A hazardous waste macro-micro feature mapping analysis method and system

Abstract

The present application relates to a kind of dangerous waste macro-micro feature mapping analysis method and system, method includes: S1, the existence of micro-component response matrix for macro-micro correlation analysis is constructed;S2, structured analysis extracts key semantic word element, carries out field weighting vectorization and dimension reduction, generates process semantic feature matrix;S3, excluding double zero term Jaccard distance is used to construct dissimilarity matrix;S4, the principal coordinate matrix is obtained by principal coordinate analysis;S5, based on the redundancy analysis and significance test of distance, and correction determination coefficient is calculated;S6, key description word element is projected to ordering space with semantic weight, the fitting relationship of micro-component vector and each ordering axis in ordering space is calculated, generates double order diagram and correlation strength matrix. Constructed from dangerous waste macroscopic description feature to the quantifiable mapping path of micro-component variation, provide scientific basis for fine classification and accurate risk control.

Description

Technical Field

[0001] This invention belongs to the field of hazardous waste management and environmental informatics technology, and in particular relates to a method and system for macro-micro feature mapping analysis of hazardous waste. Background Technology

[0002] Effective management of hazardous waste relies on an accurate understanding of its sources, technological characteristics, and material composition. Current hazardous waste classification systems are primarily based on industrial sectors at the production end, using macroscopic industry sources and processes as the basis for classification, failing to fully reflect the differences in the microscopic material composition of waste. However, the environmental risks and resource value of hazardous waste are essentially determined by its microscopic material composition (the types of heavy metal elements and mineral phase composition), and relying solely on macroscopic descriptive information makes it difficult to accurately assess the hazardousness and recyclability of waste.

[0003] The National Hazardous Waste List contains a large amount of unstructured textual descriptions, covering macroscopic characteristics such as the industry sources, generation processes, and waste properties. These characteristics are intrinsically linked to the microscopic composition of the waste, but currently, there is a lack of systematic quantitative analysis methods to explore and quantify the mapping relationship between the two. Existing research mostly relies on qualitative speculation or empirical judgment, failing to effectively quantify the extent to which macroscopic descriptive features explain variations in microscopic components, and also making it difficult to intuitively present the correlation between specific process terms and specific mineral phases or elements in the ordination space.

[0004] At the methodological level, classical redundancy analysis (RDA) relies on Euclidean distance and assumes a linear response relationship, making it unsuitable for waste micro-component data characterized by binary (presence / missing) data. For such data, Jaccard distance can more accurately measure the compositional similarity between samples, but existing research rarely reports combining distance-based redundancy analysis (db-RDA) with natural language processing techniques for the correlation analysis of macro-micro characteristics in hazardous waste.

[0005] Therefore, there is an urgent need for a quantitative method that can integrate unstructured text information processing with multivariate ranking statistical analysis to achieve a quantitative explanation of the variation in microscopic material composition based on the macroscopic characteristics of hazardous waste, and to provide scientific support for the refined classification and management of waste. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for macro-micro feature mapping analysis of hazardous waste.

[0007] The technical solution adopted in this invention is: Firstly, a method for macro-micro characteristic mapping analysis of hazardous waste is provided, including: S1. Obtain the detection data of waste samples, extract the set of heavy metal elements and mineral phases, and combine solid waste domain knowledge to perform feature screening and encoding, and construct a micro-component existence response matrix for macro-micro correlation analysis; S2, for the unstructured national hazardous waste list, uses domain-adaptive natural language processing technology to perform structured parsing to extract key semantic words, performs domain-weighted vectorization and dimensionality reduction, and generates a process semantic feature matrix; S3, for the micro-component existence response matrix, the dissimilarity matrix is ​​constructed by excluding double zero terms using the Jaccard distance; S4. Map the dissimilarity matrix to the Euclidean sort space through principal coordinate analysis to obtain the principal coordinate matrix; S5 uses the process semantic feature matrix to perform distance-based redundancy analysis and significance test on the principal coordinate matrix, and calculates the corrected coefficient of determination. The corrected coefficient of determination is used to quantify the proportion of macroscopic key semantic terms that unbiasedly explain the variation of microscopic components. S6 projects key descriptive terms onto the ranking space with semantic weights, calculates the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generates a bi-order graph and a correlation strength matrix.

[0008] Further, in S1, the detection data of waste samples are obtained, the set of heavy metal elements and the set of mineral phases are extracted, and feature screening and encoding are performed in combination with knowledge of solid waste. A micro-component existence response matrix for macro-micro correlation analysis is constructed, including: Test data for all waste samples were obtained from the hazardous waste testing database and the national hazardous waste list; Extract the set of heavy metal elements E = {e1, e2, ..., e3} from the detection data. p} and the mineral phase set M = {m1, m2, ..., m q}, p represents the total number of heavy metal elements in E, q represents the total number of mineral phases in M, and p and q are natural numbers greater than 2; Combining knowledge of the solid waste field, and based on preset rules, feature screening is performed on heavy metal elements and mineral phases to obtain a feature set containing n effective features; Construct a binary micro-component existence response matrix Y for macro-micro correlation analysis on the feature set; Y∈{0,1} (g×n) Where g is the total number of waste samples; if a valid feature j is detected in waste sample i, then Y {ij} = 1; if no valid feature j is detected in waste sample i, then Y {ij} = 0.

[0009] Furthermore, S2, for the unstructured national hazardous waste list, uses domain-adaptive natural language processing technology to perform structured parsing and extract key semantic terms, performs domain-weighted vectorization and dimensionality reduction, and generates a process semantic feature matrix, including: For all hazardous waste documents in the unstructured National Hazardous Waste List, Chinese word segmentation and stop word processing were performed using a dictionary specifically for solid waste, thereby extracting and retaining key semantic units t that have the ability to distinguish processes; Calculate the domain-weighted score (TF-IDF) for key semantic terms t in hazardous waste document d. domain(t,d) The formula for calculating the domain-weighted score is: TF-IDF domain(t,d) = TF (t,d) × IDF (t) × w domain(t) ; Among them, TF (t,d) = (Frequency of key semantic unit t in hazardous waste document d) / (Total number of terms in hazardous waste document d); IDF (t) = log(N / (1 + df(t))), where N represents the total number of hazardous waste documents, df(t) represents the number of hazardous waste documents containing the key semantic unit t, and w domain(t) This represents the pre-defined domain weight factor corresponding to the key semantic unit t; Based on the total number of waste samples g and the total number of all terms s in all hazardous waste documents, a g×s domain weighting matrix is ​​formed by combining the domain weighted score TF-IDF. Principal component analysis is performed on the g×s order neighborhood weighted matrix. The number of principal components u to be retained is determined by the cumulative variance contribution rate, resulting in the process semantic feature matrix X∈ .

[0010] Furthermore, S3, for the micro-component existence response matrix, uses the Jaccard distance excluding double-zero terms to construct a dissimilarity matrix, including: To address the high sparsity of the binary data in the micro-component existence response matrix Y, a Jaccard distance algorithm excluding double-zero terms is used to calculate the Jaccard distance d between any two hazardous waste samples f and k. fk The expression for the Jaccard distance algorithm is: ; Where, a represents the number of valid features present in both hazardous waste sample f and hazardous waste sample k, b represents the number of valid features present only in sample f, and c represents the number of valid features present only in sample k; The valid features of double-zero terms are excluded. Valid features of double-zero terms are those that do not exist in either hazardous waste sample f or hazardous waste sample k; d fk ∈ [0, 1], if d jk =0, which means that the effective feature composition of hazardous waste sample f and hazardous waste sample k are exactly the same. If d jk =1 indicates that hazardous waste sample f and hazardous waste sample k have no common valid features; Construct a dissimilarity matrix D of order g×g based on all Jaccard distances.

[0011] Furthermore, in S4, the dissimilarity matrix is ​​mapped to the Euclidean sort space through principal coordinate analysis to obtain the principal coordinate matrix, including: Perform principal coordinate analysis on the dissimilarity matrix D to construct the squared distance matrix D. (2) ; For the squared distance matrix D (2) By performing bi-centering, we obtain the bi-centering matrix G. , H is Order-centered matrix, for An identity matrix of order 1, where 1 represents all components equal to 1. 3D column vector, It is a matrix of all 1s; Perform eigenvalue decomposition on the bicentered matrix G: ; , indicating that the eigenvalues ​​are arranged in descending order; V is an orthogonal matrix, and the column vectors are the corresponding eigenvectors; Extract the positive eigenvectors of all positive eigenvalues, and scale the positive eigenvectors by the square root to obtain the principal coordinate matrix. ; r represents the number of positive eigenvalues; From positive eigenvalues The corresponding positive eigenvectors constitute the structure.

[0012] Furthermore, in S5, distance-based redundancy analysis and significance testing are performed on the principal coordinate matrix using the process semantic feature matrix, and the corrected coefficient of determination is calculated, including: Distance-based redundancy analysis is performed on the principal coordinate matrix Z using the process semantic feature matrix X. This is achieved by projecting the principal coordinate matrix Z onto the subspace spanned by the column vectors of the process semantic feature matrix X, thus obtaining the constraint fitting matrix. Constraint fitting matrix The expression is: ; Among them, H X = X(XT X) -1 X T The hat matrix of the principal coordinate matrix X; Singular value decomposition is performed on the constraint fitting matrix Ẑ to obtain the canonical sorting axes. and their corresponding constraint eigenvalues Each paradigmatic sorting axis The proportion of variance explained is , Indicates the total variation; Use a preset number of times n perm For a permutation test with a value ≥999, under the null hypothesis H0, permutation samples are generated by randomly shuffling the row order of the principal coordinate matrix Z, and the pseudo-F statistic F is calculated for each permutation. perm The expression is: F perm = (SS constrained / df constrained ) / (SS residual / df residual ); in, To constrain the amount of variation, df is the residual variation. constrained = df residual =g- -1; The value of F is in the permutation distribution. perm ≥F observed The proportion, F observed For the actual F value; when When macroscopic key descriptive terms are identified, they have statistically significant explanatory power for microscopic component variations; when At that time, it was determined that key descriptive terms at the macro level did not have statistically significant explanatory power for variations in micro-components; from The various paradigmatic sorting axes Select The two canonical sort axes with the smallest values ​​are used as the target sort axes; The correction determination coefficient is calculated according to the formula. Corrected coefficient of determination The proportion of key descriptive terms used to quantify the macroscopic level that unbiasedly explain the variation of microscopic components; The formula for calculating the corrected coefficient of determination is as follows: ; Where R²=SS constrained / SS total , .

[0013] Furthermore, in S6, key descriptive terms are projected onto the ranking space with semantic weights, and the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space is calculated to generate a bi-order graph and an association strength matrix, including: Project the key descriptive term t in waste sample i onto the ranking space, and let the ranking coordinates (scores) of key descriptive term t be... t The calculation formula is: ; Among them, z i Let i be the coordinate vector of waste sample i in the sorting space. Hazardous waste document d for waste sample i i Domain-weighted score of key descriptive term t; For each micro-component vector v in the micro-component existence response matrix Y, the micro-component vector v is projected onto the ranking space through multiple linear regression to obtain the vector fitting coordinates fit. v ; Based on vector fitting coordinates v The ranking coordinates of the keyword descriptor t: score t Calculate the correlation strength matrix A, and the elements A in the correlation strength matrix A. {tv} Define as score t with fit v Cosine similarity between them, element A {tv} The expression is: A {tv} =cos(score t fit v ) = (score t ×fit v ) / (||score t ||×||fit v ||); A {tv} ∈[-1, 1],|A {tv} The larger the value, the stronger the correlation between the macroscopic key descriptive term t and the microscopic component vector v; According to the target sorting axis Using the two canonical sort axes with the smallest values ​​as coordinate axes, a bi-order graph is plotted.

[0014] Secondly, a macro-micro characteristic mapping analysis system for hazardous waste is provided, including: The data construction module is used to acquire detection data of waste samples, extract heavy metal element sets and mineral phase sets, and combine solid waste domain knowledge to perform feature screening and encoding, and construct a micro-component existence response matrix for macro-micro correlation analysis; The text processing module is used to perform structured parsing and extract key semantic words from the unstructured national hazardous waste list using domain-adaptive natural language processing technology, and then perform domain-weighted vectorization and dimensionality reduction to generate a process semantic feature matrix. The dissimilarity matrix construction module is used to construct a dissimilarity matrix for the existence response matrix of micro-components by excluding double zero terms using the Jaccard distance. The principal coordinate matrix construction module is used to map dissimilarity matrices to Euclidean sorting space through principal coordinate analysis to obtain the principal coordinate matrix. The distance-based redundancy analysis ranking module is used to perform distance-based redundancy analysis and significance testing on the principal coordinate matrix using the process semantic feature matrix, and to calculate the corrected coefficient of determination. The corrected coefficient of determination is used to quantify the proportion of macroscopic key semantic terms that unbiasedly explain the variation of microscopic components. The visualization analysis module projects key descriptive terms onto the ranking space with semantic weights, calculates the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generates a bi-order graph and a correlation strength matrix.

[0015] The beneficial effects achieved by this invention are as follows: To address the domain specificity of descriptive text in hazardous waste, an adaptive natural language processing (NLP) method with a pre-built solid waste domain-specific dictionary and domain weight factors is introduced. This solves the problem of insufficient accuracy of general NLP tools in recognizing technical terms related to hazardous waste and improves the semantic accuracy of the process semantic feature matrix at the macro level. In response to the unique characteristics of hazardous waste micro-component data being highly sparse and containing many double zeros, Jaccard distance, which excludes double zeros, is used to replace Euclidean distance in RDA. This fundamentally eliminates the interference of the double zero problem on similarity measurement, enabling the ranking results to accurately reflect the true compositional differences of hazardous waste. By combining the distance-based redundancy analysis db-RDA ranking framework with domain-weighted semantic features, and by correcting the coefficient of determination, an unbiased quantitative estimate of the proportion of macro-level key descriptive terms that explain the variation of micro-components is provided, which makes up for the shortcomings of existing macro-micro correlations which only remain at the level of qualitative speculation. The association strength matrix, calculated based on the vector fitting coordinates and the sorted coordinates of key descriptive terms, uses cosine similarity to quantitatively characterize the association direction and strength between each pair of macroscopic key descriptive terms and microscopic component vectors. It can support batch querying and threshold filtering, providing an operable quantitative tool for the refined classification, source identification and precise risk control of hazardous waste. Attached Figure Description

[0016] Figure 1 This is a flowchart of the macro-micro feature mapping analysis method for hazardous waste according to the present invention; Figure 2 This is a structural diagram of the macro-micro feature mapping analysis system for hazardous waste of the present invention. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0018] like Figure 1 As shown, this embodiment of the invention provides a method for macro-micro feature mapping analysis of hazardous waste, including: S1. Obtain the detection data of waste samples, extract the set of heavy metal elements and mineral phases, and combine solid waste domain knowledge to perform feature screening and encoding, and construct a micro-component existence response matrix for macro-micro correlation analysis; In this embodiment, the detection data of all waste samples are obtained from the hazardous waste detection database and the national hazardous waste list; the detection records of heavy metal elements and mineral phases corresponding to the codes of multiple types of hazardous waste are included. Extract the set of heavy metal elements E = {e1, e2, ..., e3} from the detection data. p} and the mineral phase set M = {m1, m2, ..., m q}, p represents the total number of heavy metal elements in E, q represents the total number of mineral phases in M, and p and q are natural numbers greater than 2; Combining knowledge from the solid waste field, and based on preset rules, feature screening is performed on heavy metal elements and mineral phases to obtain a feature set containing n valid features; the specific screening includes: Remove heavy metal element features and mineral phase features (all zero columns) that are missing in all waste samples, and remove meaningless zero features in total; Features that are detected less than 5% of the total sample size are evaluated, and features that have no toxicity or resource value are deleted. The characteristics of heavy metal elements with clear toxic effects, such as As, Pb, Hg, Cr, Cd, Zn, Cu, and Ni, are unconditionally retained and are not subject to frequency threshold constraints. Finally, n valid features are retained to form a feature set.

[0019] Construct a binary micro-component existence response matrix Y for macro-micro correlation analysis on the feature set; Y∈{0,1} (g×n) Where g is the total number of waste samples; if a valid feature j is detected in waste sample i, then Y {ij} = 1; if no valid feature j is detected in waste sample i, then Y {ij} = 0; In actual testing, the overall sparsity of the micro-component existence response matrix was high (the proportion of non-zero elements was about 8% to 15%), and the proportion of double zero terms exceeded 75%, which verified the necessity of subsequently selecting a distance metric that excludes double zero terms.

[0020] S2, for the unstructured national hazardous waste list, uses domain-adaptive natural language processing technology to perform structured parsing to extract key semantic words, performs domain-weighted vectorization and dimensionality reduction, and generates a process semantic feature matrix; In this embodiment, for all hazardous waste documents in the unstructured National Hazardous Waste List, a Chinese word segmentation and stop word processing is performed using a dictionary specifically for solid waste, thereby extracting and retaining key semantic units t that have the ability to distinguish processes; The solid waste-specific dictionary pre-includes compound terms related to hazardous waste, including but not limited to multi-character phrases such as "hydrometallurgical processing," "pyrometallurgical processing," "wastewater treatment," "electroplating wastewater," "mercury-containing waste," and "lead-zinc smelting," to avoid general word segmentation tools breaking them down into meaningless single characters; and it removes low-information stop words from the W set. stop = {“use”, “conduct”, “generate”, “manufacture”, “produce”, “process”, “product”, “device”, “include”, “waste”, “hazardous waste”}; Calculate the domain-weighted score (TF-IDF) for key semantic terms t in hazardous waste document d. domain(t,d) The formula for calculating the domain-weighted score is: TF-IDF domain(t,d) = TF (t,d) × IDF (t) × w domain(t) ; Among them, TF (t,d) = (Frequency of key semantic unit t in hazardous waste document d) / (Total number of terms in hazardous waste document d); IDF (t) = log(N / (1 + df(t))), where N represents the total number of hazardous waste documents, df(t) represents the number of hazardous waste documents containing the key semantic unit t, and w domain(t)This represents the pre-defined domain weight factor corresponding to the key semantic unit t; Domain weight factor w domain The assignment rules include: Key semantic terms characterizing waste-generating processes (e.g., electroplating, smelting, catalysis, refining, cyanidation), w domain =1.5; Key semantic terms representing waste morphology (e.g., sludge, dust, fly ash, slag, molten slag) are assigned to w domain =1.3; Key semantic terms characterizing pollutants (e.g., mercury-containing, chromium-containing, lead-zinc-containing, arsenic-containing) are assigned to w domain =2.0; The remaining key semantics of the lexical units are assigned to w domain =1.0.

[0021] Based on the total number of waste samples g and the total number of all terms s in all hazardous waste documents, a g×s domain weighting matrix is ​​formed by combining the domain weighted score TF-IDF. Principal component analysis (PCA) is performed on the g×s order neighborhood weighted matrix, using the cumulative variance contribution rate as the criterion. For example, setting 85% as the retention criterion, if the cumulative variance contribution rate of the first 5 principal components exceeds 85%, then the number of principal components retained is u=5, resulting in the process semantic feature matrix X∈ That is, X∈ .

[0022] S3, for the micro-component existence response matrix, the dissimilarity matrix is ​​constructed by excluding double zero terms using the Jaccard distance; In this embodiment, step S1 above describes that the sparsity of the micro-component existence response matrix Y is relatively high. The Jaccard distance algorithm, which excludes double zero terms, is used to calculate the Jaccard distance d between any two hazardous waste samples f and k. fk The expression for the Jaccard distance algorithm is: ; Where, a represents the number of valid features present in both hazardous waste sample f and hazardous waste sample k, b represents the number of valid features present only in sample f, and c represents the number of valid features present only in sample k; The valid features of double-zero terms are excluded. Valid features of double-zero terms are those that do not exist in either hazardous waste sample f or hazardous waste sample k; d fk ∈ [0, 1], if d jk =0, which means that the effective feature composition of hazardous waste sample f and hazardous waste sample k are exactly the same. If d jk=1 indicates that hazardous waste sample f and hazardous waste sample k have no common valid features; Construct a dissimilarity matrix D of order g×g based on all Jaccard distances; For example, taking waste sample f as sludge containing Fe(OH)3, CaSO4, and Cr(OH)3, and waste sample k as dust containing PbSO4 and ZnO: a=0 (no common features), b=3, c=2, then =(3+2) / (0+3+2)=1.0, the two components have no overlap, and the dissimilarity is the greatest. Double zero terms (367-5=362 features missing in both) are all excluded to ensure that similarity is driven only by shared component features.

[0023] S4. Map the dissimilarity matrix to the Euclidean sort space through principal coordinate analysis to obtain the principal coordinate matrix; In this embodiment, principal coordinate analysis (PCA) is performed on the dissimilarity matrix D to construct the squared distance matrix D. (2) ; For the squared distance matrix D (2) By performing bi-centering, we obtain the bi-centering matrix G. , H is Order-centered matrix, for An identity matrix of order 1, where 1 represents all components equal to 1. 3D column vector, It is a matrix of all 1s; Perform eigenvalue decomposition on the bicentered matrix G: ; , indicating that the eigenvalues ​​are arranged in descending order; V is an orthogonal matrix, and the column vectors are the corresponding eigenvectors; Extract the positive eigenvectors of all positive eigenvalues, and scale the positive eigenvectors by the square root to obtain the principal coordinate matrix. ; r represents the number of positive eigenvalues; From positive eigenvalues The corresponding positive eigenvectors constitute the structure.

[0024] S5 uses the process semantic feature matrix to perform distance-based redundancy analysis and significance test on the principal coordinate matrix, and calculates the corrected coefficient of determination. The corrected coefficient of determination is used to quantify the proportion of macroscopic key semantic terms that unbiasedly explain the variation of microscopic components. In this embodiment, distance-based redundancy analysis is performed on the principal coordinate matrix Z using the process semantic feature matrix X. By projecting the principal coordinate matrix Z onto the subspace spanned by the column vectors of the process semantic feature matrix X, the constraint fitting matrix is ​​obtained. Constraint fitting matrix The expression is: ; Among them, H X = X(X T X) -1 X T The hat matrix of the principal coordinate matrix X; Fitting the constraint matrix Perform singular value decomposition to obtain the canonical sort axes. and their corresponding constraint eigenvalues Each paradigmatic sorting axis The proportion of variance explained is , Indicates the total variation; Use a preset number of times n perm For a permutation test with a value ≥999, under the null hypothesis H0 (macro-level key descriptive terms are not correlated with micro-level component variations), permutation samples are generated by randomly shuffling the row order of the principal coordinate matrix Z, and the pseudo-F statistic F is calculated for each permutation. perm The expression is: F perm = (SS constrained / df constrained ) / (SS residual / df residual ); in, To constrain the amount of variation, df is the residual variation. constrained = df residual =g- -1; The value of F is in the permutation distribution. perm ≥F observed The proportion, F observed For the actual F value; when When macroscopic key descriptive terms are identified, they have statistically significant explanatory power for microscopic component variations; when At that time, it was determined that key descriptive terms at the macro level did not have statistically significant explanatory power for variations in micro-components; Assuming that the results of 999 permutation tests indicate that, for mineral phase characteristics, the global F = 15.22, For heavy metal element characteristics, the global F=10.03. Both models are highly significant; the permutation test results for each canonical ordination axis show that the first two canonical ordination axes (dbRDA1, dbRDA2) are both significant. ), the remaining shafts This indicates that the effective constraint information is mainly concentrated on the first two canonical sorting axes; from The various paradigmatic sorting axes Select The two canonical sort axes with the smallest values ​​are used as the target sort axes; The correction determination coefficient is calculated according to the formula. Corrected coefficient of determination The proportion of key descriptive terms used to quantify the macroscopic level that unbiasedly explain the variation of microscopic components; The formula for calculating the corrected coefficient of determination is as follows: ; Where R²=SS constrained / SS total , .

[0025] S6 projects key descriptive terms onto the ranking space with semantic weights, calculates the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generates a bi-order graph and a correlation strength matrix.

[0026] In this embodiment, the key descriptive term t in waste sample i is projected onto the ranking space, and the ranking coordinate score of key descriptive term t is... t The calculation formula is: ; Among them, z i Let i be the coordinate vector of waste sample i in the sorting space. Hazardous waste document d for waste sample i i Domain-weighted score of key descriptive term t; For each micro-component vector v in the micro-component existence response matrix Y, the micro-component vector v is projected onto the ranking space through multiple linear regression to obtain the vector fitting coordinates fit. v ; Based on vector fitting coordinates v The ranking coordinates of the keyword descriptor t: score t Calculate the correlation strength matrix A, and the elements A in the correlation strength matrix A. {tv} Define as score t with fit v Cosine similarity between them, element A {tv} The expression is: A {tv} =cos(score t fit v ) = (scoret ×fit v ) / (||score t ||×||fit v ||); A {tv} ∈[-1, 1],|A {tv} The larger the value, the stronger the correlation between the macroscopic key descriptive term t and the microscopic component vector v; According to the target sorting axis Using the two canonical sort axes with the smallest values ​​as coordinate axes, a bi-order graph is drawn. In the bi-sequence diagram, scattered points of waste samples can be marked (colored according to waste morphology category); Statistically significant micro-component vector arrows (length reflects r) 2 (The direction reflects the associated direction). At the macro level, the projection points of key descriptive terms (labeled terms); By visually identifying the relationship between specific process descriptions and specific micro-components through the spatial relationship between the direction of arrows and the location of key descriptive terms, we can intuitively identify the correlation between specific process descriptions and specific micro-components.

[0027] The final simulation shows that different waste properties (sludge, dust and fly ash, slag, molten slag, and waste materials) exhibit a significant clustering effect in the ordination space, verifying the explanatory power of macroscopic morphological features on the differentiation of microscopic components; the spatial relationship between microscopic component vectors and macroscopic keywords is corroborated by the correlation strength matrix results.

[0028] The beneficial effects achieved by the embodiments of the present invention are as follows: To address the domain specificity of descriptive text in hazardous waste, an adaptive natural language processing (NLP) method with a pre-built solid waste domain-specific dictionary and domain weight factors is introduced. This solves the problem of insufficient accuracy of general NLP tools in recognizing technical terms related to hazardous waste and improves the semantic accuracy of the process semantic feature matrix at the macro level. In response to the unique characteristics of hazardous waste micro-component data being highly sparse and containing many double zeros, Jaccard distance, which excludes double zeros, is used to replace Euclidean distance in RDA. This fundamentally eliminates the interference of the double zero problem on similarity measurement, enabling the ranking results to accurately reflect the true compositional differences of hazardous waste. By combining the distance-based redundancy analysis db-RDA ranking framework with domain-weighted semantic features, and by correcting the coefficient of determination, an unbiased quantitative estimate of the proportion of macro-level key descriptive terms that explain the variation of micro-components is provided, which makes up for the shortcomings of existing macro-micro correlations which only remain at the level of qualitative speculation. The association strength matrix, calculated based on the vector fitting coordinates and the sorted coordinates of key descriptive terms, uses cosine similarity to quantitatively characterize the association direction and strength between each pair of macroscopic key descriptive terms and microscopic component vectors. It can support batch querying and threshold filtering, providing an operable quantitative tool for the refined classification, source identification and precise risk control of hazardous waste.

[0029] Based on the macro-micro feature mapping analysis method for hazardous waste described in the above embodiments, the macro-micro feature mapping analysis system for hazardous waste will be explained below through embodiments.

[0030] like Figure 2 As shown, this embodiment of the invention provides a macro-micro feature mapping analysis system for hazardous waste, comprising: The data construction module 201 is used to acquire the detection data of waste samples, extract the set of heavy metal elements and mineral phases, and combine solid waste domain knowledge to perform feature screening and encoding, and construct a micro-component existence response matrix for macro-micro correlation analysis; Text processing module 202 is used to perform structured parsing and extract key semantic words from the unstructured national hazardous waste list using domain-adaptive natural language processing technology, perform domain-weighted vectorization and dimensionality reduction, and generate a process semantic feature matrix. The dissimilarity matrix construction module 203 is used to construct a dissimilarity matrix for the existence response matrix of micro-components by excluding double zero terms using the Jaccard distance. The principal coordinate matrix construction module 204 is used to map the dissimilarity matrix to the Euclidean sorting space through principal coordinate analysis to obtain the principal coordinate matrix. The distance-based redundancy analysis ranking module 205 is used to perform distance-based redundancy analysis and significance test on the principal coordinate matrix using the process semantic feature matrix, and to calculate the corrected coefficient of determination. The corrected coefficient of determination is used to quantify the proportion of macroscopic key semantic terms that unbiasedly explain the variation of microscopic components. The visualization analysis module 206 is used to project key descriptive terms onto the ranking space with semantic weights, calculate the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generate a bi-order graph and a correlation strength matrix.

[0031] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0032] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0033] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0034] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0035] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for macro-micro characteristic mapping analysis of hazardous waste, characterized in that, include: S1. Obtain the detection data of waste samples, extract the set of heavy metal elements and mineral phases, and combine solid waste domain knowledge to perform feature screening and encoding, and construct a micro-component existence response matrix for macro-micro correlation analysis; S2, for the unstructured national hazardous waste list, uses domain-adaptive natural language processing technology to perform structured parsing to extract key semantic words, performs domain-weighted vectorization and dimensionality reduction, and generates a process semantic feature matrix; S3, for the micro-component existence response matrix, the dissimilarity matrix is ​​constructed by excluding double zero terms using the Jaccard distance; S4. Map the dissimilarity matrix to the Euclidean sort space through principal coordinate analysis to obtain the principal coordinate matrix; S5 uses the process semantic feature matrix to perform distance-based redundancy analysis and significance test on the principal coordinate matrix, and calculates the corrected coefficient of determination. The corrected coefficient of determination is used to quantify the proportion of macroscopic key semantic terms that unbiasedly explain the variation of microscopic components. S6 projects key descriptive terms onto the ranking space with semantic weights, calculates the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generates a bi-order graph and a correlation strength matrix.

2. The method for macro-micro characteristic mapping analysis of hazardous waste according to claim 1, characterized in that, S1. Obtain the detection data of waste samples, extract the set of heavy metal elements and mineral phases, and combine solid waste domain knowledge for feature screening and encoding to construct a micro-component existence response matrix for macro-micro correlation analysis, including: Test data for all waste samples were obtained from the hazardous waste testing database and the national hazardous waste list; Extract the set of heavy metal elements E = {e1, e2, ..., e3} from the detection data. p } and the mineral phase set M = {m1, m2, ..., m q }, p represents the total number of heavy metal elements in E, q represents the total number of mineral phases in M, and p and q are natural numbers greater than 2; Combining knowledge of the solid waste field, and based on preset rules, feature screening is performed on heavy metal elements and mineral phases to obtain a feature set containing n effective features; Construct a binary micro-component existence response matrix Y for macro-micro correlation analysis on the feature set; Y∈{0,1} (g×n) Where g is the total number of waste samples; if a valid feature j is detected in waste sample i, then Y {ij} = 1; if no valid feature j is detected in waste sample i, then Y {ij} = 0.

3. The method for macro-micro characteristic mapping analysis of hazardous waste according to claim 2, characterized in that, S2, for the unstructured national hazardous waste list, uses domain-adaptive natural language processing technology to perform structured parsing and extract key semantic terms, performs domain-weighted vectorization and dimensionality reduction, and generates a process semantic feature matrix, including: For all hazardous waste documents in the unstructured National Hazardous Waste List, Chinese word segmentation and stop word processing were performed using a dedicated dictionary for solid waste, thereby extracting and retaining key semantic units t that have the ability to distinguish processes; Calculate the domain-weighted score (TF-IDF) for key semantic terms t in hazardous waste document d. domain(t,d) The formula for calculating the domain-weighted score is: TF-IDF domain(t,d) = TF (t,d) × IDF (t) × w domain(t) ; Among them, TF (t,d) = (Frequency of key semantic unit t in hazardous waste document d) / (Total number of terms in hazardous waste document d); IDF (t) = log(N / (1 + df(t))), where N represents the total number of hazardous waste documents, df(t) represents the number of hazardous waste documents containing the key semantic unit t, and w domain(t) This represents the pre-defined domain weight factor corresponding to the key semantic unit t; Based on the total number of waste samples g and the total number of all terms s in all hazardous waste documents, a g×s domain weighting matrix is ​​formed by combining the domain weighted score TF-IDF. Principal component analysis is performed on the g×s order neighborhood weighted matrix. The number of principal components u to be retained is determined by the cumulative variance contribution rate, resulting in the process semantic feature matrix X∈ .

4. The method for macro-micro characteristic mapping analysis of hazardous waste according to claim 2, characterized in that, S3, for the microscopic component existence response matrix, uses the Jaccard distance, excluding double-zero terms, to construct a dissimilarity matrix, including: To address the high sparsity of the binary data in the micro-component existence response matrix Y, a Jaccard distance algorithm excluding double-zero terms is used to calculate the Jaccard distance d between any two hazardous waste samples f and k. fk The expression for the Jaccard distance algorithm is: ； Where, a represents the number of valid features present in both hazardous waste sample f and hazardous waste sample k, b represents the number of valid features present only in sample f, and c represents the number of valid features present only in sample k; The valid features of double-zero terms are excluded. Valid features of double-zero terms are those that do not exist in either hazardous waste sample f or hazardous waste sample k; d fk ∈ [0, 1], if d jk If d = 0, it means that the effective feature composition of hazardous waste sample f and hazardous waste sample k are exactly the same. jk =1 indicates that hazardous waste sample f and hazardous waste sample k have no common valid features; Construct a dissimilarity matrix D of order g×g based on all Jaccard distances.

5. The method for macro-micro characteristic mapping analysis of hazardous waste according to claim 4, characterized in that, S4. Map the dissimilarity matrix to the Euclidean sort space using principal coordinate analysis to obtain the principal coordinate matrix, including: Perform principal coordinate analysis on the dissimilarity matrix D to construct the squared distance matrix D. (2) ; For the squared distance matrix D (2) By performing bi-centering, we obtain the bi-centering matrix G. , H is Order-centered matrix, for An identity matrix of order 1, where 1 represents all components equal to 1. 3D column vector, It is a matrix of all 1s; Perform eigenvalue decomposition on the bicentered matrix G: ; , indicating that the eigenvalues ​​are arranged in descending order; V is an orthogonal matrix, and the column vectors are the corresponding eigenvectors; Extract the positive eigenvectors of all positive eigenvalues, and scale the positive eigenvectors by the square root to obtain the principal coordinate matrix. ; r represents the number of positive eigenvalues; From positive eigenvalues The corresponding positive eigenvectors constitute the structure.

6. The method for macro-micro characteristic mapping analysis of hazardous waste according to claim 5, characterized in that, S5, using the process semantic feature matrix, performs distance-based redundancy analysis and significance testing on the principal coordinate matrix, and calculates the corrected coefficient of determination, including: Distance-based redundancy analysis is performed on the principal coordinate matrix Z using the process semantic feature matrix X. This is achieved by projecting the principal coordinate matrix Z onto the subspace spanned by the column vectors of the process semantic feature matrix X, thus obtaining the constraint fitting matrix. Constraint fitting matrix The expression is: ； Among them, H X = X(X T X) -1 X T The hat matrix of the principal coordinate matrix X; Singular value decomposition is performed on the constraint fitting matrix Ẑ to obtain the canonical sorting axes. and their corresponding constraint eigenvalues Each paradigmatic sorting axis The proportion of variance explained is , Indicates the total variation; Use a preset number of times n perm For a permutation test with a value ≥999, under the null hypothesis H0, permutation samples are generated by randomly shuffling the row order of the principal coordinate matrix Z, and the pseudo-F statistic F is calculated for each permutation. perm The expression is: F perm = (SS constrained / df constrained ) / (SS residual / df residual ); in, To constrain the amount of variation, df is the residual variation. constrained = df residual =g- -1; The value of F is in the permutation distribution. perm ≥F observed The proportion, F observed For the actual F value; when When macroscopic key descriptive terms are identified, they have statistically significant explanatory power for microscopic component variations; when At that time, it was determined that key descriptive terms at the macro level did not have statistically significant explanatory power for variations in micro-components; from The various paradigmatic sorting axes Select The two canonical sort axes with the smallest values ​​are used as the target sort axes; The correction determination coefficient is calculated according to the formula. Corrected coefficient of determination The proportion of key descriptive terms used to quantify the macroscopic level that unbiasedly explain the variation of microscopic components; The formula for calculating the corrected coefficient of determination is as follows: ； Where R²=SS constrained / SS total , .

7. The method for macro-micro characteristic mapping analysis of hazardous waste according to claim 6, characterized in that, S6, project key descriptive terms onto the ranking space with semantic weights, calculate the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generate a bi-order graph and a correlation strength matrix, including: Project the key descriptive term t in waste sample i onto the ranking space, and let the ranking coordinates (scores) of key descriptive term t be... t The calculation formula is: ； Among them, z i Let i be the coordinate vector of waste sample i in the sorting space. Hazardous waste document d for waste sample i i Domain-weighted score of key descriptive term t; For each micro-component vector v in the micro-component existence response matrix Y, the micro-component vector v is projected onto the ranking space through multiple linear regression to obtain the vector fitting coordinates fit. v ; Based on vector fitting coordinates v The ranking coordinates of the keyword descriptor t: score t Calculate the correlation strength matrix A, and the elements A in the correlation strength matrix A. {tv} Define as score t with fit v Cosine similarity between them, element A {tv} The expression is: A {tv} =cos(score t ,fit v ) = (score t ×fit v ) / (||score t ||×||fit v ||); A {tv} ∈[-1, 1],|A {tv} The larger the value, the stronger the correlation between the macroscopic key descriptive term t and the microscopic component vector v; According to the target sorting axis Using the two canonical sort axes with the smallest values ​​as coordinate axes, a bi-order graph is plotted.

8. A macro-micro characteristic mapping analysis system for hazardous waste, characterized in that, include: The data construction module is used to acquire detection data of waste samples, extract heavy metal element sets and mineral phase sets, and combine solid waste domain knowledge to perform feature screening and encoding, and construct a micro-component existence response matrix for macro-micro correlation analysis; The text processing module is used to perform structured parsing and extract key semantic words from the unstructured national hazardous waste list using domain-adaptive natural language processing technology, and then perform domain-weighted vectorization and dimensionality reduction to generate a process semantic feature matrix. The dissimilarity matrix construction module is used to construct a dissimilarity matrix for the existence response matrix of micro-components by excluding double zero terms using the Jaccard distance. The principal coordinate matrix construction module is used to map dissimilarity matrices to Euclidean sorting space through principal coordinate analysis to obtain the principal coordinate matrix. The distance-based redundancy analysis ranking module is used to perform distance-based redundancy analysis and significance testing on the principal coordinate matrix using the process semantic feature matrix, and to calculate the corrected coefficient of determination. The corrected coefficient of determination is used to quantify the proportion of macroscopic key semantic terms that unbiasedly explain the variation of microscopic components. The visualization analysis module projects key descriptive terms onto the ranking space with semantic weights, calculates the fitting relationship between the micro-component vectors in the micro-component existence response matrix and each ranking axis in the ranking space, and generates a bi-order graph and a correlation strength matrix.

Images