to the modified category. This includes: obtaining the association weights of each category determined by the graph attention sub-model of the information self-verification model; determining the first weight of the category of the modified data in the negative raw material information sample based on a preset error type; determining the calculated weight of each category based on the first weight and the association weight; for the negative raw material information sample, determining the loss value of the negative raw material information sample based on the calculated weights, the first sample probability information, and the second sample probability information of each category; for the sample raw material information, determining the loss value of the sample raw material information based on the first sample probability information and the second sample probability information of each category; and determining the loss function of the information self-verification model based on the loss value of the negative raw material information sample and the loss value of the sample raw material information.

[0041] According to an embodiment of the present invention, the association weights of each category determined by the graph attention sub-model of the information self-verification model are obtained. The graph attention sub-model of the information self-verification model processes the verification feature information of each category, and treats each category verification feature information as a node in a graph structure. The connection weights between each node can then be obtained, which are the association weights. These weights describe the tightness of the relationship between categories; in other words, they describe the degree of influence of data from an incorrect category on data from related categories. When determining the association weights, the category verification feature information corresponding to two nodes with a connection relationship is concatenated and processed through the fully connected layer of the graph attention sub-model to obtain the corresponding value, which can be used as the association weight.

[0042] According to an embodiment of the present invention, a first weight for the category of modified data in the negative sample of raw material information is determined based on a preset error type. This first weight can be a preset weight, set according to the probability that the erroneous category data was intentionally modified. For example, in the preset error types, the probability that the unit of consumption was intentionally modified is low, so the first weight for the unit error of consumption can be set to 0.4; the probability that consumption (intentionally lowering the raw material consumption can significantly reduce carbon emissions) was intentionally modified is high, so the first weight for the consumption error can be set to 0.8. Further, by multiplying the first weight and the associated weight of each category, the calculated weight of each category can be obtained. If multiple categories are modified, the weights can be calculated as follows: For example, if the weight of the first modified category is 0.8, the weight of the second category is 0.6, and the association weight between the two categories is 0.5, then the weight of the first modified category is 0.8 + 0.6 × 0.5 = 1.1, and the weight of the second modified category is 0.6 + 0.5 × 0.8 = 1. The weights of other categories can be the sum of the weights affected by the modified categories, that is, the sum of the first weight of each modified category and the product of the association weight between the modified category and that category. For example, if the association weight between a category and the first modified category is 0.4, and the association weight between a category and the second modified category is 0.6, then the weight of that category is 0.8 × 0.4 + 0.6 × 0.6 = 0.68.

[0043] According to an embodiment of the present invention, for a negative sample of raw material information, the loss value of the negative sample of raw material information is determined based on the calculated weights, the first sample probability information and the second sample probability information of each category in the negative sample of raw material information being abnormal. The first sample annotation probability information of the data corresponding to each category name in the negative sample of raw material information being abnormal, and the second sample annotation probability information of the negative sample of raw material information being abnormal, can be manually labeled. Since the data corresponding to the modified categories in the negative sample of raw material information has been manually modified to erroneous data, the first sample probability labeling information of each modified category is 1, that is, the data is abnormal; the first sample probability labeling information of the unmodified categories is 0, and the second sample annotation probability information of the negative sample of raw material information being abnormal is also 1. Furthermore, the cross-entropy loss function based on the first sample probability labeling information and the first sample probability information corresponding to each category can be obtained, as well as the cross-entropy loss function based on the second sample probability labeling information and the second sample probability information corresponding to the negative raw material information sample. By calculating the weights of each modified category (no weight is calculated for unmodified categories), the cross-entropy loss functions based on the first sample probability labeling information and the first sample probability information corresponding to each category are weighted and summed to obtain the cross-entropy loss function based on the first sample probability labeling information and the first sample probability information corresponding to the negative raw material information sample. Then, the cross-entropy loss function based on the first sample probability labeling information and the first sample probability information corresponding to the negative raw material information sample is weighted and summed with the cross-entropy loss function based on the second sample probability labeling information and the second sample probability information corresponding to the negative raw material information sample to obtain the loss value of the negative raw material information sample, which can describe the accuracy of the information self-checking model in judging the erroneous raw material information.

[0044] According to an embodiment of the present invention, for sample raw material information, the loss value of the sample raw material information is determined based on the first sample probability information and the second sample probability information of each category. Similar to obtaining the loss value of negative samples of raw material information, since all raw material information in the sample raw material information is correct, the first sample probability labeling information of each category is 0, that is, the data is normal, and the second sample labeling probability information of abnormal raw material information is also 0. Further, the cross-entropy loss function based on the first sample probability labeling information and the first sample probability information corresponding to each category can be obtained and summed to obtain the cross-entropy loss function based on the first sample probability labeling information and the first sample probability information corresponding to the sample raw material information. Then, the cross-entropy loss function based on the first sample probability labeling information and the first sample probability information corresponding to the sample raw material information is weighted and summed with the cross-entropy loss function based on the second sample probability labeling information and the second sample probability information corresponding to the sample raw material information to obtain the loss value of the sample raw material information, which can describe the accuracy of the information self-verification model in judging correct raw material information. The loss function corresponding to the raw material information is obtained by summing the loss value of the negative samples of raw material information and the loss value of the sample raw material information.

[0045] According to an embodiment of the present invention, the information self-verification model is trained based on the loss function of the information self-verification model to obtain the trained information self-verification model. Backpropagation can be performed using the loss function of the aforementioned information self-verification model, and the parameters of the information self-verification model can be adjusted using gradient descent to train the information self-verification model. After multiple training iterations (i.e., training using multiple sets of sample raw material information and multiple sets of negative samples of raw material information), training is completed, and the trained information self-verification model is obtained. Based on the same processing method, the loss functions of the information self-verification models corresponding to product information, process information, technological information, scale information, and treatment facility information can be obtained. Furthermore, the loss functions of the information self-verification models corresponding to various types of information can be used to train information self-verification models corresponding to various types of information.

[0046] In this way, when training the information self-verification model, negative samples of raw material information can be manually set, and the loss values ​​of the negative samples and the sample raw material information can be obtained. This allows the loss function of the information self-verification model to be determined, and training can then be performed to obtain the trained information self-verification model. Considering the impact of data in an incorrect category on data in related categories, association weights are determined, and the first weight is determined based on the error type, thus determining the calculation weights. This ensures that the loss value of the negative raw material information samples accurately describes the error when the information self-verification model identifies incorrect raw material information. This improves the accuracy and relevance of training, enhancing the performance of the information self-verification model.

[0047] Example 5: According to an embodiment of the present invention, in step S3, if there is information indicating an abnormal accuracy identification result, industry pollution discharge information is filtered from the database based on at least one of the various information submitted by the enterprise. For example, if the raw material information submitted by the industry is that the raw material name is lime, etc., and the product information is that the product information is cement, etc., and the enterprise belongs to the cement manufacturing industry, then various information of the cement manufacturing industry can be filtered from the database. For example, the accurate data, raw material categories, basic processes and technologies, equipment, and various industry coefficients (such as pollution discharge coefficients) submitted by multiple enterprises in the same industry constitute the industry pollution discharge information.

[0048] Example 6: Figure 3 An exemplary flowchart illustrating the process of obtaining corrected raw material information, product information, process information, technology information, scale information, and treatment facility information according to an embodiment of the present invention is shown.

[0049] According to an embodiment of the present invention, in step S4, corrected raw material information, product information, process information, technological information, scale information, and treatment facility information are obtained through industry pollution discharge information, various information reported by enterprises, and a pollution discharge data association model. This includes: labeling category verification feature information based on accuracy identification results to obtain category correction feature information; processing the category correction feature information of each piece of information through the first self-attention mechanism of the pollution discharge data association model to obtain first-level output correction feature information corresponding to each category of each piece of information; and processing industry pollution discharge information through the feature extraction hierarchy of the pollution discharge data association model to obtain industry pollution discharge features. Information; through the cross-attention mechanism of the wastewater discharge data association model, the industry wastewater discharge characteristic information and the first-level output verification characteristic information are processed to obtain the second-level output verification characteristic information corresponding to each category; through the second self-attention mechanism of the wastewater discharge data association model, the second-level output verification characteristic information corresponding to each category is processed to obtain the third-level output verification characteristic information of each category; the third-level output verification characteristic information corresponding to the categories marked as abnormal is input into the third multi-layer perception level of the wastewater discharge data association model to obtain the corrected data; based on the corrected data, the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information are obtained.

[0050] According to embodiments of the present invention, category verification feature information is annotated based on the accuracy identification result to obtain category correction feature information. For example, in the accuracy identification result, when the data corresponding to the category name is abnormal data, the category verification feature information corresponding to that category can be annotated as 1. For example, if the category verification feature information is [1,2,3], and the accuracy identification result of the category verification feature information is that the data corresponding to the category name is abnormal data, the annotated category feature information is [1,2,3,1], which is the category correction feature information corresponding to the category verification feature information. The same processing method can be used to obtain the category correction feature information corresponding to each category of each piece of information.

[0051] According to an embodiment of the present invention, by multiplying the Q-matrix and K-matrix in the first self-attention mechanism of the sewage discharge data association model with the category-specific feature information of each piece of information, the category-specific feature information can be projected into an attention space more suitable for computation, thereby obtaining the query vector and key-value vector corresponding to each category-specific feature information. Since the Q-matrix and K-matrix have the same shape, the query vector and key-value vector also have the same dimension, thus enabling the dot product operation between the query vector and the key-value vector. Furthermore, the query vector corresponding to the y-th category correction feature information of the x-th information is multiplied (i.e., a dot product operation) by the key-value vectors corresponding to the category correction feature information of each information (including the y-th category correction feature information of the x-th information itself), and then divided by the square root of the dimension of the key-value vector. After processing by an activation function (softmax activation function), the relevant weights corresponding to the category correction feature information of each information can be obtained. These weights describe the degree of correlation between the category correction feature information of each information and the y-th category correction feature information of the x-th information, and also describe the degree of correlation between the data of each category of each information and the data of the y-th category of the x-th information. The larger the relevant weights, the higher the degree of correlation and importance of the category correction feature information of each information with the y-th category correction feature information of the x-th information. Of course, the Q-matrix and K-matrix are learnable relevant weight matrices, and their parameters can be determined through training.

[0052] According to an embodiment of the present invention, by multiplying the category-specific correction feature information of each piece of information with the V matrix of the first self-attention mechanism, the weight vector corresponding to the category-specific correction feature information of each piece of information can be obtained. The V matrix is ​​a learnable weight matrix, and its parameters can be determined through training. Furthermore, by using the relevant weights corresponding to the category-specific proofreading feature information of each piece of information, and then weighting and summing the weight vectors corresponding to the category-specific proofreading feature information of each piece of information, a new vector is obtained, which is the first-level output proofreading feature information. For example, using the relevant weights between the y-th category-specific proofreading feature information of the x-th piece of information and other category-specific proofreading feature information, the new vector corresponding to the y-th category-specific proofreading feature information of the x-th piece of information is weighted and summed with the weight vectors corresponding to the other category-specific proofreading feature information, resulting in the first-level output proofreading feature information of the y-th category-specific proofreading feature information of the x-th piece of information. This allows for the fusion of features of the category-specific data corresponding to the category-specific proofreading feature information of each piece of information, describing the relationship between the features of the category-specific data corresponding to the y-th category-specific proofreading feature information of the x-th piece of information and other category features, comprehensively describing the influence of other category data on the y-th category data of the x-th piece of information, and improving the accuracy and comprehensiveness of the feature representation of the y-th category data of the x-th piece of information. Based on the same processing method, the first-level output proofreading feature information corresponding to each category of each piece of information can be obtained.

[0053] According to an embodiment of the present invention, by processing industry pollution discharge information through the feature extraction layers of the pollution discharge data association model (e.g., including multiple fully connected layers and activation layers, where the activation layer is a feature extraction layer processed using the ReLU activation function), a high-dimensional (e.g., 512-dimensional) vector can be obtained, which is the industry pollution discharge feature information. This vector can describe the features of various categories of data (i.e., standard data of various categories of various information) of the industry's raw material information, product information, process information, technology information, scale information, and treatment facility information.

[0054] According to an embodiment of the present invention, the Q-matrix in the cross-attention mechanism of the sewage discharge data association model is the query matrix. Multiplying the first-level output correction feature information by the query matrix yields a transformed vector, which is the query vector. The K-matrix in the cross-attention mechanism is the key-value matrix. Multiplying the industry sewage discharge feature information by the key-value matrix yields a transformed vector, which is the key-value vector. Based on the same processing method, the key-value vector corresponding to each industry sewage discharge feature information can be obtained. Since the query matrix and the key-value matrix have the same shape, the query vector and the key-value vector also have the same dimension, thus allowing the dot product operation to be performed on the query vector and the key-value vector. Furthermore, by multiplying the query vector by the key-value vector corresponding to each industry's pollution discharge feature information (i.e., performing a dot product operation), and then dividing by the square root of the key-value vector's dimension, the weight of each industry's pollution discharge feature information can be obtained. This weight, or relevance weight, describes the degree of correlation between each industry's pollution discharge feature information and the first-level output calibration feature information; that is, the degree of correlation between each industry's pollution discharge information and the data of each category of information. The larger the relevance weight, the higher the relevance between the industry's pollution discharge information and the category data, and the higher its importance. Of course, both the query matrix and the key-value matrix are learnable weight matrices, and their parameters can be determined through training.

[0055] According to an embodiment of the present invention, by multiplying the pollution discharge characteristic information of each industry with the weight matrix of the cross-attention mechanism (i.e., the V matrix of the cross-attention mechanism), a weight vector corresponding to each industry's pollution discharge characteristic information can be obtained. The first weight matrix is ​​a learnable weight matrix, and its parameters can be determined through training. Further, similar to obtaining the first-level output verification characteristic information, by weighting and summing multiple weight vectors using the relevant weights corresponding to the pollution discharge characteristic information of each industry, a new vector can be obtained, which is the second-level output verification characteristic information. Based on the same processing method, the second-level output verification characteristic information corresponding to each category can be obtained. Based on the first-level output verification characteristic information, the characteristics of industry pollution discharge information can be fused to describe the characteristics of the data of each category.

[0056] According to an embodiment of the present invention, similar to obtaining the first output verification feature information, the second self-attention mechanism of the sewage discharge data association model is used to process the secondary output verification feature information corresponding to each category to obtain the tertiary output verification feature information for each category. This allows for the fusion of features from all categories of data based on the integration of industry sewage discharge information, thereby improving the accuracy and comprehensiveness of feature representation for each category of data.

[0057] According to an embodiment of the present invention, the three-level output correction feature information corresponding to the categories marked as abnormal is input into the third multi-layer perception layer of the sewage discharge data association model (e.g., including a fully connected layer and an activation layer, the activation layer being a network layer processed using the ReLU activation function) for processing, thereby obtaining the corrected data for each abnormal category, which is the corrected data. Further, based on the corrected data, the raw material information, product information, process information, technology information, scale information, and treatment facility information are modified to obtain the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. In the example, enterprises may lack some crucial information when filling out the form, making it impossible to accurately identify process information, scale information, treatment facility information, etc. For example, if a company has two or more production lines, and the first production line underreports the pollutants from clinker to cement, and the second production line underreports the pollutants from clinker production, and the thermal power plant underreports the essential boiler process, and the product output is incorrectly reported in units, then based on the above correction methods, the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information can be obtained. For example, it can automatically correct abnormally high and low values ​​of single indicator data, abnormal values ​​that cause abnormal fluctuation trends in activity levels, differences in the same indicator between different data tables, and abnormal activity level data for different pollutants.

[0058] According to an embodiment of the present invention, in the example, when determining raw material information, product information, process information, technological information, scale information, and treatment facility information, a basic table of thresholds for enterprise activity levels (e.g., product output, energy consumption, and raw material usage) for different industries and scales can be established based on historical big data and statistical methods. This table serves as part of the industry's pollution discharge information. When enterprises report their activity levels, the system can automatically match the value range and historical year data, alerting them to outliers. Abnormally high or low values ​​are replaced according to the variation of indicators such as product output, total industrial output value, and raw material quantity. Furthermore, the system can verify the product output reported in the industry-specific tables for boilers, cement, coking, and steel industries in the emission source statistics against the basic information tables reported by enterprises. For example, it can verify coal consumption, clinker output, coke output, key raw and auxiliary materials involving volatile organic compounds, and materials stored in storage tanks. When correcting for significant deviations in the same indicator across different tables, the maximum value within the threshold range is used. For products with multiple production lines, and where the sum of product outputs used in the pollutant accounting table differs significantly from the product outputs in the basic information entry form, the difference is supplemented according to the proportion of product outputs from each production line. The accounting activity levels for different pollutants within the pollutant accounting table can also be extracted and corrected. For those with significant deviations, the highest frequency data is used as the replacement; for example, if the frequencies are the same, the maximum value is used.

[0059] Example 7: According to an embodiment of the present invention, the training steps of the sewage discharge data association model include: acquiring multiple sets of sample information; modifying the data of at least one category in a portion of the sample information according to a preset error type to obtain negative information samples; processing the sample information or negative information samples through the sewage discharge data association model to obtain first-level sample output correction feature information, second-level sample output correction feature information, and third-level sample output correction feature information corresponding to each category; processing the first-level sample output correction feature information corresponding to each category through a fourth multi-layer perception layer to obtain first anomaly identification probability information for each category; and processing the second-level sample output correction feature information corresponding to each category through a fifth multi-layer perception layer to obtain the first anomaly identification probability information for each category. The second anomaly identification probability information; the third anomaly identification probability information for each category is obtained by processing the third-level sample output calibration feature information corresponding to each category through the sixth multi-layer perception layer; the third anomaly identification probability information for each category is obtained by processing the third-level sample output calibration feature information corresponding to each category through the third multi-layer perception layer; the corrected data for each category is obtained by processing the corrected data for each category, the data of the unmodified category, the data of the modified category before modification, the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information; the loss function of the sewage discharge data association model is obtained by training the sewage discharge data association model based on the loss function of the sewage discharge data association model, and the trained sewage discharge data association model is obtained.

[0060] According to embodiments of the present invention, similar to obtaining multiple sample raw material information, multiple sample information can be obtained, and the data of at least one category in some sample information of the multiple sample information can be modified according to a preset error type to obtain negative information samples. Similar to obtaining primary output calibration feature information, secondary output calibration feature information, and tertiary output calibration feature information, the sample information or negative information samples are processed through a sewage discharge data association model to obtain primary sample output calibration feature information, secondary sample output calibration feature information, and tertiary sample output calibration feature information corresponding to each category. Furthermore, by processing the first-level sample output calibration feature information corresponding to each category through the fourth multi-layer perceptron (e.g., including fully connected layers and activation layers, where the activation layers are network layers processed using the sigmoid activation function), the probability of anomalies in each category can be obtained, which is the first anomaly identification probability information. Similarly, the second-level sample output calibration feature information corresponding to each category can be processed through the fifth multi-layer perceptron to obtain the second anomaly identification probability information for each category, and the third-level sample output calibration feature information corresponding to each category can be processed through the sixth multi-layer perceptron to obtain the third anomaly identification probability information for each category. Similar to obtaining the corrected data, the corrected data corresponding to each category can be obtained by processing the third-level sample output calibration feature information corresponding to each category through the third multi-layer perceptron.

[0061] Example 8: According to an embodiment of the present invention, obtaining the loss function of the sewage discharge data association model based on the corrected data corresponding to each category, the data of the unmodified category, the data of the modified category before modification, the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information includes: obtaining a data consistency loss function based on the corrected data corresponding to each category and the data of the unmodified category; obtaining a data correction loss function based on the corrected data corresponding to each category and the data of the modified category before modification; obtaining an anomaly identification loss function based on the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information of each category; and determining the loss function of the sewage discharge data association model based on the data consistency loss function, the data correction loss function, and the anomaly identification loss function.

[0062] According to an embodiment of the present invention, a data consistency loss function is obtained based on the corrected data and the unmodified data corresponding to each category. Since the unmodified data should not be corrected if the correction is correct, it should remain unchanged. Therefore, if the unmodified data is corrected, an error will occur; that is, data for categories that do not need correction will be incorrectly corrected. For example, the unmodified data for a production volume of 1500 tons is correct and does not require correction. However, after correcting the data for each category, the production volume is incorrectly corrected to 1400 tons, resulting in an error. Therefore, the average absolute error or root mean square error between the corrected data and the unmodified data corresponding to each category can be determined as the data consistency loss function, which describes the accuracy of maintaining correct category data after correction.

[0063] According to an embodiment of the present invention, a data correction loss function is obtained based on the corrected data corresponding to each category and the uncorrected data of the modified category. Since the data of the modified category, if correctly corrected, should be corrected to the data before modification, an error will occur if the data of the modified category is corrected but not to the data before modification in the corrected data corresponding to each category. That is, the data of the category that needs correction has not been correctly corrected. For example, if the data of the modified category is a production volume of 1400 tons, the data before modification is a production volume of 1500 tons, and the corrected data is a production volume of 1450 tons, an error has occurred compared to the data before modification. Therefore, the average absolute error or root mean square error between the corrected data of each category and the data before modification of the modified category can be determined as the data correction loss function, which can describe the accuracy of correctly correcting the erroneous category data after correction.

[0064] According to embodiments of the present invention, an anomaly identification loss function is obtained based on the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information for each category. The first anomaly identification label probability information for anomalous categories can be determined as 1, and the first anomaly identification label probability information for normal categories can be determined as 0. Further, the cross-entropy loss function based on the first anomaly identification label probability information and the first anomaly identification probability information for each category can be obtained, and summed to obtain the cross-entropy loss function corresponding to the first-level output correction feature information, which can describe the accuracy of the output first-level output correction feature information. Similarly, based on the second and third anomaly identification probability information, the cross-entropy loss function corresponding to the second-level and third-level output correction feature information can be obtained, respectively describing the accuracy of the output second-level and third-level output correction feature information. Summing the above three cross-entropy loss functions yields the anomaly identification loss function, which can describe the accuracy of the data for anomaly-identified categories. Improving the accuracy of the first, second, and third anomaly identification probability information during training can promote the improvement of the accuracy of the corrected data. Furthermore, by performing a weighted summation of the aforementioned data consistency loss function, data correction loss function, and anomaly identification loss function, the loss function of the sewage discharge data association model can be obtained.

[0065] According to an embodiment of the present invention, the sewage data association model is trained based on the loss function of the sewage data association model to obtain the trained sewage data association model. Backpropagation can be performed using the loss function of the sewage data association model, and the parameters of the sewage data association model can be adjusted using gradient descent to train the model. After multiple training iterations (i.e., training using corrected data corresponding to multiple categories, data of unmodified categories, data of modified categories before modification, first anomaly identification probability information, second anomaly identification probability information, and third anomaly identification probability information), training is completed, and the trained sewage data association model is obtained.

[0066] In this way, when training the sewage discharge data association model, the data consistency loss function, data correction loss function, and anomaly identification loss function can be determined, thereby determining the loss function of the sewage discharge data association model. Considering the scenarios where data of categories that do not require correction are incorrectly corrected, and data of categories that require correction are not correctly corrected, data consistency loss functions and data correction loss functions are set. This improves the accuracy and relevance of training and enhances the performance of the information self-verification model.

[0067] Example 9: According to an embodiment of the present invention, in step S5, emission data for multiple pollutants are determined based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. The emission data for multiple pollutants can be determined using the formula: Pollutant emission data = Pollutant generation amount × (1 - Removal efficiency), where Pollutant generation amount = Product output in the corrected product information (or raw material consumption in the raw material information) × Pollution generation coefficient. For example, a thermal power plant burns coal to generate electricity and heat. The raw material information shows a coal consumption of 500,000 tons, an average dry ash-free volatile matter content greater than 37%, low-NOx combustion is used, and high-efficiency selective catalytic reduction (SCR) is used for denitrification. A query reveals that the NOx generation coefficient of this boiler is 1.78 kg / ton of fuel, and the NOx treatment facility efficiency is 83%. Therefore, the NOx generation amount is 50 × 10⁻⁶ tons. 4 The nitrogen oxide emissions are 890 tons × 1.78 kg / ton, or 890 tons, and the nitrogen oxide emissions are 890 × (1-0.83), or 151.3 tons.

[0068] Based on the same processing method, the emission data of various pollutants in multiple processes of a thermal power plant can be determined according to the corrected raw material information, product information, process information, technological information, scale information, and treatment facility information. In the example, the facilities used are also different; for example, the quality of the coal used, the size of the boiler, and the sophistication of the facilities are all different. Therefore, different companies use different coefficients, which can then be found in a coefficient database. For example, a cement company, whose raw material information is identified as cement and clinker, will have a higher pollution coefficient and generate more pollutants due to the larger pollutant content of the clinker production process compared to cement grinding. Furthermore, a monitoring and inspection correction coefficient has been added, which can be used to adjust the removal efficiency based on the daily inspections and supervision assistance provided by environmental protection departments, within the existing system.

[0069] Example 10: According to an embodiment of the present invention, in step S6, an industrial enterprise carbon emission data list is generated based on the emission data of various pollutants. Based on the emission data of various pollutants, a carbon emission data list of various industrial enterprises, including information such as the emission amount of pollutants, can be generated. In the example, various industrial enterprise carbon emission data lists can be directly imported through an EXCEL template, seamlessly integrated with historical industrial enterprise carbon emission data lists. Furthermore, based on the industrial enterprise carbon emission data list, the calculation results of pollutant generation and emission of individual industrial enterprises can be compared and verified with emission source statistics. The calculation results of pollutant generation and emission of enterprises in major, medium, and minor industries can be compared and verified with emission source statistics, industry association statistics, and other social statistical data. The calculation results of regional pollutant generation and emission can be compared and verified with emission source statistics. Furthermore, key products and pollution discharge links in important industries (e.g., power generation, coking, sintering, clinker, organic chemicals, etc.) can be compared and verified. In addition, by combining industry and product information, enterprises can automatically generate a set of forms. Using a guided filling method, based on the enterprise's industry and main products, the enterprise is guided to choose whether to produce or purchase important raw and auxiliary materials in-house, thereby determining the enterprise's production and pollution discharge links. The system automatically assigns the production and pollution discharge links to the corresponding forms for filling, and can automatically trigger greenhouse gas inventory calculations based on the products filled in by the enterprise.

[0070] In this way, based on corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, the emission data of various pollutants can be determined, thereby generating a carbon emission data list for industrial enterprises. A comprehensive identification and supplementary technology for all aspects of enterprise pollution control has been established, particularly capable of comprehensively identifying combined processes in the production process, achieving integrated accounting for combined processes. Furthermore, a technology for verifying enterprise production activity levels has been established, guiding enterprises to supplement their reporting based on accurate identification results, ensuring the completeness of major production and pollution control accounting links to the greatest extent possible. A rapid carbon pollution inventory accounting system has also been established, supplementing environmental statistics data based on a production and pollution factor database and combined processes, identifying problems in enterprise environmental statistics accounting, and quickly generating an inventory. This reduces abnormal data at the production activity level and improves the rationality of basic data.

[0071] The AI-based industrial enterprise carbon emission data inventory generation method according to embodiments of the present invention can integrate and establish a comprehensive emission source statistics fusion inventory production and emission coefficient system. It can supplement and verify the main products produced by industrial enterprises to ensure the integrity of the production process. Based on industry characteristics and big data, it establishes a corresponding matching relationship between main products and key products in different industries, enabling the review and supplementary verification of missing or omitted product items during data collection and reporting, ensuring the integrity of the production and emission accounting process. It can also supplement and verify the emissions of essential processes or equipment in the production process of industrial enterprises to ensure the integrity of the production process. It comprehensively reviews the production and emission coefficient system, and, combined with essential production processes, studies and establishes combined emission coefficients to ensure the completeness of the accounting process. Furthermore, it can determine the value range of production activity level data, eliminate outliers, and, based on historical big data, use statistical methods to determine reasonable value ranges for main products, energy consumption, and raw material usage of enterprises of different industries and sizes, providing accurate training and verification data for model training. Based on the raw material, product, process, technology, scale, and treatment facility information submitted by enterprises, the system can determine the accuracy of various information sources. It then filters industry pollution discharge information from the database to obtain corrected raw material, product, process, technology, scale, and treatment facility information, thereby determining the emission data for multiple pollutants and generating a list of industrial enterprise carbon emissions. Utilizing the production and emission coefficient system, it can establish pollutant comparison tables and combine process coefficients, reducing omissions of production and emission links and mandatory pollutants during enterprise reporting. It can also determine data thresholds for activity levels such as product and energy consumption, fully leveraging the advantages of emission source statistical data to establish a compatible carbon emission accounting system and conduct automatic calculations. Furthermore, it integrates other data to further improve the accuracy of calculation results, reduce abnormal data in production activity levels, and enhance the rationality of basic data. It can also obtain the raw material, product, process, technology, scale, and treatment facility information submitted by enterprises, providing basic data for determining the accuracy of various information sources. The existing coefficient database can be supplemented and improved. Based on the statistical coefficient system, a standardized and unified six-level classification source inventory pollution coefficient database, refined to the process level, has been established, expanding the pollutant indicators of the existing emission source statistical coefficient system. Furthermore, based on the existing "four same" combination, NH3 and PM2.5 have been added. 10 PM 2.5The model incorporates BC, OC coefficients and their corresponding efficiencies for different treatment facilities. Compared to traditional source inventory coefficients, it significantly expands the combinations of traditional source inventory coefficients. When training the information self-verification model, negative samples of raw material information can be manually set, and the loss values ​​of these negative samples and the sample raw material information can be obtained. This allows for the determination of the loss function of the information self-verification model, followed by training to obtain the trained model. Considering the impact of incorrect category data on related categories, association weights are determined. Furthermore, considering the possibility of intentionally modified incorrect category data, a first weight is determined, leading to the determination of the calculation weights. This ensures that the loss value of negative raw material information samples accurately describes the error in the information self-verification model's identification of incorrect raw material information. This improves the accuracy and relevance of training, enhancing the performance of the information self-verification model. When training the wastewater discharge data association model, data consistency loss function, data correction loss function, and anomaly identification loss function can be determined, thereby establishing the loss function of the wastewater discharge data association model. To address the issues of incorrectly correcting data in categories that do not require correction and failing to correctly correct data in categories that do require correction, a data consistency loss function and a data correction loss function were established. This improved the accuracy and relevance of training and enhanced the performance of the information self-verification model. Furthermore, based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, the emission data of various pollutants can be determined, thereby generating a carbon emission data list for industrial enterprises. A technology for identifying and supplementing the entire process of enterprise pollution discharge accounting was established, particularly capable of comprehensively identifying combined processes in the production process, achieving integrated accounting for combined processes. A technology for verifying the enterprise's production activity level was also established, guiding enterprises to supplement their reporting based on the accuracy of the identification results, ensuring the completeness of the main production and pollution discharge accounting links to the greatest extent possible. A rapid carbon pollution inventory accounting system was also established, supplementing the environmental statistics data based on the production and pollution discharge factor library and combined processes, identifying problems in the enterprise's environmental statistics accounting, and quickly generating an inventory. This reduces abnormal data at the production activity level and improves the rationality of the basic data.

[0072] Example 11: Figure 4 An exemplary schematic diagram illustrates an artificial intelligence-based industrial enterprise carbon emission inventory generation system according to an embodiment of the present invention. The system includes: The acquisition module is used to acquire information on raw materials, products, processes, technologies, scale, and treatment facilities reported by enterprises. The self-verification module is used to verify the information of raw materials, products, processes, technology, scale, and treatment facilities through the information self-verification model, and determine the accuracy of the identification results of various types of information. The filtering module is used to filter industry pollution discharge information from the database based on at least one of the various information submitted by the enterprise if there is information with abnormal accuracy identification results. The calibration module is used to obtain calibrated raw material information, product information, process information, technology information, scale information, and treatment facility information by using industry pollution discharge information, various information reported by enterprises, and pollution discharge data association models. The data module is used to determine the emission data of various pollutants based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. The inventory module is used to generate an inventory of carbon emissions from industrial enterprises based on emission data of various pollutants.

[0073] Example 12: According to an embodiment of the present invention, an artificial intelligence-based industrial enterprise carbon emission inventory generation device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the artificial intelligence-based industrial enterprise carbon emission inventory generation method.

[0074] Example 13: According to an embodiment of the present invention, a computer-readable storage medium is characterized in that it stores computer program instructions thereon, which, when executed by a processor, implement the artificial intelligence-based method for generating an industrial enterprise carbon pollution emission data list.

[0075] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0076] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments, and any modifications or variations of the embodiments of the present invention may be made without departing from the stated principles.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a data inventory of carbon emissions from industrial enterprises based on artificial intelligence, characterized in that, include: Obtain information on raw materials, products, processes, technologies, scale, and treatment facilities submitted by enterprises; The information self-verification model is used to verify the information of raw materials, products, processes, technology, scale, and treatment facilities to determine the accuracy of various types of information. If there is information indicating an abnormality in the accuracy of the identification results, then industry pollution discharge information will be screened from the database based on at least one of the multiple pieces of information submitted by the enterprise. By using industry pollution discharge information, various information reported by enterprises, and pollution discharge data correlation models, we can obtain corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. Based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information, the emission data of various pollutants are determined; Based on emission data of various pollutants, a data list of carbon emissions from industrial enterprises is generated.

2. The method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence according to claim 1, characterized in that, Using an information self-verification model, information on raw materials, products, processes, technologies, scale, and treatment facilities is verified to determine the accuracy of various types of information, including: By using the text recognition sub-model of the information self-verification model, features are extracted from multiple category names in the raw material information to obtain category name feature information; The data corresponding to the category name is concatenated with the category name feature information to obtain the category verification feature information; The graph attention sub-model of the information self-verification model is used to process the verification feature information of each category to obtain the verification feature information of each category; By splicing together the verification feature information of each category, the raw material information verification feature information is obtained; The verification feature information of each category is processed by the first multi-layer perception layer of the information self-verification model to obtain the first probability information that the data corresponding to each category name is abnormal data. The raw material information verification feature information is processed by the second multi-layer perception layer of the information self-verification model to obtain the second probability information of raw material information anomalies; Based on the first probability information and the second probability information, the accuracy of the raw material information identification result is determined.

3. The method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence according to claim 2, characterized in that, The training steps of the information self-verification model include: Obtain raw material information from multiple samples; The data of at least one category in some sample raw material information from multiple sample raw material information are modified according to a preset error type to obtain negative raw material information samples; The information self-verification model is used to process the sample raw material information or negative raw material information to obtain the first sample probability information that the data corresponding to each category name is abnormal data, and the second sample probability information that the sample raw material information or negative raw material information is abnormal. Based on the probability information of the first sample, the probability information of the second sample, and the data corresponding to the modified category, the loss function of the information self-verification model is obtained. The information self-calibration model is trained based on the loss function of the information self-calibration model to obtain the trained information self-calibration model.

4. The method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence according to claim 3, characterized in that, Based on the probability information of the first sample, the probability information of the second sample, and the data corresponding to the modified category, the loss function of the information self-verification model is obtained, including: The graph attention sub-model of the information self-verification model determines the association weights of each category; Determine the first weight of the category of the modified data in the negative sample of raw material information based on the preset error type; The calculated weights for each category are determined based on the first weight and the associated weights. For negative samples of raw material information, the loss value of the negative samples of raw material information is determined based on the calculated weights, the first sample probability information and the second sample probability information of each category. For sample raw material information, the loss value of the sample raw material information is determined based on the first sample probability information and the second sample probability information of each category; Based on the loss value of the negative sample of raw material information and the loss value of the sample raw material information, the loss function of the information self-verification model is determined.

5. The method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence according to claim 2, characterized in that, By utilizing industry-specific pollution discharge information, various information reported by enterprises, and pollution discharge data correlation models, corrected raw material information, product information, process information, technological information, scale information, and treatment facility information are obtained, including: Based on the accuracy identification results, the category verification feature information is labeled to obtain the category proofreading feature information; By using the first self-attention mechanism of the sewage discharge data association model, the verification feature information of each category of each piece of information is processed to obtain the first-level output verification feature information corresponding to each category of each piece of information. By using the feature extraction hierarchy of the pollution discharge data association model, the industry pollution discharge information is processed to obtain industry pollution discharge characteristic information; By using the cross-attention mechanism of the pollution discharge data association model, the industry pollution discharge characteristic information and the primary output verification characteristic information are processed to obtain the secondary output verification characteristic information corresponding to each category. By using the second self-attention mechanism of the sewage discharge data association model, the secondary output verification feature information corresponding to each category is processed to obtain the tertiary output verification feature information of each category. Input the three-level output correction feature information corresponding to the categories marked as abnormal into the third multi-level perception layer of the sewage data association model to obtain the corrected data; Based on the corrected data, corrected raw material information, product information, process information, technology information, scale information, and treatment facility information are obtained.

6. The method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence according to claim 5, characterized in that, The training steps for the sewage discharge data association model include: Obtain information from multiple samples; Data from at least one category in a subset of multiple sample data is modified according to a preset error type to obtain negative information samples. By processing sample information or negative samples through a sewage discharge data association model, we can obtain the first-level sample output verification feature information, second-level sample output verification feature information, and third-level sample output verification feature information corresponding to each category. The first anomaly identification probability information of each category is obtained by processing the first-level sample output verification feature information of each category through the fourth multi-layer perception layer. The fifth multi-layer perception layer processes the second-level sample output verification feature information corresponding to each category to obtain the second anomaly recognition probability information of each category. The sixth multi-layer perception layer processes the proofreading feature information of the third-level sample output corresponding to each category to obtain the third anomaly recognition probability information of each category. Through the third multi-layer perception layer, the calibration feature information of the three-level sample output corresponding to each category is processed to obtain the corrected data corresponding to each category. Based on the corrected data for each category, the data for the unmodified categories, the data for the modified categories before modification, the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information, the loss function of the sewage discharge data association model is obtained. Based on the loss function of the sewage discharge data association model, the sewage discharge data association model is trained to obtain the trained sewage discharge data association model.

7. The method for generating an inventory of carbon emissions from industrial enterprises based on artificial intelligence according to claim 6, characterized in that, Based on the corrected data for each category, the data for the unmodified categories, the data for the modified categories before modification, the first anomaly identification probability information, the second anomaly identification probability information, and the third anomaly identification probability information, the loss function of the sewage discharge data association model is obtained, including: Based on the corrected data for each category and the data for the unmodified categories, obtain the data consistency loss function; Based on the corrected data for each category and the uncorrected data for the modified category, the data correction loss function is obtained. Based on the first, second, and third anomaly identification probability information for each category, the anomaly identification loss function is obtained. Based on the data consistency loss function, data correction loss function, and anomaly identification loss function, the loss function of the sewage discharge data association model is determined.

8. An artificial intelligence-based system for generating a data inventory of carbon emissions from industrial enterprises, used to execute the method as described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire information on raw materials, products, processes, technologies, scale, and treatment facilities reported by enterprises. The self-verification module is used to verify the information of raw materials, products, processes, technology, scale, and treatment facilities through the information self-verification model, and determine the accuracy of the identification results of various types of information. The filtering module is used to filter industry pollution discharge information from the database based on at least one of the various information submitted by the enterprise if there is information with abnormal accuracy identification results. The calibration module is used to obtain calibrated raw material information, product information, process information, technology information, scale information, and treatment facility information by using industry pollution discharge information, various information reported by enterprises, and pollution discharge data association models. The data module is used to determine the emission data of various pollutants based on the corrected raw material information, product information, process information, technology information, scale information, and treatment facility information. The inventory module is used to generate an inventory of carbon emissions from industrial enterprises based on emission data of various pollutants.

9. An artificial intelligence-based device for generating an inventory of carbon emissions from industrial enterprises, characterized in that, include: processor; A memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores computer program instructions that, when executed by a processor, implement the method as described in any one of claims 1-7.

Images