Method for flocculation and defluorination of coking wastewater

Abstract

The present application relates to the technical field of coking wastewater treatment, and particularly relates to a flocculation defluorination method for coking wastewater, which comprises obtaining water quality parameter data of the coking wastewater, and constructing a flocculation defluorination knowledge model based on the data; then obtaining flocculant characteristic data, and mapping the data to the knowledge model to form a flocculation network comprising component nodes, attribute edges and reaction edges; then extracting key processing elements of the coking wastewater, and mapping the elements to the flocculation network to obtain key processing element related component nodes and form a processing element graph; then reasoning and judging the processing element graph based on the flocculation network and preset defluorination rules to form a preliminary processing scheme; then quantitatively scoring the preliminary processing scheme; finally, making parameter adjustment decisions based on the scheme score and the preliminary processing scheme to obtain a final processing decision, which can scientifically integrate water quality and flocculant information, reasonably form and optimize the scheme, improve the defluorination processing effect and efficiency of the coking wastewater, and reduce the cost.

Description

Technical Field

[0001] This invention relates to the field of coking wastewater treatment technology, and in particular to a flocculation defluorination method for coking wastewater. Background Technology

[0002] The coking industry generates a large amount of coking wastewater during production. This wastewater contains various pollutants such as fluoride. If discharged directly without effective treatment, it will seriously impact the ecological environment, including water bodies and soil, and may also endanger human health. Currently, various technologies have been developed in the industry for defluorination treatment of coking wastewater. Among them, flocculation has been applied to some extent in practical engineering due to its relatively simple operation and relatively low cost. Existing methods for flocculation-based defluoridation of coking wastewater still have many shortcomings in practical applications. Traditional flocculation-based defluoridation processes rely heavily on experience in setting operating parameters, lacking a systematic consideration of the relationship between wastewater quality parameters and flocculant characteristics. Coking wastewater from different sources and under different production conditions exhibits significant differences in water quality parameters such as fluoride concentration, pH value, and suspended solids content, while different types of flocculants also differ in composition, activity, and applicable conditions. Relying solely on experience to select flocculants and set treatment parameters often fails to achieve optimal defluoridation results, potentially leading to wastewater fluoride levels failing to meet discharge standards, or wasting flocculants and increasing treatment costs. Existing methods lack effective model support and reasoning mechanisms in the formation and optimization of treatment schemes. When faced with complex and variable coking wastewater quality, it is difficult to quickly and accurately identify key treatment elements and construct reasonable treatment schemes based on these elements. Multiple trials and adjustments are usually required to determine a suitable treatment scheme, which not only prolongs the treatment cycle but also increases experimental costs. Furthermore, the lack of scientific and quantitative evaluation methods for initially formed treatment schemes makes it impossible to accurately judge the merits of the schemes, hindering targeted parameter adjustments and scheme optimization. This results in poor stability and reliability of the treatment process, making it difficult to meet the requirements of continuous industrial treatment. Existing flocculation defluorination methods do not adequately consider the interrelationships between various stages of the treatment process. Flocculation defluorination is a complex process involving the interaction of multiple factors, including wastewater quality, flocculant characteristics, and reaction conditions, all of which are closely interconnected. Traditional methods often consider each stage in isolation, failing to form a complete treatment system. This leads to situations where optimization of one stage results in poor overall performance, further limiting the improvement of defluorination efficiency and the reduction of treatment costs. Summary of the Invention

[0003] The main objective of this invention is to provide a flocculation defluorination method for coking wastewater, aiming to solve the technical problems in the prior art.

[0004] This invention proposes a flocculation-based defluorination method for coking wastewater, comprising: Obtain water quality parameter data for coking wastewater; A knowledge model for flocculation and fluoride removal is constructed based on the aforementioned water quality parameter data; Acquire flocculant characteristic data and map the flocculant characteristic data into the flocculation and defluorination knowledge model to form a flocculation network containing component nodes, attribute edges, and reaction edges; Key elements for treating coking wastewater; The key processing elements are mapped onto the flocculation network to obtain the key processing element related component nodes, and a processing element diagram is formed based on the key processing element related component nodes. Based on the flocculation network and the preset defluorination rules, the processing element diagram is inferred and judged to form a preliminary processing plan; The preliminary processing plan is quantitatively scored to obtain a plan score; Based on the proposed solution score and the preliminary processing solution, parameter adjustment decisions are made to obtain the final processing decision.

[0005] Preferably, the step of constructing a flocculation defluoridation knowledge model based on the water quality parameter data includes: Data processing techniques are used to extract core concepts related to flocculation and fluoride removal from the water quality parameter data; Based on the core concepts and the water quality parameter data, the chemical relationships between the core concepts are determined, and the model relationships are defined according to the chemical relationships. Based on the model relationship definition, the core concepts are defined with attributes to obtain attribute definition data; The flocculation and fluoride removal knowledge model is constructed based on the attribute definition data.

[0006] Preferably, mapping the flocculant characteristic data to the flocculation and defluorination knowledge model to form a flocculation network containing component nodes, attribute edges, and reaction edges includes: The flocculant characteristic data is preprocessed to obtain preprocessed flocculant data; Based on the flocculation and defluorination knowledge model, the pre-treated flocculant data is mapped, the entities in the pre-treated flocculant data are mapped to corresponding component nodes, and the attributes in the pre-treated flocculant data are mapped to attribute edges. Based on the flocculation and defluorination knowledge model, relation extraction is performed on the pretreated flocculant data, the chemical relationships between entities in the pretreated flocculant data are extracted, and the chemical relationships are mapped as reaction edges. A flocculation network is constructed based on the component nodes, attribute edges, and reaction edges.

[0007] Preferably, the key treatment elements for extracting coking wastewater include: The coking wastewater data was structured to obtain structured coking wastewater data; Text parsing was performed on the structured coking wastewater data to obtain key information elements; Based on the key information elements, entity naming is performed to obtain key entities; The key entities are type-defined to obtain the defined key entities; Based on the defined key entities, relations are extracted to determine semantic associations; Based on the defined key entities, attribute definitions are performed to obtain attribute data; The defined key entities, attribute data, and semantic relationships are organized into a structured knowledge representation to obtain key processing elements.

[0008] Preferably, the step of mapping the key processing elements to the flocculation network to obtain key processing element-related component nodes, and forming a processing element graph based on the key processing element-related component nodes, includes: The key processing elements are mapped to component nodes in the flocculation network to obtain component nodes related to the key processing elements. Based on the relevant component nodes of the key processing elements, node expansion is performed in the flocculation network to obtain extended nodes; Subgraphs are extracted from the flocculation network based on the extended nodes and the key processing element-related component nodes. Redundancy is removed from the sub-graph to obtain the processed element graph.

[0009] Preferably, the quantitative scoring of the preliminary processing scheme to obtain a scheme score includes: Based on the defluorination requirements and standards, determine the quantitative dimensions and quantitative indicators; Assign weights to the quantification dimensions and quantification indicators; Based on the quantitative indicators and weights, the preliminary processing scheme is quantitatively scored to obtain a scheme score.

[0010] Preferably, the step of making parameter adjustment decisions based on the scheme score and the preliminary processing scheme to obtain the final processing decision includes: The proposed solution score and the preliminary processing solution will be integrated and analyzed. A decision is made based on a preset scoring threshold and the score of the proposed solution, resulting in a final processing decision.

[0011] Preferably, the method further includes: Obtain time-series data of coking wastewater parameters; Calculate the parameter variation deviation based on the time series data; Analyze the deviation of the parameter changes to obtain the parameter change values; The parameter changes are used to optimize the extraction of the key processing elements.

[0012] Preferably, the method further includes: Calculate the spatial distribution differences of coking wastewater in the reactor; By combining the changes in the parameters and the differences in their spatial distribution, the flocculation adaptability index is obtained; The formation process of the preliminary treatment scheme is adjusted based on the flocculation adaptability index.

[0013] Preferably, the method further includes: Obtain wastewater treatment condition data; When the wastewater treatment condition data meets the first condition, the weighting parameter is adjusted. The flocculation complexity index is calculated based on the adjusted weight parameters; Multiple candidate solutions are ranked based on the flocculation complexity index to optimize the generation of the final processing decision.

[0014] The beneficial effects of this invention are as follows: This flocculation-based defluoridation method for coking wastewater, by acquiring water quality parameter data of the coking wastewater and constructing a flocculation-based defluoridation knowledge model, provides a scientific framework for the formation of subsequent treatment solutions. This knowledge model can systematically integrate relevant information on wastewater water quality parameters, breaking through the limitations of traditional methods that rely on experience and lack systematic consideration. This allows subsequent treatment processes to be based on actual water quality conditions, avoiding the problem of poor treatment results due to insufficient understanding of water quality parameters. By mapping flocculant characteristic data onto a flocculation and defluorination knowledge model to form a flocculation network, which includes component nodes, attribute edges, and reaction edges, the network clearly presents the properties of each component of the flocculant and the reaction relationships between components and with wastewater quality parameters. This visualized network structure helps staff understand the correlation between flocculant characteristics and wastewater quality more intuitively, facilitating accurate assessment of the applicability of different flocculants under specific water quality conditions and enabling more rational flocculant selection. Compared to the traditional method of blindly selecting flocculants, this approach improves the accuracy of flocculant selection, reduces waste caused by inappropriate flocculant selection, and provides strong support for the subsequent development of efficient treatment solutions. Extracting key treatment elements from coking wastewater and mapping them onto a flocculation network to form a treatment element map allows for rapid identification of the core factors affecting defluorination efficiency. In complex coking wastewater treatment systems, the clear identification of key treatment elements avoids excessive focus on irrelevant factors during treatment scheme design, improving the targeting and efficiency of the design. Based on the treatment element map, staff can clearly understand the role and status of each key element in the entire treatment process, laying the foundation for subsequent reasoning and judgment to form a preliminary treatment plan, reducing the blind spots in scheme design caused by unclear key elements in traditional methods.

[0015] This approach, based on flocculation networks and pre-defined defluoridation rules, uses reasoning to determine preliminary treatment plans from a symmetric graph. This scientific reasoning mechanism eliminates reliance on empirical trial-and-error. The pre-defined defluoridation rules, derived from extensive practical experience and theoretical research, combined with the interrelationships of various factors presented in the flocculation network, ensure the rationality and feasibility of the preliminary treatment plan. This reasoning method can quickly generate treatment plans that meet current water quality conditions and flocculant characteristics, shortening the plan development cycle, avoiding the tedious process of multiple trials and adjustments required in traditional methods, and reducing experimental costs. Quantitative scoring of preliminary treatment plans, followed by parameter adjustments based on these scores and the preliminary plans, yields the final treatment decision, providing a scientific basis for optimizing the treatment plan. Quantitative scoring objectively and accurately assesses the merits of preliminary plans, offering greater fairness and reliability compared to traditional methods that rely on subjective judgment. Parameter adjustments based on the quantitative scoring results allow for targeted improvements to the shortcomings of the preliminary plans, ensuring the final treatment decision better meets actual treatment needs. This process continuously optimizes treatment parameters, enhancing defluorination efficiency while simultaneously controlling cost factors such as flocculant dosage to achieve a balance between treatment effectiveness and cost, all while maintaining treatment efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the working principle of the flocculation and defluorination method for coking wastewater described in this invention. Figure 2 A schematic diagram illustrating the working principle of constructing a knowledge model for flocculation and fluoride removal based on water quality parameter data. Figure 3 A flowchart for mapping key processing elements to form a processing element diagram.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] like Figure 1 As shown, this application provides a flocculation-based defluoridation method for coking wastewater. It achieves precise defluoridation decision-making through systematic data processing and knowledge reasoning. Obtaining water quality parameter data for the coking wastewater is the starting point. This data includes multiple indicators such as fluoride ion concentration, pH value, chemical oxygen demand (COD), suspended solids content, and water temperature. Based on this water quality parameter data, a flocculation-based defluoridation knowledge model is constructed. This model represents the chemical substances, reaction rules, and relationships between water quality characteristics involved in the flocculation-based defluoridation process in a structured form. Flocculant characteristic data is obtained and mapped into the flocculation-based defluoridation knowledge model. This flocculant characteristic data covers information such as flocculant type, molecular structure, charge characteristics, hydrolysis performance, and optimal pH range for addition. The process involves: forming a flocculation network containing component nodes, attribute edges, and reaction edges; extracting key treatment elements for coking wastewater, which are wastewater characteristics or operating conditions that significantly affect defluorination efficiency; mapping these key treatment elements onto the flocculation network to obtain related component nodes, and forming a treatment element graph; inferring and judging the treatment element graph based on the flocculation network and preset defluorination rules to generate preliminary treatment plans, including flocculant selection criteria, pH control range, and reaction time requirements; quantifying and scoring the preliminary treatment plans to obtain a plan score; and making parameter adjustment decisions based on the plan score and the preliminary treatment plan, ultimately outputting an executable final treatment decision.

[0020] In one embodiment, Example 1: See Figure 2When constructing a knowledge model for flocculation and defluorination, it is necessary to first process water quality parameter data. This data usually comes from real-time monitoring sensors or laboratory test reports, including multi-dimensional indicators such as fluoride ion concentration, pH value, chemical oxygen demand, suspended solids content, and water temperature. Data processing technology is used to extract the core concepts related to flocculation and defluorination from these raw data. The data processing technology includes data cleaning steps to remove outliers and noise interference, normalization to eliminate dimensional differences, and feature extraction algorithms to identify key parameter patterns. The core concepts focus on the fluoride existence forms such as free fluoride or complexed fluoride, coexisting ion types such as calcium and magnesium ions or sulfate ions, alkalinity level, and organic matter categories such as phenols or aromatic compounds. Based on core concepts and water quality parameter data, the chemical relationships between core concepts are further determined. Chemical relationship analysis requires combining water chemistry theory to determine the forms of interionic interactions, including precipitation reactions such as calcium fluoride formation, complexation reactions such as the formation of aluminum-fluoride complexes, adsorption mechanisms such as the adsorption of fluoride by aluminum hydroxide, and charge neutralization effects. Model relationships are defined based on these chemical relationships, clarifying the type and direction of association between each concept. For example, fluoride ions are defined as having a precipitation relationship with calcium ions and a complexation relationship with aluminum ions, and the pH conditions and energy changes for the reactions are noted. Based on the model relationship definitions, attribute definitions are obtained for the core concepts, assigning quantitative characteristics to each core concept, including ionic strength values, solubility product constants, reaction rate constants, optimal pH ranges, and temperature influence coefficients. Finally, a flocculation-based fluoride removal knowledge model is constructed based on the attribute definition data. This model uses a graph structure to store all concepts, attributes, and relationships. Nodes represent chemical entities or parameters, edges represent attributes or reaction associations, and dynamic updates and queries are achieved through a database system.

[0021] When mapping flocculant characteristic data to a flocculation and defluorination knowledge model to form a flocculation network, the flocculant characteristic data must first be preprocessed to obtain preprocessed flocculant data. This data originates from product technical manuals or experimental results and includes flocculant type (e.g., inorganic flocculant polyaluminum chloride or organic flocculant polyacrylamide), molecular structure information, charge characteristics (e.g., zeta potential), hydrolysis performance (e.g., degree of hydrolysis), and optimal pH range for addition. Preprocessing operations include missing value imputation using mean or median interpolation, unit conversion to standard units, and outlier removal based on statistical outlier detection methods. Based on the flocculation and defluorination knowledge model, the preprocessed flocculant data is mapped. This mapping process uses entity recognition algorithms to match entities in the preprocessed flocculant data, such as ferrous sulfate or polyferric sulfate, to corresponding component nodes in the knowledge model. Component nodes represent the abstract representation of flocculant chemical substances in the knowledge model. Simultaneously, attributes in the preprocessed flocculant data, such as alkalinity or molecular weight, are mapped to attribute edges. These attribute edges connect component nodes to attribute value nodes and are labeled with attribute types. Based on a knowledge model for flocculation and fluoride removal, relation extraction is performed on pretreated flocculant data. This extraction utilizes rule matching or machine learning methods to extract chemical relationships between entities in the pretreated flocculant data, such as identifying the charge neutralization reaction between aluminum salt flocculants and fluoride ions, and the net-catching and sweeping effect of iron salt flocculants. These chemical relationships are mapped to reaction edges, which connect interacting component nodes and are labeled with reaction types and condition parameters. Finally, a flocculation network is constructed based on component nodes, attribute edges, and reaction edges. This network integrates flocculant characteristics and the wastewater chemical environment into a unified graph structure. Nodes cover all flocculant components and wastewater components, edges describe attribute dependencies and reaction paths, and efficient traversal and reasoning are achieved through a graph database.

[0022] In one embodiment, Example 2: Text parsing is performed on structured coking wastewater data to obtain key information elements. Text parsing focuses on non-numerical descriptive content in the data, such as treatment process descriptions, pollutant component annotations, or abnormal situation records. The parsing process employs natural language processing technology, including word segmentation to break continuous text into lexical units, part-of-speech tagging to identify noun terms and numerical modification relationships, and syntactic analysis to determine the modification logic between parameters. Key information elements are extracted as data fragments with independent semantics, such as "the proportion of fluoride complexes increases under high temperature conditions" or "insufficient calcium ion concentration affects precipitation effect." Based on these key information elements, entity naming is performed to obtain key entities. Entity naming is achieved through named entity recognition technology, which matches professional terms in the text based on a predefined wastewater treatment dictionary. The dictionary includes standard entries such as chemical substance names, operating parameters, and equipment types. The recognition process combines rule matching with a statistical model. Rule matching relies on keyword lists and pattern rules, while the statistical model uses a conditional random field algorithm to identify contextual features. Key entities are labeled with standardized names, for example, "Ca2+" is uniformly labeled as "calcium ion," and "TOC" is labeled as "total organic carbon."

[0023] The key entities are defined by type definition. The type definition categorizes entities according to the knowledge system of wastewater treatment, including ion categories (e.g., anions, cations), organic components (e.g., phenols, polycyclic aromatic hydrocarbons), physical parameters (e.g., temperature, turbidity), and operating conditions (e.g., stirring intensity, reaction time). The definition process is completed using a type inference algorithm, which references the modifiers and numerical characteristics of entities in the text. For example, a numerical entity modified by "mg / L" is defined as a concentration parameter, and a noun used with "adsorption" is defined as a pollutant. Based on the defined key entities, relation extraction is performed to determine semantic associations. Relation extraction analyzes the interaction between entities in textual descriptions and numerical relationships. The extraction method combines syntactic dependency analysis and semantic role labeling to identify causal, comparative, and conditional logical relationships between entities. For example, from "pH decreases, leading to a decrease in fluoride precipitation efficiency," the negative impact relationship between "pH" and "efficiency" is extracted; from "turbidity decreases after adding aluminum salt," the relationship between "aluminum salt" and "turbidity reduction" is extracted. Semantic associations are stored in triplets (subject, relation type, object).

[0024] Attribute data is obtained by defining attributes based on the defined key entities. For each key entity, the attribute definition extracts its quantitative characteristics or state description. Numerical attributes are obtained by directly extracting corresponding values ​​from the data table, such as defining calcium ion concentration as "125 mg / L" and pH as "8.3". Descriptive attributes are extracted from text annotations, such as defining fluoride speciation as "predominantly complexed" and water temperature as "fluctuating significantly". A one-to-one correspondence is established between attribute data and key entities. The defined key entities, attribute data, and semantic relationships are organized into a structured knowledge representation to obtain key processing elements. The knowledge representation is integrated using a framework structure or ontology model. The framework structure designs slots for filling attribute values ​​for each type of entity; for example, for "fluoride ions", it designs concentration slots, speciation slots, and influencing factor slots. The ontology model uses OWL language to define entity categories, attributes, and relational constraints. All elements are stored and managed through a knowledge graph database, forming a queryable and reasonable structured knowledge system. The entire implementation process emphasizes the hierarchical nature of data transformation and the completeness of knowledge representation, ensuring that key processing elements can accurately capture the core factors affecting fluoride removal efficiency.

[0025] Taking a coking wastewater treatment case as an example, the wastewater treatment station of a coking plant's second-phase project needed to perform defluoridation pretreatment on the effluent from its equalization tank. The operators collected test reports and operation logs for a recent batch of wastewater samples. These raw data included ICP-MS test reports issued by the laboratory (recording fluoride ion concentration of 18.7 mg / L, calcium ion concentration of 95 mg / L, and magnesium ion concentration of 42 mg / L), gas chromatography-mass spectrometry reports (detecting phenol concentration of 68 mg / L and total polycyclic aromatic hydrocarbons of 12 μg / L), pH fluctuation data recorded by online sensors (range 6.8-7.5), and daily operating condition records filled out by the operators (noting "recently, water temperature fluctuations are large, often rising above 35℃ in the afternoon"). When structuring this batch of heterogeneous data, a pre-designed standard data template was used for format conversion. The template includes fields such as fluoride concentration (floating-point type), heavy metal concentration (array type), organic matter index (structure type), and physical parameters (key-value pair type). During the processing, obvious outliers (such as the incorrectly entered pH value 12.5) were removed through a data cleaning program. The units were standardized by converting mg / L to a unified unit of measurement. The "F-" label was mapped to the standard "fluoride ion" field through field mapping. Finally, structured coking wastewater data conforming to the JSON format was generated.

[0026] When performing text parsing on structured coking wastewater data, the focus is on processing the natural language descriptions in the operating condition record table. A rule-based word segmenter is used to break the text down into lexical units, identifying keywords such as "water temperature," "large fluctuations," "afternoon," and "increase." Dependency parsing is used to establish the modification relationship of "water temperature-fluctuation-large" and the temporal association of "afternoon-increase," extracting key information elements including water temperature fluctuation characteristics, fluctuation time patterns, and temperature extreme value trends. When naming entities based on key information elements, a pre-built wastewater treatment dictionary is used to match professional terms. The dictionary includes standardized mappings such as "water temperature"="water_temperature" and "fluctuation"="fluctuation." A conditional random field model is used to identify contextual features, recognizing "increase" as an action entity and "afternoon" as a time entity, ultimately outputting a standardized set of key entities.

[0027] When defining the types of key entities, classification inference is performed based on the ontology of wastewater treatment: "water_temperature" is classified as a physical parameter, "fluctuation" as a state feature, "afternoon" as a time condition, and "increase" as a trend of change. The type definition algorithm refers to the modification relationship of entities in the text. For example, in "large fluctuation", "large" is used as an adverb of degree to modify the fluctuation state, so the entity "fluctuation" is given an "amplitude" attribute. When extracting relations based on the defined key entities, semantic role labeling technology is used to analyze the sentence "water temperature rises in the afternoon", identifying "water temperature" as the topic, "rises" as the action, and "afternoon" as a time adverb. The triple relations (water_temperature, increase_in, afternoon) and (fluctuation, has_amplitude, large) are extracted. When defining attributes based on the defined key entities, corresponding values ​​are extracted from the structured data: the water_temperature entity is given the range attribute value of "30-38℃", the fluctuation entity is given the frequency attribute value of "daily", and the increase entity is given the slope attribute value of "steep"; qualitative attributes are extracted from the text description: the polycyclic aromatic hydrocarbon entity is given the composition attribute value of "naphthalene series".

[0028] When organizing the defined key entities, attribute data, and semantic relationships into a structured knowledge representation, an ontology-based knowledge modeling method is adopted. The `WaterQualityParameter` class and its subclasses `PhysicalParameter` and `ChemicalParameter` are defined, as are the `Relation` class and its subclasses `TemporalRelation` and `CausalRelation`. Entities are created as instances of the classes and populated with attribute slots; for example, the `hasValue` slot of a `water_temperature` instance is filled with "30-38℃", and the `hasUnit` slot is filled with "℃". Relation instances are stored using subject-predicate-object triples; for example, the `subject` slot of a `TemporalRelation` instance points to `water_temperature`, the `predicate` slot points to `occur_in`, and the `object` slot points to `afternoon`. The final knowledge representation is stored in RDF format, generating a key processing element knowledge graph containing 15 entity nodes, 22 attribute edges, and 8 relation edges. This graph fully represents the core elements affecting the defluorination effect of the coking wastewater and their inherent relationships.

[0029] In one embodiment, Example 3: See Figure 3 The process maps key processing elements to a flocculation network to form a processing element graph. Mapping these key processing elements to component nodes within the flocculation network is achieved through a node matching algorithm. This algorithm calculates the semantic similarity between key entities and nodes in the flocculation network. Semantic similarity comprehensively considers name matching, attribute overlap, and contextual relevance. Name matching uses an edit distance algorithm to calculate text similarity; attribute overlap compares the defined attribute sets between the two entities; and contextual relevance analyzes the neighborhood structure of nodes within the network. The matching process traverses all entities within the key processing elements, comparing them with the component node library of the flocculation network. When the similarity exceeds a set threshold, a mapping relationship is established, thus obtaining the relevant component nodes of the key processing elements. These nodes constitute the focus set for subsequent processing. Based on the relevant component nodes of the key processing elements, node expansion is performed in the flocculation network to obtain extended nodes. The node expansion adopts a breadth-first search strategy to traverse the connection edges of the network. The search starts from the key node, visits the first-order adjacent nodes directly connected to it, records the identifiers of these nodes and their relationship types with the key node, and then expands to second-order adjacent nodes as needed. The expansion range is controlled by a preset depth parameter, which is dynamically adjusted according to the processing complexity requirements. The set of extended nodes includes all new nodes found during the traversal. These nodes may represent auxiliary components or influencing factors indirectly related to the key elements.

[0030] Subgraphs are extracted from the flocculation network based on extended nodes and key processing element-related component nodes. The subgraph extraction operation is performed using a graph database query language, with query conditions specifying all node identifiers to be included. The extraction process preserves all attribute edges and reaction edges between these nodes, maintaining their topological relationships in the original network. The subgraph content covers node attributes, edge attributes, and edge directionality information. The extraction result constitutes a complete connected subnetwork focusing on key elements and their related environments. Redundancy pruning is performed on the subgraph to obtain a processing element graph. Redundancy pruning is based on node importance assessment and edge relevance analysis. Node importance is quantified using graph indicators such as degree centrality and betweenness centrality, prioritizing the removal of nodes with low importance and those not directly related to key processing elements. Edge relevance is determined based on edge type and weight, removing edges with weights below a threshold that do not affect overall connectivity. The pruning process iterates until the subgraph meets the requirements of simplicity and criticality, ultimately resulting in a processing element graph that retains core decision-making elements and their main associations.

[0031] When quantifying and scoring preliminary treatment plans, quantitative dimensions and indicators are determined based on defluorination requirements and standards. Defluorination requirements refer to national emission standards and internal quality control indicators of enterprises. Quantitative dimensions cover technical dimensions such as defluorination efficiency, economic dimensions such as cost input, and operational dimensions such as feasibility. Specific quantitative indicators are set under each dimension. Technical dimensions include fluoride removal rate, residual fluoride concentration, and floc settling velocity; economic dimensions include unit consumption cost of flocculant, energy consumption costs, and equipment maintenance expenses; and operational dimensions include pH adjustment frequency, degree of automation, and process compatibility. Weights are assigned to quantitative dimensions and indicators. The weight allocation is determined using the analytic hierarchy process (AHP) combined with expert consultation. First, a hierarchical structure model of dimensions and indicators is constructed. The relative importance of each element is calculated through pairwise comparison judgment matrices. Consistency checks ensure the rationality of the judgment logic. The final weight values ​​are a combination of scores from multiple field experts. The economic dimension may have a higher weight than the operational dimension, and the fluoride removal rate indicator has a higher weight in the technical dimensions.

[0032] The preliminary treatment plans are quantitatively scored based on quantitative indicators and weights to obtain a plan score. The scoring process employs a multi-attribute decision model, which normalizes the indicator values ​​of each plan to eliminate dimensional differences. The indicator value types include benefit-type indicators (such as removal rate) and cost-type indicators (such as consumption cost), which are processed using maximization normalization and minimization normalization methods, respectively. The total plan score is obtained by weighting and summing the normalized indicator values ​​with their corresponding weights. The calculation model is as follows: in: This represents the score of the j-th preliminary processing plan. This represents the weight coefficient corresponding to the i-th quantitative indicator. This represents the normalized value of the j-th scheme on the i-th index. This represents the total number of quantitative indicators. The scoring results of all options form an ordered set, providing a quantitative basis for subsequent decision-making. The entire implementation process emphasizes the accuracy of graph operations and the systematic nature of the scoring model to ensure the scientific nature and comparability of the decision-making basis.

[0033] In one embodiment, Example 4: Taking the actual operation scenario of a coking plant wastewater treatment station as an example, the station collected a set of preliminary treatment schemes and their scoring data. Scheme A suggested using polyaluminum chloride as the main flocculant, with a dosage of 85 mg / L and controlling the pH in the range of 6.8-7.2. Scheme B suggested using polyferric sulfate, with a dosage of 120 mg / L and controlling the pH in the range of 7.5-7.8. The scheme scores were obtained through multi-dimensional quantitative evaluation. Scheme A had a higher score in the technical dimension but a lower score in the economic dimension, while Scheme B had a higher score in the economic dimension but had defects in the operational dimension. When integrating the scheme scores and preliminary treatment schemes for analysis, a score-parameter correlation matrix needs to be established. The rows of the matrix correspond to different schemes, and the columns correspond to the scoring dimensions and operational parameters. The analysis process compares the balance between the technical feasibility, economy, and operational convenience of each scheme, identifies the advantages of scheme A in terms of high fluoride removal rate and its disadvantages in terms of high reagent cost, and identifies the advantages of scheme B in terms of cost control and its disadvantages in terms of pH adjustment complexity. The integrated analysis also considers actual operating constraints, such as the processing capacity of existing equipment, the technical level of operators, and the stability of reagent supply.

[0034] The final processing decision is derived based on preset scoring thresholds and scheme scores. The scoring thresholds are set according to historical operational data: 70 points for the technical dimension, 60 points for the economic dimension, and 65 points for the operational dimension. Scheme A scores 82 points in the technical dimension, 58 points in the economic dimension, and 75 points in the operational dimension; Scheme B scores 68 points in the technical dimension, 72 points in the economic dimension, and 62 points in the operational dimension. The decision-making rules require that all schemes meet the thresholds in all dimensions to be directly adopted. If any dimension fails to meet the standard, parameter adjustments are required. Scheme A enters the adjustment process because its economic dimension is below the threshold, while Scheme B enters the pending state because its technical and operational dimensions are close to the thresholds. Parameter adjustment addresses the insufficient economic dimension by reducing the dosage of polyaluminum chloride to 70 mg / L and adding 5 mg / L of coagulant aid polyacrylamide to reduce the total cost. After adjustment, the economic dimension score is re-evaluated and improved to 65 points, resulting in the final processing decision selecting the adjusted Scheme A.

[0035] Acquiring time-series data of coking wastewater parameters is a crucial component of this embodiment. The time-series data comes from online monitoring instruments installed in the equalization and reaction tanks, monitoring parameters including fluoride ion concentration, pH value, turbidity, and water temperature. Data is collected every 15 minutes, continuously for 72 hours, resulting in 288 sets of time-series data. Data is stored in a time-series database, with each data point including a timestamp, parameter type, value, and equipment status identifier. Parameter variation deviations are calculated based on the time-series data using a moving window statistical method, with a window size of 12 data points (3 hours). For the fluoride ion concentration series, the ratio of the standard deviation to the moving average within each window is calculated as the relative deviation; for the pH series, the difference between the maximum and minimum values ​​within each window is calculated as the absolute deviation. The deviation calculation reveals the parameter fluctuation patterns: fluoride ion concentration exhibits periodic peaks during shift changes, and pH value fluctuates drastically after the addition of alkali solution before gradually stabilizing.

[0036] The parameter variation values ​​are obtained by analyzing the deviation of parameter changes. These values ​​are divided into two categories: trend-based and random-based. Trend-based values ​​are obtained by linearly fitting time series data to obtain the slope value. Positive values ​​indicate a continuous increase in the parameter, while negative values ​​indicate a continuous decrease. The slope of the fitted fluoride ion concentration sequence is 0.08 mg / L / h, indicating a slow upward trend in concentration. Random-based values ​​are obtained by calculating the standard deviation of the detrended sequence. The standard deviation of the pH sequence is 0.35, indicating large fluctuations. The parameter variation values ​​are used to optimize the extraction of key processing elements. When a trend-based value is detected, the extraction of key processing elements includes descriptions of the trend direction and rate. When the random-based value is large, the extraction process focuses more on extreme fluctuations rather than average values.

[0037] Table 1: Comparison of Allocation Scores and Decision Results

[0038] Referring to Table 1, when calculating the spatial distribution differences of coking wastewater in the reactor, a multi-point sampling method was used to obtain data. Sampling points were set at the inlet, middle and outlet of the reactor, and water samples were collected every 2 hours to detect fluoride ion concentration and turbidity. The spatial distribution differences were characterized by calculating the concentration gradient at different points at the same time. The concentration difference between the inlet and the middle reflects the initial mixing effect, and the concentration difference between the middle and the outlet reflects the degree of reaction completion. The test found that the fluoride concentration at the inlet fluctuated greatly (12.5-18.3 mg / L), the concentration in the middle tended to be uniform (14.8-16.1 mg / L), and the concentration at the outlet further decreased but occasionally showed a high level (9.5-11.2 mg / L). The flocculation adaptability index was obtained by combining parameter changes and spatial distribution differences. The index was calculated using a weighted comprehensive method, with the weights allocated as follows: 0.6 for time-dimensional changes and 0.4 for spatial differences. The time-dimensional changes were calculated as the average of the absolute value of the slope of the fluoride ion concentration trend and the pH standard deviation; the spatial differences were calculated as the ratio of the maximum concentration difference between the reactor inlet and outlet to the average concentration. The calculated flocculation adaptability index was 0.72, reflecting the system's overall adaptability to flocculation treatment. The initial treatment plan was adjusted based on the flocculation adaptability index. When the index was below 0.75, pretreatment was considered in the plan formation process, such as adding a homogenization tank or increasing the mixing intensity. When the index was above 0.75, the plan focused more on optimizing the main reaction process. In this case, the index was 0.72, so pre-homogenization measures were added to the adjustment of Plan A, extending the mixing time to 5 minutes to ensure the wastewater reached a more stable state before entering the main reactor. The entire implementation process demonstrated the characteristics of real-time data-driven decision-making, improving the adaptability and reliability of the treatment plan through multi-source data fusion and dynamic adjustment.

[0039] In one embodiment, Example 5: the process of optimizing flocculation and defluorination decisions through multi-dimensional data fusion and dynamic weight adjustment is illustrated using actual operating data from a coking wastewater treatment plant as an example. Obtaining wastewater treatment condition data is the first step. This data comes from the plant's distributed control system (DCS) and daily laboratory monitoring records, covering three main categories: reactor geometric parameters, operating parameters, and water quality parameters. Reactor geometric parameters include an effective volume of 45 m³, a length-to-diameter ratio of 2.3:1, and agitator blade type (four-bladed turbine). Operating parameters include an agitation speed gradient of 85 s⁻¹, a hydraulic retention time of 42 min, and a sludge return ratio of 0.35. Water quality parameters include an influent fluoride ion concentration of 15.8 mg / L, a COD concentration of 480 mg / L, and a SS concentration of 110 mg / L. The data is collected in real-time from the control system via the OPC protocol, updated hourly, and stored in a relational database. When the wastewater treatment conditions meet the first condition, the weight parameters are adjusted. The first condition, based on historical operating experience, is a combined threshold of an effective reactor volume of less than 50 m³ and a stirring speed gradient exceeding 80 s⁻¹. In the current data, the reactor volume of 45 m³ is below the threshold, and the stirring speed gradient of 85 s⁻¹ exceeds the threshold, thus triggering the weight adjustment mechanism. The weight parameters are adjusted for the technical, economic, and operational dimensions in the scheme quantitative evaluation. The original weight ratio was 0.5 for technology, 0.3 for economy, and 0.2 for operation. After adjustment, the weight of the operation dimension is strengthened to 0.3, the weight of the technical dimension is reduced to 0.4, and the weight of the economic dimension remains at 0.3. The weight adjustment is based on the impact of volume limitations on mixing efficiency and the risk of floc destruction by high shear force, making the scoring system more in line with actual constraints.

[0040] The flocculation complexity index was calculated based on the adjusted weight parameters. A multi-factor weighted model was used, selecting four core factors: reactor volume factor (the reciprocal of the ratio to the standard volume of 50 m³), ​​stirring intensity factor (the absolute value of the difference between the actual velocity gradient and the optimal gradient of 70 s⁻¹), water quality fluctuation factor (the coefficient of variation of fluoride ion concentration over the past 8 hours), and reagent compatibility factor (the reaction compatibility score between the currently used flocculant and coexisting ions in the wastewater). Each factor was assigned the same adjusted weight: 0.4 for technology-related factors, 0.3 for economic-related factors, and 0.3 for operational-related factors. The calculated flocculation complexity index was 0.68, which falls within the medium complexity range (0.6-0.8).

[0041] Multiple candidate solutions were ranked based on the flocculation complexity index to optimize the generation of the final treatment decision. The candidate solutions included Solution C (90 mg / L polyaluminum chloride + 0.5 mg / L anionic PAM, pH=7.0), Solution D (110 mg / L polyferric sulfate, pH=7.5), and Solution E (calcium chloride + aluminum salt combined addition, Ca²⁺ 40 mg / L + Al³⁺ 60 mg / L, pH=6.8). The flocculation complexity index was incorporated as a correction coefficient into the original scoring system during the ranking process. The fitness score is obtained by multiplying the original score of the scheme by (1 + flocculation complexity index). Scheme C originally scored 82 points, and its fitness score is 82 × 1.68 = 137.8. Scheme D originally scored 78 points, and its fitness score is 78 × 1.68 = 131.0. Scheme E originally scored 85 points, and its fitness score is 85 × 1.68 = 142.8. Based on the fitness scores, the schemes are reordered as Scheme E > Scheme C > Scheme D. The sorting result prioritizes Scheme E, which can better cope with complex working conditions, as the final processing decision.

[0042] To calculate the spatial distribution differences of coking wastewater in the reactor, conductivity imaging technology was used to assist traditional sampling methods. Three monitoring sections (inlet, middle, and outlet) were set up along the reactor axis, with four conductivity sensors arranged in each section. Monitoring data showed that the conductivity fluctuation range of the inlet section was 650-820 μS / cm, the middle section tended to be stable at 720-780 μS / cm, and the outlet section decreased to 680-740 μS / cm. The spatial distribution differences were obtained by calculating the ratio of the maximum conductivity difference to the average value within the same section. The ratio was 0.23 for the inlet section, 0.08 for the middle section, and 0.09 for the outlet section, with an overall spatial non-uniformity coefficient of 0.15. The flocculation fitness index was obtained by combining parameter variation values ​​and spatial distribution differences. The parameter variation values ​​were taken from the fluoride ion trend slope (0.08 mg / L / h) and pH standard deviation (0.35) calculated in Example 4, and the spatial distribution differences were taken from the overall non-uniformity coefficient (0.15). The flocculation fitness index was calculated using a three-dimensional weighted formula, with a time dimension parameter weight of 0.5, a spatial dimension parameter weight of 0.3, and an operational constraint weight of 0.2, resulting in a score of 0.62, reflecting that the system's adaptability to flocculation treatment is at a moderately low level. Based on the flocculation fitness index, the formation process of the preliminary treatment scheme was adjusted. When the index was below 0.65, the scheme formation process added enhanced mixing measures and pretreatment unit optimization. In this case, the index was 0.62, so based on Scheme E, a pipeline static mixer design was added (the number of mixing units was increased by 2) and the residence time in the neutralization and equalization tank was extended to 25 min to ensure that the wastewater reached a higher uniformity requirement before entering the main reactor.

[0043] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for flocculation-based defluorination of coking wastewater, characterized in that, The method includes: Obtain water quality parameter data for coking wastewater; A knowledge model for flocculation and fluoride removal is constructed based on the aforementioned water quality parameter data; Acquire flocculant characteristic data and map the flocculant characteristic data into the flocculation and defluorination knowledge model to form a flocculation network containing component nodes, attribute edges, and reaction edges; Key elements for treating coking wastewater; The key processing elements are mapped onto the flocculation network to obtain the key processing element related component nodes, and a processing element diagram is formed based on the key processing element related component nodes. Based on the flocculation network and the preset defluorination rules, the processing element diagram is inferred and judged to form a preliminary processing plan; The preliminary processing plan is quantitatively scored to obtain a plan score; Based on the proposed solution score and the preliminary processing solution, parameter adjustment decisions are made to obtain the final processing decision.

2. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The construction of the flocculation and fluoride removal knowledge model based on the water quality parameter data includes: Data processing techniques are used to extract core concepts related to flocculation and fluoride removal from the water quality parameter data; Based on the core concepts and the water quality parameter data, the chemical relationships between the core concepts are determined, and the model relationships are defined according to the chemical relationships. Based on the model relationship definition, the core concepts are defined with attributes to obtain attribute definition data; The flocculation and fluoride removal knowledge model is constructed based on the attribute definition data.

3. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The process of mapping the flocculant characteristic data into the flocculation and defluorination knowledge model to form a flocculation network containing component nodes, attribute edges, and reaction edges includes: The flocculant characteristic data is preprocessed to obtain preprocessed flocculant data; Based on the flocculation and defluorination knowledge model, the pre-treated flocculant data is mapped, the entities in the pre-treated flocculant data are mapped to corresponding component nodes, and the attributes in the pre-treated flocculant data are mapped to attribute edges. Based on the flocculation and defluorination knowledge model, relation extraction is performed on the pretreated flocculant data, the chemical relationships between entities in the pretreated flocculant data are extracted, and the chemical relationships are mapped as reaction edges. A flocculation network is constructed based on the component nodes, attribute edges, and reaction edges.

4. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The key treatment elements for extracting coking wastewater include: The coking wastewater data was structured to obtain structured coking wastewater data; Text parsing was performed on the structured coking wastewater data to obtain key information elements; Based on the key information elements, entity naming is performed to obtain key entities; The key entities are type-defined to obtain the defined key entities; Based on the defined key entities, relations are extracted to determine semantic associations; Based on the defined key entities, attribute definitions are performed to obtain attribute data; The defined key entities, attribute data, and semantic relationships are organized into a structured knowledge representation to obtain key processing elements.

5. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The step of mapping the key processing elements to the flocculation network to obtain key processing element-related component nodes, and forming a processing element graph based on the key processing element-related component nodes, includes: The key processing elements are mapped to component nodes in the flocculation network to obtain component nodes related to the key processing elements. Based on the relevant component nodes of the key processing elements, node expansion is performed in the flocculation network to obtain extended nodes; Subgraphs are extracted from the flocculation network based on the extended nodes and the key processing element-related component nodes. Redundancy is removed from the sub-graph to obtain the processed element graph.

6. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The quantitative scoring of the preliminary processing scheme to obtain a scheme score includes: Based on the defluorination requirements and standards, determine the quantitative dimensions and quantitative indicators; Assign weights to the quantification dimensions and quantification indicators; Based on the quantitative indicators and weights, the preliminary processing scheme is quantitatively scored to obtain a scheme score.

7. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The step of making parameter adjustment decisions based on the scheme score and the preliminary processing scheme to obtain the final processing decision includes: The proposed solution score and the preliminary processing solution will be integrated and analyzed. A decision is made based on a preset scoring threshold and the score of the proposed solution, resulting in a final processing decision.

8. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The method further includes: Obtain time-series data of coking wastewater parameters; Calculate the parameter variation deviation based on the time series data; Analyze the deviation of the parameter changes to obtain the parameter change values; The parameter changes are used to optimize the extraction of the key processing elements.

9. The flocculation defluorination method for coking wastewater according to claim 8, characterized in that, The method further includes: Calculate the spatial distribution differences of coking wastewater in the reactor; By combining the changes in the parameters and the differences in their spatial distribution, the flocculation adaptability index is obtained; The formation process of the preliminary treatment scheme is adjusted based on the flocculation adaptability index.

10. The flocculation defluorination method for coking wastewater according to claim 1, characterized in that, The method further includes: Obtain wastewater treatment condition data; When the wastewater treatment condition data meets the first condition, the weighting parameter is adjusted. The flocculation complexity index is calculated based on the adjusted weight parameters; Multiple candidate solutions are ranked based on the flocculation complexity index to optimize the generation of the final processing decision.

Images