Agricultural microorganism regulation decision method and system based on large model

Abstract

The application discloses a kind of agricultural microbial regulation decision-making method and system based on large model, method includes: obtaining unstructured agricultural microbiology literature text dataset;Using pre-training large language model, literature text dataset is carried out semantic analysis, extracts strain identification, functional characteristic parameter, environmental adaptation parameter and interaction relationship parameter structured knowledge information;Based on structured knowledge information, environmental adaptability constraint, functional target constraint and interaction relationship constraint are constructed;Acquire the soil environment parameter and microbial community composition data of target agricultural field;With constraint condition set as constraint space, with field data as input, multi-objective optimization is solved, and compound inoculant proportioning scheme is generated.The application breaks through the barrier between unstructured microbiology literature knowledge and multi-objective optimization solving, realizes the automation quantitative decision-making from literature knowledge extraction to proportioning scheme generation, improves the knowledge systematicness and scene adaptability of agricultural microbial regulation decision-making.

Description

Technical Field

[0001] This invention relates to the field of large-scale models, and more particularly to a method and system for agricultural microbial regulation decision-making based on large-scale models. Background Technology

[0002] The application of compound microbial agents in agricultural production is becoming increasingly widespread. By introducing multiple beneficial microbial strains with complementary functions into the soil or crop rhizosphere, they achieve multiple benefits such as nitrogen fixation, phosphorus solubilization, disease resistance, and growth promotion. In practice, when formulating agent formulations for specific fields, it is necessary to comprehensively consider the physicochemical properties of the target soil, the existing microbial community structure, climatic conditions, and crop requirements. Suitable strains must be selected from numerous candidate strains, and the proportions of each strain must be determined. This decision-making process relies heavily on professional knowledge in the field of microbiology, including but not limited to the functional characteristics of the strains (such as nitrogen fixation efficiency, phosphorus solubilization capacity, and antimicrobial spectrum), environmental adaptability (such as pH tolerance range, suitable temperature range, and soil organic matter preference), and interactions between strains (such as synergistic effects, substrate competition, and antagonistic inhibition). This knowledge, accumulated over decades of research, is primarily documented in natural language in academic papers, technical specifications for microbial agent products, field trial reports, and agricultural technology extension literature, forming an indispensable knowledge base for making rational formulation decisions.

[0003] Current technologies for determining compound microbial agent formulations primarily rely on two approaches. The first is a human experience-driven approach, where agricultural technicians formulate formulations by reviewing literature and combining it with their personal experience. This approach is limited by the amount of literature an individual can access and digest, lacking the ability to transfer knowledge when facing new scenarios involving different crops and soil types, and the decision-making process lacks systematic quantitative evidence. The second approach is a data-driven approach, using historical field trial data as training samples and employing machine learning models such as random forests, extreme gradient boosting, or deep neural networks to directly predict the field performance of different formulations. This method requires a large amount of labeled experimental data as a training foundation, but constructing a complete experimental matrix covering multiple strain combinations, soil types, and climatic conditions is virtually impossible in practice, resulting in insufficient model generalization ability. The more fundamental problem lies in the fact that neither manual experience-based nor data-driven approaches have addressed a key bottleneck: the vast amount of microbiological knowledge existing in unstructured text form, including crucial constraints such as strain functional characteristics, environmental adaptation boundaries, and interspecies interactions, cannot be systematically extracted and incorporated into the mathematical framework of formulation calculations. This has resulted in compound microbial agent formulation decisions remaining in a state of separation between empirical and computational science for a long time. Therefore, the technical problem to be solved is how to automatically extract and formalize the knowledge of agricultural microbial strain functional characteristics, environmental adaptability, and interactions scattered in unstructured literature into computable formulation constraints for further agricultural microbial regulation decisions. Summary of the Invention

[0004] The purpose of this invention is to disclose an agricultural microbial regulation decision-making method and system based on a large model, so as to solve the technical problems pointed out in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an agricultural microbial regulation decision-making method based on a large model, comprising: S1. Obtain an unstructured agricultural microbiology literature text dataset; S2. Using a pre-trained large language model, semantic parsing is performed on the document text dataset to extract structured knowledge information of microbial strains. The structured knowledge information includes strain identifiers, functional characteristic parameters and environmental adaptation parameters associated with the strain identifier, and interaction parameters between the strain corresponding to the strain identifier and other strains. S3. Based on the structured knowledge information, construct a set of constraints for compound microbial agent ratio decision-making, wherein the set of constraints includes functional target constraints generated by the functional characteristic parameters, environmental adaptability constraints generated by the environmental adaptability parameters, and interaction relationship constraints generated by the interaction relationship parameters. S4. Obtain soil environmental parameters and microbial community composition data for the target agricultural field; S5. Using the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters, perform multi-objective optimization to generate a compound microbial agent ratio scheme, and make agricultural microbial regulation decisions based on the compound microbial agent ratio scheme, wherein the compound microbial agent ratio scheme includes the selected strain identifier and the ratio of each strain.

[0006] Optionally, the step of using a pre-trained large language model to perform semantic parsing on the document text dataset and extracting structured knowledge information about microbial strains includes: S21, each document text in the document text dataset is divided into multiple text segments; S22, for each of the text segments, construct a structured extraction prompt, wherein the structured extraction prompt defines extraction slots corresponding to the strain identifier, the functional characteristic parameter, the environmental adaptation parameter and the interaction relationship parameter respectively; S23, input the text fragment and the corresponding structured extraction prompt into the pre-trained large language model to obtain the structured extraction record output by the large language model; S24, Summarize all the structured extracted records corresponding to the text fragments to obtain the structured knowledge information of the microbial strain.

[0007] Optionally, the process of constructing the structured extraction prompts includes: S221, Obtain a preset slot definition template, wherein the slot definition template contains the slot name, slot description and output format constraints corresponding to the strain identifier, the functional characteristic parameters, the environmental adaptation parameters and the interaction relationship parameters respectively; S222, Select several literature texts containing at least one of the strain identifier, functional characteristic parameters, environmental adaptation parameters and interaction relationship parameters from the literature text dataset as samples, extract the standard structured extraction record corresponding to each sample text, and combine the sample text and the corresponding standard structured extraction record into a few sample examples; S223, the slot definition template is concatenated with the few sample examples to obtain the structured extraction prompt.

[0008] Optionally, the step of summarizing all the structured extracted records corresponding to the text fragments to obtain the structured knowledge information of the microbial strain includes: S241, perform synonym merging on the strain identifiers in each of the structured extraction records, and group multiple structured extraction records pointing to the same strain into the same strain record group; S242, for parameters with the same name from different structured extraction records within the same strain record group, when there is a conflict in parameter values, the conflicting parameter values ​​are selected or merged based on the timeliness weight and / or journal impact factor weight of the literature text from which each parameter value originates. S243, merge the parameter values ​​after conflict resolution in each of the strain record groups to obtain the integrated record corresponding to each strain identifier. The integrated record of all strain identifiers constitutes the structured knowledge information of the microbial strain.

[0009] Optionally, the step of constructing a set of constraints for the compound microbial agent ratio decision based on the structured knowledge information includes: S31, extract the environmental adaptation parameters corresponding to each strain identifier from the structured knowledge information, and generate environmental adaptation constraints based on the environmental adaptation parameters. The environmental adaptation constraints are used to limit the range of environmental parameter values ​​for maintaining the activity of the strain. S32, extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and generate functional target constraints based on the functional characteristic parameters. The functional target constraints are used to limit the contribution value of a unit dosage of the strain to a preset functional index. S33, extract the interaction parameters corresponding to the strain pairs with interaction relationships from the structured knowledge information, and generate interaction constraints based on the interaction parameters. The interaction constraints are used to limit the functional correction coefficient or compatibility restriction conditions of the strain pairs when they are used in combination. S34, combine all the environmental adaptability constraints, the functional objective constraints, and the interaction relationship constraints into the constraint set.

[0010] Optionally, the step of using the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters to perform multi-objective optimization to generate a compound microbial agent formulation scheme includes: S51, Match the soil environmental parameters with the environmental adaptability constraints, and screen out strains whose environmental adaptability parameters do not meet the range of environmental parameter values ​​of the target agricultural field to obtain a candidate strain set; S52, Based on the microbial community composition data, determine the microbial functional gap of the target agricultural field, and select the corresponding target optimization function from the functional target constraints according to the microbial functional gap; S53, using the ratio of each strain in the candidate strain set as the decision variable, the objective optimization function as the optimization objective, and the interaction relationship constraints and the range of values ​​of the ratio corresponding to each strain identifier as constraints, a multi-objective optimization model is constructed. S54, Solve the multi-objective optimization model and select the compound microbial agent ratio scheme from the Pareto optimal solution set obtained by the solution.

[0011] Optionally, the step of matching the soil environmental parameters with the environmental adaptability constraints and screening out strains whose environmental adaptability parameters do not meet the range of environmental parameter values ​​for the target agricultural field to obtain a candidate strain set includes: S511, for each soil environmental parameter, the value range corresponding to the parameter in the environmental adaptability constraint is divided into an optimal sub-range and a tolerance sub-range; S512, compare the environmental adaptation parameters corresponding to each strain identifier with the soil environmental parameters one by one, and record whether the comparison result of each parameter falls into the optimal sub-interval or the tolerance sub-interval; S513, filter out strain identifiers that have comparison results for at least one of the soil environmental parameters that exceed both the optimal sub-interval and the tolerance sub-interval, include the remaining strain identifiers in the candidate strain set, and retain the comparison results corresponding to each strain identifier.

[0012] Optionally, determining the microbial functional gaps in the target agricultural field based on the microbial community composition data includes: S521, Perform functional gene annotation on the microbial community composition data to obtain the current functional gene abundance spectrum of the target agricultural field; S522, extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and construct a reference functional requirement spectrum for the crop type to which the target agricultural field belongs based on the functional characteristic parameters; S523, compare the current functional gene abundance spectrum with the reference functional requirement spectrum function by function, and identify functional items whose abundance difference exceeds a preset threshold as the microbial functional gap.

[0013] Optionally, the decision-making process for agricultural microbial regulation based on the compound microbial agent formulation includes: S61, for the strain identifier selected in the compound microbial agent formulation scheme, obtain the preparation process parameters associated with the strain identifier from the literature text dataset, and generate microbial agent preparation guidelines based on the preparation process parameters; S62, based on the soil environmental parameters and the environmental adaptation parameters corresponding to each selected strain, determine the effectiveness coefficient of each selected strain in the current field environment, and based on the effectiveness coefficient and the ratio, determine the application rate per unit area of ​​each selected strain. S63, outputting agricultural microbial regulation decision results including the microbial agent preparation guidelines and the application rate per unit area.

[0014] Secondly, the present invention provides an agricultural microbial regulation decision system based on a large model, comprising: The acquisition module is used to acquire unstructured agricultural microbiology literature text datasets; The extraction module is used to perform semantic parsing on the document text dataset using a pre-trained large language model to extract structured knowledge information of microbial strains. The structured knowledge information includes strain identifiers, functional characteristic parameters and environmental adaptation parameters associated with the strain identifier, and interaction parameters between the strain corresponding to the strain identifier and other strains. The construction module is used to construct a set of constraints for compound microbial agent ratio decision based on the structured knowledge information. The set of constraints includes functional target constraints generated by the functional characteristic parameters, environmental adaptability constraints generated by the environmental adaptability parameters, and interaction relationship constraints generated by the interaction relationship parameters. The acquisition module is also used to acquire soil environmental parameters and microbial community composition data of the target agricultural field; The execution module is used to perform multi-objective optimization with the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters to generate a compound microbial agent ratio scheme, and to make agricultural microbial regulation decisions based on the compound microbial agent ratio scheme, wherein the compound microbial agent ratio scheme includes the selected strain identifier and the ratio of each strain.

[0015] Beneficial effects: This invention utilizes a pre-trained large language model to semantically analyze unstructured agricultural microbiology literature, automatically extracting strain functional characteristics, environmental adaptation parameters, and interaction parameters scattered across academic papers, product descriptions, and field reports into structured knowledge information. Based on this, three types of computable constraints—environmental adaptation constraints, functional objective constraints, and interaction constraint constraints—are constructed, thus bridging the gap between massive amounts of unstructured microbiology knowledge and multi-objective optimization solutions. Compared to experience-driven methods relying on manual literature review, this approach overcomes the limitations of individual knowledge coverage, achieving systematic knowledge aggregation across crops and soil types. Compared to data-driven methods that train machine learning models using historical experimental data, this approach does not rely on large-scale labeled experimental matrices that are difficult to construct. Instead, it directly extracts the necessary constraint information for formulation decisions from literature knowledge, combining it with field-measured soil environmental parameters and microbial community composition data for multi-objective optimization. This achieves automated quantitative decision-making from literature knowledge extraction to formulation generation, significantly improving the systematic nature and scenario adaptability of agricultural microbial regulation decisions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an agricultural microbial regulation decision-making method based on a large model according to the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the process of extracting structured knowledge information according to the present invention.

[0019] Figure 3 This is a schematic diagram illustrating the process of constructing the constraint set of the present invention.

[0020] Figure 4 This is a schematic diagram of an agricultural microbial regulation decision system based on a large model according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown in one embodiment, the present invention provides an agricultural microbial regulation decision-making method based on a large model, comprising: S1. Obtain an unstructured agricultural microbiology literature text dataset.

[0023] The agricultural microbiology literature text dataset refers to an unstructured collection of texts that record knowledge related to agricultural microbial strains in natural language. Specifically, it can include full-text or abstract texts of academic papers in the field of microbiology retrieved from academic databases (such as PubMed, Web of Science, and CNKI); field trial records and analysis texts extracted from publicly available technical reports from agricultural technology extension departments or agricultural research institutes; strain characteristic descriptions parsed from product instructions of microbial inoculant manufacturers; and microbial inoculant application guidelines collected from agricultural technology extension publications and agricultural knowledge bases. These texts contain the knowledge necessary for making compound inoculant formulation decisions, including information on the functional characteristics of strains (such as nitrogen fixation efficiency values, phosphorus solubilization capacity descriptions, and antimicrobial spectrum ranges), environmental adaptability information (such as pH tolerance range, suitable temperature range, and soil organic matter preference), and strain interaction information (such as synergistic descriptions, substrate competition descriptions, and antagonistic inhibition records). However, this information exists in a scattered, unstructured natural language format and cannot be directly utilized by computerized formulation optimization processes.

[0024] Optionally, obtain an unstructured dataset of agricultural microbiology literature texts, including: S11, retrieve literature texts related to agricultural microbial strains from a preset literature data source, wherein the literature data source includes at least one of academic paper databases, agricultural technology report databases, microbial agent product instruction manual databases, and agricultural technology promotion literature databases; S12, perform format unification processing on the retrieved literature texts, convert literature texts from different sources into plain text format, remove format marks, figure descriptions and reference lists and other content that is not related to strain knowledge extraction, and obtain the agricultural microbiology literature text dataset.

[0025] S2. Using a pre-trained large language model, perform semantic parsing on the document text dataset to extract structured knowledge information of microbial strains. The structured knowledge information includes strain identifiers, functional characteristic parameters and environmental adaptation parameters associated with the strain identifier, and interaction parameters between the strain corresponding to the strain identifier and other strains.

[0026] Among them, large language models refer to deep neural network language models that have been pre-trained on large-scale corpora and are able to understand the semantics of natural language texts and perform information extraction tasks, such as the GPT series, LLaMA series, or DeepSeek series models.

[0027] Among them, microbial strains refer to microbial cultures with clear taxonomic classification and specific agricultural functions, such as a numbered strain of Bacillus subtilis or a certain isolate of Trichoderma.

[0028] Among them, structured knowledge information refers to a standardized record set of strain-related knowledge extracted from literature texts, organized according to a preset field format. Each record contains at least four fields: strain identifier, functional characteristic parameters, environmental adaptation parameters, and interaction relationship parameters.

[0029] Among them, the strain identifier refers to the name or number used to uniquely identify a certain microbial strain, such as the Latin name Bacillus subtilis Bs-916, the strain preservation number CGMCC No.xxxx, or the trade name of a certain compound microbial agent A component, etc.

[0030] Among them, functional characteristic parameters refer to quantitative or semi-quantitative indicators that describe the agricultural functions of strains, such as nitrogen fixation efficiency, diameter of phosphorus solubilization zone, and inhibition rate against specific pathogens.

[0031] Among them, environmental adaptation parameters refer to the range of environmental conditions that describe the strain's ability to maintain activity in the soil environment, such as the suitable pH range of 5.5-8.0, the suitable temperature range of 15-35°C, and the minimum soil organic matter content requirements.

[0032] Among them, the interaction parameter refers to the parameter that describes the degree of mutual influence when the strain is applied together with other strains, such as the function correction coefficient when coexisting with another strain (e.g., 1.2 indicates 20% synergistic effect), or compatibility restriction conditions (e.g., it cannot be combined with acid-producing strains).

[0033] Optional, such as Figure 2 As shown, the process of using a pre-trained large language model to perform semantic parsing on the document text dataset and extracting structured knowledge information about microbial strains includes: S21, each document in the document text dataset is divided into multiple text segments.

[0034] Among them, "document text" refers to the complete content of a single document in the document text dataset, such as the full text of an academic paper, the main text of a field trial report, or the complete text of a fungicide product instruction manual.

[0035] Among them, a text segment refers to a continuous text paragraph obtained by dividing a document text according to preset rules, such as several paragraphs obtained by dividing it into natural paragraphs, or several substrings of equal length obtained by dividing it into a sliding window with a fixed number of characters.

[0036] Optionally, each document in the document text dataset is segmented into multiple text segments, including: Each document text is identified as a natural paragraph, and the text is divided into initial segments using paragraph boundaries as the dividing points. For initial segments with more than a preset limit on the number of characters, a second segmentation is performed based on semantically complete units (such as sentence boundaries) to obtain text segments that meet the input length limit of the large language model.

[0037] S22, for each of the text segments, construct a structured extraction prompt, wherein the structured extraction prompt defines extraction slots corresponding to the strain identifier, the functional characteristic parameter, the environmental adaptation parameter and the interaction relationship parameter respectively.

[0038] Among them, structured extraction prompts refer to the instruction text used to guide the large language model to extract specified information from the input text according to a preset format. It usually includes two parts: slot definition and output format description, to ensure that the large language model outputs extraction results with a unified structure that can be parsed by the program.

[0039] Here, the extraction slot refers to the pre-defined position of the field to be filled in the structured extraction prompts, and each slot corresponds to a type of knowledge to be extracted. For example, the strain identification slot requires the large language model to output the names of strains appearing in the text, and the functional characteristic parameter slot requires the output of quantitative values ​​such as nitrogen fixation efficiency.

[0040] Optionally, the process of constructing the structured extraction prompts includes: S221, Obtain a preset slot definition template, wherein the slot definition template contains the slot name, slot description and output format constraints corresponding to the strain identifier, the functional characteristic parameters, the environmental adaptation parameters and the interaction relationship parameters respectively.

[0041] Among them, the slot definition template refers to a pre-compiled configuration file used to specify the structure and format requirements of the output fields of a large language model. For example, a JSON format template file contains the metadata definition of each slot.

[0042] The slot name refers to the unique identifier assigned to each extracted field in the template, such as "strain_id", "function_params", "env_params", "interaction_params", etc.

[0043] The slot description refers to the natural language explanation of the semantics of each slot in the template, which is used to tell the large language model what type of information should be extracted from the slot, such as "extract the Latin name or strain number of the microbial strain mentioned in the text".

[0044] Among them, the output format constraint refers to the format requirements for the output value of each slot in the template. For example, it specifies the output string type for the strain identifier slot and the output "parameter name: value: unit" triple format for the functional characteristic parameter slot.

[0045] Optionally, obtain a preset slot definition template, including: Construct a template structure containing four types of slots. After configuring the name, description and format constraints of each type of slot, store the template in the configuration file. During the actual extraction process, load the template structure from the configuration file to obtain the preset slot definition template.

[0046] S222, Select several literature texts from the literature text dataset that contain at least one of the strain identifier, the functional characteristic parameter, the environmental adaptation parameter and the interaction relationship parameter as samples, extract the standard structured extraction record corresponding to each sample text, and combine the sample text and the corresponding standard structured extraction record into a few sample examples.

[0047] Among them, the standard structured extraction record refers to the standardized extraction result of the sample text being labeled and filled in according to the slot definition template. For example, the complete record of strain identification (e.g., Bacillus subtilis Bs-916) and functional characteristic parameters (e.g., nitrogen fixation efficiency: 12.5 mg N / g) manually extracted and filled in from a paper related to Bacillus subtilis.

[0048] Optionally, extract the standard structured extraction record corresponding to each sample text, including: The sample text is queried, and the strain information appearing in the text is filled into the corresponding slot in the slot definition template one by one. After review and confirmation, a standard structured extraction record is formed.

[0049] S223, the slot definition template is concatenated with the few sample examples to obtain the structured extraction prompt.

[0050] Optionally, the slot definition template is concatenated with the few-sample example to obtain the structured extraction prompt, including: The slot definition template is placed at the beginning of the prompt text as the format instruction part, the few sample examples are immediately following it as the example demonstration part, and the input area of ​​the text fragment to be processed is appended after the example demonstration part to form a complete structured extraction prompt.

[0051] S23, input the text fragment and the corresponding structured extraction prompt into the pre-trained large language model to obtain the structured extraction record output by the large language model.

[0052] Optionally, the text fragment and the corresponding structured extraction prompt are input together into the pre-trained large language model to obtain the structured extraction record output by the large language model, including: Replace the reserved input area of ​​the current text fragment to be processed in the structured extraction prompt with the text fragment, submit an inference request through the large language model application interface; receive the response text returned by the large language model, parse the response text into JSON according to the output format constraints, and obtain the structured extraction record.

[0053] By employing the structured extraction prompting method based on slot definition templates and a small number of sample examples, this embodiment leverages the semantic understanding capabilities of a large language model to uniformly convert multi-source heterogeneous knowledge existing in free text form in agricultural microbiology literature into structured extraction records with fixed slot structures. Compared to traditional information extraction methods based on rule matching or named entity recognition (NER), this approach eliminates the need to write extraction rules separately for each type of parameter or label large-scale training corpora. It requires only a small number of labeled samples to achieve consistent knowledge extraction across documents and expression formats, significantly reducing the engineering cost of converting unstructured documents into structured knowledge.

[0054] S24, Summarize all the structured extracted records corresponding to the text fragments to obtain the structured knowledge information of the microbial strain.

[0055] Optionally, the step of summarizing all the structured extracted records corresponding to the text fragments to obtain the structured knowledge information of the microbial strain includes: S241, perform synonym merging on the strain identifiers in each of the structured extraction records, and group multiple structured extraction records pointing to the same strain into the same strain record group.

[0056] Among them, a strain record group refers to a collection of multiple structured extraction records that point to the same strain after being synonymously merged. For example, extraction records of Bacillus subtilis Bs-916 mentioned in several different documents are grouped into the same strain record group.

[0057] Optionally, the strain identifiers in each of the structured extraction records are grouped by synonyms, and multiple structured extraction records pointing to the same strain are grouped into the same strain record group, including: A strain synonym mapping table is constructed, which records the multiple referential forms (such as full Latin name, abbreviation of Latin name, strain preservation number, and trade name) that the same strain may appear in different literatures and their mutual mapping relationships. Traverse each of the structured extraction records, query the strain synonym mapping table, map the strain identifier in the structured extraction record to a unified standard strain identifier, and group the structured extraction records with the same standard strain identifier into the same strain record group.

[0058] S242, for parameters with the same name from different structured extraction records within the same strain record group, when there is a conflict in parameter values, the conflicting parameter values ​​are selected or merged based on the timeliness weight and / or journal impact factor weight of the literature text from which each parameter value originates.

[0059] Among them, the same-name parameter refers to the parameter that belongs to the same slot type and describes the same attribute in different structured extraction records within the same strain record group, such as the nitrogen fixation efficiency parameter contained in both records.

[0060] Among them, the timeliness weight refers to the weight value determined based on the publication time of the literature text. The more recent the publication time, the higher the timeliness weight, reflecting the priority of accepting newer research results.

[0061] Among them, the journal impact factor weight refers to the weight value determined based on the impact factor of the journal from which the text of the article is sourced. The higher the impact factor of the journal, the greater its weight, reflecting the priority of accepting articles with higher academic credibility.

[0062] Optionally, for parameters with the same name from different structured extraction records within the same strain record group, when parameter values ​​conflict, the conflicting parameter values ​​are selected or merged based on the timeliness weight and / or journal impact factor weight of the literature text from which each parameter value originates, including: The publication year and impact factor of the journal from which each structured extraction record originates are obtained, normalized, and then combined into a comprehensive weight according to a preset ratio. When the conflicting parameter values ​​are numerical, the parameter value from the structured extraction record with the highest overall weight is selected as the best result. When the conflicting parameter values ​​are interval types, calculate the intersection of the intervals of each conflicting parameter value. If there is a non-empty intersection, the intersection interval is used as the merging result. If there is no intersection, the parameter value interval in the structured extracted record with the highest comprehensive weight is selected as the best result.

[0063] S243, merge the parameter values ​​after conflict resolution in each of the strain record groups to obtain the integrated record corresponding to each strain identifier. The integrated record of all strain identifiers constitutes the structured knowledge information of the microbial strain.

[0064] Among them, the integrated record refers to the result of summarizing the parameter values ​​of the same strain record group after conflict resolution into a complete record. For example, all environmental adaptation parameters, functional characteristic parameters and interaction parameters of Bacillus subtilis Bs-916 are summarized into a JSON record containing all three dimensions of information.

[0065] Optionally, the parameter values ​​after conflict resolution in each of the strain record groups are merged to obtain an integrated record corresponding to each strain identifier, including: Using the strain identifier as the key, the functional characteristic parameters, environmental adaptation parameters, and interaction relationship parameters after conflict resolution are filled into the corresponding parameter fields. Missing parameters are assigned default null values ​​to form an integrated record in the form of key-value pairs. The integrated records of all strains are stored in the form of a dataset to obtain the structured knowledge information of the microbial strain.

[0066] By introducing timeliness and journal impact factor weights to optimize or merge parameter conflicts in multi-source literature, this embodiment effectively solves the problem of inconsistent strain parameters caused by differences in experimental conditions, measurement methods, or publication dates among different literature. Compared to simple mean fusion or equal-weighted voting, this method prioritizes literature data with higher timeliness or academic credibility, avoiding interference from outdated or low-quality literature data on subsequent constraint construction and optimization solutions, and improving the credibility of structured knowledge information and the reliability of decision-making schemes.

[0067] S3. Based on the structured knowledge information, construct a set of constraints for the compound microbial agent ratio decision, wherein the set of constraints includes functional target constraints generated by the functional characteristic parameters, environmental adaptability constraints generated by the environmental adaptability parameters, and interaction relationship constraints generated by the interaction relationship parameters.

[0068] Among them, compound microbial agent formulation decision refers to the process of selecting suitable strains from candidate strains and determining the proportion of each strain in the microbial agent for a specific target agricultural field. For example, the microbial agent formulation for a certain acidic phosphorus-deficient field is determined to be: Bacillus subtilis 40%, phosphate-solubilizing bacteria 35%, and Trichoderma 25%.

[0069] Optional, such as Figure 3 The set of constraints for constructing the compound microbial agent ratio decision based on the structured knowledge information includes: S31, extract the environmental adaptation parameters corresponding to each strain identifier from the structured knowledge information, and generate environmental adaptation constraints based on the environmental adaptation parameters. The environmental adaptation constraints are used to limit the range of environmental parameter values ​​for maintaining the activity of the strain.

[0070] Among them, environmental adaptability constraints refer to the survival conditions of the strain expressed in the form of mathematical inequalities, such as pH ∈ [5.5, 8.0] or soil organic matter ≥ 1.5%. The strain cannot maintain its activity in field environments outside of these ranges.

[0071] Optionally, generating environmental adaptability constraints based on the environmental adaptability parameters includes: The environmental adaptation parameters corresponding to each strain identifier are extracted from the structured knowledge information. Each environmental adaptation parameter is converted into an environmental adaptation constraint expression with a triple structure of "parameter name: lower limit: upper limit". For parameters with only one-sided limits (such as those that only record the minimum temperature requirement), the missing side is set to an infinite boundary.

[0072] S32, extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and generate functional target constraints based on the functional characteristic parameters. The functional target constraints are used to limit the contribution value of a unit dosage of the strain to a preset functional index.

[0073] Among them, the functional objective constraint refers to the strain functional contribution model expressed in a quantitative relation, such as the strain contributing 12.5 mg N nitrogen fixation per gram or 8.0 units of phosphorus solubilization activity per gram, which is used to calculate the overall functional achievement degree of different ratio schemes in multi-objective optimization.

[0074] Optionally, generating functional target constraints based on the functional characteristic parameters includes: Extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and convert each functional characteristic parameter into a functional objective constraint expression with a triple structure of "strain identifier: function item: contribution value per unit dosage", wherein the contribution value per unit dosage represents the quantitative contribution of the strain to the specified function item (such as nitrogen fixation, phosphorus solubilization, disease resistance) per unit dosage (such as per gram of bacterial agent).

[0075] S33, extract the interaction parameters corresponding to the strain pairs with interaction relationships from the structured knowledge information, and generate interaction constraints based on the interaction parameters. The interaction constraints are used to limit the functional correction coefficient or compatibility restriction conditions of the strain pairs when they are used in combination.

[0076] Interaction constraints refer to the coexistence rules of strains expressed in mathematical formulas. For example, when strain A and strain B coexist, the nitrogen fixation contribution of strain A is multiplied by a correction factor of 1.2 (synergistic effect), or strain A and strain C cannot appear in the same formulation at the same time (incompatibility).

[0077] Optionally, generating interaction relationship constraints based on the interaction relationship parameters includes: Interaction parameters of strain pairs with interactions are extracted from the structured knowledge information. If the interaction parameter is a quantitative correction value, it is converted into a corrected interaction constraint with a quadruple structure of "strain A: strain B: function item: correction coefficient". If the interaction parameter is a compatibility restriction description, it is converted into an exclusionary interaction constraint with a triple structure of "strain A: strain B: mutual exclusion".

[0078] S34, combine all the environmental adaptability constraints, the functional objective constraints, and the interaction relationship constraints into the constraint set.

[0079] Optionally, all the environmental adaptability constraints, functional objective constraints, and interaction relationship constraints are combined into the constraint set, including: The system detects whether there is a conflict between the exclusionary interaction constraints and the functional objective constraints in the interaction relationship constraints; when a pair of strains has a mutually exclusive relationship, but at least one of the strains is identified as a key functional contributing strain by the microbial functional gap, the exclusionary interaction constraint is downgraded to a modified interaction constraint, and the modification coefficient after downgrading is determined based on the antagonistic mechanism data recorded in the structured knowledge information (e.g., downgrading "incompatible" to "function reduced to 40% when coexisting").

[0080] S4. Obtain soil environmental parameters and microbial community composition data for the target agricultural field.

[0081] Among them, soil environmental parameters refer to the index data that describe the physical and chemical properties of the soil in the target agricultural field, such as soil pH, temperature, water content, organic matter content, and electrical conductivity.

[0082] Among them, microbial community composition data refers to data describing the types and abundance of microorganisms in the soil of the target agricultural field, such as OTU / ASV abundance tables obtained by 16S rRNA amplicon sequencing or metagenomic sequencing.

[0083] Optionally, obtain soil environmental parameters and microbial community composition data for the target agricultural plot, including: In-situ environmental parameters such as soil pH, temperature, and moisture content are collected by soil sensors deployed in the target agricultural field; soil samples are collected from the target agricultural field, and indicators such as organic matter content and electrical conductivity are obtained through laboratory analysis; metagenomic sequencing is performed on the soil samples, and the microbial community composition data are obtained after sequence quality control and species annotation.

[0084] S5. Using the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters, perform multi-objective optimization to generate a compound microbial agent ratio scheme, and make agricultural microbial regulation decisions based on the compound microbial agent ratio scheme, wherein the compound microbial agent ratio scheme includes the selected strain identifier and the ratio of each strain.

[0085] Among them, the constraint space refers to the set of boundary conditions of the solution space, which is composed of environmental adaptability constraints, functional objective constraints, and interaction relationship constraints. The candidate solutions generated during the optimization process must satisfy all the constraints in this space. For example, all selected strains in a certain candidate ratio scheme must satisfy the environmental adaptability constraint of field pH.

[0086] Among them, the compound microbial agent formulation scheme refers to the microbial agent formulation result output after multi-objective optimization solution, which includes a list of selected strain identifiers and the proportion of each strain in the total mass of the microbial agent. For example: selected strains: [Bs-916, Ps-307, Tm-112], formulation ratio: [40%, 35%, 25%].

[0087] Optionally, the step of using the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters to perform multi-objective optimization to generate a compound microbial agent formulation scheme includes: S51, the soil environmental parameters are matched with the environmental adaptability constraints, and strains whose environmental adaptability parameters do not meet the environmental parameter value range of the target agricultural field are screened out to obtain a candidate strain set.

[0088] Among them, the range of environmental parameters refers to the upper and lower boundaries of various environmental parameters defined in the environmental adaptability constraints. For example, the pH range [5.5, 8.0] indicates that the strain can only maintain its activity within this pH range.

[0089] Among them, the candidate strain set refers to the set of strains that can adapt to the current field environment after environmental adaptability screening. For example, among all 50 strains extracted from the literature knowledge base, 12 candidate strains that meet the current field environment remain after screening based on pH and organic matter conditions.

[0090] Optionally, the step of matching the soil environmental parameters with the environmental adaptability constraints and screening out strains whose environmental adaptability parameters do not meet the range of environmental parameter values ​​for the target agricultural field to obtain a candidate strain set includes: S511, for each soil environmental parameter, the value range corresponding to the parameter in the environmental adaptability constraint is divided into an optimal sub-range and a tolerance sub-range.

[0091] The optimal sub-interval refers to the range within which the bacterial activity is highest. For example, Bacillus subtilis has the highest nitrogen fixation efficiency in the pH range of 6.5-7.5, and this range is the optimal sub-interval.

[0092] The tolerance sub-range refers to the sub-range within which the strain can survive but its activity is reduced. For example, the strain can survive in the pH range of 5.5-6.5 and pH 7.5-8.0 but its nitrogen fixation efficiency drops to less than 60% of the optimal level. This range is the tolerance sub-range.

[0093] Optionally, the value range corresponding to the parameter in the environmental adaptability constraint is divided into an optimal sub-range and a tolerance sub-range, including: The optimal range description and tolerance boundary description of the strain for this parameter are read from the structured knowledge information. The optimal range is mapped to the optimal sub-interval, and the difference region between the tolerance boundary and the optimal range is mapped to the tolerance sub-interval. If only a single boundary is recorded in the literature, the optimal sub-interval is the middle 60% sub-region of the value range, and the tolerance sub-interval is the 20% on each side.

[0094] S512, compare the environmental adaptation parameters corresponding to each strain identifier with the soil environmental parameters one by one, and record whether the comparison result of each parameter falls into the optimal sub-interval or the tolerance sub-interval.

[0095] S513, filter out strain identifiers that have comparison results for at least one of the soil environmental parameters that exceed both the optimal sub-interval and the tolerance sub-interval, include the remaining strain identifiers in the candidate strain set, and retain the comparison results corresponding to each strain identifier.

[0096] By subdividing the range of environmental adaptability constraints into optimal and tolerant sub-ranges and retaining the comparison results, this embodiment avoids the problem of excessive loss of candidate strains that may be caused by traditional binary screening (pass / reject). Even if a strain is in the tolerant sub-range (with reduced activity) for a single environmental parameter in the current field, if it performs well in other parameters and its contribution is precisely the main functional gap in the field, it can still be retained through quantitative trade-offs in subsequent multi-objective optimization. This hierarchical screening strategy ensures the lower limit of environmental adaptability of the candidate strain set without sacrificing optimization flexibility.

[0097] S52, Based on the microbial community composition data, determine the microbial functional gap of the target agricultural field, and select the corresponding target optimization function from the functional target constraints according to the microbial functional gap.

[0098] Among them, the microbial functional gap refers to the gap between the current functional status of the microbial community in the target agricultural field and the ideal needs of the crop. For example, if the abundance of nitrogen-fixing functional genes in the soil is only 30% of the ideal level, then the field has a nitrogen-fixing functional gap.

[0099] In this context, the objective optimization function refers to the mathematical objective expression that needs to be maximized or satisfied in multi-objective optimization. For example, maximizing the overall nitrogen fixation capacity of the compound microbial agent is a function composed of a weighted combination of functional objective constraints of the strains corresponding to the functional gaps.

[0100] Optionally, determining the microbial functional gaps in the target agricultural field based on the microbial community composition data includes: S521, Perform functional gene annotation on the microbial community composition data to obtain the current functional gene abundance spectrum of the target agricultural field.

[0101] Among them, the current functional gene abundance spectrum refers to the relative abundance distribution of various agricultural-related functional genes obtained from functional gene annotation in the current field microbial community. For example, the relative abundance of nitrogen fixation-related gene nifH is 0.8%, and the relative abundance of phosphorus solubilization-related gene gcd is 0.3%.

[0102] Optionally, functional gene annotation is performed on the microbial community composition data to obtain the current functional gene abundance spectrum of the target agricultural field, including: The sequences in the microbial community composition data are compared with functional gene databases (such as KEGG, eggNOG, or CAZy) to obtain functional annotation information corresponding to each sequence. The annotation results are classified and summarized according to preset agricultural function categories (such as nitrogen fixation, phosphorus solubilization, and disease resistance). The relative abundance of each type of functional gene in the total sequence is calculated to obtain the current functional gene abundance spectrum.

[0103] S522, extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and construct a reference functional requirement spectrum for the crop type to which the target agricultural field belongs based on the functional characteristic parameters.

[0104] Among them, the reference functional requirement spectrum refers to the benchmark reference data formed by summarizing the ideal abundance levels of each functional item for the type of crop planted in the target agricultural field. For example, in the reference functional requirement spectrum of corn planting field, the ideal abundance of nitrogen fixation is 1.2% and the ideal abundance of phosphorus solubilization is 0.8%.

[0105] Optionally, a reference functional requirement spectrum for the crop type of the target agricultural field is constructed based on the functional characteristic parameters, including: The functional characteristic parameters corresponding to each strain identifier are extracted from the structured knowledge information. The functional characteristic parameters of strains associated with the crop type of the target agricultural field are screened. The functional characteristic parameter values ​​of similar functional items are summarized by weighted average to obtain the ideal reference value of each function, thus forming the reference functional requirement spectrum.

[0106] S523, compare the current functional gene abundance spectrum with the reference functional requirement spectrum function by function, and identify functional items whose abundance difference exceeds a preset threshold as the microbial functional gap.

[0107] The preset threshold refers to the critical value of abundance difference for determining functional gaps. For example, it is set to 20% of the reference abundance value, meaning that functional items with a current abundance lower than 80% of the reference abundance are considered to have gaps.

[0108] Among them, the functional item refers to a specific category of agricultural microbial function, such as nitrogen fixation function, phosphorus solubilization function, disease resistance function (such as resistance to Fusarium), growth promotion function (such as secretion of indoleacetic acid), etc.

[0109] Optionally, a corresponding objective optimization function is selected from the functional objective constraints based on the microbial functional gap, including: The process iterates through all functional items included in the microbial functional gap, searching for functional objective constraints that match the functional items from the functional objective constraints. Using the proportion of each strain in the candidate strain set as a variable and the contribution of each strain per unit dosage to the functional item as a coefficient, a target optimization function is constructed to maximize the overall contribution of the functional item. For example, if the microbial functional gap includes nitrogen fixation and disease resistance, then two target optimization functions are constructed: one maximizing the total nitrogen fixation of the compound microbial agent and the other maximizing the comprehensive disease resistance of the compound microbial agent.

[0110] S53, using the ratio of each strain in the candidate strain set as the decision variable, the objective optimization function as the optimization objective, and the interaction relationship constraints and the range of values ​​of the ratio corresponding to each strain identifier as constraints, a multi-objective optimization model is constructed.

[0111] Optionally, a multi-objective optimization model can be constructed, including: Define the decision variable vector X = [x1, x2, ..., x...] n ], where xᵢ represents the ratio of the i-th strain in the candidate strain set, and the sum of all ratios is 1; Construct an objective function vector F(X) = [f1(X), f2(X), ..., f...] using the objective optimization function. m [X], where each component corresponds to a functional gap objective optimization function, and each component is obtained by weighted summation of the functional objective constraints of the corresponding strain; The interaction constraint is introduced to correct the contribution coefficient of the relevant strain pairs in the objective function. The corrected objective function and the ratio range constraint together constitute the multi-objective optimization model.

[0112] For example, constructing a multi-objective optimization model includes: Define the decision variable vector X = [x1, x2, ..., x...] n ], where xᵢ represents the ratio of the i-th strain in the candidate strain set, and the sum of all ratios is 1, i.e., ∑xᵢ=1.

[0113] Construct an objective function vector F(X) = [f1(X), f2(X), ..., f2(X)] using the objective optimization function. m [X], where m is the number of functional items contained in the microbial functional gap, and each objective function fⱼ(X) is defined as: fⱼ(X)=∑ᵢ(Cⱼᵢ·xᵢ·δᵢⱼ); in: Cⱼᵢ represents the contribution value per unit dosage of the i-th strain to the j-th function in the functional objective constraints (e.g., the nitrogen fixation contribution value of strain Bs-916, C_nitrogen fixation, Bs-916 = 12.5 mgN / g). δᵢⱼ is the environmental attenuation correction factor for the j-th function of the i-th strain, which is determined based on the comparison results of various environmental parameters of the i-th strain recorded in S511: if all parameters fall into the optimal sub-interval, then δᵢⱼ=1.0; if some parameters fall into the tolerance sub-interval, then δᵢⱼ=0.8-0.2×Ntol, where Ntol is the number of parameters that fall into the tolerance sub-interval.

[0114] The interaction constraint is introduced to correct the contribution coefficients of related strain pairs in the objective function: For the strain pair (p,q) recorded in the corrected interaction constraint, its contribution term in the objective function fⱼ(X) is corrected as follows: Corrected contribution item = Cⱼ p ·x p ·δ p ⱼ+Cⱼq·xq·δqⱼ+γ p qⱼ·min(x p ·δ p ⱼ,xq·δqⱼ); Where γ p qⱼ is the synergistic correction coefficient between strain p and strain q for functional term j in the modified interaction constraint (positive value indicates synergistic enhancement, negative value indicates antagonistic reduction).

[0115] Simultaneously, the exclusionary interaction constraints (those not downgraded after conflict resolution in S34) are introduced as hard constraints: for mutually exclusive strain pairs (p,q) recorded in the exclusionary interaction constraints, the constraint condition x is added. p xq=0.

[0116] The modified objective function, the range of the proportion values ​​(0≤xᵢ≤1), the normalization constraint that the sum of all proportions is 1, and the exclusionary interaction constraint together constitute the multi-objective optimization model: minimize-F(X)=[-f1(X),-f2(X),...,-f m (X)] subjectto: 0 ≤ xᵢ ≤ 1 (i = 1, 2, ..., n); ∑xᵢ=1; x p xq=0 (for all exclusionary interaction-constrained strain pairs (p,q)).

[0117] S54, Solve the multi-objective optimization model and select the compound microbial agent ratio scheme from the Pareto optimal solution set obtained by the solution.

[0118] Optionally, the multi-objective optimization model is solved, and the compound microbial agent formulation scheme is selected from the Pareto optimal solution set, including: The multi-objective evolutionary algorithm (such as NSGA-II or MOEA / D) is used to iteratively solve the multi-objective optimization model, generating a set of non-dominated solutions that constitute the Pareto optimal solution set; Calculate the normalized weighted sum of the objective function values ​​of each candidate solution in the Pareto optimal solution set, sort them according to the weighted sum, and select the candidate solution with the highest weighted sum as the compound microbial agent formulation scheme; The weight of each objective function is dynamically determined based on the abundance gap ratio of each functional item in the microbial functional gap, with higher weights assigned to functional items with larger gaps.

[0119] For example, solving the multi-objective optimization model and selecting the compound microbial agent formulation scheme from the Pareto optimal solution set includes: The multi-objective evolutionary algorithm (such as NSGA-II or MOEA / D) is used to iteratively solve the multi-objective optimization model, generating a set of non-dominated solutions that constitute the Pareto optimal solution set P={P1,P2,...,P...}. k}, where each candidate solution P k It is an n-dimensional matching vector.

[0120] Calculate each candidate solution P in the Pareto optimal solution set. k Normalized weighted composite score S(P) k ): S(P k )=∑ⱼwⱼ·(fⱼ(P k )-fⱼ_min) / (fⱼ_max-fⱼ_min) in: fⱼ_max and fⱼ_min are the maximum and minimum values ​​of all candidate solutions in the Pareto optimal solution set on the j-th objective function, respectively. wⱼ is the dynamic weight of the j-th objective function, determined by the following formula: wⱼ=Δⱼ / ∑ⱼΔⱼ, where Δⱼ=max(0,Refⱼ-Curⱼ) / Refⱼ, Refⱼ is the ideal abundance value of the j-th function in the reference functional demand spectrum, and Curⱼ is the actual abundance value of the j-th function in the current functional gene abundance spectrum.

[0121] According to the comprehensive score S(P) k Sort from highest to lowest, and select S(P) k The highest candidate solution is taken as the compound microbial agent formulation scheme.

[0122] By dynamically determining the optimization target weights based on the abundance gap ratio of each functional item in the microbial functional gap, this embodiment achieves automatic adaptation of the formulation scheme selection to the actual functional needs of the field: the functional item with the larger the gap receives higher weight in the multi-objective trade-off, ensuring that the final output compound microbial agent formulation scheme prioritizes addressing the most pressing functional shortcomings of the current field. Compared to Pareto selection methods with fixed weights or equal weight compromises, this dynamic weighting mechanism enables the same optimization framework to produce differentiated formulation schemes for fields with different soil conditions, significantly improving the scenario-specificity and practical application value of the decision-making process.

[0123] Optionally, the decision-making process for agricultural microbial regulation based on the compound microbial agent formulation includes: S61, for the strain identifier selected in the compound microbial agent formulation scheme, obtain the preparation process parameters associated with the strain identifier from the literature text dataset, and generate microbial agent preparation guidelines based on the preparation process parameters.

[0124] Among them, the preparation process parameters refer to the process conditions data of each strain in the production process of the inoculant recorded in the literature text dataset, such as the fermentation medium formula, fermentation temperature and time, drying method, carrier type and inoculant storage conditions, etc.

[0125] Among them, the bacterial agent preparation guide refers to the operation guidance document that organizes the preparation process parameters of each selected strain according to the production process. For example, it includes the process description of "Strain Bs-916: use LB medium, ferment at 30°C for 48 hours, freeze dry and mix with diatomaceous earth carrier at a ratio of 1:3".

[0126] Optionally, a guide for preparing the microbial agent is generated based on the preparation process parameters, including: The preparation process parameters for each selected strain were categorized and organized into five stages: fermentation culture, cell collection, drying, carrier mixing, and packaging and storage. These stages were then assembled into a guide document for strain preparation, containing operational parameters and control conditions for each stage.

[0127] S62, based on the soil environmental parameters and the environmental adaptation parameters corresponding to each selected strain, determine the effectiveness coefficient of each selected strain in the current field environment, and based on the effectiveness coefficient and the ratio, determine the application rate per unit area of ​​each selected strain.

[0128] The effective coefficient refers to the correction factor that takes into account the activity decline of the strain under the current field environment when converting the theoretical ratio into the actual field application. For example, the effective coefficient of a certain strain is 1.0 in the optimal sub-range of pH 6.0, and the effective coefficient drops to 0.6 in the tolerance sub-range of pH 5.5.

[0129] The application rate per unit area refers to the actual amount of each strain of inoculant required per unit area of ​​field. For example, the application rate of Bacillus subtilis Bs-916 per unit area is 200g per mu.

[0130] Optionally, based on the soil environmental parameters and the environmental adaptation parameters corresponding to each selected strain, the effectiveness coefficient of each selected strain in the current field environment is determined, including: The soil environmental parameters are compared with the results recorded in step S511 for each selected strain. If all parameters fall into the optimal sub-interval, the effective coefficient is set to 1.0. If any parameters fall into the tolerance sub-interval, the effective coefficient is calculated using a preset decay function based on the number of parameters falling into the tolerance sub-interval and the degree of deviation.

[0131] Optionally, based on the effectiveness coefficient and the ratio, the application rate per unit area for each selected strain is determined, including: Obtain the recommended total amount of microbial agent per acre for the target crop type. Multiply the ratio of the microbial agent to the recommended total amount of microbial agent to obtain the theoretical application amount of each selected strain. Divide the theoretical application amount of each selected strain by its corresponding effective coefficient to obtain the application amount per unit area of ​​each selected strain.

[0132] S63, outputting agricultural microbial regulation decision results including the microbial agent preparation guidelines and the application rate per unit area.

[0133] like Figure 4 As shown in one embodiment, the present invention also provides an agricultural microbial regulation decision system based on a large model, comprising: The acquisition module is used to acquire unstructured agricultural microbiology literature text datasets; The extraction module is used to perform semantic parsing on the document text dataset using a pre-trained large language model to extract structured knowledge information of microbial strains. The structured knowledge information includes strain identifiers, functional characteristic parameters and environmental adaptation parameters associated with the strain identifier, and interaction parameters between the strain corresponding to the strain identifier and other strains. The construction module is used to construct a set of constraints for compound microbial agent ratio decision based on the structured knowledge information. The set of constraints includes functional target constraints generated by the functional characteristic parameters, environmental adaptability constraints generated by the environmental adaptability parameters, and interaction relationship constraints generated by the interaction relationship parameters. The acquisition module is also used to acquire soil environmental parameters and microbial community composition data of the target agricultural field; The execution module is used to perform multi-objective optimization with the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters to generate a compound microbial agent ratio scheme, and to make agricultural microbial regulation decisions based on the compound microbial agent ratio scheme, wherein the compound microbial agent ratio scheme includes the selected strain identifier and the ratio of each strain.

[0134] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A decision-making method for agricultural microbial regulation based on a large model, characterized in that, include: S1. Obtain an unstructured agricultural microbiology literature text dataset; S2. Using a pre-trained large language model, semantic parsing is performed on the document text dataset to extract structured knowledge information of microbial strains. The structured knowledge information includes strain identifiers, functional characteristic parameters and environmental adaptation parameters associated with the strain identifier, and interaction parameters between the strain corresponding to the strain identifier and other strains. S3. Based on the structured knowledge information, construct a set of constraints for compound microbial agent ratio decision-making, wherein the set of constraints includes functional target constraints generated by the functional characteristic parameters, environmental adaptability constraints generated by the environmental adaptability parameters, and interaction relationship constraints generated by the interaction relationship parameters. S4. Obtain soil environmental parameters and microbial community composition data for the target agricultural field; S5. Using the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters, perform multi-objective optimization to generate a compound microbial agent ratio scheme, and make agricultural microbial regulation decisions based on the compound microbial agent ratio scheme, wherein the compound microbial agent ratio scheme includes the selected strain identifier and the ratio of each strain.

2. The method according to claim 1, characterized in that, The process of using a pre-trained large language model to perform semantic parsing on the document text dataset and extracting structured knowledge information about microbial strains includes: S21, each document text in the document text dataset is divided into multiple text segments; S22, for each of the text segments, construct a structured extraction prompt, wherein the structured extraction prompt defines extraction slots corresponding to the strain identifier, the functional characteristic parameter, the environmental adaptation parameter and the interaction relationship parameter respectively; S23, input the text fragment and the corresponding structured extraction prompt into the pre-trained large language model to obtain the structured extraction record output by the large language model; S24, Summarize all the structured extracted records corresponding to the text fragments to obtain the structured knowledge information of the microbial strain.

3. The method according to claim 2, characterized in that, The process of constructing the structured extraction prompts includes: S221, Obtain a preset slot definition template, wherein the slot definition template contains the slot name, slot description and output format constraints corresponding to the strain identifier, the functional characteristic parameters, the environmental adaptation parameters and the interaction relationship parameters respectively; S222, Select several literature texts containing at least one of the strain identifier, functional characteristic parameters, environmental adaptation parameters and interaction relationship parameters from the literature text dataset as samples, extract the standard structured extraction record corresponding to each sample text, and combine the sample text and the corresponding standard structured extraction record into a few sample examples; S223, the slot definition template is concatenated with the few sample examples to obtain the structured extraction prompt.

4. The method according to claim 2, characterized in that, The structured extracted records corresponding to all the text fragments are summarized to obtain the structured knowledge information of the microbial strain, including: S241, perform synonym merging on the strain identifiers in each of the structured extraction records, and group multiple structured extraction records pointing to the same strain into the same strain record group; S242, for parameters with the same name from different structured extraction records within the same strain record group, when there is a conflict in parameter values, the conflicting parameter values ​​are selected or merged based on the timeliness weight and / or journal impact factor weight of the literature text from which each parameter value originates. S243, merge the parameter values ​​after conflict resolution in each of the strain record groups to obtain the integrated record corresponding to each strain identifier. The integrated record of all strain identifiers constitutes the structured knowledge information of the microbial strain.

5. The method according to claim 1, characterized in that, The set of constraints for constructing the compound microbial agent ratio decision based on the structured knowledge information includes: S31, extract the environmental adaptation parameters corresponding to each strain identifier from the structured knowledge information, and generate environmental adaptation constraints based on the environmental adaptation parameters. The environmental adaptation constraints are used to limit the range of environmental parameter values ​​for maintaining the activity of the strain. S32, extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and generate functional target constraints based on the functional characteristic parameters. The functional target constraints are used to limit the contribution value of a unit dosage of the strain to a preset functional index. S33, extract the interaction parameters corresponding to the strain pairs with interaction relationships from the structured knowledge information, and generate interaction constraints based on the interaction parameters. The interaction constraints are used to limit the functional correction coefficient or compatibility restriction conditions of the strain pairs when they are used in combination. S34, combine all the environmental adaptability constraints, the functional objective constraints, and the interaction relationship constraints into the constraint set.

6. The method according to claim 1, characterized in that, The process of using the constraint set as the constraint space, and the soil environmental parameters and the microbial community composition data as input parameters, to perform multi-objective optimization and generate a compound microbial agent formulation scheme includes: S51, Match the soil environmental parameters with the environmental adaptability constraints, and screen out strains whose environmental adaptability parameters do not meet the range of environmental parameter values ​​of the target agricultural field to obtain a candidate strain set; S52, Based on the microbial community composition data, determine the microbial functional gap of the target agricultural field, and select the corresponding target optimization function from the functional target constraints according to the microbial functional gap; S53, using the ratio of each strain in the candidate strain set as the decision variable, the objective optimization function as the optimization objective, and the interaction relationship constraints and the range of values ​​of the ratio corresponding to each strain identifier as constraints, a multi-objective optimization model is constructed. S54, Solve the multi-objective optimization model and select the compound microbial agent ratio scheme from the Pareto optimal solution set obtained by the solution.

7. The method according to claim 6, characterized in that, The process involves matching the soil environmental parameters with the environmental adaptability constraints, filtering out strains whose environmental adaptability parameters do not meet the range of environmental parameter values ​​for the target agricultural field, and obtaining a candidate strain set, including: S511, for each soil environmental parameter, the value range corresponding to the parameter in the environmental adaptability constraint is divided into an optimal sub-range and a tolerance sub-range; S512, compare the environmental adaptation parameters corresponding to each strain identifier with the soil environmental parameters one by one, and record whether the comparison result of each parameter falls into the optimal sub-interval or the tolerance sub-interval; S513, filter out strain identifiers that have comparison results for at least one of the soil environmental parameters that exceed both the optimal sub-interval and the tolerance sub-interval, include the remaining strain identifiers in the candidate strain set, and retain the comparison results corresponding to each strain identifier.

8. The method according to claim 6, characterized in that, The process of determining the microbial functional gaps in the target agricultural field based on the microbial community composition data includes: S521, Perform functional gene annotation on the microbial community composition data to obtain the current functional gene abundance spectrum of the target agricultural field; S522, extract the functional characteristic parameters corresponding to each strain identifier from the structured knowledge information, and construct a reference functional requirement spectrum for the crop type to which the target agricultural field belongs based on the functional characteristic parameters; S523, compare the current functional gene abundance spectrum with the reference functional requirement spectrum function by function, and identify functional items whose abundance difference exceeds a preset threshold as the microbial functional gap.

9. The method according to claim 1, characterized in that, The decision-making process for agricultural microbial regulation based on the compound microbial agent formulation includes: S61, for the strain identifier selected in the compound microbial agent formulation scheme, obtain the preparation process parameters associated with the strain identifier from the literature text dataset, and generate microbial agent preparation guidelines based on the preparation process parameters; S62, based on the soil environmental parameters and the environmental adaptation parameters corresponding to each selected strain, determine the effectiveness coefficient of each selected strain in the current field environment, and based on the effectiveness coefficient and the ratio, determine the application rate per unit area of ​​each selected strain. S63, outputting agricultural microbial regulation decision results including the microbial agent preparation guidelines and the application rate per unit area.

10. A decision-making system for agricultural microbial regulation based on a large model, characterized in that, include: The acquisition module is used to acquire unstructured agricultural microbiology literature text datasets; The extraction module is used to perform semantic parsing on the document text dataset using a pre-trained large language model to extract structured knowledge information of microbial strains. The structured knowledge information includes strain identifiers, functional characteristic parameters and environmental adaptation parameters associated with the strain identifier, and interaction parameters between the strain corresponding to the strain identifier and other strains. The construction module is used to construct a set of constraints for compound microbial agent ratio decision based on the structured knowledge information. The set of constraints includes functional target constraints generated by the functional characteristic parameters, environmental adaptability constraints generated by the environmental adaptability parameters, and interaction relationship constraints generated by the interaction relationship parameters. The acquisition module is also used to acquire soil environmental parameters and microbial community composition data of the target agricultural field; The execution module is used to perform multi-objective optimization with the constraint set as the constraint space and the soil environmental parameters and the microbial community composition data as input parameters to generate a compound microbial agent ratio scheme, and to make agricultural microbial regulation decisions based on the compound microbial agent ratio scheme, wherein the compound microbial agent ratio scheme includes the selected strain identifier and the ratio of each strain.

Images