A multi-agent and knowledge-enhanced engineering report generation method and system

Abstract

The application discloses a kind of engineering report generation method and system based on multi-agent and knowledge enhancement, specifically related to engineering design intelligent generation field, including receiving conditional commission, experimental report, topographic map and other materials by preface input module, synchronous archiving, analysis and integrity check are carried out to received data, and project input dataset is obtained;Through AI processing module, document parameter extraction, knowledge enhancement retrieval and field normalization processing are carried out to project input dataset, the corresponding relationship between parameter value and source fragment is established, and structured parameter set is obtained. By unifying preface input data, structured parameters, chart results and report content into intermediate achievement link, and performing relevant writing and regeneration on change content, it is helpful to relatively inhibit the inconsistency of chart, table and text facts.

Description

Technical Field

[0001] This invention relates to the field of intelligent generation technology for engineering design, and more specifically, to a method and system for generating engineering reports based on multi-agent and knowledge enhancement. Background Technology

[0002] In the design of metal mine beneficiation engineering, the existing work usually revolves around shortening the design cycle and improving the efficiency of the results preparation. The common practice is to first organize and analyze the conditions and entrustment documents, experimental reports, topographic maps and other project data by the designers, and then use calculation software, parametric tabulation tools, process management tools or knowledge base retrieval methods to complete the calculation of process parameters, equipment selection, flow diagram and report generation, and prepare feasibility study reports and design specifications based on these. Taking the feasibility study preparation scenario of a mineral processing project as an example, the design process not only needs to handle the continuous supplementation of data from multiple sources, multiple rounds of parameter revisions, and multi-disciplinary collaborative flow, but also requires that the values ​​be consistent, the sources be traceable, and the chapter formats conform to the standards from the design index table, equipment table, material consumption table to the flowchart, connection diagram and report text. Under this condition, the existing practices can often only improve efficiency in some parts, and it is difficult to incorporate the extraction of preliminary data, calculation results, chart generation and text preparation into the same continuous data thread. As a result, after the parameters are adjusted, the same project is prone to the phenomenon that the table has been updated but the drawings still use the old values, the chart has been revised but the report chapters still retain another version of the expression, and the personnel need to repeatedly check the source documents and intermediate results before the final draft can be completed. The root cause is that the results generated in each stage are stored separately, the mutual reference relationship is unclear, and there is a lack of a unified intermediate result acceptance and change transmission mechanism. The technical problem this application aims to solve is: how to unify the preceding input data, processing results, charts and graphs, and report content in the process of generating engineering reports on a traceable and continuously updated intermediate output chain, so as to avoid inconsistencies in the factual statements of the same project's charts, tables, and texts. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an engineering report generation method and system based on multi-agent and knowledge enhancement. By unifying the preceding input data, structured parameters, charts and graphs, and chapter content in the same intermediate output chain, and driving chart generation, chapter writing, consistency verification, and rollback regeneration based on task association relationships, the problems mentioned in the background art are solved.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for generating engineering reports based on multi-agent and knowledge enhancement, comprising: S1. Receive the conditional authorization letter, experimental report, topographic map and other data through the pre-input module, and perform synchronous archiving, parsing and integrity checks on the received data to obtain the project input dataset; S2. The AI ​​processing module performs document parameter extraction, knowledge enhancement retrieval, and field normalization on the project input dataset to establish the correspondence between parameter values ​​and source fragments, thereby obtaining a structured parameter set. S3. The intelligent agent collaboration module generates chapter task sets and workflow task sets based on the structured parameter set and the standard catalog of the engineering report, and assigns the chapter task sets and workflow task sets to the AI-assisted platform - workflow module to obtain the task association table; S4. Through the AI-assisted platform - workflow module, the structured parameter set is calculated based on the task association table to generate calculation tables, calculate flowchart nodes, select equipment and output reports, forming an intermediate engineering result set that includes design index table, equipment table, material consumption table and plant employee number table. S5. Using the AI-assisted platform - Smart Drawing module, flowcharts, workshop configuration diagrams and building connection diagrams are generated based on the intermediate project deliverables set. Key annotation items in the drawings and their source relationships are written back to the intermediate project deliverables set to obtain a chart-related deliverables set. S6. The AI-assisted platform - intelligent writing module generates a draft of the engineering report according to the chapter generation rules based on the chart association result set, chapter task set and knowledge enhancement retrieval results. The intelligent agent collaboration module performs consistency verification between the draft of the engineering report and the chart association result set. When changes or conflicts in the associated content are detected, the system performs step-by-step or skip-level return to regenerate according to the task association table and outputs the engineering report.

[0005] In a preferred embodiment, S1 includes the following steps: S1-1. Write data type identifier, source identifier, project identifier and receiving sequence number into the received data through the pre-input module, extract the title field, text fragment, table fragment and graphic fragment from each received data, and generate input data fragment set; S1-2. Based on the input data fragment set, establish the field correspondence between the conditional entrustment form, experimental report, topographic map and other data according to the field name and field position. Write conflict flags for fragments with the same field name but different field content, write missing flags for fragments that only appear in some of the received data, and generate a data verification result set. S1-3. Based on the data verification result set, input data fragments with the same project identifier and no conflict markers are grouped into the same archiving unit according to the receiving sequence number, and input data fragments with conflict markers or missing markers are written into the supplementary recording unit to generate the project input dataset.

[0006] In a preferred embodiment, S2 includes the following steps: S2-1. Using the AI ​​processing module, extract parameter names, parameter values, units of measurement, object names, and location indexes from title segments, body segments, table segments, and graphic annotation segments in the project input dataset according to the data type identifier and segment sequence number. Extraction results with the same parameter name, object name, and location index are written into the same candidate parameter record. The candidate parameter record is then bound to the data type identifier, source identifier, and location index of the corresponding source segment to generate a candidate parameter record set. S2-2. Based on the candidate parameter record set and the knowledge-enhanced retrieval results, standard name replacement is performed on the parameter names in each candidate parameter record, unit conversion is performed on parameter values ​​with different units of measurement, and object merging is performed on candidate parameter records with different object names but the same location index or the same object name but adjacent location index. Candidate parameter records with the same parameter name after replacement, the same object after merging, and the same parameter value after conversion are written into the same parameter merging group. Candidate parameter records with the same parameter name after replacement, the same object after merging, but different parameter value after conversion are written into the same conflict parameter group, generating a parameter merging result set. S2-3. Based on the parameter merging result set, sort each parameter merging group in descending order of the number of source fragments, in order of priority of conditional authorization documents over experimental reports and experimental reports over other documents in the data type identifier, and in ascending order of position index. Take the candidate parameter record at the top of the sorted list as the target parameter record, and write the standard parameter name, parameter value, uniform unit of measurement, object name and source fragment set of the target parameter record into the structured parameter table. Write all candidate parameter records and their source fragment sets in each conflict parameter group into the conflict parameter table to generate a structured parameter set.

[0007] In a preferred embodiment, S3 includes the following steps: S3-1. Based on the directory hierarchy, title order, and title name in the standard directory of the engineering report, the intelligent agent collaboration module performs chapter affiliation matching on the standard parameter name, object name, source fragment set, and conflict status of the structured parameter set item by item. Structured parameter records with the same title name affiliation are written into the same chapter task unit, and structured parameter records with calculation dependency, drawing dependency, or report dependency are written into the corresponding workflow task unit respectively, generating chapter task set and workflow task set. S3-2. Based on the chapter task set and workflow task set, extract the workflow task unit number corresponding to the structured parameter record contained in each chapter task unit, write the chapter task units that reference the same workflow task unit into the same task association group, arrange the workflow task units with a previous and subsequent reference relationship in the order of being referenced first and referencing later, write the chapter task units and workflow task units corresponding to the structured parameter record with conflict status into the pending task group, and generate a task association result set. S3-3. Based on the task association result set, write the chapter task units and workflow task units that do not belong to the task group to be verified into the task association table according to the task association group number, the workflow task unit arrangement order, and the chapter task unit title order. Then, add the conflict status and corresponding source fragment set to the chapter task units and workflow task units that belong to the task group to be verified and write them into the task association table to obtain the task association table.

[0008] In a preferred embodiment, S4 includes the following steps: S4-1. Using the AI-assisted platform - workflow module, based on the workflow task units in the task association table, the structured parameter records in the structured parameter set are grouped into rows by object name and into columns by standard parameter name. The parameter value with the smallest receiving sequence number in the source fragment set is written into the corresponding cell, and the cell group consisting of the same object name and different standard parameter names is written into the same calculation table page to generate a calculation table set. S4-2. Based on the calculation table set, extract the cell groups with adjacent object names according to the page order and row order. Connect the cell groups with the same output object name of the previous cell group and the input object name of the next cell group as adjacent nodes. When multiple previous cell groups are connected to the same next cell group, add the corresponding flow values ​​as the input flow value of the next node. When multiple previous cell groups are connected to the same next cell group, multiply the corresponding ratio value by their respective flow values, sum them, and then divide by the total input flow to obtain the input ratio value of the next node. Generate the flowchart node result set. S4-3. Based on the node name, input flow rate value, and input ratio value in the flowchart node result set, read the applicable node name, lower limit of processing capacity, upper limit of processing capacity, and applicable ratio range for each device record in the device library. Keep the device records with the same applicable node name, input flow rate value falling between the lower limit of processing capacity and the upper limit of processing capacity, and input ratio value falling within the applicable ratio range as candidate device records. Sort them in ascending order by the difference between the upper limit of processing capacity and the input flow rate value, in ascending order by device specification value, and in ascending order by device number. Then, take the first candidate device record and write it into the device table to generate the device table set. S4-4. Based on the calculation table set, flowchart node result set, and equipment table set, write the structured parameter records whose standard parameter names belong to the design index names into the design index table by name and object name. Merge the structured parameter records whose standard parameter names belong to the material names and consumption quantities by material name and sum them with the consumption quantities, and write them into the material consumption table. Merge the structured parameter records whose standard parameter names belong to the job names and number of employees by job name and sum them with the number of employees, and write them into the ore dressing plant employee count table. Combine and store the design index table, equipment table, material consumption table, and ore dressing plant employee count table to generate the intermediate engineering result set.

[0009] In a preferred embodiment, S5 includes the following steps: S5-1. Using the AI-assisted platform - Smart Drawing module, perform primitive matching on the design index table and equipment table in the intermediate engineering deliverables set according to object name. Merge design index records and equipment records with the same object name into the same process node. Write the node pairs with the same output object name of the previous process node and the same input object name of the next process node into candidate connection edges. Calculate the edge cost of each candidate connection edge, which is composed of the number of shared object names and the difference in record position. In order of edge cost from smallest to largest, retain the candidate connection edges that do not form a loop and do not cause duplicate input edges or duplicate output edges in the same process node to form a process directed graph. Then, write column coordinates layer by layer into the process directed graph with the process node with zero in-degree as the starting layer. In each column, write row coordinates in the order of the fewest number of connection edge intersections and the smallest sum of length of adjacent connection edges to generate a flowchart. S5-2. Based on the flowchart, perform the first merging of equipment records in the equipment table according to the connected branches of the directed graph, the second splitting according to the equipment category in the equipment records, and then merge the split equipment records into workshop units in descending order of the number of connecting edges. For each workshop unit, calculate the rectangular boundary obtained by accumulating the number of equipment, the equipment dimensions, and the maintenance interval. Project each workshop unit onto the workshop configuration map area according to the column coordinate order in the flowchart. For units in the same train, write the vertical position by accumulating the rectangular boundary height. For workshop units with overlapping rectangles, translate them sequentially along the vertical position according to their maintenance interval until they no longer overlap, and generate the workshop configuration map.

[0010] In a preferred embodiment, S5 further includes the following step: S5-3. Based on the workshop layout diagram, write each workshop unit as a building node, and write the connecting edges in the flowchart as connection requirement edges. In the grid diagram with the workshop layout diagram area as the plane coordinate system, write the rectangular boundaries of each building node into the inaccessible area. For each connection requirement edge, take the center of the outer boundary of the starting building node as the starting point and the center of the outer boundary of the target building node as the ending point, calculate the path length, number of turns, and number of intersections with the generated building connection lines of each passable grid path, and sum the path length, number of turns, and number of intersections to obtain the path cost. Select the passable grid path with the minimum path cost and write it into the building connection diagram to obtain the building connection line set. Extract the node names and coordinates from the flowchart, the workshop unit names and rectangle boundaries from the workshop configuration diagram, and the starting point, ending point, and path coordinates of the connecting lines from the building connection diagram. Combine these with the corresponding source table names, source field names, and source record locations and write them back to the intermediate project deliverables set to generate a chart-related deliverables set.

[0011] In a preferred embodiment, S6 includes the following steps: S6-1. Using the AI-assisted platform - intelligent writing module, read the source table items, graphic element labels and source fragments in the chart-related results set according to the chapter titles in the chapter task set. Connect the source table items, graphic element labels and source fragments with the same object name and corresponding source record positions to form a chapter evidence chain. Calculate the parameter coverage, graphic co-occurrence, and fragment repetition for each chapter evidence chain. Keep the chapter evidence chain with the largest parameter coverage, the largest graphic co-occurrence, and the smallest fragment repetition as the target evidence chain to generate a chapter evidence set. S6-2. Based on the chapter evidence set and chapter generation rules, generate parameter sentences for the source entries in each target evidence chain according to the record order, generate relation sentences for the graphic element labels according to the graphic element connection order, and generate basis sentences for the source fragments according to the fragment order. Write the parameter sentences, relation sentences, and basis sentences into the draft of the engineering report in the order of the chapter titles. Then, extract the parameter values, object names, and relation pairs from the text of each chapter and compare them with the corresponding source entries, graphic element labels, and source fragments one by one. Write the text of chapters with the same parameter values, the same object names, and the same relation pairs into the pass set, and write the text of the remaining chapters into the conflict set to generate the draft of the engineering report and the verification result set.

[0012] In a preferred embodiment, S6 further includes the following step: S6-3. The intelligent agent collaboration module calculates the number of parameter conflicts, name conflicts, and relationship conflicts for each chapter text in the conflict set. When the number of parameter conflicts and name conflicts only appears in the current chapter text, the chapter text is written to the hierarchical return task. When the number of relationship conflicts involves multiple chapter texts or the corresponding source table entries change, the chapter text is written to the hierarchical return task. After regeneration according to the task association table, the comparison is repeated item by item until only the pass set is retained in the verification result set, and the engineering report is output.

[0013] An engineering report generation system based on multi-agent and knowledge enhancement includes: The pre-input module is used to receive conditional authorization forms, experimental reports, topographic maps and other materials, and performs synchronous archiving, parsing and integrity checks on the received materials to obtain the project input dataset. The AI ​​processing module is used to perform document parameter extraction, knowledge enhancement retrieval, and field normalization on the project input dataset, establish the correspondence between parameter values ​​and source fragments, and obtain a structured parameter set. The intelligent agent collaboration module is used to generate chapter task sets and workflow task sets based on the structured parameter set and the standard catalog of engineering reports, and to allocate the chapter task sets and workflow task sets to the AI-assisted platform - workflow module to obtain a task association table; The AI-assisted platform - workflow module is used to perform calculation table generation, flowchart node calculation, equipment selection and report output on the structured parameter set based on the task association table, forming an intermediate engineering result set including design index table, equipment table, material consumption table and plant employee number table; The AI-assisted platform - Smart Drawing Module is used to generate flowcharts, workshop layout diagrams and building connection diagrams based on the intermediate project deliverables set, and write back the key annotation items in the drawings and their source relationships to the intermediate project deliverables set to obtain a chart-related deliverables set; The AI-assisted platform's intelligent writing module generates a draft of the engineering report based on the chart-related result set, chapter task set, and knowledge-enhanced retrieval results, according to chapter generation rules. It works with the intelligent agent collaboration module to perform consistency checks between the draft engineering report and the chart-related result set. When changes or conflicts in related content are detected, it performs step-by-step or skip-level regeneration according to the task association table, and outputs the engineering report.

[0014] The technical effects and advantages of this invention are as follows: 1. By integrating the preceding input data, structured parameters, charts and graphs, and report content into the intermediate deliverables chain, and performing associated write-back and regeneration on the changed content, it helps to relatively suppress inconsistencies in the factual statements of charts, tables, and texts. 2. By performing segmented extraction, field verification, and parameter merging on conditional authorization forms, experimental reports, topographic maps, and other data, a structured parameter set with a clear source can be formed, thereby relatively improving the consistency of subsequent calculations and writing; 3. By generating calculation tables, flowchart node results, equipment tables, and various reports based on the task association table, the parameter processing results can be continuously transferred to the drawing and text stages, thereby relatively improving the connection between intermediate results. 4. By generating the initial draft of the project report according to the evidence chain of each chapter and performing item-by-item verification, step-by-step return, and skip-level return for regeneration, the burden of repeated manual verification can be reduced and the efficiency of finalizing the results can be improved. Attached Figure Description

[0015] Figure 1 This is a diagram of the overall system functional architecture of the present invention. Detailed Implementation

[0016] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1 The present invention provides a method for generating engineering reports based on multi-agent and knowledge enhancement, comprising: S1. Receive the conditional authorization letter, experimental report, topographic map and other data through the pre-input module, and perform synchronous archiving, parsing and integrity checks on the received data to obtain the project input dataset; In this implementation process, the pre-input module's role is to organize the conditional authorization form, experimental report, topographic map, and other materials into a unified input result that is verifiable, merging, and traceable. This allows the subsequent AI processing module to perform document parameter extraction, knowledge enhancement retrieval, and field normalization based on the same input criteria. To this end, each received document is first assigned a unified identifier and its content is segmented. Then, field correspondences are established between different documents to identify conflicting and missing content. Finally, archiving and supplementary data are distributed according to the verification results, forming the project input dataset. This implementation process includes the following steps: S1-1 is used to convert the received data into a set of input data fragments that can be read piece by piece. The implementation method is to first write a unified identifier, and then divide the fragment records according to the content format. In practice, after the pre-input module receives the conditional authorization letter, experimental report, topographic map, and other materials, it first writes a material type identifier, source identifier, project identifier, and receiving sequence number for each received document. The material type identifier distinguishes between conditional authorization letters, experimental reports, topographic maps, and other materials; the source identifier indicates the upload source or interface source; the project identifier is the current project number; and the receiving sequence number is determined sequentially within the same project. Subsequently, each received document is segmented: chapter titles, table headers, and independent title rows are extracted as title fields; continuous text is segmented into text segments based on paragraph boundaries; table areas are segmented into table segments based on row and column structure; and legends, captions, and independent graphic annotations in figures are segmented... The data is divided into graphic segments; for each segment record, the segment sequence number and segment position are written simultaneously. The segment position includes at least the page number and the area number within the page, which are used for subsequent field correspondence; when there is no table area or graphic area in the received data, the corresponding segment is written to an empty set, but the data type identifier, source identifier, project identifier and receiving sequence number of the received data are retained; finally, the input data segment set is generated and written to the project cache for subsequent steps to read; taking a mineral processing project as an example, the "design scale" header row in the condition entrustment form can be extracted as a header field, and the continuous description of the processing capacity below it can be divided into text segments, the grade analysis table in the experimental report can be divided into table segments, and the elevation and coordinate labels on the topographic map can be divided into graphic segments; S1-2 is used to identify consistent, conflicting, and missing fields among different data. It is implemented by establishing a field correspondence based on field name and field location, and then determining the verification result based on the field content. Specifically, after the preceding input module reads the input data fragment set, it extracts field names and field locations from title fields, body text fragments, table fragments, and graphic fragments. Field names are taken from title text, table header text, graphic label names, or explicit references in the body text; field locations are taken from the page number, page area number, and fragment sequence number corresponding to the fragment. Then, using identical project identifiers as a constraint, field correspondence groups are established among the conditional authorization form, experimental report, topographic map, and other data. The establishment method is as follows: fragments with the same field name are preferentially assigned to the same field correspondence group; fragments with different field names but located in the same title field are assigned to the same field correspondence group. The fragments are grouped into the same candidate field corresponding group; for each field corresponding group, if there are two or more fragments with the same field name but different field content, a conflict mark is written on the relevant fragment record, and the conflicting field name, fragment position and source identifier are written in the data verification result set; if a field name only appears in some data, the missing field name, the data type identifier that has appeared and the missing data type identifier are written in the data verification result set; if the field name is the same and the field content is consistent, a consistent status is written; when the explicit field name cannot be directly extracted from the text fragment, the text fragment is attached to the nearest title field before participating in the field correspondence, so as to avoid the text fragment from the title context; finally, the data verification result set is generated, and the correspondence between the fragment records and the verification results is written into the verification mapping table for subsequent steps to read; For S1-2, taking "annual processing capacity" as an example, if the conditional entrustment form records 600,000 tons and the experimental report summary records 500,000 tons, then the corresponding segment will be marked with a conflict flag; if the conditional entrustment form contains a "surface elevation" field but the experimental report does not contain this field, then this field will be marked as a missing field in the data verification result set. S1-3 is used to form a project input dataset that can be directly entered into subsequent processing based on the verification results. This is achieved by writing conflict-free fragments into the archiving unit and fragments with conflict or missing markers into the supplementary recording unit. Specifically, after the preceding input module reads the data verification result set and the verification mapping table, it first groups all input data fragments according to the project identifier. Within each project group, it filters out input data fragments without conflict markers and arranges them in ascending order of reception sequence number, then merges them into archiving units. Each archiving unit must at least write the project identifier, archiving unit number, fragment sequence number set, data type identifier set, and source identifier set. For input data with conflict or missing markers... Data is segmented and written to supplementary recording units instead of archiving units. Each supplementary recording unit must contain at least the project identifier, supplementary recording unit number, segment position, field name, conflict or missing flag, and source identifier. When the same input data segment has both conflict and missing flags, both types of flags are retained in the same supplementary recording unit. When the same project has multiple rounds of data reception, the archiving units formed in the previous round remain unchanged, and the newly added segments in the subsequent round form subsequent archiving units according to the new reception sequence number, so as to ensure that the input source is traceable and does not overwrite previous results. Finally, the archiving units and supplementary recording units are combined to generate the project input dataset and written to the input interface of the AI ​​processing module for S2 to read. For S1-3, taking the mineral processing project as an example, the sections that have been verified to be consistent in the main text of the condition entrustment letter, the experimental report form and the topographic map annotation are entered into the archiving unit, while sections with conflicts in "design scale" or missing "reagent system" are entered into the supplementary recording unit. The subsequent AI processing module will perform normal parameter extraction and abnormal supplementary recording processing accordingly. Through the above processing, the pre-processing input module converts the original received data into a project input dataset with unified identification, unified fragmentation structure, unified verification status, and unified archiving method. This allows the subsequent AI processing module to directly read the input content according to the data type identifier, fragmentation sequence number, and field correspondence. Simultaneously, conflicting and missing content is separated in advance, avoiding repeated retrieval of the original data in subsequent steps. In practical applications: For a mineral processing project, after the designer uploads the conditional authorization form, experimental report, topographic map, and supplementary instructions, the pre-processing input module first writes the project identifier and receiving sequence number, then segments the title field, body fragments, table fragments, and graphic fragments. Subsequently, it identifies conflicts in the "processing scale" field and missing "some equipment parameters" fields. Finally, it writes conflict-free fragments into the archiving unit, and conflicting and missing fragments into the supplementary recording unit, combining them to form the project input dataset for direct reading in subsequent structured parameter extraction and task allocation.

[0018] S2. The AI ​​processing module performs document parameter extraction, knowledge enhancement retrieval, and field normalization on the project input dataset to establish the correspondence between parameter values ​​and source fragments, thereby obtaining a structured parameter set. In this implementation process, the AI ​​processing module transforms various input data in the project input dataset into a structured parameter set that can be used for task allocation, calculation table generation, and image-text linkage. This is based on first extracting parameter elements from different types of fragments, then using knowledge-enhanced retrieval results to achieve name unification, unit unification, and object unification, and finally performing optimal retention and conflict separation among records from multiple sources. To ensure consistency in subsequent steps, this implementation process uses a unified recording method for parameter names, parameter values, units of measurement, object names, and location indexes, and writes candidate parameter records, parameter merging groups, and conflict parameter groups into corresponding result tables. This implementation process includes the following steps: S2-1 is used to form mergeable candidate parameter records from the project input dataset. Its working mechanism is to extract the parameter information from the title fragment, body fragment, table fragment and graphic annotation fragment and write it into a unified field, and then complete the first aggregation by parameter name, object name and position index. In practice, the AI ​​processing module reads the data type identifier, source identifier, segment sequence number, and segment location from the project input dataset. Then, according to the data type identifier, it sequentially reads the title segment, body segment, table segment, and graphic annotation segment from the conditional authorization letter, experimental report, topographic map, and other data. For the title segment, it extracts the parameter name and object name contained in the title, leaving the parameter value and unit of measurement blank. For the body segment, it extracts the parameter name, parameter value, unit of measurement, and object name from the body. For the table segment, it extracts the parameter name corresponding to the table header, the parameter value in the cell, the unit of measurement corresponding to the table header or cell, and the object name corresponding to the row and column headers. For the graphic annotation segment, it extracts the parameter name and object name from the annotation name, extracting the parameter value and unit of measurement when numerical annotations exist, and retaining only the parameter name and object name when no numerical annotations exist. The location index is generated by combining the page number, the area number within the page, and the segment sequence number, used to uniquely identify the location of the extracted result. When multiple parameter names appear within the same segment, multiple extraction results are generated separately. When the same parameter name is extracted repeatedly under the same object name and the same location index, these extraction results are written into the same candidate parameter record, and the candidate parameter record is bound to the data type identifier, source identifier, and location index of the corresponding source fragment; when the extraction result is missing a parameter value or unit of measurement, the null value is retained but the candidate parameter record is not deleted, so that it can be completed later by combining other source fragments; finally, a candidate parameter record set is generated and written into the parameter processing cache for subsequent steps to read. Taking the grade analysis table in the experimental report as an example, the AI ​​processing module can extract the "selected grade" in the table header as a parameter name, extract the value in the corresponding cell as a parameter value, extract the percentage sign as a unit of measurement, extract the "raw ore" in the row title as an object name, and then write it into the candidate parameter record along with the position index of the table segment. S2-2 is used to unify different expressions in the candidate parameter record set to the same parameter caliber. Its working mechanism is to use knowledge-enhanced search results to complete standard name replacement, unit conversion and object merging, and then distinguish consistent parameters and conflicting parameters according to the merged results. In practice, after reading the candidate parameter record set, the AI ​​processing module calls the standard parameter name mapping item, standard unit of measurement mapping item, and object name mapping item in the knowledge-enhanced retrieval results for each candidate parameter record. For parameter names, if the parameter name in the candidate parameter record matches the standard parameter name mapping item, the original parameter name is replaced with the mapped standard parameter name; if not, the original parameter name is retained and marked as pending verification. For parameter values ​​and units of measurement, if the units of measurement in the candidate parameter record are different from the standard units of measurement, the unit conversion is performed according to the conversion relationship given in the knowledge-enhanced retrieval results. The converted value is written into the unified parameter value, and the standard unit of measurement is written into the unified unit of measurement. If there is no conversion relationship, the original parameter value is retained and marked as pending verification. For object names, if the object names are different but the position indices are the same, object merging is performed according to the object name mapping item. If the object names are the same but the position indices are adjacent, object merging is also performed. Adjacent position indices mean that the page number and the intra-page area number are the same and the difference in the fragmentation sequence number is one. After completing the above processing, candidate parameter records with the same standard parameter name, the same object merging result, and the same unified parameter value are written into the same parameter merging group. Candidate parameter records with the same standard parameter name, the same object merging result, but different unified parameter values ​​are written into the same conflict parameter group. When a candidate parameter record has only its name replaced but its parameter value is empty, the record only participates in object merging and does not participate in the determination of whether the unified parameter value is the same or different. Finally, a parameter merging result set is generated, and the parameter merging group and the conflict parameter group are written into the merging index table and the conflict index table, respectively, for subsequent steps to read. Taking the "processing capacity" in the condition entrustment form and the "processing scale" in the experimental report as an example, if the knowledge-enhanced retrieval results are all mapped to the same standard parameter name "annual processing volume", and one parameter value is "600,000 tons / year" and the other parameter value is "500,000 tons / year", then they are first uniformly converted to the same unit of measurement, and then it is determined whether they are written into the same parameter merging group or the same conflict parameter group. S2-3 is used to determine the target parameter records to be directly read in subsequent processing from the parameter merging result set. Its mechanism is to complete a stable sorting within the same parameter merging group according to the sufficiency of source, the order of data credibility, and the order of position, and write consistent records and conflicting records into the structured parameter table and conflicting parameter table, respectively. In specific implementation, after the AI ​​processing module reads the parameter merging result set, it counts the number of source fragments for the corresponding candidate parameter record in each parameter merging group. The number of source fragments is taken as the number of fragments in the source fragment set bound to the candidate parameter record. Then, it sorts them in descending order of the number of source fragments. If the number of source fragments is the same, it sorts them by data type identifier, with conditional authorization forms placed before experimental reports, and experimental reports placed before other data. If the data type identifiers are still the same, it sorts them in ascending order of position index. The process proceeds in the following order: page number first, then intra-page region number, and finally fragment sequence number. The candidate parameter record at the top of the sorted list is taken as the target parameter record for that parameter merging group. The standard parameter name, parameter value, unified unit of measurement, object name, and source fragment set of this target parameter record are written into the structured parameter table. For each conflicting parameter group, the top record is not retained; instead, all candidate parameter records and their source fragment sets are written into the conflicting parameter table for later allocation into the pending task group. When all candidate parameter records in a parameter merging group are marked as pending, the top record is still taken and written into the structured parameter table in the same order, while retaining the pending status on that record. Finally, the structured parameter table and the conflicting parameter table are combined to generate a structured parameter set, which is then written into the input interface of the intelligent agent collaboration module for S3 to read. For example, in S2-3, if the "grade of raw ore" in the same project is given in both the table and the main text of the experimental report, and the converted values ​​are consistent, the candidate parameter record with more source fragments will be retained as the target parameter record; if the corresponding values ​​in the main text and the table are inconsistent, they will be written together into the conflict parameter table for subsequent task association and pending verification. Through the above processing, the AI ​​processing module transforms the scattered, heterogeneous, and heterogenous parameter information in the project input dataset into a structured parameter set with a unified name, unified unit of measurement, unified object definition, and clear source relationship. This enables the subsequent intelligent agent collaboration module to directly generate chapter task sets and workflow task sets based on the standard parameter names and object names. At the same time, it retains numerical conflicts that cannot be directly resolved in the conflict parameter table to avoid the introduction of inconsistent parameters in the subsequent calculation and writing stages. In practical applications: For a mineral processing project, the AI ​​processing module first extracts candidate parameter records such as annual processing capacity, raw ore grade, and building elevation from the condition entrustment form, experimental report, and topographic map. Then, based on the knowledge-enhanced retrieval results, it uniformly replaces the processing capacity and processing scale with the annual processing capacity, converts "ten thousand tons / year" and "tons / year" into the same unit of measurement, and merges records of the same object in adjacent location indexes under the same object name. Subsequently, the target parameter record is selected from the consistent records and written into the structured parameter table, and a conflict parameter table is formed from the inconsistent records. Finally, a structured parameter set that can be directly used for task allocation, calculation table generation, and image-text linkage reading is generated.

[0019] S3. The intelligent agent collaboration module generates chapter task sets and workflow task sets based on the structured parameter set and the standard catalog of the engineering report, and assigns the chapter task sets and workflow task sets to the AI-assisted platform - workflow module to obtain the task association table; In this implementation process, the role of the intelligent agent collaboration module is to convert the structured parameter set into chapter task sets, workflow task sets, and task association tables that can be directly scheduled in subsequent workflows. This prevents the structured parameter records from participating in subsequent calculations, drawing generation, and text generation in a scattered state, and instead completes chapter attribution, dependency attribution, and association arrangement first. To this end, the structured parameter records are first assigned to corresponding chapter task units and workflow task units according to the standard catalog of engineering reports. Then, reference relationships, sequential order, and pending verification relationships are established between chapter task units and workflow task units. Finally, the arrangement results are written into the task association table for unified reading by the AI-assisted platform - workflow module and the AI-assisted platform - intelligent writing module. This implementation process includes the following steps: S3-1 is used to assign structured parameter records to corresponding chapter task units and workflow task units. Its working mechanism is to use the standard directory of the engineering report to determine the chapter affiliation of the parameters, and then use the parameter dependency type to determine the processing destination of the parameters. In practice, the intelligent agent collaboration module reads the directory hierarchy, title order, and title name in the standard directory of the engineering report, and reads the standard parameter name, object name, source fragment set, and conflict status in the structured parameter set one by one; first, it performs chapter affiliation matching based on the correspondence between the title name and the standard parameter name and object name, and writes the structured parameter records that match the same title name into the same chapter task unit, wherein the chapter task unit shall at least write the chapter title, directory hierarchy, title order, and structured parameter record number; Next, the dependency type of each structured parameter record is determined. Structured parameter records that need to be generated by the calculation table are written to the workflow task unit corresponding to the calculation dependency; structured parameter records that need to be generated by the flowchart, workshop configuration diagram, or building connection diagram are written to the workflow task unit corresponding to the drawing dependency; and structured parameter records that need to be output by the design index table, equipment table, material consumption table, or plant employee number table are written to the workflow task unit corresponding to the report dependency. When the same structured parameter record has two or more types of dependencies, the structured parameter record is written to multiple corresponding workflow task units simultaneously, and the same structured parameter record number is retained in each workflow task unit. When the structured parameter record has a conflict state, chapter affiliation matching and dependency allocation are still performed, but the conflict state is retained for subsequent steps to determine. Finally, chapter task sets and workflow task sets are generated and written to the task orchestration cache for subsequent steps to read. Here, taking the structured parameter record of annual processing volume as an example, the structured parameter record can be matched with the chapter task unit corresponding to the general overview or design scale, and written into the calculation-dependent workflow task unit and the report-dependent workflow task unit, so that the subsequent calculation table generation and design index table output can read it together. S3-2 is used to establish the association and processing order between chapter task units and workflow task units. Its mechanism is to first form task association groups based on the workflow task units referenced by the chapter task units, then determine the arrangement order of the workflow task units according to the referencing direction, and simultaneously separate tasks with conflicting states into pending task groups. In specific implementation, after the intelligent agent collaboration module reads the chapter task set and the workflow task set, it extracts all structured parameter record numbers from each chapter task unit, and then looks up the corresponding workflow task unit number based on the structured parameter record number to obtain the set of workflow task units referenced by that chapter task unit. Chapter task units referencing the same set of workflow task units are written into the same task association group, and a task association group number is written to this task association group. Subsequently, each task unit within the same task association group is compared sequentially. The workflow task units are arranged in a reference relationship, with those whose output results are read by other workflow task units listed first, followed by those that read the output results of other workflow task units. When there is no direct reference relationship between two workflow task units, they are arranged in ascending order by their workflow task unit numbers. Chapter task units and workflow task units containing conflict state structured parameter records are not included in the normal task association group sorting but are simultaneously written into the pending task group. The pending task group must contain at least the chapter task unit number, workflow task unit number, structured parameter record number, and conflict state. Finally, a task association result set is generated, and the task association group, the workflow task unit arrangement order, and the pending task group are written together into the association result cache for subsequent steps to read. For example, in S3-2, the structured parameter record with conflicting "raw ore grade" can be written into the relevant chapter of "process conditions" first. At the same time, its corresponding calculation dependency workflow task unit is written into the pending task group and does not directly enter the normal sorting, thereby avoiding conflicting parameters from entering the subsequent calculation first. S3-3 is used to organize the task association result set into a task association table that can be directly called by subsequent modules. Its mechanism is to organize directly executable tasks and pending tasks into the same table according to different writing methods, allowing subsequent modules to directly distinguish between the normal execution path and the pending processing path when reading. In specific implementation, after the intelligent agent collaboration module reads the task association result set, it writes chapter task units and workflow task units that do not belong to the pending task group into the task association table in ascending order of task association group number. Within the same task association group, it first writes workflow task units in the order of their arrangement, and then writes chapter task units in the order of their chapter task unit titles. For chapter task units and workflow task units that belong to the pending task group, the writing process... The association table includes conflict status and corresponding source fragment set, where the source fragment set is taken from the source fragment set bound in the structured parameter table or conflict parameter table corresponding to the structured parameter record. The task association table includes at least the task association group number, chapter task unit number, workflow task unit number, sorting order, title order, conflict status, and source fragment set. If the same chapter task unit corresponds to both a normal workflow task unit and a workflow task unit to be processed, two separate task association table records are written to ensure that subsequent modules can read both the normal execution path and the processing path to be processed. Finally, the task association table is generated and written to the shared task interface of the AI-assisted platform - workflow module and the AI-assisted platform - intelligent writing module for subsequent steps to read. Taking a design scale chapter of a project as an example, the corresponding calculation-dependent workflow task units and report-dependent workflow task units will first be written into the task association table in the order of arrangement, and then the task units of the chapter will be written into the same task association group in the order of the title; if the raw ore grade-related tasks have conflict states, the corresponding additional conflict states and source fragment sets will be recorded for subsequent processing to be read. Through the above processing, the intelligent agent collaboration module organizes the structured parameter set into a task association table with chapter affiliation, task dependency, arrangement order, and pending verification identifier. This enables the subsequent AI-assisted platform - workflow module to execute calculation table generation, flowchart node calculation, equipment selection, and report output in a clear order. It also enables the AI-assisted platform - intelligent writing module to stably read the corresponding workflow results by chapter task unit. At the same time, conflict states and source fragment sets are retained in the task association table, facilitating subsequent execution of step-by-step or skip-level return. In practical applications: For a mineral processing project, the intelligent agent collaboration module first records structured parameters such as annual processing capacity, raw ore grade, and equipment model, and assigns them to the corresponding chapter task units of the general overview, process conditions, and main equipment, and writes them into the workflow task units of calculation dependencies, drawing dependencies, and report dependencies, respectively; then, chapter task units that reference the same workflow task unit are grouped into the same task association group, workflow task units with reference relationships are arranged in order of priority, and tasks with conflicting states are written into the pending task group separately; finally, a task association table is generated for subsequent workflow execution and project report writing to read together.

[0020] S4. Through the AI-assisted platform - workflow module, the structured parameter set is calculated based on the task association table to generate calculation tables, calculate flowchart nodes, select equipment and output reports, forming an intermediate engineering result set that includes design index table, equipment table, material consumption table and plant employee number table. In this implementation process, the AI-assisted platform's workflow module transforms the workflow task units in the task association table and the structured parameter records in the structured parameter set into a set of intermediate engineering deliverables that can be directly read for subsequent drawing and text generation. The processing sequence is as follows: first, the structured parameter records are written into the calculation table set based on the workflow task units; then, the flowchart node result set is calculated based on the calculation table set; subsequently, equipment selection is completed based on the flowchart node result set, and an equipment table set is generated; finally, the design index table, material consumption table, and plant employee headcount table are generated based on the calculation table set, flowchart node result set, and equipment table set, and combined with the equipment table to form the intermediate engineering deliverable set. This implementation process includes the following steps: S4-1 is used to organize structured parameter records into a set of computable calculation tables. Its working mechanism is to limit the range of the table page by the workflow task unit and determine the coordinates in the table by the object name and standard parameter name, thereby converting the scattered parameters into a table structure that can be read by rows and columns. In practice, the AI-assisted platform's workflow module first reads the workflow task unit number from the task association table, and then reads the structured parameter records from the structured parameter set according to the workflow task unit number. For structured parameter records belonging to the same workflow task unit, they are grouped into rows by object name and columns by standard parameter name. The object name is taken from the object name in the structured parameter table, and the standard parameter name is taken from the standard parameter name in the structured parameter table. When the same object name and the same standard parameter name correspond to multiple parameter values, instead of re-checking the source fragment set for values, the parameter value of the already determined target parameter record in the structured parameter table is read first as the corresponding cell value to maintain consistency with S... 2. Consistent output standards; when the structured parameter record corresponding to the cell comes from the conflict parameter table, it is not written to the normal calculation table page, but to the pending calculation table page and the conflict status is retained; when the same object name corresponds to multiple different standard parameter names, these cells are written to the same calculation table page, and the workflow task unit number, object name set, standard parameter name set and cell value are written in the calculation table page; when a certain standard parameter name is missing under the current object name, the corresponding cell is written with an empty value and the standard parameter name column is retained so that subsequent steps can identify the missing item; finally, a calculation table set is generated, and the calculation table page number, row number, column number and cell value are written to the calculation table cache area for subsequent steps to read; Taking the grinding workflow task unit as an example, the AI-assisted platform - workflow module can write the names of objects such as raw ore and grinding products into different rows, write the names of standard parameters such as flow rate, ratio, and particle size into different columns, and then write the parameter values ​​in the target parameter record into the corresponding cells, thereby forming the calculation table page corresponding to grinding. S4-2 is used to generate a flowchart node result set based on the calculation table set. Its mechanism involves connecting cell groups with sequential relationships into flowchart nodes according to page and row order, and performing merge calculations on multiple-source inputs for the same node. In specific implementation, the AI-assisted platform's workflow module first reads the calculation table set in ascending order of page number, then reads cell groups in ascending order of row number within each calculation table page. A cell group is the set of all cells in the row containing the same object name. Subsequently, within the same calculation table page, preceding and following cell groups are compared sequentially. If the output object name corresponding to the preceding cell group is the same as the input object name corresponding to the following cell group, they are connected as adjacent nodes. Both the output object name and the input object name are taken from... In the task association table, the object inheritance relationship record corresponding to the workflow task unit is used. If the current workflow task unit is not written to an object inheritance relationship record separately, the object name in the previous row is used as the output object name and the object name in the next row is used as the input object name. For a node with only one previous cell group connected, the flow value in the previous cell group is directly written as the input flow value of the current node, and the corresponding ratio value is written as the input ratio value of the current node. In the case of multiple previous cell groups connected to the same node, the flow values ​​of each previous cell group are first added to obtain the total input flow, then the ratio value of each previous cell group is multiplied by its respective flow value and summed, and finally the sum is divided by the total input flow to obtain the input ratio value of the current node. When a previous cell group lacks a flow value, it does not participate in the flow summation and weighted ratio calculation, and a missing item mark is written on the current node; when the total input flow is empty, the current node does not write the input ratio value, but only writes the pending status; finally, a flowchart node result set is generated, and the node name, input flow value, input ratio value, upstream node set and downstream node set are written to the node result cache for subsequent steps to read; For example, in S4-2, taking the merging of two flotation concentrates into the thickening node, the flow rates of the two previous cell groups are first added together to obtain the input flow rate of the thickening node. The corresponding proportional values ​​of the two channels are multiplied by their respective flow rates and then summed. Finally, the sum is divided by the total input flow rate to obtain the input proportional value of the thickening node. S4-3 is used to determine the equipment set based on the flowchart node result set. Its mechanism is to determine the applicable scope of equipment by node name, limit the selectable range of equipment by input flow rate and input ratio values, and then prioritize equipment selection based on minimum redundancy. In specific implementation, the AI-assisted platform - workflow module reads the node name, input flow rate, and input ratio values ​​from the flowchart node result set one by one, and then reads the equipment records in the equipment library one by one. Each equipment record includes at least the applicable node name, lower limit of processing capacity, upper limit of processing capacity, applicable ratio range, equipment specification value, and equipment number. For each node, only equipment records with the same applicable node name as the current node name, input flow rate values ​​greater than or equal to the lower limit of processing capacity and less than or equal to the upper limit of processing capacity, and input ratio values ​​greater than or equal to the lower limit of the applicable ratio range and less than or equal to the upper limit of the applicable ratio range are retained as candidates. Select device records; when there are more than one candidate device record, first calculate the difference between the upper limit of the processing volume and the input flow value of each candidate device record, then sort them in ascending order according to the difference. If the differences are the same, sort them in ascending order according to the device specification value. If the device specification values ​​are still the same, sort them in ascending order according to the device number. Take the first candidate device record in the sorted list and write it into the device table. When the candidate device record is empty, write the current node name, input flow value, input ratio value, and pending status into the device table, but do not write the device number. When the input ratio value in the flowchart node result set is empty, filter the candidate device records only by node name and input flow value, and add a pending status mark to the generated device table record. Finally, generate a device table set and write the node name, device number, device specification value, input flow value, input ratio value, and filtering status into the device table cache for subsequent steps to read. Here, taking the concentration node as an example, if the input flow rate is 100 tons per hour and the input ratio is 25%, then only the equipment records with the applicable node name "concentration", the processing capacity range covering 100 tons per hour and the applicable ratio range covering 25% are retained, and then the equipment with the smallest processing capacity redundancy is selected from them and written into the equipment table. S4-4 is used to generate intermediate engineering result sets based on calculation tables, flowchart node result sets, and equipment tables. Its mechanism involves distributing intermediate results from different sources according to result type and writing them into corresponding reports, then combining them into a unified output result. In specific implementation, the AI-assisted platform's workflow module first reads the structured parameter record mapping relationships from the calculation tables, the node calculation results from the flowchart node result sets, and the equipment records from the equipment tables. Then, it reads the standard parameter name attribution table, which is written into the rule base during system deployment to indicate whether each standard parameter name belongs to a design indicator name, material name, consumption quantity, job title, or number of employees. For structured parameter records whose standard parameter names belong to design indicator names, they are written into the design indicator table according to the standard parameter name and object name. For structured parameter records whose standard parameter names belong to material names and consumption quantities, they are first assigned to the material name... Then, the consumption corresponding to the same material name is summed and written into the material consumption table; for structured parameter records whose standard parameter names belong to both job titles and number of employees, they are first merged by job title, and then the number of employees corresponding to the same job title is summed and written into the ore dressing plant employee count table; the equipment table set is directly written into the equipment table without repeated calculation; when a structured parameter record appears in both the design index table and the material consumption table, it is written according to its unique belonging in the standard parameter name table without repeated writing; when there are records in the design index table, material consumption table, or ore dressing plant employee count table that are pending verification, the status is retained and written into the intermediate engineering result set; finally, the design index table, equipment table, material consumption table, and ore dressing plant employee count table are combined and stored to generate the intermediate engineering result set, and written into the shared interface of the AI ​​assisted platform - smart drawing module and AI assisted platform - intelligent writing module for subsequent steps to read; Taking the reagent consumption results as an example, the AI-assisted platform - workflow module can merge and sum the consumption of the same reagent in different calculation sheets and write it into the material consumption sheet. Then, it can combine the records of the flotation machine, thickener and feed pump selected in the equipment sheet into the intermediate engineering results set. Through the above processing, the AI-assisted platform - workflow module transforms the structured parameter set into a set of calculation tables, a set of flowchart node results, a set of equipment tables, as well as a set of design index tables, a set of material consumption tables, and a set of plant employee count tables, ultimately forming an intermediate engineering result set. This allows subsequent drawing and text generation to no longer directly rely on the original parameter records, but instead read the engineering result objects that have been uniformly calculated and filtered. At the same time, the pending status is continuously retained in various tables, facilitating subsequent rollback processing and anomaly correction. In practical applications: For a mineral processing project, the AI-assisted platform's workflow module first records the structured parameters corresponding to the names of objects such as raw ore, grinding products, and flotation concentrate into different calculation tables. Then, based on the object's hierarchical relationship, it connects grinding, flotation, and thickening-related objects into process nodes. It calculates the input flow rate and input ratio values ​​for multiple incoming nodes. Subsequently, based on the node name, input flow rate value, and input ratio value, it filters out mills, flotation machines, and thickeners from the equipment library and writes them into the equipment table respectively. Finally, it writes the parameters corresponding to the design index names into the design index table, merges the reagent and steel ball consumption into the material consumption table, merges the number of positions into the plant employee count table, and combines them with the equipment table to form an intermediate engineering deliverable set, which can be directly read for subsequent workshop layout diagrams, building connection diagrams, and engineering report writing.

[0021] S5. Using the AI-assisted platform - Smart Drawing module, flowcharts, workshop configuration diagrams and building connection diagrams are generated based on the intermediate project deliverables set. Key annotation items in the drawings and their source relationships are written back to the intermediate project deliverables set to obtain a chart-related deliverables set. In this implementation process, the AI-assisted platform - Smart Drawing module further converts the tabular results in the intermediate project deliverables set into flowcharts, workshop configuration diagrams, and building connection diagrams. It also re-establishes the correspondence between the nodes, units, and connecting lines formed in the diagrams and their source table entries, so that the AI-assisted platform - Intelligent Writing module can generate an engineering report based on the consistent source of the diagrams, tables, and text. The processing sequence is as follows: first, generate flowcharts based on the design index table and equipment table; then, generate workshop configuration diagrams based on the flowcharts and equipment table; subsequently, generate building connection diagrams based on the workshop configuration diagrams and flowcharts; finally, write back the key annotations in the diagrams and their source relationships to the intermediate project deliverables set. This implementation process includes the following steps: S5-1 is used to convert design index tables and equipment tables into flowcharts. Its mechanism is to first complete the matching of process nodes and the screening of candidate connection edges, and then complete the construction of the directed graph of the process and the arrangement of node coordinates. In specific implementation, the AI-assisted platform - smart drawing module reads the design index tables and equipment tables in the intermediate engineering results set, and extracts the object name, source table name, source field name and source record position for each item. For design index records and equipment records with the same object name, primitive matching is performed, and the matching results are merged into the same process node. Each process node is written with at least the node name, the set of source table names, the set of source field names and the set of source record positions. Then, the process nodes are compared one by one according to the record position of the process node in the design index table and equipment table. When the input object name of the later process node is the same as the output object name of the earlier process node, the node pair is written into the candidate connection edge. The edge cost is calculated for each candidate connection edge. The edge cost is determined by the number of shared object names and the difference in record positions. The more shared object names there are, the smaller the edge cost. The smaller the difference in record positions, the smaller the edge cost. After calculating all candidate edges, retain them in ascending order of edge cost. Before each retention, two conditions are checked: first, whether retaining the candidate edge will form a cycle; if a cycle is formed, the candidate edge is deleted; second, whether retaining the candidate edge will result in duplicate input or output edges in the current process node; if duplicate input or output edges are found, the candidate edge is deleted. The filtered candidate edges form a directed process graph. After the directed process graph is constructed, process nodes with an in-degree of zero are used as the starting layer, and column coordinates are written layer by layer. Process nodes in the same layer are written in the same column. In each column, the order of all process nodes is tried one by one, and the number of edge intersections and the sum of the lengths of adjacent edges under the corresponding arrangement are counted. The arrangement with the fewest number of edge intersections and the smallest sum of the lengths of adjacent edges is written in the row coordinates. If multiple arrangements satisfy the above conditions, they are written in the row coordinates in ascending order of node name. Finally, a flowchart is generated, and the node names, column coordinates, row coordinates, and set of edges are written in the flowchart record table for subsequent steps to read. Taking raw ore, grinding products, and flotation concentrate as examples, the AI-assisted platform - smart drawing module can first merge the corresponding design index records and equipment records into three process nodes, then generate connection edges according to the object acceptance relationship, and form a flowchart from raw ore to grinding products to flotation concentrate. S5-2 is used to convert flowcharts and equipment lists into workshop layout diagrams. Its mechanism involves first forming equipment merging results based on the connectivity of the directed flowchart, then forming workshop units based on equipment categories and connection densities, and finally completing the planar layout based on rectangular boundaries and column coordinates. In specific implementation, the AI-assisted platform – the smart drawing module – reads the directed flowchart and equipment records from the equipment list. It first merges the equipment records according to the connectivity branches of the directed flowchart, writing equipment records in the same connectivity branch into the same equipment merging group. Then, within each equipment merging group, it performs a second split based on the equipment category in the equipment records, splitting equipment records of different categories into different equipment category groups. After each equipment category group is formed, the number of connecting edges between the split equipment records is counted, and they are merged into workshop units in descending order of the number of connecting edges. Priority is given to merging the equipment category group with the most connecting edges into the same workshop unit. If a newly added equipment category group has no connecting edge with the current workshop unit, the merging of the current workshop unit is stopped, and the process moves to the next workshop unit. After the unit is formed, the rectangular boundary is calculated for each workshop unit. The width of the rectangular boundary is obtained by summing the lateral dimensions of all equipment in the workshop unit and the maintenance spacing, and the height is obtained by summing the longitudinal dimensions of all equipment and the maintenance spacing. The equipment dimensions are taken from the corresponding equipment record in the equipment table, and the maintenance spacing is taken from the maintenance spacing field in the equipment table. If the equipment table lacks equipment dimensions, the equipment record is written to the pending verification state and is not included in the current workshop unit boundary calculation. After the rectangular boundary calculation is completed, each workshop unit is projected onto the workshop configuration map area according to the column coordinates in the flowchart. Units in the same workshop are written to the longitudinal position by summing the rectangular boundary heights. When rectangular overlap occurs between units in the same workshop, the later-written workshop unit is translated along the longitudinal position according to its maintenance spacing until it no longer overlaps with the previous workshop unit. If the longitudinal translation still exceeds the configuration range of the current column, the workshop unit is transferred to the next column and the longitudinal position is written again. Finally, the workshop configuration map is generated, and the workshop unit name, rectangular boundary, column coordinates, and longitudinal position are written to the workshop configuration record table for subsequent steps to read. For example, in S5-2, taking the grinding equipment group and the flotation equipment group as examples, the AI-assisted platform - smart drawing module can first classify the equipment on the same production chain into the same equipment group according to the process connection relationship, and then split and recombine them into grinding workshop units and flotation workshop units according to equipment category and the number of connecting edges, and then write them into the workshop configuration diagram. S5-3 is used to convert workshop configuration diagrams and flowcharts into building connection diagrams, and write back the key annotations and their source relationships in the diagrams to the intermediate engineering deliverables set. Its mechanism is to first convert workshop units into building nodes, then perform path search according to connection requirement edges, and finally write back the key results from the three types of diagrams in a unified manner. In specific implementation, the AI-assisted platform - Smart Drawing module reads the workshop unit names, rectangular boundaries, and location coordinates from the workshop configuration diagram and writes each workshop unit as a building node; simultaneously, it reads the connection edges in the flowchart and writes each connection edge as a connection requirement edge; then, a grid diagram is established using the workshop configuration diagram area as a plane coordinate system, with the grid edge length taken as the minimum maintenance interval in the current workshop configuration diagram, serving as the path search step size for the building connection diagram; in this grid diagram, the rectangular boundaries of each building node are written into the inaccessible area, and grid points within the inaccessible area do not participate in the path search; for each connection requirement edge, all traversable grid paths are enumerated, with the starting point being the center of the outer boundary of the initial building node and the ending point being the center of the outer boundary of the target building node; for each traversable grid path, the path length, number of turns, and... The path cost is calculated by summing the number of intersections with existing building connection lines, where path length is the number of grid steps, turning number is the number of direction changes, and intersection number is the number of intersections between the current path and existing connection lines. Then, the accessible grid path with the lowest path cost is selected and written into the building connection diagram. If multiple accessible grid paths have the same path cost, the one with the shortest path length is retained first; if the path lengths are still the same, the one with the fewest turning numbers is retained. After all connection requirement edges are processed, a set of building connection lines is obtained, and the start and end points of the connection lines and path coordinates are written into the building connection record table. Finally, the node names and coordinates in the flowchart, the workshop unit names and rectangle boundaries in the workshop configuration diagram, and the start and end points of the connection lines and path coordinates in the building connection diagram are extracted and combined with the corresponding source table names, source field names, and source record locations, and written back to the intermediate engineering result set to form a graphic element source relationship table and a graphic element annotation table, generating a chart association result set. When there is no accessible grid path between a certain building node, the corresponding connection requirement edge is written to the pending status, and the start, end, and failure markers are retained. For example, in S5-3, the material transfer between the grinding workshop unit and the flotation workshop unit can be written as building nodes by the AI-assisted platform - smart drawing module. Then, a connection line that avoids other building boundaries and has the minimum path cost can be searched in the grid map and written into the building connection map. Through the above processing, the AI-assisted platform - Smart Drawing module further converts the tabular results in the intermediate engineering deliverables set into flowcharts, workshop configuration diagrams, and building connection diagrams. It also re-establishes the correspondence between process nodes, workshop units, and building connection lines and their source table entries, enabling the subsequent AI-assisted platform - Intelligent Writing module to simultaneously read both graphical and tabular results and generate engineering reports based on consistent sources. Meanwhile, anomalies such as unreachable paths and missing equipment dimensions are retained as pending verification, facilitating subsequent rollback processing. In practical applications: For a mineral processing project, the AI-assisted platform - Smart Drawing module first generates a flowchart based on the design index table and equipment table. Then, it generates a workshop configuration diagram based on the connected branches and equipment categories of the flowchart. Subsequently, it generates a building connection diagram based on the connection requirements of the workshop configuration diagram and the flowchart. The module then writes back the process node coordinates, workshop unit rectangle boundaries, and connection line path coordinates, along with the corresponding source table names, source field names, and source record locations, to the intermediate engineering deliverables set, forming a chart-related deliverables set for direct reading in subsequent engineering report writing.

[0022] S6. The AI-assisted platform - intelligent writing module generates a draft of the engineering report according to the chapter generation rules based on the chart association result set, chapter task set and knowledge enhancement retrieval results. The intelligent agent collaboration module performs consistency verification between the draft of the engineering report and the chart association result set. When changes or conflicts in the associated content are detected, the system performs step-by-step or skip-level return according to the task association table to regenerate the report and output the engineering report. In this implementation process, the AI-assisted platform's intelligent writing module and intelligent agent collaboration module are used to convert the chart-related result set, chapter task set, and knowledge-enhanced retrieval results into a verifiable, rollback-able, and regenerable engineering report. During processing, firstly, chapter evidence chains are constructed based on chapter titles, and target evidence chains are filtered out. Then, a draft engineering report is generated according to chapter generation rules, and item-by-item comparisons are performed. Finally, conflicting chapters are returned either step-by-step or skipped-step until a verified engineering report is obtained. This implementation process includes the following steps: S6-1 is used to form a chapter evidence set for each chapter title that can be directly written and verified. Its mechanism is to re-associate source table items, graphic element labels, and source fragments according to the same object name and the same source location, and then select the most complete and least repetitive chain of evidence from multiple chapter evidence chains as the target evidence chain. In specific implementation, the AI-assisted platform - intelligent writing module first reads the chapter titles in the chapter task set, and then reads the source table items, graphic element labels, and source fragments in the chart association result set. For source table items, graphic element labels, and source fragments under the same chapter title, they are connected into a chapter evidence chain according to the condition that the object name is the same and the source record position is corresponding. The corresponding source record position means that the source table item, graphic element label, and source fragment point to the same source table name and the same source record position, or point to the same source table name and the source record positions are adjacent. Then, the evidence chain of each chapter is calculated separately. The evidence chain is determined by the number of parameters covered, the number of text-image co-occurrences, and the number of fragment repetitions. The parameter coverage is the number of standard parameter names in the source table entries, the text-image co-occurrence is the number of object names that appear simultaneously in the graphic element labels and the source fragments, and the fragment repetition is the number of source fragments with the same content. The evidence chains are then sorted in descending order of parameter coverage, descending order of text-image co-occurrences, and ascending order of fragment repetitions. The chapter evidence chain with the highest number of repetitions is taken as the target evidence chain for that chapter title and written into the chapter evidence set. If there is no chapter evidence chain that meets the connection conditions under a certain chapter title, the chapter title is written into the chapter table to be verified and the empty evidence status is retained. Taking the design scale chapter as an example, the annual processing volume source table entries, the corresponding graphic element labels, and the processing scale source fragments can be connected into the same chapter evidence chain. Then, the chapter evidence chain with the largest parameter coverage and the smallest fragment repetition is selected from the candidate chapter evidence chains as the target evidence chain. S6-2 is used to generate a draft engineering report and form a verification result set based on the chapter evidence set. Its mechanism involves first generating parameter sentences, relation sentences, and basis sentences according to different evidence types in the target evidence chain, then reversing the generation of the chapter text and comparing it item by item with the source evidence. Specifically, the AI-assisted platform's intelligent writing module reads the chapter evidence set and chapter generation rules, then generates parameter sentences for each source entry in the target evidence chain according to the record order. The parameter sentences are formed by combining the object name, standard parameter name, and parameter value. For element annotations, relation sentences are generated according to the element connection order. The relation sentences are formed by combining the starting element name, ending element name, and connection relationship. For source fragments, basis sentences are generated according to the fragment order. The basis sentences are extracted from the original description content of the source fragments. Finally, the parameter sentences, relation sentences, and basis sentences are written into the draft engineering report according to the chapter titles. The corresponding chapter texts are generated. After the initial draft of the engineering report is formed, the AI-assisted platform - intelligent writing module extracts parameter values, object names, and relationship pairs from each chapter text. The relationship pairs are taken from the pairs of related object names in the chapter text. The extracted results are then compared with the corresponding source table items, graphic element annotations, and source fragments item by item. Chapter texts with consistent parameter values, consistent object names, and consistent relationship pairs are written into the pass set. Chapter texts with any inconsistency are written into the conflict set. The inconsistency item type and corresponding chapter title are written into the verification result set. If the chapter evidence chain corresponding to a certain chapter text is empty, it is directly written into the conflict set and the status of missing evidence is written. Taking the process flow chapter as an example, the raw ore processing capacity parameter sentence, node connection relationship sentence, and test basis sentence can be generated first. Then, the processing capacity, object name, and connection relationship in the chapter text are read back item by item and checked against the source evidence. S6-3 is used to roll back and regenerate conflicting chapters and output an engineering report. Its mechanism involves first differentiating the scope of the conflict, then sending the conflicting chapters back to the corresponding task level for regeneration according to the task association table, until only the pass set remains in the verification result set. In practice, the intelligent agent collaboration module reads the conflict set and the verification result set, and then counts the number of parameter conflicts, name conflicts, and relationship conflicts for each chapter in the conflict set. The parameter conflict count is the number of items where the parameter value is inconsistent with the source table entry, and the name conflict count is the number of items where the object name is inconsistent with the graphic element label or source fragment. The number of consistent items and the number of relationship conflicts are taken as the number of items whose connection relationships with the relationship pairs are inconsistent with those in the element labels. When the number of parameter conflicts and the number of name conflicts only appear in the current chapter text and the corresponding source table items have not changed, the chapter text is written to the hierarchical return task and returned to the corresponding chapter task unit according to the task association table to regenerate the chapter text. When the number of relationship conflicts involves multiple chapter texts, or the corresponding source table items have changed, the chapter text is written to the hierarchical return task and returned to the corresponding workflow task unit according to the task association table to regenerate the source table items, element labels, or source fragments. After regeneration, the AI-assisted platform - intelligent writing module reads the updated chapter evidence set again and repeats the generation of parameter sentences, relational sentences, basis sentences, and item-by-item comparison. When there are still conflicting chapters in the verification result set, it continues to perform step-by-step return or skip-level return until only the pass set is retained in the verification result set, and then outputs the engineering report. If the same chapter text still retains a conflicting state after two consecutive rounds of skip-level return, the chapter title is written into the manual verification table and the current chapter text is retained as the version to be verified. Taking the simultaneous occurrence of node relationship conflicts in the process flow chapter and the main equipment chapter as an example, the intelligent agent collaboration module can write both into the skip-level return task, return to the corresponding workflow task unit to regenerate the equipment table and element annotations, and then the AI-assisted platform - intelligent writing module regenerates the corresponding chapter text. Through the above processing, the AI-assisted platform's intelligent writing module and intelligent agent collaboration module transform the chart association result set, chapter task set, and knowledge-enhanced retrieval results into an engineering report with evidence chain constraints, item-by-item verification capabilities, and rollback and regeneration capabilities. This ensures that the parameter values, object names, and relationship descriptions in the engineering report remain consistent with the chart association result set. Simultaneously, chapters lacking evidence, single-chapter conflicts, and cross-chapter conflicts are processed through different paths, avoiding repeated rewriting of the entire engineering report. In practical applications: for a mineral processing engineering project, the AI-assisted platform's intelligent writing module first processes the report according to the general overview and process flow. Establish a chapter evidence chain with the chapter titles such as main equipment, then generate parameter sentences, relation sentences, and basis sentences and write them into the initial draft of the engineering report; subsequently, extract the processing volume, object name, and connection relationship from the text of each chapter, and compare them item by item with the design index table, graphic element annotation, and source fragment; if the process flow chapter only has inconsistent parameter values, write it into the hierarchical return task to regenerate the text of that chapter; if the process flow chapter and the main equipment chapter have conflicting relationships, write it into the hierarchical return task and return to the corresponding workflow task unit to regenerate the source table items and graphic element annotations; output the engineering report after all chapters have passed the verification.

[0023] Furthermore, it also includes an engineering report generation system based on multi-agent and knowledge enhancement, the system comprising: The pre-input module is used to receive conditional authorization forms, experimental reports, topographic maps and other materials, and performs synchronous archiving, parsing and integrity checks on the received materials to obtain the project input dataset. The AI ​​processing module is used to perform document parameter extraction, knowledge enhancement retrieval, and field normalization on the project input dataset, establish the correspondence between parameter values ​​and source fragments, and obtain a structured parameter set. The intelligent agent collaboration module is used to generate chapter task sets and workflow task sets based on the structured parameter set and the standard catalog of engineering reports, and to allocate the chapter task sets and workflow task sets to the AI-assisted platform - workflow module to obtain a task association table; The AI-assisted platform - workflow module is used to perform calculation table generation, flowchart node calculation, equipment selection and report output on the structured parameter set based on the task association table, forming an intermediate engineering result set including design index table, equipment table, material consumption table and plant employee number table; The AI-assisted platform - Smart Drawing Module is used to generate flowcharts, workshop layout diagrams and building connection diagrams based on the intermediate project deliverables set, and write back the key annotation items in the drawings and their source relationships to the intermediate project deliverables set to obtain a chart-related deliverables set; The AI-assisted platform's intelligent writing module generates a draft of the engineering report based on the chart-related result set, chapter task set, and knowledge-enhanced retrieval results, according to chapter generation rules. It works with the intelligent agent collaboration module to perform consistency checks between the draft engineering report and the chart-related result set. When changes or conflicts in related content are detected, it performs step-by-step or skip-level regeneration according to the task association table, and outputs the engineering report.

[0024] Working Principle: This solution first receives the conditional authorization form, experimental report, topographic map, and other materials. It then archives, segments, and verifies these materials to form a unified project input dataset. Next, it extracts parameter names, parameter values, units of measurement, and object names from the project input dataset. Combining this with knowledge enhancement results, it unifies names, converts units, and merges objects, generating a structured parameter set. Based on this, the structured parameter set is allocated into chapter tasks and workflow tasks according to the standard engineering report directory. This further generates calculation tables, process nodes, equipment tables, design index tables, material consumption tables, and a plant employee count table. Based on these, it generates flowcharts, workshop layout diagrams, and building connection diagrams, forming a set of chart-related results. Finally, based on the chart-related results set, chapter tasks, and source fragments, it generates a draft engineering report. Then, it verifies each parameter value, object name, and relationship description in the report, performing a backtracking and regeneration for any conflicting content, until an engineering report consistent with the figures, tables, and source data is output. For example, in the scenario of preparing a feasibility study report for a mineral processing project, the system first reads the project's processing scale, ore properties, test indicators, equipment conditions, and topographic information, organizing the content originally scattered in the condition entrustment letter, test report, and topographic map into unified parameters. Then, based on these parameters, it automatically generates design indicator tables, equipment tables, and material consumption tables, and generates corresponding flowcharts, workshop configuration diagrams, and building connection diagrams. Subsequently, the system generates report content in sequence according to chapters such as general overview, process flow, main equipment, and investment estimation, and checks the processing volume, equipment names, and process relationships in the report against the previously generated tables and drawings item by item. If any data in a certain chapter is found to be inconsistent with the equipment table or flowchart, it returns to the corresponding step to regenerate until the diagrams, tables, and text are consistent, finally obtaining a result that can be directly used for engineering design and report submission.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating engineering reports based on multi-agent and knowledge enhancement, characterized in that, include: S1. Receive the conditional authorization letter, experimental report, topographic map and other data through the pre-input module, and perform synchronous archiving, parsing and integrity checks on the received data to obtain the project input dataset; S2. The AI ​​processing module performs document parameter extraction, knowledge enhancement retrieval, and field normalization on the project input dataset to establish the correspondence between parameter values ​​and source fragments, thereby obtaining a structured parameter set. S3. The intelligent agent collaboration module generates chapter task sets and workflow task sets based on the structured parameter set and the standard catalog of the engineering report, and assigns the chapter task sets and workflow task sets to the AI-assisted platform - workflow module to obtain the task association table; S4. Through the AI-assisted platform - workflow module, the structured parameter set is calculated based on the task association table to generate calculation tables, calculate flowchart nodes, select equipment and output reports, forming an intermediate engineering result set that includes design index table, equipment table, material consumption table and plant employee number table. S5. Using the AI-assisted platform - Smart Drawing module, flowcharts, workshop configuration diagrams and building connection diagrams are generated based on the intermediate project deliverables set. Key annotation items in the drawings and their source relationships are written back to the intermediate project deliverables set to obtain a chart-related deliverables set. S6. The AI-assisted platform - intelligent writing module generates a draft of the engineering report according to the chapter generation rules based on the chart association result set, chapter task set and knowledge enhancement retrieval results. The intelligent agent collaboration module performs consistency verification between the draft of the engineering report and the chart association result set. When changes or conflicts in the associated content are detected, the system performs step-by-step or skip-level return to regenerate according to the task association table and outputs the engineering report.

2. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 1, characterized in that: S1 includes the following steps: S1-1. Write data type identifier, source identifier, project identifier and receiving sequence number into the received data through the pre-input module, extract the title field, text fragment, table fragment and graphic fragment from each received data, and generate input data fragment set; S1-2. Based on the input data fragment set, establish the field correspondence between the conditional entrustment form, experimental report, topographic map and other data according to the field name and field position. Write conflict flags for fragments with the same field name but different field content, write missing flags for fragments that only appear in some of the received data, and generate a data verification result set. S1-3. Based on the data verification result set, input data fragments with the same project identifier and no conflict markers are grouped into the same archiving unit according to the receiving sequence number, and input data fragments with conflict markers or missing markers are written into the supplementary recording unit to generate the project input dataset.

3. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 2, characterized in that: S2 includes the following steps: S2-1. Using the AI ​​processing module, extract parameter names, parameter values, units of measurement, object names, and location indexes from title segments, body segments, table segments, and graphic annotation segments in the project input dataset according to the data type identifier and segment sequence number. Extraction results with the same parameter name, object name, and location index are written into the same candidate parameter record. The candidate parameter record is then bound to the data type identifier, source identifier, and location index of the corresponding source segment to generate a candidate parameter record set. S2-2. Based on the candidate parameter record set and the knowledge-enhanced retrieval results, standard name replacement is performed on the parameter names in each candidate parameter record, unit conversion is performed on parameter values ​​with different units of measurement, and object merging is performed on candidate parameter records with different object names but the same location index or the same object name but adjacent location index. Candidate parameter records with the same parameter name after replacement, the same object after merging, and the same parameter value after conversion are written into the same parameter merging group. Candidate parameter records with the same parameter name after replacement, the same object after merging, but different parameter value after conversion are written into the same conflict parameter group, generating a parameter merging result set. S2-3. Based on the parameter merging result set, sort each parameter merging group in descending order of the number of source fragments, in order of priority of conditional authorization documents over experimental reports and experimental reports over other documents in the data type identifier, and in ascending order of position index. Take the candidate parameter record at the top of the sorted list as the target parameter record, and write the standard parameter name, parameter value, uniform unit of measurement, object name and source fragment set of the target parameter record into the structured parameter table. Write all candidate parameter records and their source fragment sets in each conflict parameter group into the conflict parameter table to generate a structured parameter set.

4. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 3, characterized in that: S3 includes the following steps: S3-1. Based on the directory hierarchy, title order, and title name in the standard directory of the engineering report, the intelligent agent collaboration module performs chapter affiliation matching on the standard parameter name, object name, source fragment set, and conflict status of the structured parameter set item by item. Structured parameter records with the same title name affiliation are written into the same chapter task unit, and structured parameter records with calculation dependency, drawing dependency, or report dependency are written into the corresponding workflow task unit respectively, generating chapter task set and workflow task set. S3-2. Based on the chapter task set and workflow task set, extract the workflow task unit number corresponding to the structured parameter record contained in each chapter task unit, write the chapter task units that reference the same workflow task unit into the same task association group, arrange the workflow task units with a previous and subsequent reference relationship in the order of being referenced first and referencing later, write the chapter task units and workflow task units corresponding to the structured parameter record with conflict status into the pending task group, and generate a task association result set. S3-3. Based on the task association result set, write the chapter task units and workflow task units that do not belong to the task group to be verified into the task association table according to the task association group number, the workflow task unit arrangement order, and the chapter task unit title order. Then, add the conflict status and corresponding source fragment set to the chapter task units and workflow task units that belong to the task group to be verified and write them into the task association table to obtain the task association table.

5. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 4, characterized in that: S4 includes the following steps: S4-1. Using the AI-assisted platform - workflow module, based on the workflow task units in the task association table, the structured parameter records in the structured parameter set are grouped into rows by object name and into columns by standard parameter name. The parameter value with the smallest receiving sequence number in the source fragment set is written into the corresponding cell, and the cell group consisting of the same object name and different standard parameter names is written into the same calculation table page to generate a calculation table set. S4-2. Based on the calculation table set, extract the cell groups with adjacent object names according to the page order and row order. Connect the cell groups with the same output object name of the previous cell group and the input object name of the next cell group as adjacent nodes. When multiple previous cell groups are connected to the same next cell group, add the corresponding flow values ​​as the input flow value of the next node. When multiple previous cell groups are connected to the same next cell group, multiply the corresponding ratio value by their respective flow values, sum them, and then divide by the total input flow to obtain the input ratio value of the next node. Generate the flowchart node result set. S4-3. Based on the node name, input flow rate value, and input ratio value in the flowchart node result set, read the applicable node name, lower limit of processing capacity, upper limit of processing capacity, and applicable ratio range for each device record in the device library. Keep the device records with the same applicable node name, input flow rate value falling between the lower limit of processing capacity and the upper limit of processing capacity, and input ratio value falling within the applicable ratio range as candidate device records. Sort them in ascending order by the difference between the upper limit of processing capacity and the input flow rate value, in ascending order by device specification value, and in ascending order by device number. Then, take the first candidate device record and write it into the device table to generate the device table set. S4-4. Based on the calculation table set, flowchart node result set, and equipment table set, write the structured parameter records whose standard parameter names belong to the design index names into the design index table by name and object name. Merge the structured parameter records whose standard parameter names belong to the material names and consumption quantities by material name and sum them with the consumption quantities, and write them into the material consumption table. Merge the structured parameter records whose standard parameter names belong to the job names and number of employees by job name and sum them with the number of employees, and write them into the ore dressing plant employee count table. Combine and store the design index table, equipment table, material consumption table, and ore dressing plant employee count table to generate the intermediate engineering result set.

6. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 5, characterized in that: S5 includes the following steps: S5-1. Using the AI-assisted platform - Smart Drawing module, perform primitive matching on the design index table and equipment table in the intermediate engineering deliverables set according to object name. Merge design index records and equipment records with the same object name into the same process node. Write the node pairs with the same output object name of the previous process node and the same input object name of the next process node into candidate connection edges. Calculate the edge cost of each candidate connection edge, which is composed of the number of shared object names and the difference in record position. In order of edge cost from smallest to largest, retain the candidate connection edges that do not form a loop and do not cause duplicate input edges or duplicate output edges in the same process node to form a process directed graph. Then, write column coordinates layer by layer into the process directed graph with the process node with zero in-degree as the starting layer. In each column, write row coordinates in the order of the fewest number of connection edge intersections and the smallest sum of length of adjacent connection edges to generate a flowchart. S5-2. Based on the flowchart, perform the first merging of equipment records in the equipment table according to the connected branches of the directed graph, the second splitting according to the equipment category in the equipment records, and then merge the split equipment records into workshop units in descending order of the number of connecting edges. For each workshop unit, calculate the rectangular boundary obtained by accumulating the number of equipment, the equipment dimensions, and the maintenance interval. Project each workshop unit onto the workshop configuration map area according to the column coordinate order in the flowchart. For units in the same train, write the vertical position by accumulating the rectangular boundary height. For workshop units with overlapping rectangles, translate them sequentially along the vertical position according to their maintenance interval until they no longer overlap, and generate the workshop configuration map.

7. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 6, characterized in that: S5 further includes the following steps: S5-3. Based on the workshop layout diagram, write each workshop unit as a building node, and write the connecting edges in the flowchart as connection requirement edges. In the grid diagram with the workshop layout diagram area as the plane coordinate system, write the rectangular boundaries of each building node into the inaccessible area. For each connection requirement edge, take the center of the outer boundary of the starting building node as the starting point and the center of the outer boundary of the target building node as the ending point, calculate the path length, number of turns, and number of intersections with the generated building connection lines of each passable grid path, and sum the path length, number of turns, and number of intersections to obtain the path cost. Select the passable grid path with the minimum path cost and write it into the building connection diagram to obtain the building connection line set. Extract the node names and coordinates from the flowchart, the workshop unit names and rectangle boundaries from the workshop configuration diagram, and the starting point, ending point, and path coordinates of the connecting lines from the building connection diagram. Combine these with the corresponding source table names, source field names, and source record locations and write them back to the intermediate project deliverables set to generate a chart-related deliverables set.

8. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 7, characterized in that: S6 includes the following steps: S6-1. Using the AI-assisted platform - intelligent writing module, read the source table items, graphic element labels and source fragments in the chart-related results set according to the chapter titles in the chapter task set. Connect the source table items, graphic element labels and source fragments with the same object name and corresponding source record positions to form a chapter evidence chain. Calculate the parameter coverage, graphic co-occurrence, and fragment repetition for each chapter evidence chain. Keep the chapter evidence chain with the largest parameter coverage, the largest graphic co-occurrence, and the smallest fragment repetition as the target evidence chain to generate a chapter evidence set. S6-2. Based on the chapter evidence set and chapter generation rules, generate parameter sentences for the source entries in each target evidence chain according to the record order, generate relation sentences for the graphic element labels according to the graphic element connection order, and generate basis sentences for the source fragments according to the fragment order. Write the parameter sentences, relation sentences, and basis sentences into the draft of the engineering report in the order of the chapter titles. Then, extract the parameter values, object names, and relation pairs from the text of each chapter and compare them with the corresponding source entries, graphic element labels, and source fragments one by one. Write the text of chapters with the same parameter values, the same object names, and the same relation pairs into the pass set, and write the text of the remaining chapters into the conflict set to generate the draft of the engineering report and the verification result set.

9. The engineering report generation method based on multi-agent and knowledge enhancement according to claim 8, characterized in that: S6 further includes the following steps: S6-3. The intelligent agent collaboration module calculates the number of parameter conflicts, name conflicts, and relationship conflicts for each chapter text in the conflict set. When the number of parameter conflicts and name conflicts only appears in the current chapter text, the chapter text is written to the hierarchical return task. When the number of relationship conflicts involves multiple chapter texts or the corresponding source table entries change, the chapter text is written to the hierarchical return task. After regeneration according to the task association table, the comparison is repeated item by item until only the pass set is retained in the verification result set, and the engineering report is output.

10. An engineering report generation system based on multi-agent and knowledge enhancement, used to implement the method of any one of claims 1-9, characterized in that, include: The pre-input module is used to receive conditional authorization forms, experimental reports, topographic maps and other materials, and performs synchronous archiving, parsing and integrity checks on the received materials to obtain the project input dataset. The AI ​​processing module is used to perform document parameter extraction, knowledge enhancement retrieval, and field normalization on the project input dataset, establish the correspondence between parameter values ​​and source fragments, and obtain a structured parameter set. The intelligent agent collaboration module is used to generate chapter task sets and workflow task sets based on the structured parameter set and the standard catalog of engineering reports, and to allocate the chapter task sets and workflow task sets to the AI-assisted platform - workflow module to obtain a task association table; The AI-assisted platform - workflow module is used to perform calculation table generation, flowchart node calculation, equipment selection and report output on the structured parameter set based on the task association table, forming an intermediate engineering result set including design index table, equipment table, material consumption table and plant employee number table; The AI-assisted platform - Smart Drawing Module is used to generate flowcharts, workshop layout diagrams and building connection diagrams based on the intermediate project deliverables set, and write back the key annotation items in the drawings and their source relationships to the intermediate project deliverables set to obtain a chart-related deliverables set; The AI-assisted platform's intelligent writing module generates a draft of the engineering report based on the chart-related result set, chapter task set, and knowledge-enhanced retrieval results, according to chapter generation rules. It works with the intelligent agent collaboration module to perform consistency checks between the draft engineering report and the chart-related result set. When changes or conflicts in related content are detected, it performs step-by-step or skip-level regeneration according to the task association table, and outputs the engineering report.

Images