An AI-based project table account intelligent generation method and system

Abstract

The application relates to an AI-based project account intelligent generation method and system, which comprises the following steps: in response to an account generation instruction, extracting a target project identifier and account range information; obtaining a project data set, extracting project entities and a planning index set, and constructing an initial knowledge graph; extracting planning adjustment event information, incrementally updating the initial knowledge graph, and generating a compliance knowledge graph; performing time sequence reasoning on the compliance knowledge graph, generating a compliance state time sequence chain; obtaining a structured account template and filling it, and generating a project account report; the application constructs a knowledge graph by automatically integrating multi-source project data, performs incremental updating and time sequence reasoning based on planning adjustment events, realizes dynamic deduction of a compliance state and intelligent generation of an account, effectively solves the problems of information fragmentation and difficulty in deducing a dynamic compliance state in traditional manual account compilation, and has the effect of significantly improving the automation level of account generation and the accuracy of compliance verification.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of engineering projects, in particular to an AI-based project account intelligent generation method and system. BACKGROUND

[0002] In engineering project management, especially in the field of urban planning and construction, the preparation of a project account is the core work to ensure that the project complies with various regulations and planning requirements throughout its life cycle. Traditionally, this work is highly dependent on manual completion. Management personnel need to manually search, sort out, and check various constraint indicators related to the project, such as land use, plot ratio, building height limit, etc., from a large amount of planning documents, design drawings, approval files, and project-related materials, and determine whether the project is compliant at different time nodes, and finally form a structured account report.

[0003] However, this manual approach has significant limitations. First, planning information is highly fragmented, with diverse document sources, inconsistent formats, and evolving planning systems over time, resulting in multiple historical versions with clear time effectiveness. Long-term engineering projects often span multiple system cycles, making it difficult for manual operations to efficiently integrate complex document systems and accurately locate the actual effective planning provisions at a specific time point, significantly increasing the risk of missing or misjudging provisions. Second, the project compliance status has dynamic evolution characteristics, which is affected not only by external planning system adjustments but also by real-time changes in project design schemes. Manual methods can only perform static checks at isolated time points and lack the ability to systematically deduce the temporal evolution of compliance status, making it difficult to visually present the dynamic trajectory of compliance from project initiation to completion and acceptance, making it difficult for management personnel to fully grasp the evolution of project compliance risks. SUMMARY

[0004] To solve the above-mentioned defects, the application provides an AI-based project account intelligent generation method and system.

[0005] The above-mentioned application objectives of the application are achieved through the following technical solutions:

[0006] An AI-based project account intelligent generation method, comprising the steps of:

[0007] In response to an account generation instruction from a user terminal, extracting a target project identifier and account range information, the account range information including time range information, spatial range information, and content depth level;

[0008] Based on the target project identifier and the account range information, obtaining a project data set, and based on the project data set, extracting a project entity and a set of planning indicators, the project data set including project basic information and planning document information;

[0009] An initial knowledge graph is constructed according to a project entity and a planning index set, and planning adjustment event information is extracted based on planning document information;

[0010] The initial knowledge graph is incrementally updated based on the planning adjustment event information to generate a compliance knowledge graph;

[0011] The compliance knowledge graph is sequentially reasoned based on account range information to deduce and generate a compliance state time sequence chain;

[0012] A structured account template is obtained, and the structured account template is filled according to the compliance state time sequence chain and the compliance knowledge graph to generate a project account report.

[0013] In a preferred example, the application can be further configured to: the step of obtaining a project data set based on a target project identifier and account range information, and extracting a project entity and a planning index set based on the project data set, the project data set including project basic information and planning document information, includes the steps of:

[0014] The project data set is input into a pre-trained multi-channel extraction model to obtain a planning index set, a spatial element information set, and a dynamic event information set;

[0015] The planning index set, the spatial element information set, and the dynamic event information set are input into a pre-trained multi-modal fusion model to obtain a fusion representation vector;

[0016] The fusion representation vector is processed by a pre-configured entity recognition head to obtain a project entity.

[0017] In a preferred example, the application can be further configured to: the multi-channel extraction model includes a natural language processing channel, a spatial analysis channel, and a time sequence event channel, and the step of inputting the project data set into the pre-trained multi-channel extraction model to obtain the planning index set, the spatial element information set, and the dynamic event information set includes the steps of:

[0018] The natural language processing channel identifies structured text and unstructured text in the project data set, extracts constraint clause text, and instantiates it as a planning index set;

[0019] The spatial analysis channel identifies planning drawing information and GIS data in the project data set, extracts geometric features, topological relationships, and attribute information of spatial elements, and generates a spatial element information set;

[0020] The time sequence event channel identifies planning version history information and revision record information in the project data set, extracts version change information, strategy adjustment information, and effectiveness state change information, and generates a dynamic event information set.

[0021] The application can be further configured in a preferred example: the step of performing sequence labeling processing on the fusion feature vector through the pre-configured entity recognition head to obtain the project entity includes the steps of:

[0022] serializing and decoding the fusion feature vector to generate a text sequence and context-aware features of basic language units in the text sequence;

[0023] respectively calculating a start probability distribution of each basic language unit in the text sequence as a start boundary of an entity and an end probability distribution as an end boundary of the entity;

[0024] generating a candidate entity boundary position set based on the start probability distribution and the end probability distribution, and generating a candidate entity semantic feature vector based on the candidate entity boundary position set and the context-aware features;

[0025] performing global sequence labeling on the candidate entity semantic feature vector to generate an entity label sequence, and generating a project entity according to the candidate entity boundary position set and the entity label sequence.

[0026] The application can be further configured in a preferred example: the step of constructing an initial knowledge graph according to the project entity and the set of planning indicators, and extracting planning adjustment event information based on the planning document information includes the steps of:

[0027] performing semantic relationship classification on the fusion feature vector through the pre-configured relationship extraction head to obtain semantic relationships between the project entity and the set of planning indicators, the semantic relationships including spatial constraint relationships, attribute association relationships, and temporal influence relationships;

[0028] Based on the project entity, the set of planning indicators, and the semantic relationships, an initial knowledge graph is constructed, including entity nodes and multi-dimensional semantic relationship edges, the entity nodes including project nodes and planning indicator nodes;

[0029] Monitor the version flow of the planning document information, and use a pre-set temporal difference algorithm to identify structured difference information between different versions in the version flow;

[0030] Map the structured difference information to a graph structure evolution operation on the entity nodes, the graph structure evolution operation including attribute updates, relationship additions and deletions, and node state changes;

[0031] Attach event timestamps, validity times, and change basis information to the graph structure evolution operation to generate planning adjustment event information.

[0032] The application can be further configured in a preferred example: the step of incrementally updating the initial knowledge graph based on the planning adjustment event information to generate a compliance knowledge graph includes the steps of:

[0033] extract an operation type, a target node and associated parameters based on the adjustment event information, and match a preset corresponding knowledge graph update strategy based on the operation type;

[0034] obtain an event timestamp and a validity time based on the adjustment event information, and determine an operation effective timing based on the event timestamp and the validity time, to establish an incremental update sequence based on a time window;

[0035] execute the knowledge graph update strategy according to the incremental update sequence, to generate a compliance knowledge graph.

[0036] In a preferred example, the application can be further configured to: the step of extracting an operation type, a target node and associated parameters based on the adjustment event information, and matching a preset corresponding knowledge graph update strategy based on the operation type, comprises the steps of:

[0037] extracting an operation type based on the adjustment event information, and matching a preset atomization strategy function according to the operation type, wherein the operation type includes node attribute update, relationship topology adjustment and node state migration;

[0038] based on the operation type, parsing a unique identifier of the target node and the associated parameters from the adjustment event information;

[0039] binding the unique identifier of the target node and the associated parameters to the matched atomization strategy function to generate a knowledge graph update strategy.

[0040] In a preferred example, the application can be further configured to: the step of performing time sequence reasoning on the compliance knowledge graph based on the account range information, to deduce and generate a compliance state time sequence chain, comprises the steps of:

[0041] discretizing time range information in the account range information into a plurality of sequential time slices;

[0042] based on the event timestamp and the validity time recorded in the compliance knowledge graph, reconstructing a graph subset corresponding to the effective time of each sequential time slice, and taking it as a compliance judgment benchmark for the sequential time slice, wherein the graph subset includes a planning indicator node and its associated original source information;

[0043] extracting attributes and relationships of the project entity, comparing the extracted attributes and relationships with the compliance judgment benchmark, calculating the compliance deviation degree of the project entity and each planning indicator node, and obtaining the compliance state of the project entity;

[0044] setting a planning indicator node with an absolute value of the compliance deviation degree greater than a preset deviation threshold as a key indicator node in the comparison process;

[0045] generating a key evidence link based on the original source information associated with the key indicator node;

[0046] Serially connecting all the project entity compliance states of the sequential time slices in a preset time sequence, and attaching a timestamp and a key evidence link to each project entity compliance state, to generate a compliance state time sequence chain.

[0047] In a preferred example, the application can be further configured to: the step of extracting the attributes and relationships of the project entity, comparing the extracted attributes and relationships with the compliance judgment criteria, calculating the compliance deviation degree of the project entity and each planning indicator node, and obtaining the project entity compliance state, includes the steps of:

[0048] Identifying the constraint type of the planning indicator node, the constraint type including numerical interval constraint, enumeration option constraint, spatial relationship constraint and logical dependency constraint;

[0049] Based on the constraint type, match and call the preset corresponding deviation degree calculation function to obtain the compliance deviation degree of the project entity and each planning indicator node;

[0050] Based on the preset deviation degree processing strategy, aggregate the compliance deviation degree of the project entity and each planning indicator node to obtain the project entity compliance state.

[0051] The above-mentioned second application object of the application is achieved by the following technical scheme:

[0052] An AI-based project account intelligent generation system, comprising:

[0053] A first information extraction module for extracting target project identifiers and account range information including time range information, spatial range information and content depth level in response to account generation instructions issued by the user end;

[0054] A second information extraction module for obtaining a project data set based on the target project identifiers and the account range information, and extracting project entities and planning indicators based on the project data set, the project data set including project basic information and planning document information;

[0055] A graph construction module for constructing an initial knowledge graph according to the project entities and the planning indicators, and extracting planning adjustment event information based on the planning document information;

[0056] A graph updating module for incrementally updating the initial knowledge graph based on the planning adjustment event information to generate a compliance knowledge graph;

[0057] A time sequence reasoning module for performing time sequence reasoning on the compliance knowledge graph based on the account range information to deduce and generate a compliance state time sequence chain;

[0058] A report generation module is configured to acquire a structured account template, and fill the structured account template according to the compliance state time sequence chain and the compliance knowledge graph to generate a project account report.

[0059] In summary, the AI-based project account intelligent generation method and system provided by the present application automatically integrates multi-source project data to construct a knowledge graph, and performs incremental updating and time sequence reasoning based on planning adjustment events, thereby realizing dynamic deduction of compliance states and intelligent generation of accounts, effectively solving the problems of information fragmentation and difficulty in deducing dynamic compliance states in traditional manual account preparation, and significantly improving the automation level of account generation and the accuracy of compliance verification. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is a flowchart of an embodiment of the AI-based project account intelligent generation method of the present application;

[0061] Figure 2 is a flowchart of step S20 in an embodiment of the AI-based project account intelligent generation method of the present application;

[0062] Figure 3 is a flowchart of step S21 in an embodiment of the AI-based project account intelligent generation method of the present application. DETAILED DESCRIPTION

[0063] The following will be described in detail with reference to the accompanying drawings. Figures 1-3 The present application will be further described in detail.

[0064] In an embodiment, as shown in Figure 1 The present application discloses an AI-based project account intelligent generation method, which specifically includes the following steps:

[0065] S10: In response to an account generation instruction issued by a user terminal, a target project identifier and account range information are extracted, the account range information including time range information, spatial range information, and content depth level;

[0066] In the embodiment, the account generation instruction refers to a structured command issued by the user end, used to start the entire account generation process. Generally, the account generation instruction encapsulates the core parameters required for the account generation task. The account range information refers to a set of parameter information defining the boundaries of the account generation. The account range information includes time range information, spatial range information, and content depth level. The time range information specifies the time window that the project account needs to cover, which can be implemented by using the start and end dates specified by the user, such as January 1, 2020 to December 31, 2023, etc. The spatial range information limits the geographical area of interest of the project account, which is used to filter planning indicators related to this space and ensure the relevance of the content. It can be implemented by using the geographical coordinate range, such as the latitude and longitude bounding box. The content depth level can be implemented by using the pre-set level of detail, such as summary level, standard level, or detailed level. It is mainly used to achieve accurate filtering and granularity control of project data. The target project identifier refers to a code or keyword used to uniquely specify a corresponding project. The target project identifier serves as a key value for data retrieval, ensuring that the system accurately calls data related to a specific target project from databases such as the project library and document library.

[0067] S20: Obtain a project data set based on the target project identifier and the account range information, and extract project entities and a planning indicator set based on the project data set. The project data set includes project basic information and planning document information.

[0068] In the embodiment, the project data set refers to a collection of all original data related to the target project. The project data set includes project basic information and planning document information. The project basic information is data describing the attributes of the target project itself, such as project name, location, design scheme, technical and economic indicators, etc. The planning document information is external regulations and constraint files that the target project needs to comply with, such as overall planning files, detailed planning files, and their historical versions. The project entity refers to a specific object identified from the project basic information, such as a building, a road, or a land parcel. The project entity can be obtained by querying a structured database, such as extracting records from a project element table through an SQL statement, or identifying named entities in unstructured documents through a text analysis tool, such as using regular expressions to match specific keywords. The main purpose is to achieve a structured representation of project elements. The planning indicator set refers to a collection of specific constraint requirements extracted from the planning document information, such as volume rate, building height limit, etc. The planning indicator set can be implemented by parsing the table data of the planning document, such as reading the constraint clauses in the Excel file. The main purpose is to achieve quantitative definition of planning constraints.

[0069] S30: Construct an initial knowledge graph according to the project entity and the planning indicator set, and extract planning adjustment event information based on the planning document information.

[0070] In the embodiment, the initial knowledge graph refers to an initial static knowledge network constructed based on project entities, a set of planning indicators and their mutual relationships. The initial knowledge graph can be implemented by using a graph database technology, for example, using Neo4j to store nodes and relationships, or being constructed based on RDF triples, for example, generating a semantic network through SPARQL query. The initial knowledge graph is used to represent the corresponding relationship between project entities and planning requirements in a certain reference state. The planning adjustment event information refers to structured change records identified by analyzing version differences of planning documents. The planning adjustment event information can be implemented by using a version control system extraction method, for example, obtaining document change logs from a Git repository. The planning adjustment event information describes specific changes in planning content, time points at which the changes take effect, and planning regulations on which the changes are based.

[0071] S40: incrementally updating the initial knowledge graph based on the planning adjustment event information to generate a compliance knowledge graph;

[0072] In the embodiment, the incremental update refers to applying a change operation on the basis of the existing initial knowledge graph. The incremental update can be implemented by using a transaction log playback method, for example, replaying a database transaction log, or by using a timestamp-based merge update method, for example, comparing graph snapshots at different time points. The main purpose is to implement efficient maintenance of the knowledge graph. The compliance knowledge graph refers to a knowledge base formed by incrementally updating the initial knowledge graph based on the planning adjustment event information and having time sequence information. The compliance knowledge graph not only records current effective compliance requirements, but also retains historical evolution records, and supports querying compliance status by time point.

[0073] S50: performing time sequence reasoning on the compliance knowledge graph based on the account range information to deduce and generate a compliance state time sequence chain;

[0074] In this embodiment, time sequence reasoning refers to a process of deducing the compliance state of a project based on time, which can be implemented in the form of time series database query, such as retrieving historical state using InfluxDB, or in the form of state machine model, such as defining state transition rules and simulating execution, which is mainly to realize the dynamic deduction of the evolution of compliance state; the compliance state time sequence chain refers to the state sequence generated after time sequence reasoning on the compliance knowledge graph based on the time range in the account range information, which represents the sequence of compliance state changes over time, and the compliance state time sequence chain can be implemented in the form of time slicing calculation, such as dividing the time range into fixed intervals and calculating the state at each point, or in the form of event sourcing technology, such as reconstructing the state history based on event flow, the compliance state time sequence chain records the compliance state of the project at different time points and its judgment basis in chronological order, to realize the structured presentation of the compliance track.

[0075] S60: Obtain a structured account template, and fill the structured account template according to the compliance state time sequence chain and the compliance knowledge graph to generate a project account report.

[0076] In this embodiment, the structured account template refers to a report framework with pre-defined format, chapters and data fields, and the final project account report is generated by filling the compliance state time sequence chain and the content in the compliance knowledge graph into the structured account template.

[0077] Specifically, in response to the account generation instruction issued by the user terminal, the target project identifier and the account range information are extracted, wherein the account range information includes time range information, space range information and content depth level. This process effectively avoids the blindness of manual retrieval by precisely constraining the project boundary and range parameters; the time range information is used to focus on a specific system period, the space range information is used to limit the geographical area, and the content depth level is used to control the information granularity, thereby solving the problem of retrieval difficulty caused by information fragmentation; further, based on the target project identifier and the account range information, the project data set is obtained, and the project entity and the planning index set are extracted therefrom; the project data set covers the project basic information and the planning document information, this step ensures that only the data matching the specified project, space-time range is processed, reduces irrelevant information interference, and converts unstructured documents into structured semantic units, laying a foundation for subsequent knowledge modeling; subsequently, based on the project entity and the planning index set, the initial knowledge graph is constructed, and the planning adjustment event information is extracted from the planning document information; this process dynamically associates discrete planning clauses with project elements using a semantic association network, while capturing the time sequence traces of system version changes; on this basis, the initial knowledge graph is incrementally updated by the planning adjustment event information to generate a compliance knowledge graph; this incremental update mechanism determines the operation effective time sequence based on the event timestamp and the effective time, efficiently integrates historical changes and current state, and avoids the reconstruction of the knowledge graph; further, based on the account range information, the compliance knowledge graph is subjected to time sequence reasoning, and a compliance state time sequence chain is derived and generated; this time sequence reasoning process discretizes the time range information into sequential time slices, reconstructs the sub-set of the graph corresponding to each slice as the compliance judgment benchmark, and intuitively presents the time sequence evolution of the compliance state by comparing the compliance deviation degree of the project entity and the planning index node; finally, a structured account template is obtained, and the template is filled according to the compliance state time sequence chain and the compliance knowledge graph to generate a project account report; this automated filling process ensures that the report contains the compliance evolution track and key evidence, avoiding manual omission or misjudgment.

[0078] For example, in a new urban district development project, after the user issues an account generation instruction, the target project identifier "New District A-2024" is extracted, and the account range information is set as the time range from January 2024 to December 2026, the spatial range is the area from 116.2 degrees east longitude to 116.5 degrees east longitude, and the content depth level is detailed; based on this, the project data set is filtered and obtained, including project basic information such as project name, geographic location, and planning document information such as land use planning map and volume rate document; project entities such as "Residential Area B" and "Public Green Space C" and planning index set such as "Volume rate ≤ 2.0" and "Building density ≤ 30%" are extracted; the initial knowledge graph is constructed, and the planning adjustment event information such as the volume rate adjustment record in March 2025 is extracted; the initial knowledge graph is incrementally updated through the planning adjustment event information to generate a compliance knowledge graph; based on the account range information, the compliance knowledge graph is time-series inferred to deduce and generate a compliance state time-series chain, showing the compliance state changes of the project in time slices such as 2024 and 2025; a structured account template is obtained and filled to generate a project account report containing compliance evolution track and key evidence link.

[0079] Therefore, the scheme significantly reduces the interference of information fragmentation on account compilation through the precise constraint mechanism of account range information, ensures the accurate clarification of effective planning clauses at a specific time point, and realizes the dynamic deduction of project compliance state based on the time-series evolution and incremental update mechanism of the knowledge graph, which intuitively presents the complete process of compliance evolution over time, thereby effectively solving the technical problems of low efficiency and difficulty in dynamically tracking compliance state in manual mode, and having the effect of improving the automation level and result reliability of account compilation.

[0080] In an embodiment, as shown in Figure 2 Step S20 includes:

[0081] S21: input the project data set into the pre-trained multi-channel extraction model to obtain the planning index set, the spatial element information set, and the dynamic event information set;

[0082] In this embodiment, the pre-trained multi-channel extraction model refers to a feature extraction architecture designed for multi-source heterogeneous data, which can be implemented by combining neural networks with parallel processing channels, aiming to process text semantics, spatial geometric features, and temporal dynamic changes respectively, to avoid misjudgment of heterogeneous data by a single model; the planning index set refers to a structured constraint clause set extracted from the project data set by natural language processing technology and other means, aiming to clarify the project compliance judgment benchmark; the spatial element information set refers to a geometric feature and topological relationship set extracted from planning drawings and GIS data and other data, which can be implemented by using vector data structure or raster data structure encoding, aiming to represent spatial constraint relationships; the dynamic event information set refers to a change record set extracted from version history information, revision record information and other information, which can be implemented by using a time series database or event stream format, aiming to capture the temporal evolution of planning adjustments.

[0083] S22: input the planning index set, the spatial element information set, and the dynamic event information set into the pre-trained multi-modal fusion model to obtain a fusion representation vector;

[0084] In this embodiment, the pre-trained multi-modal fusion model refers to a neural network architecture that integrates different modal features, which receives the heterogeneous information set including the planning index set, the spatial element information set, and the dynamic event information set output by the multi-channel extraction model, and maps it into a unified vector space to generate a comprehensive vector representation containing multiple aspects of information, i.e., a fusion representation vector; the fusion representation vector refers to a unified feature representation generated by the multi-modal fusion model, which can be implemented by using semantic representation in a high-dimensional embedding vector space, aiming to provide global context support.

[0085] S23: perform sequence labeling processing on the fusion representation vector through a pre-configured entity recognition head to obtain a project entity.

[0086] In this embodiment, the pre-configured entity recognition head refers to a neural network layer or small network structure used for sequence labeling tasks, which can be implemented by using conditional random fields or bidirectional LSTM structures, and its function is to decode the fusion representation vector into sequence labels to identify the entity boundaries and types in the text represented by the vector sequence; sequence labeling processing refers to a technique that labels each unit of a text sequence to identify entities, which can be implemented by using boundary detection mechanisms or probability distribution calculations, aiming to improve the robustness of entity recognition.

[0087] Specifically, the scheme realizes accurate extraction of project entities and planning indicators through a hierarchical multi-modal data processing mechanism. First, the project data set is input into a pre-trained multi-channel extraction model. The multi-channel extraction model processes different modal data in parallel for multi-source heterogeneous characteristics, respectively outputs a planning indicator set, a spatial element information set, and a dynamic event information set, and ensures targeted extraction of text, spatial, and temporal information. Then, the extracted feature set is input into a pre-trained multi-modal fusion model to generate a unified fusion feature vector through cross-modal feature interaction. The fusion feature vector encodes the associated features of text semantics, spatial relationships, and temporal evolution. Finally, the pre-configured entity recognition head accurately identifies the boundaries and types of project entities based on the deep context features of the fusion feature vector through a sequence labeling mechanism. This hierarchical processing mechanism makes each step an organic whole: multi-channel targeted and specialized extraction provides a structured foundation for fusion, multi-modal fusion eliminates data fragmentation, and sequence labeling improves recognition accuracy using global context, thereby systematically analyzing multi-source data and providing high-precision input for knowledge graph construction.

[0088] Through the above scheme, the application effectively solves the problem of multi-source heterogeneous information fusion and accurate entity extraction, realizes the integrity of project entity extraction and the systematicness of planning indicator identification, avoids fuzzy entity boundaries and missing key indicators, thereby providing high-precision input for subsequent knowledge graph construction and improving the reliability of compliance deduction in project account generation.

[0089] In an embodiment, the multi-channel extraction model includes a natural language processing channel, a spatial analysis channel, and a temporal event channel, as shown in Figure 3 Step S21 includes:

[0090] S211: The natural language processing channel identifies structured text and unstructured text in the project data set, extracts constraint clause text, and instantiates it as a planning indicator set.

[0091] In this embodiment, the natural language processing channel refers to a dedicated module component for processing text data, which can be implemented in a combination of pre-trained language models based on the Transformer architecture or rule engines, aiming to accurately capture semantic constraint information in the text; structured text refers to text data with fixed format and fields, such as tables in databases, configuration files in JSON or XML format, whose content can be directly parsed by programs; unstructured text refers to continuous text data without preset format, such as paragraph descriptions or clause content in planning documents, whose semantics need to be parsed by natural language understanding technology; constraint clause text refers to the information of sentences or paragraphs representing restrictive requirements identified from planning documents, which usually contains words representing restrictions or requirements and involves specific planning indicators; instantiation refers to the process of converting constraint descriptions extracted from text into structured objects or data records containing specific attribute values, such as converting the text "volume rate not greater than 2.5" into a data object named "volume rate" with constraint operator "≤" and target value "2.5".

[0092] S212: The spatial analysis channel identifies planning drawing information and GIS data in the project data set, extracts geometric features, topological relationships and attribute information of spatial elements, and generates a spatial element information set;

[0093] In this embodiment, the spatial analysis channel refers to a component for processing geographic spatial data, which can be implemented in an integrated solution of computer vision algorithms and geographic information system processing techniques, such as combining convolutional neural networks with spatial topology analysis tools, aiming to quantitatively express the geometric and relationship features of spatial elements; geometric features are mathematical features that describe the shape and spatial properties of spatial elements, such as area, perimeter, center point coordinates, shape contour, etc. for a land parcel element; topological relationships are spatial relationships between spatial elements that are not affected by continuous deformation, such as adjacency, containment, intersection, and separation, etc., which are crucial for analyzing spatial constraints; attribute information is non-geometric description data associated with spatial elements, which is stored in the form of key-value pairs to describe the semantic features of spatial elements such as category, specification, and state, such as a spatial element representing "land parcel" whose attribute information may include "land parcel number", "land use nature", "property owner", "current state" and other fields and specific values.

[0094] S213: The time sequence event channel identifies planning version history information and revision record information in the project data set, extracts version change information, policy adjustment information and validity state change information, and generates a dynamic event information set.

[0095] In the embodiment, the time sequence event channel refers to a unit for processing time sequence change information, which can be implemented by a fusion method of time sequence pattern recognition algorithm and time sequence database technology, and the purpose is to systematically track the dynamic evolution process of planning system; the planning version history information refers to the sequence data formed by each formal release of the planning document in the life cycle, and the planning version history information records the identifier of the version (such as V1.0, V2.0), the releasing organ, the approval date and the effective date and other metadata, and reflects the overall context of the evolution of the planning document; the revision record information refers to the data that specifically describes the content modification of the planning document from one version to the next version, and the revision record information usually exists in the form of revision notes, change lists or difference comparison results, and accurately records the details of addition, deletion and modification of clauses, drawings and other contents, and is the direct basis for identifying specific adjustment events; the version change information is a structured record describing the differences between different versions of the planning document, for example, indicating that a clause is added in V2.0 version or deleted in V3.0 version; the strategy adjustment information refers to the change record of the planning content related to macro guidance, development target, implementation path and other non-quantitative constraints, such information does not directly reflect the numerical change of specific technical indicators, but the adjustment of planning principles, development direction or management requirements, for example, adjusting the “development strategy” of a certain region from “industrial dominance” to “industrial and urban integration”, or updating the “encouragement measures” for a certain type of project, which all belong to strategy adjustment information, and they reflect the evolution of the planning in the overall idea or strategy level; the validity state change information is an event record describing the change of validity of planning clauses or their revised contents, such as validity, invalidity or suspension, and it focuses on the time attribute of the validity of the clauses.

[0096] Specifically, the scheme realizes the directional extraction and structured output of the multi-modal information of the project data set through the cooperative operation of the natural language processing channel, the spatial analysis channel and the time sequence event channel; the natural language processing channel focuses on analyzing the constraint clauses in the text, and converts the unstructured document into a standardized planning index set; the spatial analysis channel deconstructs the features of the planning drawings and GIS data to generate a spatial element information set containing geometric attributes and topological relationships; the time sequence event channel identifies key change events from the version history to form a dynamic event information set; after the three channels process different modal data in parallel, the structured data set is output as the input of the subsequent multi-modal fusion model, ensuring the complete capture and independent processing of the text constraints, spatial elements and time sequence features, thereby providing an accurate and non-interfering data basis for the construction of the knowledge graph.

[0097] Exemplarily, the natural language processing channel can perform semantic parsing on the planning document by using a pre-trained BERT model, identify constraint provisions such as “the building height shall not exceed 50 meters” and instantiate them as planning indicators; the spatial analysis channel can call the GDAL library to process the planning drawing, extract the land use boundary polygon and its topological adjacency relationship, and generate a spatial feature information set; the time sequence event channel can parse the planning document revision record based on the version control system, capture change events such as “the volume rate standard is adjusted from 2.0 to 2.5 in 2023”, and generate a dynamic event information set containing a timestamp and a validity state.

[0098] Through the above technical solutions, the application effectively solves the fragmentation problem of multi-modal information extraction, realizes accurate separation and targeted processing of structured text, unstructured text, planning drawing, GIS data and time sequence change information, significantly improves the integrity of project entity and planning indicator extraction, provides reliable data support for subsequent incremental update of knowledge graph and dynamic deduction of compliance state time sequence chain, and thus improves the compliance judgment accuracy of project account report in cross-institutional period scenarios.

[0099] In an embodiment, step S23 comprises:

[0100] S231: serializing and decoding the fusion feature vector to generate a text sequence and context-aware features of basic language units in the text sequence;

[0101] In this embodiment, the serialization and decoding refers to the process of converting the fusion feature vector into a text sequence and its context features, which can be implemented by using a bidirectional long short-term memory network or a Transformer decoder architecture, aiming to dynamically capture the semantic dependency relationship of language units in the sequence, and provide accurate context semantic basis for boundary recognition; the basic language unit is the smallest unit for processing in the text sequence, which usually refers to a single character or word (word after segmentation) in Chinese, and usually refers to a word or sub-word unit in English; the context-aware feature refers to a numerical vector representation calculated for each basic language unit, which encodes not only the semantics of the unit itself, but also its meaning in the entire sentence or context.

[0102] S232: respectively calculating a starting probability distribution of each basic language unit in the text sequence as an entity starting boundary, and an ending probability distribution as an entity ending boundary;

[0103] In the embodiment, the entity start boundary refers to the position where a named entity starts in a text sequence, which is identified by the index of the basic language unit (such as a word or a character) corresponding to the named entity in the sequence, and which marks the start of an entity segment; the entity end boundary refers to the position where a named entity ends in a text sequence, which is identified by the index of the basic language unit (such as a word or a character) corresponding to the named entity in the sequence, and which marks the end of an entity segment; the start probability distribution and the end probability distribution can be understood as a probabilistic representation of each basic language unit as an entity start boundary and an entity end boundary, which can be generated based on a conditional random field model or a neural network classifier, and the purpose is to explicitly distinguish the start and end roles of the entity, and to avoid the boundary confusion problem caused by a single probability model.

[0104] S233: generating a set of candidate entity boundary positions based on the start probability distribution and the end probability distribution, and generating a candidate entity semantic representation vector based on the set of candidate entity boundary positions and the context-aware feature;

[0105] In the embodiment, the set of candidate entity boundary positions refers to the combination of all possible entity start positions and end positions filtered out based on the start probability distribution and the end probability distribution by setting a probability threshold or a specific rule, and each (start position, end position) pair identifies a candidate entity segment, and the purpose is to reduce invalid boundary interference and improve the quality of candidate boundaries; the candidate entity semantic representation vector refers to the semantic representation of the entity generated by fusing the boundary position and the context feature, which can be implemented by using an attention mechanism or a vector splicing technology, and the purpose is to strengthen the semantic integrity of the candidate entity and avoid semantic distortion caused by boundary errors.

[0106] S234: performing global sequence labeling on the candidate entity semantic representation vector to generate an entity label sequence, and generating an item entity according to the set of candidate entity boundary positions and the entity label sequence.

[0107] In the embodiment, global sequence labeling refers to the process of assigning consistent labels to the candidate entity semantic representation vector, which can be completed by using a sequence labeling algorithm, and the purpose is to ensure the logical coherence of the entity label in the text sequence and to solve the local labeling conflict problem, and at the same time, when assigning labels, the global sequence labeling not only considers the local features of the current unit, but also considers the label selection of all other units in the sequence to seek the globally optimal label sequence; the entity label sequence is the output result of the global sequence labeling process, which is a sequence equal in length to the input text sequence, and the label at each position indicates the role of the corresponding basic language unit in the entity, for example, using the BIO labeling system to identify the start, internal or external of the entity.

[0108] Specifically, the scheme first generates a text sequence and a context-aware feature by serializing and decoding the fusion representation vector, thereby constructing a dynamic semantic representation of the language unit; then, the starting probability distribution and the ending probability distribution of each basic language unit are calculated respectively, the boundary role transition is explicitly modeled, and the entity boundary position is accurately located; subsequently, a candidate entity boundary position set is generated based on the probability distribution, and a candidate entity semantic representation vector is generated in combination with the context-aware feature, a low-confidence boundary is excluded through a probability screening mechanism, and the semantic expression of the candidate entity is enhanced; finally, the candidate entity semantic representation vector is globally sequence-labeled to generate an entity label sequence, thereby realizing the optimization from local boundary recognition to global semantic consistency; the above phased processing mechanism, through the synergistic effect of context-aware feature extraction, boundary probability explicit modeling, candidate boundary screening optimization, and global consistency labeling, solves the problem of inaccurate entity boundary recognition, improves the reliability of project entity extraction, avoids the deviation of subsequent compliance knowledge graph construction caused by entity boundary recognition errors, and thereby improves the reliability of the compliance state inference process.

[0109] In an embodiment, step S30 comprises:

[0110] S31: performing semantic relation classification on the fusion representation vector through a pre-configured relation extraction head to obtain semantic relations between the project entity and the set of planning indicators, the semantic relations including spatial constraint relations, attribute association relations, and time sequence influence relations;

[0111] In the embodiment, the pre-configured relation extraction head refers to a neural network component for automatically identifying semantic relations from the fusion representation vector, which can be implemented by a sequence labeling model based on Transformer or a graph convolution network, and its function is to analyze the input fusion representation vector and determine the specific semantic relation category between the entities involved; semantic relation classification refers to the process of classifying the association between project entities and planning indicator set into a specific type, which can be supervised learning based on a pre-defined relation category set, and the purpose is to structure the constraint logic of planning rules; semantic relations include spatial constraint relations, attribute association relations and time sequence influence relations, which can be implemented by using the relation mode defined by ontology, and the purpose is to cover the multi-dimensional constraint characteristics in planning rules; wherein, the spatial constraint relation represents that the project entity must comply with the restrictions specified by the planning indicator in space, for example, there is such a relation between a "building" entity and a "building control line" indicator, which means that the building cannot exceed the control line; the attribute association relation represents that a certain attribute of a project entity needs to be directly compared with a specific value of a planning indicator, for example, there is such a relation between the "plot ratio" attribute of a "plot" entity and the "plot ratio upper limit" value in the planning indicator; the time sequence influence relation represents that the adjustment of a certain planning indicator may affect the compliance state of one or more project entities in the time dimension, which is used to infer the dynamic change of compliance.

[0112] S32: based on the project entity, the planning indicator set and the semantic relation, an initial knowledge graph including entity nodes and multi-dimensional semantic relation edges is constructed, the entity nodes include project nodes and planning indicator nodes;

[0113] In the embodiment, the entity node is a vertex representing an independent entity or concept in the initial knowledge graph, and in the scheme, the entity node mainly refers to the project node representing a specific project element and the planning indicator node representing a planning requirement; the multi-dimensional semantic relation edge is a directed edge with types and attributes connecting two entity nodes in the initial knowledge graph, which is used to represent the specific type of semantic relation between the nodes, wherein "multi-dimensional" means that different attributes (such as relation type, strength, effective time, etc.) can be attached to the edge.

[0114] S33: monitoring the version flow of the planning document information, and using a pre-set time sequence difference algorithm to identify the structured difference information between different versions in the version flow;

[0115] In this embodiment, version flow refers to a version sequence formed by arranging the same planning document in chronological order according to its release or revision time, which records the complete evolution history of the document over time; monitoring the version flow refers to continuously tracking the historical change record of the planning document information, which can be realized through a version control system or a document management API, and the purpose is to capture the complete trajectory of the evolution of the document; the time series difference algorithm refers to an algorithm for comparing the differences between different versions of the document, which can be realized by using a text similarity-based difference algorithm or a structured analysis difference tool, and the purpose is to accurately identify changes at the semantic level; structured difference information refers to the standardized representation of the differences between versions, which can be represented as a set of added, deleted or modified sentences, and the purpose is to provide an operational change description.

[0116] S34: Map the structured difference information to a graph structure evolution operation on the entity node, the graph structure evolution operation including attribute updating, relationship adding or deleting, and node state changing;

[0117] In this embodiment, the graph structure evolution operation refers to an operation instruction for modifying the structure of the initial knowledge graph, and the graph structure evolution operation directly acts on the nodes and edges of the initial knowledge graph, such as modifying the attribute value of the node (i.e., attribute updating), adding or deleting an edge between nodes (i.e., relationship adding or deleting), changing the state identifier of the node itself (i.e., node state changing), etc., which can be mapped to the CRUD operation of the graph database, and the purpose is to dynamically update the knowledge graph to reflect the changes in reality.

[0118] S35: Attach event timestamp, validity time and change basis information to the graph structure evolution operation to generate planning adjustment event information.

[0119] In this embodiment, the event timestamp refers to information recording the time when the planning adjustment event occurs; the validity time refers to the time point when the content specified by the planning adjustment event starts to have validity or starts to take effect, which can be different from the event timestamp (release approval time), for example, there is a case of “validity after 30 days of release”; the change basis information refers to the source information of the rules or system documents used to explain or prove the compliance and reason of a certain graph structure evolution operation.

[0120] Specifically, the scheme classifies the semantic relationship of the fusion representation vector through the pre-configured relationship extraction head, accurately identifies the multi-dimensional semantic relationship between the project entity and the planning index set, and ensures the semantic integrity of the knowledge graph construction; when constructing the initial knowledge graph based on the project entity, the planning index set and the semantic relationship, the entity node is subdivided into a project node and a planning index node, and the constraint logic such as spatial constraint, attribute association and time sequence influence is explicitly expressed through the multi-dimensional semantic relationship edge, so that the initial knowledge graph has the ability to express complex planning rules; at the same time, the version flow of the planning document information is monitored, and a preset time sequence difference algorithm is used to identify the structured difference information, the semantic changes between the historical versions are automatically compared through the algorithm, and the accurate capture of the difference information is ensured; the structured difference information is mapped to the graph structure evolution operation, the abstract document change is converted into executable operations such as attribute update, relationship addition and deletion, and node state change, so that the knowledge graph can reflect the substantial influence of planning adjustment in real time; finally, the event timestamp, validity time and change basis information are added to the graph structure evolution operation, and the planning adjustment event information with time sequence logic and traceability is generated, providing reliable time anchor points and evidence support for subsequent compliance state time sequence chain deduction.

[0121] Exemplarily, in the urban planning management scene, for a certain plot development project, the version update of the urban planning regulations is continuously monitored; when a new version of the planning document is detected, the time sequence difference algorithm automatically compares the new and old versions, identifies the change content of the building height limit, parses the change content into an attribute update operation on the corresponding planning index node, and adds an event timestamp, validity time and change basis information, and generates planning adjustment event information which is used for subsequent incremental update of the knowledge graph, ensuring that the graph accurately reflects the current planning requirements.

[0122] Through the above scheme, the application automatically identifies the structured difference in the version iteration of the planning system, enables the knowledge graph to dynamically evolve, thereby providing accurate event-driven basis for compliance state time sequence deduction, and solving the problem that the static knowledge representation cannot adapt to the time sequence change.

[0123] In an embodiment, step S40 comprises:

[0124] S41: extracting the operation type, target node and associated parameters based on the adjustment event information, and matching the preset corresponding knowledge graph update strategy based on the operation type;

[0125] In the embodiment, the operation type refers to the action category for modifying the knowledge graph parsed from the planning adjustment event, which defines the basic nature of the update, such as modification of node attributes, addition or deletion of relationship edges between nodes, change of node state itself, and the like, aiming to accurately adapt to different change scenarios and avoid the risk of misoperation caused by a general strategy; the target node is a unique identifier of one or more nodes in the knowledge graph, which accurately specifies the specific object to which the current update operation is applied, for example, a specific "volume rate index" node to which an attribute update operation is directed; the associated parameter is the supporting data required for performing a specific update operation, the content of which is determined by the operation type, for example, for an "attribute update" operation, the associated parameter includes the attribute name and the new attribute value, and for a "relationship addition or deletion" operation, the associated parameter includes the relationship type and the associated another node; the knowledge graph update strategy refers to a predefined and executable program logic or function, each knowledge graph update strategy is bound to at least one operation type, responsible for receiving the target node and the associated parameter, and specifically performing the addition, deletion, modification and the like of the knowledge graph data.

[0126] S42: Obtain the event timestamp and the validity time based on the adjustment event information, and determine the operation effective timing based on the event timestamp and the validity time, establish a time window-based incremental update sequence;

[0127] In the embodiment, the event timestamp refers to the actual time point of the change, which can be implemented by using a system log recording mechanism or a version control system timestamp, and the purpose is to identify the change time; the validity time refers to the effective period of the rule, which can be implemented by using a time interval representation method or a version validity period definition, and the purpose is to define the rule application period; the operation effective timing refers to an ordered sequence formed by sorting a plurality of planning adjustment events according to the order of their validity times, and the operation effective timing ensures that the state evolution of the knowledge graph correctly reflects the actual effective order of the regulations; the time window refers to a continuous time period defined by the time range information in the account range information, and the incremental update sequence is processed and takes effect within the time window; the incremental update sequence refers to an update operation sequence arranged in time sequence, which can be implemented by using a priority queue sorting algorithm or a timeline scheduling mechanism, and the purpose is to ensure that the update operation is strictly executed according to the real historical context.

[0128] S43: Execute the knowledge graph update strategy according to the incremental update sequence, and generate a compliant knowledge graph.

[0129] Specifically, the scheme is used for analyzing and adjusting event information to extract an operation type, a target node and associated parameters, and matching a corresponding knowledge graph updating strategy according to the operation type, so as to ensure that the updating logic is consistent with the event semantics; then, the operation effective time sequence is determined by using an event timestamp and a validity time, an incremental updating sequence based on a time window is established, and the updating sequence disorder problem caused by the fuzzy event time attribute is solved; finally, the knowledge graph updating strategy is executed according to the incremental updating sequence, changes are accumulated gradually, the operation in each time window is accurately applied, and finally, a compliance knowledge graph is generated, which completely retains the dynamic trajectory of the project compliance state, and provides a reliable data basis for the time sequence deduction of the account report.

[0130] Through the above scheme, the time attribute and the operation type of the event can be accurately processed, the updating sequence disorder can be effectively avoided, the compliance knowledge graph can accurately reflect the compliance state evolution of the project at different time points, and the reliability of the time sequence deduction of the account report is improved.

[0131] In an embodiment, step S41 comprises:

[0132] S411: extracting an operation type based on the adjustment event information, and matching a preset atomization strategy function according to the operation type, the operation type comprising node attribute updating, relationship topology adjustment and node state migration;

[0133] In this embodiment, the atomization strategy function refers to an independent and smallest executable code unit predefined for a specific operation type, each atomization strategy function only performs a specific type of indivisible graph modification task, for example, a function specially used for modifying a certain attribute of a node, or a function specially used for establishing a certain relationship edge between two points; the operation type comprises node attribute updating, relationship topology adjustment and node state migration, which can be understood as different dimension change operation categories in the knowledge graph, and the purpose is to classify and process the differentiated needs of heterogeneous operation types; wherein, the node attribute updating refers to modifying the attribute value of an existing node in the knowledge graph, for example, modifying the “numerical upper limit” attribute of a certain “planning indicator” node from 2.0 to 2.5; the relationship topology adjustment refers to modifying the connection relationship between nodes in the knowledge graph, including establishing a new relationship edge between two nodes that have no relationship before, deleting an existing relationship edge, or modifying the attribute of a certain relationship edge; the node state migration refers to changing the life cycle state identifier of a certain node in the knowledge graph, for example, changing the state of a “planning indicator” node from “effective” to “abandoned”, or from “draft” to “official”.

[0134] S412: based on the operation type, the unique identifier of the target node and the associated parameters are parsed from the adjustment event information;

[0135] In this embodiment, the unique identifier of the target node refers to a string or code that can uniquely and explicitly locate a specific node in the knowledge graph, which can be implemented by UUID or hash value, and the purpose is to realize the accurate specification of the operation object; the association parameter refers to the context constraint information related to the operation, which can include attribute values, relationship types and other data, and the purpose is to provide necessary parameter support for policy execution.

[0136] S413: Bind the unique identifier of the target node and the association parameter to the matched atomic policy function to generate a knowledge graph update policy.

[0137] In this embodiment, the binding is the process of associating the unique identifier of the target node and the association parameter with the matched atomic policy function.

[0138] Specifically, the scheme determines the semantic category of the change by extracting the operation type based on the adjustment event information, and accordingly matches the highly corresponding or customized atomic policy function, ensuring that the policy strictly corresponds to the operation type; then, based on the operation type, the unique identifier of the target node and the association parameter are parsed, the unique identifier provides the precise addressing ability of the node, and the association parameter carries the context information required by the operation; finally, the unique identifier and the association parameter are bound to the matched atomic policy function, the instantiation of the policy is realized, the update logic accurately acts on the specific node, thereby maintaining the structural consistency and semantic accuracy in the knowledge graph incremental update process.

[0139] Through the above scheme, the present application effectively solves the problem of insufficient accuracy caused by coarse-grained policy matching in knowledge graph incremental update, improves the pertinence and reliability of the update logic, reduces the occurrence of update errors caused by policy mismatch, and improves the execution efficiency.

[0140] In an embodiment, step S50 includes:

[0141] S51: Discretize the time range information in the account range information into a plurality of sequential time slices;

[0142] In this embodiment, the sequential time slice is a series of continuous and non-overlapping time periods divided by fixed time intervals (such as every day, every month, every year) or key event points, and each time period is an independent time slice for independent compliance analysis.

[0143] S52: Based on the event timestamp and validity time recorded in the compliance knowledge graph, reconstruct the graph subset corresponding to each sequential time slice, and use it as the compliance judgment benchmark for the sequential time slice, wherein the graph subset includes planning indicator nodes and their associated original source information;

[0144] In the present embodiment, the atlas subset refers to a set of nodes and edges that are filtered from the complete compliance knowledge graph at a certain specific time point or time period, which represents all effective compliance rules within the time slice; the compliance judgment benchmark refers to a set of effective planning indicators for evaluating project compliance at a specific time point, which can be dynamically filtered from the effective provisions in the historical version based on the event timestamp and the validity time, with the purpose of ensuring that the compliance judgment is strictly based on the rules that are effective at that time, and avoiding confusion with historical versions; the planning indicator node is an entity node in the atlas subset that represents specific planning constraint requirements, and usually, the planning indicator node is the direct standard and object of compliance comparison; the original source information is metadata associated with the planning indicator node for explaining the source of the indicator, which, in the present embodiment, associates the structured indicators in the atlas with the original documents that are effective, ensuring traceability.

[0145] S53: Extract the attributes and relationships of the project entity, compare the extracted attributes and relationships with the compliance judgment benchmark, calculate the compliance deviation degree of the project entity with each planning indicator node, and obtain the project entity compliance state;

[0146] In the present embodiment, the attributes of the project entity refer to numerical or textual data used to describe the characteristics and state of the project entity itself, and usually, the attributes are the intrinsic characteristics of the project entity and the specific content that needs to be checked in the compliance comparison, for example: for a project entity representing "building", its attributes may include building area, building height, and use function; for a project entity representing "land parcel", its attributes may include land use area, land use nature, and plot ratio; the relationship of the project entity refers to the spatial, functional, or logical interrelation between different project entities or between the project entity and external references, and usually, the relationship is also the object of compliance check, for example: spatial relationship: such as "building A" is located in "land parcel B", "parking lot" is adjacent to "main building", etc.; functional relationship: such as "pipe X" is connected to "processing facility Y", etc.; composition relationship: such as "Building 1" and "Building 2" belong to "XX project", etc.; the compliance deviation degree refers to the quantitative difference between the attributes of the project entity and the requirements of the planning indicator node, which can be achieved by distance measurement, similarity calculation, or domain-specific scoring function, with the purpose of objectively evaluating the compliance degree and reducing human subjective errors; the project entity compliance state refers to the comprehensive judgment result of whether the project as a whole complies with all relevant planning indicators within a certain specific time slice, and the project entity compliance state is usually a classification result, such as "completely compliant", "partially compliant", or "non-compliant", etc.

[0147] S54: Set the planning indicator node with an absolute value of compliance deviation degree greater than a preset deviation threshold as a key indicator node;

[0148] In the embodiment, the deviation threshold is a preset numerical critical point, and when the absolute value of the compliance deviation of the project entity with respect to a certain planning indicator node exceeds the deviation threshold, the planning indicator node is determined as a key indicator node; the key indicator node refers to a planning indicator node whose calculated compliance deviation exceeds the preset allowable range in the compliance comparison process.

[0149] S55: generating a key evidence link based on the original source information associated with the key indicator node;

[0150] In the embodiment, the key evidence link is a directly accessible pointer or identifier to the original source information associated with the key indicator node, which can be implemented in the form of a hyperlink, a document reference identifier, or a blockchain storage technology, etc., aiming to provide verifiable evidence support and enhance the credibility of compliance determination.

[0151] S56: concatenating all the project entity compliance states of the sequential time slices in a preset time sequence, and appending a time stamp and a key evidence link to each project entity compliance state, to generate a compliance state time sequence chain.

[0152] In the embodiment, the time stamp is a group of characters or coded data that identifies the time of occurrence of an event, which usually records the date and time, and can optionally include more precise time zone information; the compliance state time sequence chain is a data structure arranged in time sequence, in which each element records the project entity compliance state in the corresponding sequential time slice, the valid time stamp of the state, and the key evidence link leading to the state, which describes the complete history of the compliance state change over time.

[0153] Specifically, the scheme divides the continuous time axis into discrete time slices to provide a structured framework for time sequence reasoning; then, using the event time stamps and validity times recorded in the compliance knowledge graph, dynamically reconstructs the effective graph subset corresponding to each time slice as a compliance determination benchmark, ensuring the timeliness and authority of the benchmark; on this basis, extracts the attributes and relationships of the project entity and performs quantitative comparison with the determination benchmark to calculate the compliance deviation and evaluate the compliance state; further, focuses on key indicator nodes with significant deviation through a threshold screening mechanism, and generates key evidence links by associating original source information; finally, concatenates the compliance states of each slice in time sequence and appends time stamps and evidence links to form a complete compliance state time sequence chain; the above process realizes the accurate capture and systematic presentation of the time sequence evolution of project compliance state through the organic combination of time discretization, dynamic benchmark reconstruction, quantitative deviation evaluation, key problem focusing, and evidence association.

[0154] For example, in a certain urban new district development project, the account scope information specifies the time range from January 2020 to December 2023, and the time range is discretized into quarterly time slices; for each quarter, based on the event timestamps (such as the time of planning adjustment events) and the validity time in the compliance knowledge graph, the planning index subset that takes effect in the quarter is reconstructed, for example, the subset of the second quarter of 2021 contains the volume rate requirement and land property requirements that are effective at that time; the building area and land type attributes of the project entity (such as plot A) are extracted and compared with the planning index subset, and the deviation degree is calculated; when the volume rate deviation degree exceeds the threshold, the index node is set as a key index node, and an evidence link pointing to the original planning approval file is generated; finally, the compliance status of each quarter is concatenated in chronological order to form a compliance status time sequence chain containing timestamps and evidence links, which intuitively shows the dynamic changes of project compliance.

[0155] Through the above scheme, the application can accurately capture the compliance state changes of the project at different time nodes, reduce the omission or misjudgment of compliance determination, and systematically present the dynamic deduction result of the compliance state, thereby improving the accuracy and traceability of the project account generation.

[0156] In an embodiment, step S53 comprises:

[0157] S531: identifying the constraint type of the planning index node, the constraint type comprising a numerical interval constraint, an enumerated option constraint, a spatial relationship constraint, and a logical dependency constraint;

[0158] In the embodiment, the constraint type is a classification of the compliance requirements specified by the planning index node, which can be implemented by numerical interval definition, preset option list, geometric topological relationship or conditional logic expression, and the purpose is to provide a classification basis for deviation degree calculation to avoid the blindness of general calculation methods when dealing with heterogeneous constraints. The constraint type includes a numerical interval constraint, an enumerated option constraint, a spatial relationship constraint, and a logical dependency constraint. The numerical interval constraint specifies the effective range of numerical values, for example, the "volume rate" index requires its value to be between [1.0, 2.5]. The enumerated option constraint specifies a finite and discrete set of optional values, for example, the "building style" index requires its value to be one of ["Chinese style", "modern"]. The spatial relationship constraint specifies the specific topological or geometric relationship that the project entity must satisfy in space, for example, "building" must be located within the "construction land boundary". The compliance of the logical dependency constraint depends on the satisfaction state of one or more conditions, which is manifested as a logical rule limitation, for example, "if the land property is 'education', a'social parking lot' must be provided".

[0159] S532: Based on the constraint type, match and call the preset corresponding deviation calculation function to obtain the compliance deviation of the project entity and each planning index node;

[0160] In the embodiment, the deviation calculation function is a predefined algorithm for quantifying the difference between the actual state and the standard requirement, wherein the deviation calculation function can be understood as a quantitative evaluation algorithm designed for a specific constraint type, and can be implemented by using an interval difference quantification algorithm, an option matching verification mechanism, a geometric relationship evaluation model or a condition satisfaction degree analyzer. The purpose is to ensure that the calculation process accurately adapts to the essential characteristics of the constraint, so as to generate a deviation result that conforms to the actual constraint logic.

[0161] S533: Based on the preset deviation processing strategy, aggregate the compliance deviation of the project entity and each planning index node to obtain the project entity compliance state.

[0162] In the embodiment, the deviation processing strategy is a rule for defining how to aggregate the compliance deviations calculated on multiple planning index nodes into the overall project entity compliance state, such as a weighted comprehensive score strategy, a hierarchical counting strategy, etc. Aggregation is a data processing operation, which refers to the process of combining multiple compliance deviation values into a comprehensive judgment result (i.e. project entity compliance state) according to the preset deviation processing strategy.

[0163] Specifically, the present scheme dynamically associates the constraint type with a specially designed calculation function by identifying the constraint type of the planning index node, forming a type-driven calculation path selection mechanism. First, the constraint type of the planning index node is classified to determine whether it belongs to one of the numerical interval constraint, the enumerated option constraint, the spatial relationship constraint or the logical dependency constraint. Second, based on the classification result, the preset deviation calculation function is automatically matched, such as calling the interval difference quantification algorithm for numerical constraints and the option matching verification mechanism for enumerated constraints. Finally, the project entity attribute and the compliance judgment benchmark are compared by the called deviation calculation function, and the quantitative result reflecting the actual deviation degree is output. The above scheme ensures that the evaluation process of different types of constraints strictly follows their inherent logical characteristics, avoiding the misalignment problem of a single calculation model when processing multi-modal constraints.

[0164] Exemplarily, when processing the volume rate index in the urban planning project, it is identified that the index belongs to the numerical interval constraint, the interval difference quantization algorithm is automatically matched, the actual volume rate value of the project is compared with the interval range permitted by the planning, and the absolute deviation value exceeding the permitted range is calculated; when evaluating the land use property index, it is identified that it is an enumerated option constraint, the option matching verification mechanism is called, it is checked whether the project land use property is in the preset option list specified in the planning document, and if there is a mismatch, a deviation identifier is generated; for the building setback distance index, after identifying the spatial relationship constraint, the spatial topological relationship between the building contour and the planning boundary is analyzed by using the geometric relationship evaluation model, and the difference between the actual setback distance and the specified value is quantized; for the time sequence dependent index involving staged development, it is identified as a logic dependent constraint, and whether each development stage satisfies the precondition is verified by using the condition satisfaction degree analyzer, and the evaluation result of the condition satisfaction degree is output.

[0165] Through the above technical solutions, the application can use corresponding calculation methods for different constraint types, ensure that the compliance deviation calculation result accurately reflects the true compliance state of the project entity, effectively solve the problem of inaccurate calculation caused by the diversity of constraint types, and thus provide reliable compliance judgment basis for the project account report.

[0166] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0167] In an embodiment, an AI-based project account intelligent generation system is provided, which corresponds one-to-one to the AI-based project account intelligent generation method in the above embodiment. The AI-based project account intelligent generation system comprises:

[0168] A first information extraction module is configured to extract a target project identifier and account range information in response to an account generation instruction issued by a user terminal, wherein the account range information includes time range information, spatial range information, and content depth level;

[0169] A second information extraction module is configured to obtain a project data set based on the target project identifier and the account range information, and extract a project entity and a planning index set based on the project data set, wherein the project data set includes project basic information and planning document information;

[0170] A graph construction module is configured to construct an initial knowledge graph according to the project entity and the planning index set, and extract planning adjustment event information based on the planning document information;

[0171] The atlas updating module is configured to perform incremental updating on the initial knowledge graph based on the planning adjustment event information, and generate a compliance knowledge graph;

[0172] The time sequence reasoning module is configured to perform time sequence reasoning on the compliance knowledge graph based on the account range information, deduce and generate a compliance state time sequence chain;

[0173] The report generation module is configured to obtain a structured account template, and fill the structured account template according to the compliance state time sequence chain and the compliance knowledge graph, and generate a project account report.

[0174] For specific limitations of the AI-based project account intelligent generation system, refer to the limitations of the AI-based project account intelligent generation method described above, which will not be repeated here. Each module in the AI-based project account intelligent generation system described above can be realized by software, hardware, or a combination thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0175] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An AI-based project table account intelligent generation method, characterized in that, The method comprises the steps of: in response to a user terminal issuing an account generation instruction, extracting a target project identifier and account range information, wherein the account range information comprises time range information, space range information, and content depth level; based on the target project identifier and the account range information, obtaining a project data set, and based on the project data set, extracting a project entity and a planning index set, wherein the project data set comprises project basic information and planning document information; constructing an initial knowledge graph according to the project entity and the planning index set, and extracting planning adjustment event information based on the planning document information; based on the planning adjustment event information, incrementally updating the initial knowledge graph to generate a compliance knowledge graph; based on the account range information, performing time sequence reasoning on the compliance knowledge graph to deduce and generate a compliance state time sequence chain; obtaining a structured account template, and filling the structured account template according to the compliance state time sequence chain and the compliance knowledge graph to generate a project account report; wherein the step of based on the account range information, performing time sequence reasoning on the compliance knowledge graph to deduce and generate a compliance state time sequence chain comprises the steps of: discretizing the time range information in the account range information into a plurality of sequential time slices; based on the event timestamps and validity times recorded in the compliance knowledge graph, reconstructing a graph subset corresponding to each sequential time slice, and taking the graph subset as a compliance judgment benchmark for the sequential time slice, wherein the graph subset comprises planning index nodes and their associated original source information; extracting attributes and relationships of the project entity, comparing the extracted attributes and relationships with the compliance judgment benchmark, calculating compliance deviation degrees of the project entity and each planning index node, and obtaining a project entity compliance state; setting a planning index node whose absolute value of the compliance deviation degree is greater than a preset deviation threshold as a key index node in the comparison process; based on the original source information associated with the key index node, generating a key evidence link; in a preset time sequence, concatenating project entity compliance states of all sequential time slices, and appending a timestamp and a key evidence link to each project entity compliance state to generate a compliance state time sequence chain.

2. The AI-based project table account intelligent generation method according to claim 1, characterized in that: The step of based on the target project identifier and the account range information, obtaining a project data set, and based on the project data set, extracting a project entity and a planning index set, wherein the project data set comprises project basic information and planning document information, comprises the steps of: inputting the project data set into a pre-trained multi-channel extraction model to obtain a planning index set, a spatial element information set, and a dynamic event information set; inputting the planning index set, the spatial element information set, and the dynamic event information set into a pre-trained multi-modal fusion model to obtain a fusion representation vector; performing sequence labeling processing on the fusion representation vector through a preconfigured entity recognition head to obtain a project entity.

3. The AI-based project table account intelligent generation method according to claim 2, characterized in that: The multi-channel extraction model comprises a natural language processing channel, a spatial analysis channel, and a time sequence event channel, and the step of inputting the project data set into a pre-trained multi-channel extraction model to obtain a planning index set, a spatial element information set, and a dynamic event information set comprises the steps of: The natural language processing channel identifies structured text and unstructured text in the project dataset, extracts constraint clause text, and instantiates it as a set of planning indicators; The spatial analysis channel identifies planning drawing information and GIS data in the project dataset, extracts geometric features, topological relationships, and attribute information of spatial elements, and generates a set of spatial element information; The time sequence event channel identifies planning version history information and revision record information in the project dataset, extracts version change information, strategy adjustment information, and validity state change information, and generates a set of dynamic event information.

4. The AI-based project table account intelligent generation method according to claim 2, characterized in that: The step of performing sequence labeling on the fusion feature vector by the pre-configured entity recognition head to obtain the project entity includes the steps of: serializing and decoding the fusion feature vector to generate a text sequence and context-aware features of basic language units in the text sequence; calculating a starting probability distribution of each basic language unit in the text sequence as a starting boundary of an entity and an ending probability distribution as an ending boundary of the entity; generating a set of candidate entity boundary positions based on the starting probability distribution and the ending probability distribution, and generating a candidate entity semantic feature vector based on the set of candidate entity boundary positions and the context-aware features; performing global sequence labeling on the candidate entity semantic feature vector to generate an entity label sequence, and generating a project entity based on the set of candidate entity boundary positions and the entity label sequence.

5. The AI-based project table account intelligent generation method according to claim 2, characterized in that: The step of constructing an initial knowledge graph based on the project entity and the set of planning indicators, and extracting planning adjustment event information based on the planning document information includes the steps of: performing semantic relation classification on the fusion feature vector by the pre-configured relation extraction head to obtain semantic relations between the project entity and the set of planning indicators, the semantic relations including spatial constraint relations, attribute association relations, and time sequence influence relations; constructing an initial knowledge graph including entity nodes and multi-dimensional semantic relation edges based on the project entity, the set of planning indicators, and the semantic relations, the entity nodes including project nodes and planning indicator nodes; monitoring the version flow of the planning document information, and identifying structured difference information between different versions in the version flow using a pre-configured time difference algorithm; mapping the structured difference information to graph structure evolution operations on the entity nodes, the graph structure evolution operations including attribute updates, relation additions and deletions, and node state changes; appending event timestamps, validity times, and change basis information to the graph structure evolution operations to generate planning adjustment event information.

6. The AI-based project table account intelligent generation method according to claim 1, characterized in that: The step of incrementally updating the initial knowledge graph based on the planning adjustment event information to generate a compliance knowledge graph includes the steps of: extracting operation types, target nodes, and associated parameters based on the adjustment event information, and matching a pre-configured knowledge graph update strategy based on the operation type; obtaining event timestamps and validity times based on the adjustment event information, and determining operation effective timing based on the event timestamps and the validity times to establish an incremental update sequence based on a time window; executing the knowledge graph update strategy according to the incremental update sequence to generate a compliance knowledge graph.

7. The AI-based project table account intelligent generation method according to claim 6, characterized in that: The step of extracting the operation type, the target node, and the associated parameter based on the adjustment event information and matching the preset corresponding knowledge graph update strategy based on the operation type comprises the steps of: Based on the operation type, the atomic strategy function is matched, and the operation type includes node attribute update, relationship topology adjustment, and node state migration. Based on the operation type, the unique identifier of the target node and the associated parameter are parsed from the adjustment event information; The unique identifier of the target node and the associated parameter are bound to the matched atomic strategy function to generate a knowledge graph update strategy.

8. The AI-based project table account intelligent generation method according to claim 1, characterized in that: The step of extracting the attribute and relationship of the project entity, comparing the extracted attribute and relationship with the compliance judgment benchmark, calculating the compliance deviation of the project entity and each planning index node, and obtaining the compliance state of the project entity comprises the steps of: Identify the constraint type of the planning index node, and the constraint type includes numerical interval constraint, enumeration option constraint, spatial relationship constraint, and logical dependency constraint. Based on the constraint type, the preset corresponding deviation calculation function is matched and called to obtain the compliance deviation of the project entity and each planning index node. Based on the preset deviation processing strategy, the compliance deviation of the project entity and each planning index node is aggregated to obtain the compliance state of the project entity.

9. An AI-based project account intelligent generation system for performing the steps of an AI-based project account intelligent generation method according to any one of claims 1-8. It includes: The first information extraction module is used to extract the target project identifier and the account range information in response to the account generation instruction issued by the user end, and the account range information includes time range information, space range information, and content depth level. The second information extraction module is used to obtain the project data set based on the target project identifier and the account range information, and extract the project entity and the planning index set based on the project data set, wherein the project data set includes project basic information and planning document information. The graph construction module is used to construct an initial knowledge graph according to the project entity and the planning index set, and extract planning adjustment event information based on the planning document information. The graph update module is used to incrementally update the initial knowledge graph based on the planning adjustment event information to generate a compliance knowledge graph. The time sequence reasoning module is used to perform time sequence reasoning on the compliance knowledge graph based on the account range information, and deduce and generate a compliance state time sequence chain. The report generation module is used to obtain a structured account template, and fill the structured account template according to the compliance state time sequence chain and the compliance knowledge graph to generate a project account report.

Images