A mine soil pollution source treatment technology auxiliary decision-making method and system

Abstract

The present application provides a mine soil pollution source treatment technology auxiliary decision-making method and system, relating to the field of soil pollution treatment technology. In view of the problem that the prior art lacks self-adaptive optimization capability and the decision-making is easy to fall into a stalemate when facing severe constraints, the present application constructs a distributed directional crawler to automatically collect multi-source heterogeneous knowledge, extracts entities and relationships to construct an inferable knowledge graph; then converts the target area characteristics and treatment requirements into queries, performs multi-hop logical retrieval and matching in the graph, and quantitatively scores and sorts the feasible schemes through a multi-criteria decision analysis model; when the initial scheme is not ideal, the parameter optimization based on Bayesian optimization and the cross-validation of the knowledge graph are carried out, and when the optimization has no solution, the constraints are intelligently relaxed, the backtracking is re-evaluated, and a decision-making closed loop is formed. The present application significantly improves the quality, global optimality, practicality and robustness of the treatment scheme decision-making when facing complex, severe or special treatment requirements.

Description

Technical Field

[0001] This invention relates to the field of soil pollution remediation technology, and in particular to a method and system for assisting decision-making in the remediation of soil pollution sources in mining areas. Background Technology

[0002] Soil pollution remediation in mining areas is a crucial link in ecological environmental protection and restoration. Its complexity lies in the diversity of pollutants, the heterogeneity of geological conditions, and the multiplicity of remediation objectives. Utilizing intelligent decision support systems can systematically integrate massive amounts of knowledge to quickly and scientifically match and optimize feasible technical solutions for specific pollution scenarios, thereby significantly improving the success rate, cost-effectiveness, and timeliness of remediation projects. This has become an important development direction in the field of environmental engineering. Constructing an intelligent decision support system capable of integrating multi-source knowledge, understanding complex constraints, and simulating the impact of solutions holds broad prospects for promoting the precision and intelligent development of the soil remediation industry.

[0003] However, existing decision support methods have significant shortcomings in achieving scientifically and economically optimal solutions. First, most methods rely on structured databases or static expert rules, failing to fully utilize unstructured technical documents, project reports, and other diverse and heterogeneous information. This results in a narrow knowledge base with outdated updates, making it difficult to guarantee the comprehensiveness and cutting-edge nature of recommended solutions. Second, the recommendation logic of existing solutions often remains at a shallow matching level based on key conditions, lacking the ability to mine and reason about the deep relationships between technology, conditions, and performance. This prevents systematic quantitative comparison and ranking of candidate solutions, making it difficult to ensure the global optimality of the output. Furthermore, when a perfectly matching existing solution cannot be found, existing methods typically lack systematic parameter optimization and adaptive adjustment mechanisms. Especially when facing multiple hard constraints, they cannot automatically perform effective constraint trade-offs and solution iterations, leading to decision-making processes stalling and failing to provide a feasible solution approaching the optimal solution. These limitations challenge the effectiveness of existing decision support technologies in terms of scientific rigor, economic efficiency, and the achievability of the final solution.

[0004] Therefore, there is an urgent need for a new type of decision support method that can automatically construct knowledge systems from massive amounts of information, perform in-depth reasoning and quantitative evaluation, and has the ability to optimize in a closed loop and make adaptive adjustments. Summary of the Invention

[0005] To address these issues, the present invention provides a technology-assisted decision-making method and system for the treatment of soil pollution sources in mining areas. This system aims to overcome the problems in existing technologies, such as the failure to fully utilize multi-source heterogeneous information, the lack of ability to explore and reason about the deep correlation between technology, conditions, and performance, and the inability to provide feasible solutions that approach the optimal solution when the decision-making process stalls.

[0006] To achieve the above objectives, the present invention provides a technology-assisted decision-making method for the treatment of soil pollution sources in mining areas, comprising: S1, Construct a distributed, targeted crawler to collect information from predetermined soil remediation knowledge sources and obtain the original text dataset; S2, clean and preprocess the original text dataset to obtain a standardized text dataset; S3 uses a deep neural network model to extract technical entities, conditional entities, and performance entities from a standardized text dataset, generating triple entity information. S4 simultaneously extracts the relationships between the three entity information to obtain structured information; S5. Based on the structured information, construct a knowledge graph for soil pollution control; S6. Using the regional characteristic data of the target area and the target governance requirements as query conditions, perform multi-hop retrieval and matching in the soil pollution governance knowledge graph. If a matching governance plan exists, the governance plan is quantitatively scored and ranked through a multi-criteria decision analysis model. S7 outputs a sequence of recommended governance solutions based on the score ranking, and attaches the historical case with the highest matching degree, expected effect and key references for each solution; The nodes of the soil pollution control knowledge graph are technical entities, conditional entities, and performance entities, and the edges of the soil pollution control knowledge graph are the relationships between entities.

[0007] Furthermore, in S7, if the scores of all schemes in the recommended governance scheme sequence are lower than a preset threshold, then the parameter optimization process is initiated, including: S71, extract the set of adjustable parameters P of candidate solutions and the feasible domain of the adjustable parameters from the knowledge graph, as well as the associated multi-objective performance prediction proxy model. S72, with the primary objective of satisfying the hard constraints in the regional characteristic data of the target region and the secondary objective of approximating the target governance requirements, a loss function is constructed; S73, using the Bayesian optimization algorithm, iteratively searches for the optimal parameter vector within the feasible region to generate the optimal parameter set. The Bayesian optimization algorithm uses a Gaussian process as a surrogate model and uses a collection function to guide the parameter selection for each iteration. S74, perform cross-validation and evaluation on the optimal parameter set to generate optimized decision results.

[0008] Furthermore, the step of performing cross-validation and evaluation on the optimal parameter set to generate optimization decision results includes: S741, Using the optimal parameter set as the query condition, perform multi-hop retrieval and reasoning again in the soil pollution control knowledge graph; S742, based on the relationships between entities in the knowledge graph, identify and extract other technical entities, conditional entities or performance entities that are affected when the optimal parameter set is applied, and form a comprehensive impact assessment list; S743, compare all entities and their expected performance in the impact assessment list with the target governance requirements of the target region; S744, If all related effects can meet the target governance requirements, the verification is deemed successful, and an optimized governance scheme containing the optimal parameter set is generated and output. S745 If a negative impact or conflict that fails to meet the target governance requirements is identified, the verification is deemed unsuccessful, the feasible domain of the adjustable parameters is narrowed, and the process returns to step S72.

[0009] Furthermore, after step S745 determines that the verification failed, the process also includes: S745a, determine whether a complete traversal search has been performed on the feasible domain of the adjustable parameter set; S745b, if the feasible region has been completely traversed and no validated optimization solution has been generated, then the original timeliness constraint value is extracted, and the modified timeliness constraint value is calculated according to the preset relaxation ratio. S745c replaces the target governance requirement in the original query conditions with the corrected timeliness constraint value; S745d, based on the replaced query conditions, return and execute step S6, re-perform multi-hop retrieval and matching in the soil pollution control knowledge graph, and quantify and rank the matched control schemes through a multi-criteria decision analysis model.

[0010] Furthermore, the predetermined sources of knowledge for soil remediation include databases of Chinese and foreign technical literature, patent databases, publicly available case studies of soil pollution remediation projects and technical standards, publicly available soil pollution survey and remediation reports from government departments, environmental monitoring data platforms, technical documents from international environmental protection organizations, proceedings of authoritative academic conferences, and databases of practical guidelines.

[0011] Furthermore, the multi-hop retrieval and matching in the soil pollution remediation knowledge graph includes: S61, parse the query conditions and extract the specific condition entity features of the target region and the performance entities in the target governance requirements; S62, using the specific conditional entity features as initial nodes, search and match all conditional entity nodes that conform to or are close to the specific conditional entity features in the soil pollution control knowledge graph to form an initial matching set; S63, Starting from each conditional entity node in the initial matching set, perform a one-hop search along the edge that is restricted to or applicable to the relationship to locate all technical entity nodes directly associated with it, forming a technical candidate set; S64, Starting from each technical entity node in the set of technical candidates, perform a two-hop search along the edge that generates or causes the relationship to locate the performance entity node associated with it; S65, compare and filter the performance indicators represented by the retrieved performance entity nodes with the performance entity indicators in the target governance requirements; eliminate technologies and related paths whose associated performance cannot meet the target governance requirements.

[0012] Furthermore, the regional feature data of the target area specifically includes: Natural geographical conditions include climate type, soil type, soil pH value, groundwater level depth, topographic slope and geomorphic units; Site engineering conditions, including site area, current land use, accessibility, and infrastructure conditions; Environmental and social constraints, including the distribution of surrounding sensitive targets, local environmental regulations and emission standards, and a preliminary assessment of community acceptance.

[0013] Furthermore, the aforementioned target governance requirements specifically include: Core performance indicators include target pollutant removal rate or compliance concentration, soil ecological function restoration level, and post-treatment land use targets; Process and timeliness requirements, including the expected total governance cycle, time nodes for each stage, and time limits for construction and monitoring; Economic requirements include a total cost budget ceiling and cost control targets per unit area or per unit of pollutant; Technology and risk control requirements include the permissible range of technology maturity, acceptable levels of secondary environmental risk, and the burden of long-term monitoring and maintenance.

[0014] This invention also provides a decision support system for the treatment of soil pollution sources in mining areas, wherein the system employs any of the decision support methods for the treatment of soil pollution sources in mining areas described herein, and the system specifically includes: The information collection module is used to build a distributed, targeted crawler to collect information from predetermined soil remediation knowledge sources and obtain raw text datasets. The preprocessing module, connected to the information acquisition module, is used to clean and preprocess the original text dataset to obtain a standardized text dataset. The entity extraction module, connected to the preprocessing module, is used to extract technical entities, conditional entities, and performance entities from the standardized text dataset using a deep neural network model, generating triple entity information. The relationship extraction module, connected to the information extraction module, is used to simultaneously extract the relationships between the three entity information to obtain structured information. The graph construction module, connected to the relationship extraction module, is used to construct a knowledge graph for soil pollution control based on the structured information. The retrieval and matching module, connected to the graph construction module, is used to perform multi-hop retrieval and matching in the soil pollution control knowledge graph by using regional feature data of the target area and target governance requirements as query conditions. If a matching governance plan exists, the governance plan is quantitatively scored and ranked by a multi-criteria decision analysis model. The solution output module, connected to the retrieval and matching module, is used to output a sequence of recommended governance solutions based on the score ranking, and to attach the historical case with the highest matching degree, expected effect and key references to each solution.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: Firstly, this invention, through a knowledge graph-based multi-hop retrieval mechanism, can verify the complete logical chain, ensuring the technical feasibility of the recommended solutions. Combined with a multi-criteria decision analysis model, feasible solutions are quantitatively scored and ranked based on multiple objectives, transforming the decision-making basis from subjective experience-based judgment to objective data-driven decision-making. This systematically outputs a sequence of governance solutions with superior overall performance, significantly improving the quality and global optimality of the decisions. Secondly, this invention constructs a closed-loop decision-making method with self-optimization and dynamic adjustment capabilities. When the existing solution library cannot meet the requirements, this invention can automatically initiate a parameter optimization process, using a Bayesian optimization algorithm to search for optimal parameters within the knowledge-defined feasible domain, and conducting comprehensive cross-validation through a knowledge graph to avoid potential risks. If no solution is found, the system can intelligently relax core constraints and re-trigger global solution retrieval and evaluation, ensuring that it can still provide realistic and relatively optimal decision suggestions when facing complex, demanding, or special governance needs, thus enhancing the practicality, robustness, and problem-solving capabilities of the auxiliary decision-making system. Thirdly, this invention encodes the complex process of massive literature research, experience summarization, scheme comparison and parameter optimization, which are highly dependent on expert knowledge, into an automatically executable algorithm flow. This invention constructs a one-stop decision-making system from input problem to output solution, which greatly reduces the absolute dependence on the user's personal professional experience, is conducive to the popularization and standardized application of advanced governance knowledge, and significantly saves time and economic costs of manual research, evaluation and trial and error. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a technology-assisted decision-making method for the treatment of soil pollution sources in mining areas, provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of a technical auxiliary decision-making system for the treatment of soil pollution sources in mining areas, provided as an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0018] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0020] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] Example 1 like Figure 1 As shown, this invention provides a technology-assisted decision-making method for the treatment of soil pollution sources in mining areas, comprising: S1, Construct a distributed, targeted crawler to collect information from predetermined soil remediation knowledge sources and obtain the original text dataset; The planned sources of knowledge for soil remediation include Chinese and foreign technical literature databases, patent databases, publicly available case studies and technical standards for soil pollution remediation projects, publicly available soil pollution survey and remediation reports from government departments, environmental monitoring data platforms, technical documents from international environmental organizations, proceedings of authoritative academic conferences, and best practice guidelines.

[0022] In one possible implementation, a master-slave distributed architecture is used to build the targeted web crawler system, which is deployed on a cluster consisting of at least one master node server and multiple child node servers. The master node is responsible for task scheduling, URL seed management, deduplication control, and status monitoring; the child nodes are responsible for executing the specific web page crawling tasks.

[0023] Configure a dedicated crawler parsing template and collection strategy for each knowledge source or its sub-repository: For Chinese and foreign language technical literature databases, such as CNKI, Web of Science, and Elsevier ScienceDirect, configure crawlers to simulate academic searches, perform searches based on keyword combinations such as soil remediation, mining pollution, heavy metals, and organic matter, and capture bibliographic information from the search results, such as title, author, abstract, keywords, and full-text links. For open access literature, directly download the full text. For patent databases, such as those of the China National Intellectual Property Administration, the United States Patent and Trademark Office, and the European Patent Office, configure a web crawler to use the International Patent Classification (IPC) combined with technical keywords to search and retrieve patent titles, abstracts, claims, specifications, and figures. For publicly available project case libraries and technical standard libraries, a web crawler is configured to traverse sites and identify and download specific file formats from government environmental protection departments and research institutions, including acceptance reports of remediation projects and summaries of technical demonstration projects. For other knowledge sources, such as government reports, monitoring data platforms, and documents from international organizations, configure appropriate crawler parsers based on their website structure and data interfaces to extract structured or semi-structured data and text content.

[0024] The master node maintains a global priority queue of URLs to be crawled and a collection of crawled URLs based on Bloom filters and persistent storage. During system initialization, the entry URLs of each knowledge source are injected into the priority queue. During scheduling, the master node assigns priorities based on factors such as the importance and update frequency of the knowledge source to which the URL belongs. Before crawling, child nodes must first pass through the master node's deduplication check to avoid duplicate crawling. After receiving a crawling task, a child node requests page content via HTTP / HTTPS protocol, handles anti-crawler policies, and then uses XPath, CSS selectors, or regular expressions to accurately extract target text information and related metadata from the returned HTML, JSON, or XML content. Newly discovered related URLs parsed from the current page are submitted back to the master node's queue, thus achieving targeted breadth-first or depth-first traversal within a specific domain.

[0025] The child nodes will temporarily store the successfully crawled and parsed raw text content and its metadata in a structured or semi-structured format. The crawling results from all child nodes are ultimately aggregated and stored in a distributed file system or a high-performance NoSQL database, forming the original text dataset.

[0026] This invention, by employing a master-slave distributed architecture and parallel crawling technology, effectively solves the bottleneck of collection efficiency and single-point-of-failure risk inherent in single crawlers when facing massive, distributed online knowledge sources, significantly improving data collection throughput and system robustness. By pre-configuring dedicated parsing templates and collection strategies for different types of knowledge sources, it addresses the technical problem of traditional crawlers' difficulty in accurately and structurally acquiring specialized information from heterogeneous sources, significantly improving the relevance, relevance, and accuracy of content extraction of the original data. Furthermore, by combining Bloom filters and persistent storage to achieve global URL deduplication management, it effectively solves the problem of URL deduplication caused by circular or cross-referenced links in a distributed environment. The problem of redundant data collection is addressed by reducing the resource overhead of redundant data storage and subsequent processing. A comprehensive anti-crawler strategy, including simulating real requests, using proxy IP pools, and adhering to crawler protocols, solves the technical challenge of being blocked or restricted when collecting publicly accessible but access-protected knowledge sources, ensuring long-term, stable, and compliant data acquisition capabilities. Targeted crawling and domain keyword-driven approaches address the issues of significant interference from irrelevant information and low signal-to-noise ratio in information collection by general web crawlers, improving the relevance and knowledge density of the acquired raw text dataset to the professional field of mining area soil pollution remediation, thus laying a reliable data foundation for constructing a high-quality knowledge graph.

[0027] S2, cleans and preprocesses the original text dataset to obtain a standardized text dataset; In one possible implementation, the acquired raw text dataset, containing data from different sources and in different formats, undergoes preliminary parsing and standardization. Based on predefined rules and thresholds, it automatically identifies and removes advertising text, navigation links, copyright notices, and duplicate content unrelated to soil pollution remediation technologies. Regular expressions are used to remove or replace meaningless special characters, invisible characters, and obvious garbled sequences in the text. During the cleaning process, the association between each piece of text and its metadata is preserved or reconstructed. Metadata includes at least: source knowledge base type, such as patents or papers; unique identifiers, such as DOI, patent number, original URL, and crawling time. The cleaned and standardized text, along with this metadata, is encapsulated into independent, structured document records. All processed document records are then integrated to output a well-organized, uniformly formatted standardized text dataset.

[0028] S3 uses a deep neural network model to extract technical entities, conditional entities, and performance entities from a standardized text dataset, generating triple entity information. S4 simultaneously extracts the relationships between the three entity information to obtain structured information; S5, based on structured information, constructs a knowledge graph for soil pollution control; In the knowledge graph of soil pollution control, the nodes are technical entities, conditional entities, and performance entities, and the edges are the relationships between entities.

[0029] One possible implementation uses a large oracle model as a foundation, such as ERNIE3.0 or RoBERTa-wwm-ext, which segments each document record in the standardized text dataset into appropriate text fragments according to its natural paragraphs or semantic units, and uses these fragments as the sentence sequence input to the model.

[0030] A sequence labeling architecture based on the Big Oracle model, such as the BERT-BiLSTM-CRF model, is employed for named entity recognition. The input text fragments are labeled word-by-word. Text from the standardized text dataset obtained in S2 is input into the Big Oracle model, which performs word-by-word analysis and automatically identifies and labels all phrases belonging to technical, conditional, and performance entities. For example, from the sentence "In acidic clay, chemical oxidation technology can achieve a removal rate of over 90% for petroleum hydrocarbons, but the cost is high," the model will extract: technical entity "chemical oxidation," conditional entities "acidic clay" and "petroleum hydrocarbon pollution," and performance entities "removal rate can reach over 90%" and "high cost." All extracted entities, along with their position and type information in the original text, constitute the three-fold entity information set of the document.

[0031] For relation extraction, a joint learning or pipeline method based on a pre-trained model is employed to fine-tune a relation classification model. This model is used to determine the specific relations implied by a given entity pair and its context. Extracted entity pairs within the same context, along with their corresponding original text segments, are input into the fine-tuned relation classification model. The model analyzes the semantic context of the entity pairs, determines whether a predefined relation exists, and identifies the relation type. Each valid relation is output as a triple (head entity, relation, tail entity), serving as a structured piece of information. For example, (acidic clay, limited by, chemical oxidation) indicates that acidic clay conditions limit chemical oxidation technology; (petroleum hydrocarbon pollution, suitable for, chemical oxidation) indicates that chemical oxidation technology is suitable for treating petroleum hydrocarbon pollution; (chemical oxidation, produces, removal rate can reach over 90%) indicates that chemical oxidation technology produces high removal rate performance; and (chemical oxidation, leads to, higher cost) indicates that chemical oxidation technology leads to high cost performance.

[0032] Entities expressing the same meaning but with inconsistent wording in different texts are merged. For example, "chemical oxidation," "chemical oxidation method," and "Chemical Oxidation" are standardized into "chemical oxidation technology" and assigned a unique ID. Triples from different data sources that express the same fact are merged, and confidence levels are calculated based on the authority of their source and frequency of occurrence. All standardized technical entities, conditional entities, and performance entities are stored as nodes in a graph database. All standardized and merged triples are stored as directed edges connecting nodes in the database. The type of the edge is the relation type. Nodes and edges can be assigned attributes, such as entity aliases, relation confidence levels, and evidence sources.

[0033] Ultimately, a machine-readable, queryable, and reasonable knowledge graph for soil pollution remediation is formed. This graph, in network form, systematically encodes the domain knowledge logic under what conditions, using what technologies, and what expected performance results are generated or led to, providing core support for subsequent intelligent matching and decision-making.

[0034] This invention employs a deep neural network model for end-to-end entity recognition, effectively addressing the technical challenges of low recall and accuracy in traditional methods when dealing with technical texts characterized by diverse terminology and complex expressions. It significantly improves the accuracy and coverage of extracting key entities across three categories: technology, conditions, and performance. Through entity normalization and knowledge fusion technologies, it resolves the issues of ambiguity, information redundancy, and fragmentation in knowledge extracted from multi-source, heterogeneous texts, integrating scattered knowledge points into unified and standardized entity nodes. This significantly reduces noise and contradictions in the knowledge graph, enhancing its purity and credibility. By structurally storing massive amounts of triples in a graph database to construct a knowledge graph, it overcomes the fundamental limitations of traditional databases or document repositories in directly expressing and efficiently querying complex entity relationships. This achieves systematic encoding and intuitive network presentation of the deep logical relationships between conditions, technology, and performance in governance knowledge, greatly enhancing the interpretability, reasonability, and efficiency of related retrieval.

[0035] S6 uses the regional characteristic data of the target area and the target governance requirements as query conditions to perform multi-hop retrieval and matching in the soil pollution governance knowledge graph. If a matching governance plan exists, the governance plan is quantitatively scored and ranked through a multi-criteria decision analysis model. Multi-hop retrieval and matching are performed on the knowledge graph of soil pollution remediation, including: S61, parse the query conditions and extract the specific condition entity features of the target region and the performance entities in the target governance requirements; S62, using specific conditional entity features as initial nodes, search and match all conditional entity nodes that meet or are close to the specific conditional entity features in the soil pollution control knowledge graph to form an initial matching set; S63, Starting from each conditional entity node in the initial matching set, perform a one-hop search along the edge that is restricted to or applicable to the relationship to locate all technical entity nodes directly associated with it, forming a technical candidate set; S64. Starting from each technical entity node in the technical candidate set, perform a two-hop search along the edge that generates or causes the relationship to locate the performance entity node associated with it. S65, compare and filter the performance indicators represented by the retrieved performance entity nodes with the performance entity indicators in the target governance requirements; eliminate technologies and related paths whose associated performance cannot meet the target governance requirements.

[0036] The specific regional characteristic data of the target area includes: Natural geographical conditions include climate type, soil type, soil pH value, groundwater level depth, topographic slope and geomorphic units; Site engineering conditions, including site area, current land use, accessibility, and infrastructure conditions; Environmental and social constraints, including the distribution of surrounding sensitive targets, local environmental regulations and emission standards, and preliminary assessment of community acceptance.

[0037] Target governance requirements specifically include: Core performance indicators include target pollutant removal rate or compliance concentration, soil ecological function restoration level, and post-treatment land use targets; Process and timeliness requirements, including the expected total governance cycle, time nodes for each stage, and time limits for construction and monitoring; Economic requirements include a total cost budget ceiling and cost control targets per unit area or per unit of pollutant; Technology and risk control requirements include the permissible range of technology maturity, acceptable levels of secondary environmental risk, and the burden of long-term monitoring and maintenance.

[0038] In one possible implementation, the system inputs regional characteristic data of the target area, such as soil type being clay, pH value being 5.2, main pollutant being lead with a concentration of 500 mg / kg, and target remediation requirements, such as removal rate greater than 85%, cost less than 800 yuan / ton, and cycle less than 12 months. The system parses the input content and maps it to entities in a knowledge graph. For example, the soil type being clay is mapped to the condition entity node "clay"; the pH value being 5.2 is mapped to the condition entity node "acidic soil"; and the main pollutant being lead with a concentration of 500 mg / kg is mapped to the condition entity node "heavy metal lead pollution".

[0039] In the graph, we search for conditional entity nodes that are highly similar to clay, acidic soil, and heavy metal lead pollution to form an initial matching set. Starting from these conditional nodes, we perform a one-hop search along the "constrained by" and "applicable to" relations. For example, starting from the clay node, we find the "constrained by" relation pointing to soil leaching, and the "applicable to" relation pointing to solidification; starting from the acidic soil node, we find the "constrained by" relation pointing to microbial remediation. Combining the search results of all conditional nodes, we remove technologies constrained by most conditions and retain technologies that are applicable to at least one condition and are not strictly constrained by the key conditions, forming a technology candidate set.

[0040] Starting from each candidate technology node, a two-hop search is performed along the generation and causation relationship edges to obtain its associated performance entity nodes. For example, for solidification technology, the search yields (solidification, generation, leaching toxicity reduction >90%), (solidification, generation, treatment cycle: 6-9 months), (solidification, causation, cost: 600-1000 yuan / ton), and (solidification, causation, soil volume enhancement rate: 10-15%). The retrieved performance indicators are compared with the target. If a technology's key performance indicator, such as cost exceeding 1000 yuan / ton in this example, does not meet the hard target, then that technology and all its associated paths are eliminated. After screening, each remaining complete path constitutes a matching preliminary remediation plan.

[0041] Based on the target governance requirements, a comprehensive set of evaluation indicators is determined. For example, governance effectiveness corresponds to removal rate or compliance concentration, economic cost corresponds to unit cost, implementation cycle corresponds to total governance cycle, technical reliability corresponds to technical maturity, and environmental risk corresponds to secondary environmental risk. The weight of each indicator is determined. For each governance scheme to be evaluated, the performance entity information associated with it in the knowledge graph is converted into a specific score. For numerical performance, such as a removal rate of 95% and a cost of 500 yuan / ton, it is used directly or normalized. For example, the cost of all schemes is converted into a score between 0 and 1 according to the high and low values. For descriptive performance, such as high technical maturity and medium secondary risk, it is converted into a grade score according to the preset rules.

[0042] A standardized decision matrix is ​​constructed to determine the positive ideal solution (optimal value) and negative ideal solution (worst value) for each indicator. The Euclidean distance between each scheme and the positive and negative ideal solutions is calculated, as well as the relative closeness of each scheme to the ideal solution. The higher the closeness, the better the overall performance of the scheme. The quantitative scores and final ranking of all schemes are output to form a recommended remediation scheme sequence. For example, the chemical passivation scheme has a closeness of 0.85; the solidification scheme has a closeness of 0.72; and the phytoremediation scheme has a closeness of 0.65.

[0043] This invention effectively solves the core problems of traditional solution retrieval, which relies solely on keyword matching or single-condition filtering, leading to broken logical chains and insufficient feasibility verification. By performing multi-hop retrieval starting from regional features, it ensures that each preliminary retrieved solution possesses a complete and knowledge graph-verified technical logical loop, greatly improving the reliability and inherent rationality of solution recommendations. Furthermore, by setting a hard filter based on target performance at the end of the retrieval process, it addresses the problems of information overload and invalid solution contamination in the early stages of decision-making. This proactively eliminates all technical paths that do not meet core constraints, thereby significantly focusing on the set of feasible solutions and improving the efficiency and relevance of subsequent decision analysis. By seamlessly integrating the logical verification of graph retrieval with the quantitative optimization of the decision-making model, a progressive decision-making process is constructed, solving the disconnect problem of existing methods. This systematically elevates the selection of solutions from the question of whether they are usable to the level of which is better, significantly improving the refinement and practical value of decision support. Overall, by deeply integrating domain knowledge with decision science, this method provides decision-makers with a decision-making tool that combines strong knowledge reasoning support with rigorous mathematical analysis. This enables them to efficiently, clearly, and logically identify the governance solution with the best overall performance when faced with complex constraints, thereby significantly improving the persuasiveness, feasibility, and final decision quality of the entire decision support system.

[0044] S7 outputs a sequence of recommended governance solutions based on the score ranking, and attaches the historical case with the highest matching degree, expected effect and key references for each solution; In S7, if the scores of all solutions in the recommended governance solution sequence are lower than a preset threshold, the parameter optimization process is initiated, including: S71, extract the set of adjustable parameters P of candidate solutions and the feasible domain of the adjustable parameters from the knowledge graph, as well as the associated multi-objective performance prediction proxy model. S72, with the primary objective of satisfying the hard constraints in the regional characteristic data of the target area and the secondary objective of approximating the target governance requirements, constructs a loss function; S73 employs a Bayesian optimization algorithm to iteratively search for the optimal parameter vector within the feasible region, generating the optimal parameter set. The Bayesian optimization algorithm uses a Gaussian process as a surrogate model and uses a collection function to guide the parameter selection for each iteration. If the overall score of all recommended solutions is lower than the preset threshold, it indicates that the solutions in the existing solution library cannot meet the target well. The system will automatically start the parameter optimization process to generate an optimized solution.

[0045] In one possible implementation, one or more solutions with the highest scores are selected as the optimization basis. The system extracts the set of adjustable parameters P and their feasible domains from the knowledge graph. For example, for phytoremediation solutions, the adjustable parameters P include plant varieties such as Sedum aizoon and Pteris vittata, planting density (10-30 plants / m²), remediation agent dosage (0-5%), and maintenance period (12-36 months). Simultaneously, the system invokes or trains a multi-objective performance prediction proxy model associated with the current technical solution. This model can predict key performance indicators (KPIs) such as final remediation efficiency, total cost, and cycle time based on the input parameter vector.

[0046] Construct a loss function with the primary goal of satisfying hard constraints and the secondary goal of approximating ideal governance requirements. Hard constraints are the uncompromising parts from regional characteristic data and governance requirements, such as a total budget not exceeding 1 million yuan and the soil lead concentration after remediation must be lower than the national standard limit.

[0047] A Bayesian optimization algorithm is employed, using a Gaussian process as a surrogate model to fit the black-box relationship between parameters and the loss function. Expected improvement is used as the acquisition function to balance exploration and development. Specifically, a small number of initial points are randomly selected within the feasible region of the parameters and evaluated using a performance prediction surrogate model. Based on these evaluation points, the Gaussian process performance prediction surrogate model updates its estimate of the loss function distribution. The acquisition function calculates which point within the feasible region offers the best chance of improvement in the next evaluation; this point becomes the new candidate parameter vector. The candidate parameters are input into the performance prediction surrogate model to predict their performance and loss. The new parameter-loss data points are added to the historical dataset, and the surrogate model is updated. The iteration stops when a preset number of iterations is reached or the loss function converges. The parameter vector with the minimum loss throughout the search process is output and added to the optimal parameter set as the optimal parameters.

[0048] This invention fundamentally solves the fatal problem of optimization results violating core constraints in multi-objective optimization by constructing a loss function with hard constraints as the primary objective. It ensures that the optimization search process always takes place within the feasible solution space, strictly guaranteeing the compliance and basic feasibility of the generated optimization schemes, and ensuring that the optimization direction always aligns with the bottom line of actual engineering. By employing a Bayesian optimization algorithm with a Gaussian process as the surrogate model for iterative search, it effectively overcomes the inefficiencies, difficulty in handling black-box functions, and inability to balance exploration and development inherent in traditional parametric optimization methods when dealing with computationally expensive surrogate models. It can efficiently approach the global optimum with the fewest evaluations, significantly reducing the computational resource consumption and time cost of the optimization process, and achieving rapid and intelligent optimization. Furthermore, it constructs a complete parametric optimization closed loop, deeply integrating domain knowledge with modern optimization algorithms, achieving a qualitative leap from matching and selecting existing solutions to creating and generating customized new solutions. This significantly improves the problem-solving capabilities, flexibility, and performance potential of the final solution in decision support systems facing complex, special, or high-standard governance needs.

[0049] S74, performs cross-validation and evaluation on the optimal parameter set to generate optimization decision results, including: S741 uses the optimal parameter set as the query condition to perform multi-hop retrieval and reasoning again in the soil pollution control knowledge graph; S742, based on the relationships between entities in the knowledge graph, identify and extract other technical entities, conditional entities or performance entities that may be affected when applying the optimal parameter set, and form a comprehensive impact assessment list; S743 will comprehensively compare all entities and their expected performance in the impact assessment list with the target governance requirements of the target region. S744, If all related effects can meet the target governance requirements, the verification is deemed successful, and an optimized governance scheme containing the optimal parameter set is generated and output. S745 If a negative impact or conflict that fails to meet the target governance requirements is identified, the verification is deemed unsuccessful, the feasible domain of the adjustable parameters is narrowed, and the process returns to step S72.

[0050] In one possible implementation, the optimal parameter set is transformed into conditional entities recognizable by the knowledge graph. Starting with these newly generated conditional entities representing the optimized working condition, a multi-hop search is re-executed within the knowledge graph. This search goes beyond the direct paths of the original solution, exploring all possible associations to uncover potential indirect influences not considered in the original optimization objective. Reasoning is then based on the rich relationships within the knowledge graph, for example: Starting from the high phosphate addition condition node, in addition to pointing to the target technology chemical passivation, it also relates to the new performance entity soil phosphorus accumulation risk and groundwater phosphorus potential migration risk through possible relationships. Starting from the intermediate maintenance cycle condition nodes, and through potential influencing relationships, we can link the soil pretreatment time window of the technical entity to the subsequent phytoremediation conditions, implying the impact on the subsequent bioremediation connection.

[0051] Summarize all newly retrieved technologies, conditions, and performance entities that may be related to the application of the optimal parameter set, forming an impact assessment list that not only includes positive impacts but also highlights potential neutral or negative impacts. Each performance expectation in the list is then rigorously compared with the target governance requirements, such as determining whether the impact on the connection of subsequent phytoremediation conflicts with the established post-governance land use goals or technology combination requirements.

[0052] If all identified impacts in the list are deemed acceptable, conflict-free, or within tolerance after comparison, the verification is considered successful, and the final optimized decision result is generated. If the list contains negative impacts that fail to meet the target requirements, such as a high risk of phosphorus accumulation violating the requirement of no significant secondary pollution, or if irreconcilable conflicts are identified, such as a conflict between the optimized maintenance cycle and the required overall project duration, the verification is considered unsuccessful, and a correction mechanism is automatically initiated to locate the parameter that caused the verification failure. For example, if it is determined that the 5% addition amount caused the high risk, the feasible region of the corresponding parameter is dynamically narrowed. For example, the feasible region of the phosphate-based passivating agent addition amount is modified from [1%, 8%] to [1%, 3.5%] to exclude the high-risk parameter range. With the narrowed feasible region, the process returns to step S72, where the loss function is reconstructed based on the updated constraints, and a new round of Bayesian optimization search is initiated to find a suboptimal but globally feasible set of parameters within a new, more stringent safety boundary.

[0053] This invention addresses the technical blind spot of traditional parameter optimization verification, which focuses solely on preset target indicators and fails to systematically evaluate the unknown side effects or chain reactions that may be triggered by scheme adjustments, by transforming the optimal parameter set into conditions for multi-hop retrieval and reasoning within a knowledge graph. This expands the verification scope from points to the network, enabling a forward-looking and panoramic exploration of the potential impact of the optimization scheme. By establishing clear verification pass / fail judgment logic and correspondingly generating the final scheme or triggering parameter feasible domain adjustments, a self-correcting decision-making closed loop is constructed. This ultimately delivers a stress-tested, highly reliable optimization scheme. Furthermore, when conflicts are detected, it intelligently locates problematic parameters and narrows their feasible domain, guiding the optimization search to avoid infeasible areas, significantly improving the efficiency of subsequent optimization iterations and the overall robustness and engineering safety of the final generated scheme. The domain knowledge graph is transformed from a static information base into a dynamic verifier and navigator, reducing the technical risks of customized schemes and enhancing the intelligence level of the decision system in handling complex real-world constraints and the reliability of the output results.

[0054] After step S745 determines that the verification failed, the process also includes: S745a, determine whether a complete traversal search has been performed on the feasible domain of the adjustable parameter set; S745b, if the feasible region has been completely traversed and no validated optimization solution has been generated, then the original timeliness constraint value is extracted, and the modified timeliness constraint value is calculated according to the preset relaxation ratio. S745c replaces the target governance requirement in the original query conditions with the corrected timeliness constraint value; S745d, based on the replaced query conditions, return and execute step S6, re-perform multi-hop retrieval and matching in the soil pollution control knowledge graph, and quantify and rank the matched control schemes through a multi-criteria decision analysis model.

[0055] In one possible implementation, multiple rounds of Bayesian optimization and graph verification have been performed on the chemical passivation scheme for the target region. After each verification failure, the feasible region of the problem parameters was narrowed down according to rules. For example, the feasible region of the addition amount has been gradually narrowed down from [1%, 8%] to [1%, 2%]. The optimization history is checked, and the criteria for determining complete traversal are interval convergence, iteration saturation, and no new intervals. Specifically, interval convergence means that the feasible region of the key parameters has been narrowed down to below a preset minimum width threshold; iteration saturation means that for the currently significantly narrowed feasible region, Bayesian optimization has performed a preset maximum number of iterations, and no parameter set has been verified in multiple consecutive iterations; no new intervals means that further narrowing of the feasible region would lead to it not conforming to physical or engineering common sense. When any of the above criteria are met, the system determines that the feasible region of the current set of adjustable parameters has been completely traversed and there is no solution.

[0056] After confirming that a feasible solution cannot be generated under the current objective through parameter optimization, the system initiates objective backtracking. From the original objective governance requirements, it identifies the most flexible and usually most likely to be negotiated time constraints, extracts specific time constraint values ​​from the original requirements, such as a total governance cycle of less than 12 months, and calculates the corrected constraint values ​​according to a preset relaxation ratio, such as 25%.

[0057] The system automatically generates an updated set of target remediation requirements, where all requirements remain unchanged except for the timeliness constraint value, which is replaced with a revised constraint value. The updated target remediation requirements replace the corresponding parts of the original query conditions. Based on the original regional feature data and the updated target remediation requirements, the system re-inputs these requirements, jumps to and fully executes step S6, and performs a multi-hop search and matching again in the soil pollution remediation knowledge graph. Due to the relaxed timeliness requirement, remediation technologies that were previously filtered out due to their longer timeframes but may be more cost-effective or effective, such as some slow but low-cost phytoremediation technologies, are now included in the candidate set. A multi-criteria decision analysis model is used to quantitatively score and rank the new, broader set of candidate solutions, generating a new recommended remediation solution sequence. All solutions in the sequence meet the relaxed 15-month timeframe requirement, and effectiveness, cost, and other objectives are re-evaluated. For example, the new ranking becomes: phytoremediation combination technology solution, comprehensive score 0.78; slow-release chemical passivation solution, comprehensive score 0.70.

[0058] This invention employs a degradation and replanning strategy to strive for optimization under the original stringent objectives. When a path proves impassable, it appropriately relaxes the most likely loosened constraints and, based on the relaxed premises, restarts the global solution retrieval and comparison process. This ensures that the invention always provides decision-makers with a realistically feasible and relatively optimal governance path suggestion under current cognitive and technological conditions. This significantly enhances the practicality and robustness of the entire decision-making system.

[0059] Example 2 like Figure 2 As shown, the present invention also provides a decision support system for the treatment of soil pollution sources in mining areas, used to implement the decision support method for the treatment of soil pollution sources in mining areas according to any one of Embodiment 1, specifically including: The information collection module is used to build a distributed, targeted crawler to collect information from predetermined soil remediation knowledge sources and obtain raw text datasets. The preprocessing module, connected to the information acquisition module, is used to clean and preprocess the original text dataset to obtain a standardized text dataset. The entity extraction module, connected to the preprocessing module, is used to extract technical entities, conditional entities, and performance entities from a standardized text dataset using a deep neural network model, generating triple entity information. The relation extraction module, connected to the information extraction module, is used to simultaneously extract the relationships between the three entities to obtain structured information. The graph construction module, connected to the relation extraction module, is used to construct a knowledge graph for soil pollution control based on structured information. The retrieval and matching module, connected to the graph construction module, is used to perform multi-hop retrieval and matching in the soil pollution control knowledge graph by using regional feature data of the target area and target governance requirements as query conditions. If a matching governance solution exists, the governance solution is quantitatively scored and ranked through a multi-criteria decision analysis model. The solution output module, connected to the retrieval and matching module, is used to output a sequence of recommended governance solutions based on the score ranking, and to attach the historical case with the highest matching degree, expected effect and key references to each solution.

[0060] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A technology-assisted decision-making method for the treatment of soil pollution sources in mining areas, characterized in that, include: S1, Construct a distributed, targeted crawler to collect information from predetermined soil remediation knowledge sources and obtain the original text dataset; S2, clean and preprocess the original text dataset to obtain a standardized text dataset; S3 uses a deep neural network model to extract technical entities, conditional entities, and performance entities from a standardized text dataset, generating triple entity information. S4 simultaneously extracts the relationships between the three entity information to obtain structured information; S5. Based on the structured information, construct a knowledge graph for soil pollution control; S6. Using the regional characteristic data of the target area and the target governance requirements as query conditions, perform multi-hop retrieval and matching in the soil pollution governance knowledge graph. If a matching governance plan exists, the governance plan is quantitatively scored and ranked through a multi-criteria decision analysis model. S7 outputs a sequence of recommended governance solutions based on the score ranking, and attaches the historical case with the highest matching degree, expected effect and key references for each solution; The nodes of the soil pollution control knowledge graph are technical entities, conditional entities, and performance entities, and the edges of the soil pollution control knowledge graph are the relationships between entities.

2. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 1, characterized in that, In S7, if the scores of all schemes in the recommended governance scheme sequence are lower than a preset threshold, then the parameter optimization process is initiated, including: S71, extract the set of adjustable parameters P of candidate solutions and the feasible domain of the adjustable parameters from the knowledge graph, as well as the associated multi-objective performance prediction proxy model. S72, with the primary objective of satisfying the hard constraints in the regional characteristic data of the target region and the secondary objective of approximating the target governance requirements, a loss function is constructed; S73, using the Bayesian optimization algorithm, iteratively searches for the optimal parameter vector within the feasible region to generate the optimal parameter set. The Bayesian optimization algorithm uses a Gaussian process as a surrogate model and uses a collection function to guide the parameter selection for each iteration. S74, perform cross-validation and evaluation on the optimal parameter set to generate optimized decision results.

3. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 2, characterized in that, The process of cross-validating and evaluating the optimal parameter set to generate optimization decision results includes: S741, Using the optimal parameter set as the query condition, perform multi-hop retrieval and reasoning again in the soil pollution control knowledge graph; S742, based on the relationships between entities in the knowledge graph, identify and extract other technical entities, conditional entities or performance entities that are affected when the optimal parameter set is applied, and form a comprehensive impact assessment list; S743, compare all entities and their expected performance in the impact assessment list with the target governance requirements of the target region; S744, If all related effects can meet the target governance requirements, the verification is deemed successful, and an optimized governance scheme containing the optimal parameter set is generated and output. S745 If a negative impact or conflict that fails to meet the target governance requirements is identified, the verification is deemed unsuccessful, the feasible domain of the adjustable parameters is narrowed, and the process returns to step S72.

4. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 3, characterized in that, After step S745 determines that the verification failed, the process also includes: S745a, determine whether a complete traversal search has been performed on the feasible domain of the adjustable parameter set; S745b, if the feasible region has been completely traversed and no validated optimization solution has been generated, then the original timeliness constraint value is extracted, and the modified timeliness constraint value is calculated according to the preset relaxation ratio. S745c replaces the target governance requirement in the original query conditions with the corrected timeliness constraint value; S745d, based on the replaced query conditions, return and execute step S6, re-perform multi-hop retrieval and matching in the soil pollution control knowledge graph, and quantify and rank the matched control schemes through a multi-criteria decision analysis model.

5. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 1, characterized in that, The predetermined sources of knowledge for soil remediation include databases of Chinese and foreign technical literature, patent databases, publicly available case studies and technical standards for soil pollution remediation projects, publicly available soil pollution survey and remediation reports from government departments, environmental monitoring data platforms, technical documents from international environmental protection organizations, proceedings of authoritative academic conferences, and a database of practical guidelines.

6. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 1, characterized in that, The multi-hop retrieval and matching in the soil pollution remediation knowledge graph includes: S61, parse the query conditions and extract the specific condition entity features of the target region and the performance entities in the target governance requirements; S62, using the specific conditional entity features as initial nodes, search and match all conditional entity nodes that conform to or are close to the specific conditional entity features in the soil pollution control knowledge graph to form an initial matching set; S63, Starting from each conditional entity node in the initial matching set, perform a one-hop search along the edge that is restricted to or applicable to the relationship to locate all technical entity nodes directly associated with it, forming a technical candidate set; S64, Starting from each technical entity node in the set of technical candidates, perform a two-hop search along the edge that generates or causes the relationship to locate the performance entity node associated with it; S65, compare and filter the performance indicators represented by the retrieved performance entity nodes with the performance entity indicators in the target governance requirements; eliminate technologies and related paths whose associated performance cannot meet the target governance requirements.

7. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 1, characterized in that, The regional feature data of the target area specifically includes: Natural geographical conditions include climate type, soil type, soil pH value, groundwater level depth, topographic slope and geomorphic units; Site engineering conditions, including site area, current land use, accessibility, and infrastructure conditions; Environmental and social constraints, including the distribution of surrounding sensitive targets, local environmental regulations and emission standards, and a preliminary assessment of community acceptance.

8. The technology-assisted decision-making method for the treatment of soil pollution sources in mining areas according to claim 1, characterized in that, The specific governance requirements include: Core performance indicators include target pollutant removal rate or compliance concentration, soil ecological function restoration level, and post-treatment land use targets; Process and timeliness requirements, including the expected total governance cycle, time nodes for each stage, and time limits for construction and monitoring; Economic requirements include a total cost budget ceiling and cost control targets per unit area or per unit of pollutant; Technology and risk control requirements include the permissible range of technology maturity, acceptable levels of secondary environmental risk, and the burden of long-term monitoring and maintenance.

9. A decision support system for the treatment of soil pollution sources in mining areas, characterized in that, The system is used to implement the technology-assisted decision-making method for the treatment of soil pollution sources in mining areas as described in any one of claims 1 to 9, and the system specifically includes: The information collection module is used to build a distributed, targeted crawler to collect information from predetermined soil remediation knowledge sources and obtain raw text datasets. The preprocessing module, connected to the information acquisition module, is used to clean and preprocess the original text dataset to obtain a standardized text dataset. The entity extraction module, connected to the preprocessing module, is used to extract technical entities, conditional entities, and performance entities from the standardized text dataset using a deep neural network model, generating triple entity information. The relationship extraction module, connected to the information extraction module, is used to simultaneously extract the relationships between the three entity information to obtain structured information. The graph construction module, connected to the relationship extraction module, is used to construct a knowledge graph for soil pollution control based on the structured information. The retrieval and matching module, connected to the graph construction module, is used to perform multi-hop retrieval and matching in the soil pollution control knowledge graph by using regional feature data of the target area and target governance requirements as query conditions. If a matching governance plan exists, the governance plan is quantitatively scored and ranked by a multi-criteria decision analysis model. The solution output module, connected to the retrieval and matching module, is used to output a sequence of recommended governance solutions based on the score ranking, and to attach the historical case with the highest matching degree, expected effect and key references to each solution.

Images