Method for determining dioxin emission factors from waste incineration flue gas based on multi-modal analysis

Abstract

The present invention relates to the technical field of dioxin emission accounting, and specifically to a method for determining dioxin emission factors of waste incineration flue gas based on multimodality, which includes the following steps: collecting papers and reports related to waste incineration and establishing a semantic library; completing the retraining of the BERT model according to the collected semantic library; investigating and monitoring actual waste incineration plants; natural language description feature mapping and evaluation; PCA principal component extraction; feature combination and hierarchical clustering; emission factor estimation; and emission factor model application process. The present invention, represented by the waste incineration industry, proposes a method for determining dioxin emission factors of waste incineration flue gas based on multimodality, and uses a variety of influencing factors as the basis for classifying emission factors. At the same time, through this method, the limitations of human selection and judgment when using emission factors can be better solved, and a more accurate description of the emission situation of waste incineration plants can be achieved, providing a new perspective for more accurate and scientific evaluation of dioxin emissions.

Description

Technical Field

[0001] The present invention relates to the technical field of dioxin emission accounting, and in particular to a method for determining dioxin emission factors of waste incineration flue gas based on multi-modality. Background Art

[0002] Dioxins mainly come from combustion and various industrial by-products, among which the waste incineration industry (mainly including domestic waste incineration, hazardous waste incineration, and medical waste incineration) is a key dioxin emission industry. Since dioxins are unintentionally produced persistent organic pollutants, they cannot be controlled by banning production and prohibiting use. The key to controlling them lies in reducing emissions at the source. Scientific assessment of emission sources and identification of key emission sources are the key to formulating a dioxin emission reduction plan.

[0003] Dioxin emission factors are important quantitative characterization tools for assessing dioxin emissions. They refer to the dioxin emissions per unit of production (processing volume) from a specific source or activity. However, the sources of dioxins are extremely wide, and the dioxin generation mechanisms and emission influencing factors of different sources vary greatly. Therefore, there are many technical difficulties in the formulation of emission factors. At present, the more commonly used dioxin emission factors internationally are the default emission factors for different source categories provided in the "Toolkit for Identification and Quantification of Dioxin Emission Standards" (hereinafter referred to as the "Toolkit") issued by the United Nations Environment Programme (UNEP). The derivation of this emission factor is mainly based on data with good quality control assurance obtained from laboratory experiments, peer-reviewed literature, special experimental projects, and government or agency reports. Then, results / processes with similar characteristics are aggregated into an emission factor, and the corresponding categories are briefly described. No detailed description is given on the specific aggregation method and process of the emission factor.

[0004] After consulting relevant literature, the methods for formulating dioxin emission factors in existing literature are basically consistent with the toolkit. Among them, factors affecting dioxin emission levels, such as raw materials, furnace type, and processing (production) volume, are the main basis for the classification of emission factors, but the influencing factors considered are relatively limited. Dioxin emission levels are the result of the combined effects of multiple factors. The linear combination of 2 to 3 influencing factors cannot well summarize the basic situation of the emission source. This leads to the fact that when selecting a dioxin emission factor for the emission accounting of a certain incinerator, the selected emission factor may not represent its actual emission level, resulting in an over- or under-calculated emission amount. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a method for determining dioxin emission factors from waste incineration flue gas based on multimodality, which solves the problems in the prior art of insufficient quantitative basic data such as waste incineration process detection, and excessive subjective factors in the qualitative data judgment rules described in natural language, resulting in large deviations in emission factor estimation.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] A method for determining dioxin emission factors of waste incineration flue gas based on multimodal analysis is provided, which comprises the following steps:

[0008] S1. Collect literature related to waste incinerator types, flue gas treatment processes, raw material types, flue gas parameters, and dioxin emission levels, and establish a semantic database of waste incinerator characteristics;

[0009] S2. Based on the existing pre-trained model, learn the semantics of different waste incinerator types, flue gas treatment processes, raw material types, and flue gas parameters in natural language descriptions that affect dioxin emission levels in a self-supervised manner, and establish a mapping relationship between dioxin emissions from waste incineration and its description features in natural language descriptions;

[0010] S3. Conduct investigation and monitoring of actual waste incineration plants to obtain information on the types of waste incinerators, flue gas treatment processes, types of raw materials, flue gas parameters, and dioxin emission levels of the investigated waste incinerators;

[0011] S4, mapping the natural language description features in the actual waste incineration plant survey data obtained to the semantic space through the model trained in S2 to obtain feature vectors, and extracting principal components from the feature vectors using PCA;

[0012] S5. Calculate the principal component scores to obtain the natural language description feature scores of waste incineration, i.e., the “semantic” factor;

[0013] S6, combining the natural language description feature scores and quantitative factors of waste incineration into a feature vector;

[0014] S7, using hierarchical clustering or k-means method to combine samples into feature clusters;

[0015] S8. Draw a tree diagram based on the clustering results; define the classification categories based on practicality, calculate the dioxin emission levels of samples in the same category, and obtain the dioxin emission factor of the category;

[0016] S9. Input the data and information related to the type of waste incinerator, flue gas treatment process, raw material type, and flue gas parameters that need to be evaluated into the model trained in S2 to determine its category and its corresponding dioxin emission factor.

[0017] Furthermore, in step S2, the pre-trained BERT large model is loaded and fine-tuned on the dataset prepared in S1, and the model parameters are adjusted to adapt to the relationship between the description of waste incineration and its dioxin emissions.

[0018] Furthermore, the specific method of step S3 is:

[0019] The natural language description features of the waste incinerator type, flue gas treatment process, raw material type, and flue gas parameters are sent to the retrained BERT model to obtain its escape vector D 16×768 ; Escape vector D 16×768 Semantic coding is used to evaluate the cosine similarity between the results. The cosine similarity formula is as follows:

[0020]

[0021] Where: A represents the semantic vector of sample A; B represents the semantic vector of sample B.

[0022] Furthermore, in step S4, the semantic space D of the z-score standard encoding is used. 16×768 Get the standardized StandardD 16×768 , using PCA to encode the semantic space StandardD 16×768 Extract the principal components 5×768 ; z-score standardized disclosure:

[0023]

[0024] Where: x represents D 16×768 A sample in the data sample set, μ represents D 16×768 The mean of the data sample, σ represents D 16×768 The standard deviation of the data sample;

[0025] The specific method of PCA principal component extraction is:

[0026] S3-1: Calculate the covariance matrix:

[0027]

[0028] Where: C 768×768 is the covariance matrix, m represents the number of samples, and T represents the matrix transpose;

[0029] S3-2: Use singular value decomposition SVD to calculate the eigenvalues ​​and eigenvectors of the covariance matrix; the formula is as follows:

[0030] U,S,V=SVD(C 768×768 )

[0031] Where: U is the left singular matrix, S is the diagonal matrix, the square of the diagonal elements is the eigenvalue, V is the right singular matrix, and is the set of eigenvectors;

[0032] S3-3: Arrange the eigenvectors in V into a matrix V′ in rows from top to bottom according to the corresponding eigenvalues;

[0033] S3-4: Calculate the principal component contribution rate, the formula is as follows:

[0034]

[0035] Where: S′ is the square of the diagonal elements of S;

[0036] S3-5: Select the top 5 principal component contribution rates and obtain the required principal component contribution rate W 5×1 ;

[0037] S3-6: Select the first 5 eigenvectors in V′ to obtain the required principal components

[0038] Furthermore, the specific method of step S5 is:

[0039] According to the principal component score And the principal component contribution rate W 5×1 Calculate the "semantic" factor R 16×1 ; The formula is as follows:

[0040]

[0041] Furthermore, in step S6, the "semantic" factor and the numerical features of service life, design scale, annual processing capacity, incinerator type, production efficiency, and dioxin emission level are combined into a feature vector of the waste incineration plant.

[0042] Furthermore, in step S7, the hierarchy method in the scipy package is used to implement hierarchical clustering.

[0043] Furthermore, the calculation formula for calculating the annual dioxin emissions from waste incinerators based on the dioxin emission factor is:

[0044] Q=EF 废气 ×P×10

[0045] Where: Q: annual emission of dioxins in waste gas, unit is mgTEQ;

[0046] EF 废气 : is the dioxin emission factor, i.e. the amount of dioxins emitted through exhaust gas when 1 ton of product is incinerated, in μgTEQ / t;

[0047] P: is the annual output, in units of 10,000 tons. The beneficial effects of the present invention are:

[0048] The method provided by the present invention is used to solve the current problem of insufficient information on the estimation of dioxin emission factors from waste incineration, which leads to large estimation deviations. The present invention uses a multimodal model and comprehensively adopts various characteristics of waste incineration plants. This method helps to more comprehensively understand and optimize the estimation of dioxin emission factors from waste incineration plants. Compared with the "Toolkit for Identifying and Quantifying Dioxin Emission Standards" and the linear combination method of 2 to 3 influencing factors, the dioxin emission factor determined by this method represents a more comprehensive incinerator operating status and dioxin emission level, and can provide a more accurate dioxin emission accounting capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of the method of the present invention;

[0050] Figure 2 It is a dendrogram of the model obtained according to the samples with known dioxin emission levels in the embodiment, the horizontal axis is the enterprise number, the vertical axis is the classification level, and the dotted line is the selected classification level;

[0051] Figure 3 Schematic diagram of the categories of samples with unknown dioxin emission levels in the embodiment after being calculated by the model, the horizontal axis is the enterprise number, the vertical axis is the classification level, the dotted line is the selected classification level, and the circle is the category of the samples with unknown dioxin emission levels. DETAILED DESCRIPTION

[0052] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0053] Example

[0054] Reference Figure 1 , a method for determining dioxin emission factors of waste incineration flue gas based on multimodal analysis is provided, which specifically includes the following steps:

[0055] 1. Model retraining

[0056] Retrain the BERT model based on the collected semantic library.

[0057] 2. Investigation and monitoring of actual waste incineration plants

[0058] The survey and monitoring results of actual waste incineration plants are shown in Table 1.

[0059] Table 1 Investigation and monitoring of actual waste incineration plants

[0060]

[0061]

[0062]

[0063] 3. Natural language description feature mapping and evaluation

[0064] The natural language description features such as incinerator type, flue gas treatment process, and raw materials are sent to the retrained BERT model to obtain its escape vector D 16×768 The escape vectors are semantically encoded and the results are evaluated pairwise using cosine similarity. The cosine similarity formula is as follows:

[0065]

[0066] Where A represents the semantic vector of sample A; B represents the semantic vector of sample B; similarity is the similarity.

[0067] Taking raw materials as an example to evaluate the mapping effect, the similarity between copper-containing and nickel-containing sludge produced by wastewater treatment in circuit board factories and surface treatment waste, copper-containing waste and nickel-containing waste (about 0.8) is significantly higher than the similarity between them and paint residue, film residue and ink residue (about 0.9), indicating that the mapping results are effective.

[0068] 4. PCA principal component extraction

[0069] Semantic space D using z-score standard encoding 16×768 Get the standardized StandardD 16×768 PCA (Principal Component Analysis) is used to encode the semantic space StandardD 16×768 Extract the principal components 5×768 , based on the principal component scores And the principal component contribution rate W 5×1 Calculate the "semantic" factor R 16×1 The formula is as follows

[0070] Where: x represents D 16×768 A sample in the data sample set, μ represents D 16×768 The mean of the data sample, σ represents D 16×768 The standard deviation of the data sample;

[0071] The specific method of PCA principal component extraction is:

[0072] (1): Calculate the covariance matrix:

[0073]

[0074] Where: C 768×768 is the covariance matrix, m represents the number of samples, and T represents the matrix transpose;

[0075] (2): Use singular value decomposition (SVD) to calculate the eigenvalues ​​and eigenvectors of the covariance matrix; the formula is as follows:

[0076] U,S,V=SVD(C 768×768 )

[0077] Where: U is the left singular matrix, S is the diagonal matrix, the square of the diagonal elements is the eigenvalue, V is the right singular matrix, and is the set of eigenvectors;

[0078] (3): Arrange the eigenvectors in V in rows from top to bottom according to the corresponding eigenvalues ​​to form a matrix V′;

[0079] (4): Calculate the principal component contribution rate, the formula is as follows:

[0080]

[0081] Where: S′ is the square of the diagonal elements of S;

[0082] (5): Select the top 5 principal component contribution rates to obtain the required principal component contribution rate W 5×1 ;

[0083] (6): Select the first five eigenvectors in V′ to obtain the required principal components

[0084] 5. Feature combination and hierarchical clustering

[0085] The "semantic" factor and the service life, design scale, annual processing capacity, incinerator type, production efficiency, and dioxin emission numerical features are combined into the feature vector of the waste incineration plant. The hierarchy method in the scipy package is used to implement hierarchical clustering. The clustering results are as follows: Figure 2 shown.

[0086] 6. Emission factor estimation

[0087] Based on the clustering results, a tree diagram is drawn. Classification categories are defined based on practicality, and the dioxin emission levels of samples in the same category are calculated to obtain the emission factors for that category.

[0088] 7. Emission factor model application process

[0089] The survey focused on the time when the incinerator was put into use, the designed amount of hazardous waste incinerated (t / d), the actual amount of hazardous waste incinerated (t / d), the annual processing capacity (10,000 tons / year), the type of incinerator, the process combination of waste gas treatment facilities, the main sources and types of hazardous waste, the average production efficiency of the facilities (t / h), the benchmark flue gas volume (m3 / h), input the above information into the model to obtain its classification and dioxin emission factor. The information of a certain enterprise is shown in Table 2 below.

[0090] Table 2

[0091]

[0092] After the above steps 1 to 6, it belongs to Figure 3 The emission factor of the point class marked in can be estimated as the emission factor of this class.

[0093] In summary, the present invention takes the waste incineration industry as a representative, and proposes a method for determining dioxin emission factors from waste incineration flue gas based on multimodality, taking multiple influencing factors as the basis for classifying emission factors. At the same time, through this method, the limitations of human selection and judgment when using emission factors can be better solved, and a more accurate description of the emission situation of waste incineration plants can be achieved, providing a new perspective for more accurate and scientific evaluation of dioxin emissions.

[0094] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.

Claims

1. A method for determining dioxin emission factors from waste incineration flue gas based on multimodality, characterized in that: The following steps are involved: S1. Collect literature related to waste incinerator types, flue gas treatment processes, raw material types, flue gas parameters, and dioxin emission levels, and establish a semantic database of waste incinerator characteristics; S2. Based on the existing pre-trained model, learn the semantics of different waste incinerator types, flue gas treatment processes, raw material types, and flue gas parameters in natural language descriptions that affect dioxin emission levels in a self-supervised manner, and establish a mapping relationship between dioxin emissions from waste incineration and its description features in natural language descriptions; S3. Conduct investigation and monitoring of actual waste incineration plants to obtain information on the types of waste incinerators, flue gas treatment processes, types of raw materials, flue gas parameters, and dioxin emission levels of the investigated waste incinerators; S4, mapping the natural language description features in the actual waste incineration plant survey data obtained to the semantic space through the model trained in S2 to obtain feature vectors, and extracting principal components from the feature vectors using PCA; S5. Calculate the principal component scores to obtain the natural language description feature scores of waste incineration, i.e., the "semantic" factor; S6, combining the natural language description feature scores and quantitative factors of waste incineration into a feature vector; S7, using hierarchical clustering or k-means method to combine samples into feature clusters; S8. Draw a tree diagram based on the clustering results; define the classification categories based on practicality, calculate the dioxin emission levels of samples in the same category, and obtain the dioxin emission factor of the category; S9. Input the information related to the type of waste incinerator, flue gas treatment process, raw material type, and flue gas parameters to be evaluated into the model trained in S2 to determine its classification and its corresponding dioxin emission factor; In step S2, the pre-trained BERT large model is loaded and fine-tuned on the dataset prepared in S1, and the model parameters are adjusted to fit the relationship between the description of waste incineration and its dioxin emissions; In step S6, the "semantic" factor and the numerical features of service life, design scale, annual processing capacity, incinerator type, production efficiency, and dioxin emission level are combined into a feature vector of the waste incineration plant.

2. The method for determining dioxin emission factors of waste incineration flue gas based on multimodality according to claim 1, characterized in that: The specific method of step S3 is: The natural language description features of the waste incinerator type, flue gas treatment process, raw material type, and flue gas parameters are sent to the retrained BERT model to obtain its escape vector D 16×768 ; Escape vector D 16×768 Semantic coding is used to evaluate the cosine similarity between the results. The cosine similarity formula is as follows: ； Where: A represents the semantic vector of sample A; B represents the semantic vector of sample B.

3. The method for determining dioxin emission factors of waste incineration flue gas based on multimodality according to claim 2, characterized in that: In step S4, the encoded semantic space is standardized using z-score D 16×768 After obtaining the standardized StandardD 16×768 , using PCA to encode the semantic space StandardD 16×768 Perform principal component extraction Components 5×768 ; z-score standardized disclosure: ； Where: x express D 16×768 A sample in the data sample set, express D 16×768 The mean of the data sample, express D 16×768 The standard deviation of the data sample; The specific method of PCA principal component extraction is: S3-1: Calculate the covariance matrix: ； Where: is the covariance matrix, m represents the number of samples, and T represents the matrix transpose; S3-2: Use singular value decomposition SVD to calculate the eigenvalues ​​and eigenvectors of the covariance matrix; the formula is as follows: ； Where: U is the left singular matrix, S is the diagonal matrix, the square of the diagonal elements is the eigenvalue, V is the right singular matrix, and is the set of eigenvectors; S3-3: Arrange the eigenvectors in V into a matrix from top to bottom according to the corresponding eigenvalues. ; S3-4: Calculate the principal component contribution rate, the formula is as follows: ； Where: is the square of the diagonal elements of S; S3-5: Select the top 5 principal component contribution rates to obtain the required principal component contribution rates ; S3-6: Select The first five eigenvectors in the equation are used to obtain the required principal components. .

4. The method for determining dioxin emission factors of waste incineration flue gas based on multimodality according to claim 3, characterized in that: The specific method of step S5 is: According to the principal component score and principal component contribution W 5×1 Calculating the "semantic" factor R 16×1 ; The formula is as follows: 。 5. The method for determining dioxin emission factors of waste incineration flue gas based on multimodality according to claim 1, characterized in that: In step S7, the hierarchy method in the scipy package is used to implement hierarchical clustering.

6. A method for calculating emission conditions based on the method for determining dioxin emission factors from waste incineration flue gas based on multimodal conditions according to any one of claims 1 to 5, characterized in that: The emission calculation formula is: ； Where: Q: annual emission of dioxins in waste gas, unit is mg TEQ; : is the dioxin emission factor, i.e. the amount of dioxins emitted through exhaust gas when 1 ton of product is incinerated, in μgTEQ / t; P: is the annual output, in ten thousand tons.

Images