A method for dynamic evaluation and multi-dimensional operation monitoring of land parcel full life cycle compliance

Abstract

This invention relates to the field of land resource management technology, and in particular to a method for dynamic assessment of compliance and multi-dimensional operational monitoring of land parcels throughout their entire lifecycle. The method involves acquiring multi-source heterogeneous data throughout the entire lifecycle of land parcels, constructing an initial land parcel state map through spatiotemporal benchmark alignment and entity relationship extraction; outputting dynamic compliance state vectors and compliance risk gradients through differentiable rule embedding and a spatiotemporal graph attention network; calculating the time-series sequence of operational dimension deviations based on cross-modal feature alignment to identify abnormal operational patterns; constructing a bidirectional coupled optimization module for compliance operations, establishing a joint optimization model, and iteratively solving to generate an adaptive control instruction set; and fusing execution feedback with compliance risk gradient residuals to trigger dynamic updates of node features and edge weights in the land parcel state map, completing a closed-loop iteration of the state. This invention achieves integrated management of the entire lifecycle of land parcels, provides early warning of compliance risks, balances compliance bottom lines with operational efficiency, and improves the intelligence level and comprehensive utilization efficiency of land resource management.

Description

Technical Field

[0001] This invention relates to the field of land resource management technology, specifically a method for dynamic assessment of compliance and multi-dimensional operational monitoring of land parcels throughout their entire life cycle. Background Technology

[0002] Existing land management systems generally suffer from severe data fragmentation. Data from different departments is stored in independent business systems, creating numerous information silos. Data formats vary widely, including structured database records, semi-structured document files, and unstructured text and images. The temporal and spatial benchmarks of the data are inconsistent, making it impossible to accurately correlate and match different data for the same plot. It is difficult to construct a unified data view covering the entire lifecycle of a plot, resulting in a lack of comprehensive and accurate data support for management decisions.

[0003] Traditional compliance assessments primarily rely on static, post-hoc checks. These checks depend on manual review of paper and electronic documents, resulting in low efficiency and susceptibility to human error. Inspections are only conducted at critical junctures such as planning verification and final acceptance, failing to achieve continuous, dynamic monitoring. Many violations are only discovered long after they occur, leading to high rectification costs and potentially irreversible environmental damage and economic losses. Furthermore, frequent updates to regulations and policies make it difficult for personnel to comprehensively and promptly grasp the latest compliance requirements, increasing the risk of compliance loopholes.

[0004] In the existing management model, compliance management and operational monitoring are disconnected. Compliance requirements cannot be effectively translated into actionable operational control indicators, and abnormal situations during operation cannot be promptly reported to the compliance assessment system. Land operators often prioritize economic benefits over compliance requirements, leading to frequent violations. Meanwhile, management departments, in an effort to mitigate compliance risks, often adopt blanket, strict control measures, severely impacting operational efficiency and economic benefits. Finding the optimal balance between compliance requirements and operational efficiency remains difficult.

[0005] Traditional land management lacks a closed-loop dynamic optimization mechanism. The management process exhibits a unidirectional linear characteristic, failing to form an effective feedback loop from approval to operation. The effectiveness of regulatory measures cannot be evaluated in a timely manner, nor can regulatory strategies be dynamically adjusted based on the results. The system cannot self-optimize based on changes in the operational status of land parcels and updates to laws and policies, leading to a disconnect between management measures and actual needs, and hindering the continuous improvement of management effectiveness.

[0006] The existing management system does not provide comprehensive coverage of the entire lifecycle of land parcels. Management focus is primarily on the planning approval and land transfer stages, with insufficient oversight of the construction process and long-term operation. Many problems arising during the operation phase cannot be detected and addressed in a timely manner. This not only affects the rational use of land resources but may also lead to safety accidents and environmental problems.

[0007] In summary, existing land management technologies have significant shortcomings in areas such as data integration, dynamic compliance assessment, collaborative compliance operations, closed-loop optimization, and full lifecycle coverage. There is an urgent need to develop a technological approach that enables integrated management of land parcels throughout their entire lifecycle, addressing the various problems inherent in traditional models and improving the intelligence and overall efficiency of land resource management. Summary of the Invention

[0008] The purpose of this invention is to provide a method for dynamic assessment of compliance and multi-dimensional operational monitoring of land parcels throughout their entire lifecycle, in order to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for dynamic assessment of land parcel compliance and multi-dimensional operational monitoring throughout its entire lifecycle includes the following steps:

[0011] S1: Obtain multi-source heterogeneous data of the land parcel at each stage of its entire life cycle, perform spatiotemporal benchmark alignment and entity relationship extraction on the multi-source heterogeneous data, and construct an initial land parcel state diagram. The initial land parcel state diagram includes land parcel entity nodes, life cycle stage nodes, compliance rule nodes and operation indicator nodes. The nodes are connected through temporal dependency edges and business logic edges.

[0012] S2: Input the initial land parcel state map into the compliance dynamic assessment model, map the discrete regulatory policy text into a continuous compliance constraint vector through a differentiable rule embedding algorithm, and aggregate compliance features along the temporal dependency edge based on the spatiotemporal graph attention network to output the dynamic compliance state vector and compliance risk gradient of the land parcel at the current stage.

[0013] S3: Based on dynamic compliance state vector and compliance risk gradient, drive multi-dimensional operation monitoring model, adopt cross-modal feature alignment mechanism to map compliance constraint vector to real-time operation monitoring data, calculate the deviation time series of each operation dimension, and generate operation anomaly pattern identifier through time series anomaly detection algorithm;

[0014] S4: Construct a compliance-operation bidirectional coupling optimization module. When the operational anomaly mode identifier meets the preset trigger conditions, the deviation time series is transformed into a constraint penalty term. Establish a joint optimization model with the dynamic compliance state vector as the hard constraint boundary and the multi-dimensional operational comprehensive efficiency as the objective function. Generate an adaptive control instruction set by iteratively solving the dynamic penalty coefficient.

[0015] S5: The execution feedback data of the adaptive control instruction set is residually fused with the compliance risk gradient to trigger the node feature reset and edge weight dynamic update of the initial land parcel state graph, thus completing the closed-loop iteration of the state for the full life cycle compliance assessment and operation monitoring of the land parcel.

[0016] As can be seen from the technical solution provided by the present invention above, the beneficial effects of the method for dynamic assessment of land parcel compliance and multi-dimensional operation monitoring throughout its entire life cycle provided by the present invention are:

[0017] This invention achieves spatiotemporal benchmark alignment and unified expression of entity relationships for multi-source heterogeneous data by constructing an initial land parcel state map; it integrates approval data, monitoring data and business data throughout the entire life cycle of land parcels into a unified graph computing foundation, eliminating data silos, ensuring data consistency and traceability, and providing comprehensive and accurate data support for subsequent dynamic evaluation and monitoring.

[0018] This invention employs a differentiable rule embedding algorithm and a spatiotemporal graph attention network to construct a dynamic compliance assessment model; it transforms discrete regulatory and policy texts into continuous compliance constraint vectors, realizing the quantitative expression and dynamic updating of compliance rules; by aggregating compliance features along temporal dependency edges, it can capture the evolution trend of compliance status, identify potential compliance risks in advance, change the passive situation of traditional ex-post compliance inspections, and realize forward-looking early warning of compliance risks.

[0019] This invention proposes a cross-modal feature alignment mechanism and an adaptive temporal anomaly detection algorithm; it establishes a semantic mapping relationship between compliance constraints and operational data, realizing real-time monitoring of multi-dimensional operational status; and it dynamically adjusts the detection threshold sensitivity based on the compliance risk gradient, which can accurately identify various abnormal operational modes, effectively reducing false alarm rate and false negative rate, and improving the accuracy and timeliness of operational monitoring.

[0020] This invention constructs a compliance-operation bidirectional coupling optimization module; establishes a closed-loop channel for forward mapping and reverse correction, and performs joint optimization with the goal of maximizing multi-dimensional operational efficiency while ensuring compliance bottom line; and generates an adaptive control instruction set by iteratively solving dynamic penalty coefficients, thereby achieving a dynamic balance between compliance constraints and operational benefits, avoiding operational efficiency losses caused by excessive compliance, and preventing behaviors that exceed compliance boundaries in pursuit of benefits.

[0021] This invention establishes a closed-loop iterative update mechanism for the state; it integrates the feedback data of regulation execution with the compliance risk gradient through residual fusion, and automatically updates the node characteristics and edge weights of the land parcel state diagram; the system can continuously optimize itself according to the actual operation, adapt to changes in laws and policies and the dynamic adjustment of the land parcel operation status, and maintain long-term effectiveness and adaptability.

[0022] This invention covers the entire lifecycle of a land parcel, from planning permit to operational termination, enabling dynamic compliance assessment and multi-dimensional operational monitoring throughout the process. It can provide scientific decision-making basis for natural resource management departments, provide precise operational guidance for land parcel operators, effectively reduce labor costs and compliance risks in land management, and improve the utilization efficiency and comprehensive value of land resources. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the steps of a method for dynamic assessment of compliance and multi-dimensional operation monitoring of land parcels throughout their entire life cycle, as described in this invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0026] like Figure 1 As shown, this embodiment of the invention provides a method for dynamic assessment of land parcel compliance and multi-dimensional operational monitoring throughout its entire lifecycle, including the following steps:

[0027] S1: Obtain multi-source heterogeneous data of the land parcel at each stage of its entire life cycle, perform spatiotemporal benchmark alignment and entity relationship extraction on the multi-source heterogeneous data, and construct an initial land parcel state diagram. The initial land parcel state diagram includes land parcel entity nodes, life cycle stage nodes, compliance rule nodes and operation indicator nodes. The nodes are connected through temporal dependency edges and business logic edges.

[0028] S2: Input the initial land parcel state map into the compliance dynamic assessment model, map the discrete regulatory policy text into a continuous compliance constraint vector through a differentiable rule embedding algorithm, and aggregate compliance features along the temporal dependency edge based on the spatiotemporal graph attention network to output the dynamic compliance state vector and compliance risk gradient of the land parcel at the current stage.

[0029] S3: Based on dynamic compliance state vector and compliance risk gradient, drive multi-dimensional operation monitoring model, adopt cross-modal feature alignment mechanism to map compliance constraint vector to real-time operation monitoring data, calculate the deviation time series of each operation dimension, and generate operation anomaly pattern identifier through time series anomaly detection algorithm;

[0030] S4: Construct a compliance-operation bidirectional coupling optimization module. When the operational anomaly mode identifier meets the preset trigger conditions, the deviation time series is transformed into a constraint penalty term. Establish a joint optimization model with the dynamic compliance state vector as the hard constraint boundary and the multi-dimensional operational comprehensive efficiency as the objective function. Generate an adaptive control instruction set by iteratively solving the dynamic penalty coefficient.

[0031] S5: The execution feedback data of the adaptive control instruction set is residually fused with the compliance risk gradient to trigger the node feature reset and edge weight dynamic update of the initial land parcel state graph, thus completing the closed-loop iteration of the state for the full life cycle compliance assessment and operation monitoring of the land parcel.

[0032] In this embodiment, the core function of step S1 is to comprehensively collect multi-dimensional raw data from all stages of the land parcel's entire lifecycle, complete data cleaning and verification and unify spatiotemporal benchmarks, extract core entities and relationships through semantic parsing, and construct an initial land parcel status map containing all elements and relationships of the land parcel, providing a unified graph computing data foundation for subsequent compliance dynamic assessment and operational monitoring; the detailed steps are as follows:

[0033] Step S1-1: Collection and classification of multi-source heterogeneous data throughout the entire life cycle of the land parcel:

[0034] Establish a data collection interface system covering the entire life cycle of land parcels, and connect with public service information platforms, land parcel operators' IoT platforms and business management systems; divide the collection scope according to the life cycle stages of land parcels, and sequentially obtain all original data from the planning permit stage, land transfer stage, engineering construction stage, completion and acceptance stage, and long-term operation stage;

[0035] Data collected during the planning permit stage includes the land parcel's detailed control planning documents, land use pre-approval opinions, construction land planning permits, and attached drawings; data collected during the land transfer stage includes the announcement of the transfer of state-owned construction land use rights, the transfer contract, land transfer fee payment vouchers, and the coordinates of the land parcel boundary points; data collected during the construction stage includes the construction project planning permit, construction permit, construction drawing design document review approval certificate, and monthly project progress report; data collected during the completion and acceptance stage includes the construction project completion and acceptance filing form, planning verification opinion, fire protection acceptance opinion, and real estate ownership certificate; data collected during the operation stage includes real-time IoT sensor data, daily inspection records, equipment operation logs, energy consumption statistics, and environmental monitoring data;

[0036] All collected raw data were classified into three levels according to data structure type. The first level was divided into structured data, semi-structured data, and unstructured data. The second level further subdivided structured data into relational database records, spreadsheet data, and time-series data; semi-structured data into XML documents, JSON files, and PDF forms; and unstructured data into text documents, scanned images, and audio / video files. The third level was further subdivided according to the data source department and business type, forming a multi-level, multi-source, heterogeneous dataset. The classified datasets were then linked and stored according to the unique identifier of each land parcel, and a data index table was established to record the source, collection time, data format, and storage path of each data entry.

[0037] Step S1-2: Preprocessing and quality verification of multi-source heterogeneous data:

[0038] Standardized preprocessing operations are performed on various types of data in multi-source heterogeneous datasets; for structured data, missing value imputation, outlier removal, and data format standardization are performed; missing value imputation uses the mean or median of data of the same type in the same plot; outlier removal adopts the 3σ principle, and data exceeding three times the standard deviation of the mean are marked as outliers and deleted; data format standardization includes standardizing the number of decimal places retained for numerical data, standardizing the format of date data, and standardizing the capitalization of text data;

[0039] For semi-structured data, a structured parsing algorithm is used to extract core field information; for XML documents and JSON files, the content of specified fields is extracted by node traversal; for PDF forms, form field values ​​are extracted by form template matching; the extracted core fields are converted into structured data format and merged with the original structured data.

[0040] For unstructured data, optical character recognition technology is used to convert scanned images into editable text; speech recognition technology is used to convert audio and video files into text content; the converted text content is then processed by word segmentation and noise reduction to remove meaningless stop words and special characters.

[0041] Perform a comprehensive quality check on the preprocessed data. The check includes data integrity, data consistency, and data accuracy. Data integrity check examines whether required fields are missing. Data consistency check examines whether the attribute values ​​of the same entity are consistent across different data sources. Data accuracy check examines whether the data conforms to business logic rules and numerical range requirements. Data that fails the quality check is marked as problematic data, a data quality report is generated, and feedback is sent to the data provider for correction. The corrected data is then reprocessed and the preprocessing and quality check process is repeated until all data passes the check.

[0042] Step S1-3: Spatiotemporal reference alignment of multi-source heterogeneous data:

[0043] Perform spatiotemporal benchmark alignment on multi-source heterogeneous data that has passed quality verification; for time benchmark alignment, extract the timestamp information carried by each data; convert time strings of different formats into Unix timestamp format; for data with missing timestamps, infer and fill them according to the time range of the business stage to which they belong; for data with multiple timestamps, use the business occurrence time as the main timestamp; unify the timestamps of all data to UTC+8 time zone to ensure the comparability of time dimensions;

[0044] For spatial reference alignment, the spatial coordinate information carried by each data point is extracted; spatial coordinates in different coordinate systems are uniformly converted to latitude and longitude coordinates in the WGS84 coordinate system; the coordinate transformation adopts a seven-parameter transformation model, and the transformation formula is as follows: ,in, , , These are the spatial rectangular coordinates in the original coordinate system; , , These are spatial rectangular coordinates in the WGS84 coordinate system. , , These are the translation parameters in the three coordinate axes; For scale parameters; , , These are the rotation parameters along the three coordinate axes;

[0045] For data containing only the coordinates of land parcel boundary points, the center point coordinates and boundary polygons of the land parcels are calculated using the boundary point coordinates; for data containing only address information, the addresses are converted into latitude and longitude coordinates using a geocoding service; all spatial data are unified into GeoJSON format to ensure the consistency and computability of spatial location.

[0046] The data that have been aligned to both temporal and spatial references are integrated to generate a spatiotemporally aligned dataset. Each data entry in the spatiotemporally aligned dataset contains a unified timestamp and spatial coordinate information, providing standardized data input for subsequent entity relation extraction.

[0047] Steps S1-4: Spatiotemporal aligned data semantic parsing and entity recognition:

[0048] Deep semantic parsing is performed on each data record in the spatiotemporal aligned dataset. An entity recognition algorithm based on a pre-trained language model is used to automatically extract five core entity categories from the data records: The first category is land parcel entities, with extracted features including land parcel spatial code, parcel number, land parcel name, and spatial coordinate range; the second category is stage entities, with extracted features including stage name, stage number, stage start and end time, and stage status; the third category is rule entities, with extracted features including regulation clause number, regulation name, issuing department, and effective date; the fourth category is indicator entities, with extracted features including indicator name, indicator unit, indicator value, and indicator threshold; and the fifth category is entity entities, with extracted features including the name of the construction unit, the construction contractor, the supervision unit, and the operation unit.

[0049] The entity recognition process adopts a model architecture combining a bidirectional long short-term memory network and a conditional random field. First, the text data is converted into word vector representations. Then, contextual features are extracted through the bidirectional long short-term memory network. Finally, sequence labeling is performed through the conditional random field layer to identify the boundaries and types of entities. The identified entities are deduplicated. When the same entity appears multiple times in different data sources, its attribute information is merged, and the most complete and accurate attribute values ​​are retained.

[0050] All extracted entities are standardized and coded; each entity is assigned a globally unique identifier; a mapping relationship between entity codes and original data records is established, recording the source data and extraction time of each entity; the standardized entity information is stored in an entity database to provide an entity basis for subsequent relationship extraction.

[0051] Step S1-5: Extraction and verification of relationships between entities:

[0052] Based on entity recognition results, a relation extraction algorithm is used to automatically extract the relationships between entities. These relationships are categorized into three core types: the first type is temporal relationships, existing between land parcel entities and stage entities, indicating that the land parcel is at different lifecycle stages at different times; the second type is constraint relationships, existing between stage entities and rule entities, indicating the legal and policy requirements that must be followed at a specific lifecycle stage; the third type is monitoring relationships, existing between land parcel entities and indicator entities, indicating that specific indicators are used to monitor the operational status of the land parcel; and the fourth type is subject relationships, existing between land parcel entities and subject entities, indicating the ownership or responsibility relationship between the subject and the land parcel.

[0053] Relation extraction employs an attention-based relation classification model. First, entity pairs and their context text are converted into vector representations. Then, the semantic association strength between entity pairs is calculated using an attention mechanism. Finally, a fully connected layer is used to classify the relationships and determine the specific relationship type between entity pairs. The extracted relationships are evaluated for confidence, and relationships with confidence scores below a preset threshold are marked as relationships to be verified.

[0054] The relationships to be verified are manually reviewed and verified; the reviewers judge the correctness of the relationships to be verified based on business knowledge and original data records; erroneous relationships are deleted; incomplete relationships are supplemented with relevant attribute information; missing relationships are manually added; verified relationships are stored in the relationship database, and an association index of entities and relationships is established.

[0055] Step S1-6: Generation of land parcel state map nodes and assignment of attribute standardization values:

[0056] Based on standardized entity information in the entity database, four types of core nodes are generated for the initial land parcel status map: land parcel entity nodes, lifecycle stage nodes, compliance rule nodes, and operation indicator nodes. Each node is assigned a node number that is the same as the entity's unique identifier to ensure a one-to-one correspondence between nodes and entities.

[0057] Each node is assigned a standardized set of attributes; the attribute set for a land parcel entity node includes the unique identifier of the land parcel, spatial coordinate range, land use type, current ownership status, land parcel area, and plot ratio; the attribute set for a lifecycle stage node includes the stage name, stage start and end time, stage status identifier, stage manager, and stage completion progress; the attribute set for a compliance rule node includes the content of the legal clause, legal level identifier, constraint strength level, effective time, and expiration time; the attribute set for an operational indicator node includes the indicator name, indicator unit, current indicator value, indicator threshold range, monitoring frequency, and monitoring location.

[0058] Standardize the attribute values ​​of all nodes; retain two decimal places for numerical attributes; adopt a unified terminology for text attributes; limit enumerated attributes to a preset list of selectable values; store nodes and their attribute sets in a graph data structure, with each node as a vertex in the graph and node attributes as the attribute values ​​of the vertex.

[0059] Step S1-7: Construction of edge relationships and weight initialization of land parcel state graph:

[0060] Based on the verified relationships in the relational database, the edge relationships of the initial land parcel state graph are constructed; each edge connects two related nodes, and the type of the edge corresponds to the type of the relationship; the temporal relationship between land parcel entity nodes and lifecycle stage nodes is mapped to temporal dependency edges; the constraint relationship between lifecycle stage nodes and compliance rule nodes is mapped to constraint logic edges; the monitoring relationship between land parcel entity nodes and operational indicator nodes is mapped to monitoring logic edges; the principal relationship between land parcel entity nodes and principal entity nodes is mapped to principal responsibility edges; the temporal dependency edges, constraint logic edges, monitoring logic edges, and principal responsibility edges are uniformly merged into business logic edges;

[0061] Assign an initial weight to each edge; the initial weight of time-dependent edges is determined based on the time interval between stages, with shorter intervals resulting in higher weights; the initial weight of constraint logic edges is determined based on the constraint strength level of compliance rules, with higher constraint strength levels resulting in higher weights; the initial weight of monitoring logic edges is determined based on the importance of operational indicators, with higher importance levels resulting in higher weights; the initial weight of principal responsibility edges is determined based on the closeness of the connection between the principal and the land parcel, with closer connections resulting in higher weights.

[0062] The edge weights range from 0 to 1; all weights are normalized to ensure that the sum of the weights of edges of the same type is 1; the edges and their weights are stored in a graph data structure, with each edge as a directed edge in the graph, the direction of the edge indicating the direction of the relationship, and the edge weight as the attribute value of the edge.

[0063] Steps S1-8: Initial land parcel state map verification and persistent storage:

[0064] Perform a comprehensive verification on the completed initial land parcel state diagram; the verification includes graph structure integrity verification, node attribute correctness verification, and edge relationship rationality verification; graph structure integrity verification checks for the existence of isolated nodes and isolated edges; node attribute correctness verification checks whether the node attribute values ​​meet the standardization requirements and business logic rules; edge relationship rationality verification checks whether the connection relationship of the edges conforms to business common sense and whether there are circular dependencies and contradictory relationships;

[0065] The issues that failed the verification were categorized and marked; for graph structure issues, missing nodes and edges were added, and isolated nodes and edges were deleted; for node attribute issues, incorrect attribute values ​​were corrected, and missing attribute information was added; for edge relationship issues, unreasonable edge connection relationships were adjusted, and incorrect edge weight values ​​were corrected; the corrected initial parcel state graph was re-verified until all verification items passed.

[0066] The verified initial land parcel state map is persistently stored in a graph database format. The stored content includes information on all nodes, all edges, and the relationships between nodes and edges. A version management mechanism for the initial land parcel state map is established to record the time, content, and operator of each update. The initial land parcel state map is used as input data for the compliance dynamic assessment model in step S2, and is also backed up to an off-site storage system to ensure data security and reliability.

[0067] In this embodiment, the core function of step S2 is to input the initial land parcel state diagram into the compliance dynamic assessment model, map discrete regulatory policy texts into continuous compliance constraint vectors through a differentiable rule embedding algorithm, and aggregate compliance features along temporal dependency edges based on a spatiotemporal graph attention network to output the dynamic compliance state vector and compliance risk gradient of the land parcel at the current stage; the detailed steps are as follows:

[0068] S2-1: Semantic parsing of compliance rule text and generation of initial semantic feature sequence:

[0069] The legal clause content attributes of each compliance rule node in the initial land parcel status diagram are extracted, and a pre-trained semantic coding model is used to perform structured parsing of discrete legal policy texts. The pre-trained semantic coding model adopts a multi-layer bidirectional Transformer architecture, which can capture long-distance semantic dependencies in the text. The parsing process sequentially extracts rule condition features and threshold constraint features. Rule condition features include the subject scope, time scope, spatial scope, and behavioral requirements of the applicable regulations. Threshold constraint features include the upper limit, lower limit, and allowable fluctuation range of various quantitative indicators. The extracted rule condition features and threshold constraint features are concatenated in the order of text appearance to generate an initial semantic feature sequence. Each element of the initial semantic feature sequence corresponds to a semantic unit in the legal text, containing the contextual semantic information and constraint attribute information of that semantic unit.

[0070] S2-2: Differentiable rule embedding and continuous compliance constraint vector generation:

[0071] The initial semantic feature sequence is input into the differentiable rule embedding algorithm module. A constraint relaxation mapping mechanism and a differentiable logic approximation operator are used to project the rule condition features into a continuous space. The constraint relaxation mapping mechanism transforms discrete Boolean logic judgments into continuous probabilistic judgments, converting originally non-differentiable logical operations into differentiable numerical operations. The differentiable logic approximation operator uses the Sigmoid function to approximate the logical AND operation and the Softplus function to approximate the logical OR operation. Through these operations, discrete regulatory logic constraints are transformed into smooth, continuous compliance constraint vectors. The dimensions of the continuous compliance constraint vectors are consistent with the preset number of compliance dimensions, with each dimension corresponding to a specific compliance requirement. The continuous compliance constraint vectors are then dynamically updated to the corresponding compliance rule nodes, completing the attribute update of the initial land parcel state diagram.

[0072] S2-3: Activation of temporal information propagation pathways and calculation of dynamic attention weights in spatiotemporal graph attention networks:

[0073] Using the updated initial parcel state map as graph computation input, the temporal information propagation path of the spatiotemporal graph attention network is activated. The spatiotemporal graph attention network is specifically designed to process graph-structured data with temporal dependencies, capable of simultaneously capturing spatial topological relationships and temporal evolution relationships. Dynamic attention weights between adjacent lifecycle stage nodes are calculated along temporally dependent edges. These dynamic attention weights are jointly determined by node feature similarity, stage time span decay coefficient, and the semantic matching degree between compliance constraint vectors and stage states. The formula for calculating dynamic attention weights is as follows: ,in, For nodes For nodes Dynamic attention weights, For nodes With nodes The overall similarity score, For nodes The set of all adjacent nodes, For nodes Its any adjacent node The overall similarity score;

[0074] The formula for calculating the overall similarity score is: ,in, For nodes With nodes Feature similarity, For nodes With nodes The corresponding time span of the life cycle stage, For nodes Corresponding compliance constraint vectors and nodes The semantic matching degree of the corresponding stage state vector, The weighting coefficients are and satisfy the following conditions: , This is the time decay coefficient;

[0075] S2-4: Multi-channel temporal compliance feature aggregation and generation of temporal aggregated compliance feature representation:

[0076] Based on the calculated dynamic attention weights, a multi-channel feature aggregation operation is performed in the spatiotemporal graph attention network. This operation sets up independent feature aggregation channels for the compliance status representation of historical lifecycle stages, the compliance rule representation activated in the current stage, and the operational benchmark representation of land parcel entity nodes. Each channel is assigned different aggregation weights based on the importance of its corresponding feature. The aggregation results of the three channels are weighted and fused to generate a temporal aggregated compliance feature representation that characterizes the compliance evolution of the current stage. This temporal aggregated compliance feature representation simultaneously includes accumulated information on historical compliance status, constraint information on current compliance rules, and benchmark information on land parcel operation, comprehensively reflecting the overall compliance status of the land parcel in the current stage.

[0077] S2-5: Compliance Status Decoding and Dynamic Compliance Status Vector Output:

[0078] The temporal aggregated compliance feature representation is input into the compliance status decoding layer. This layer, composed of multiple fully connected layers and activation function layers, maps the high-dimensional temporal aggregated compliance feature representation to a low-dimensional compliance status representation. A multi-dimensional compliance discriminant function calculates the compliance status satisfaction of each dimension of the land parcel at the current lifecycle stage. This function uses the Sigmoid function to map the feature value of each compliance dimension to a value between 0 and 1. A value closer to 1 indicates a better compliance status for that dimension, while a value closer to 0 indicates a worse compliance status. The status satisfaction of each dimension is vectorized and assembled according to a preset compliance dimension sequence, outputting the dynamic compliance status vector of the land parcel at the current stage. Each element of the dynamic compliance status vector corresponds to a specific compliance dimension, and the element value is the status satisfaction of that dimension.

[0079] S2-6: Compliance Constraint Boundary Sensitivity Assessment and Compliance Risk Gradient Generation:

[0080] Based on the dynamic compliance state vector and the corresponding continuous compliance constraint vector, a compliance constraint boundary sensitivity assessment model is constructed. This model quantifies the impact of small changes in the dynamic compliance state vector at the compliance constraint boundary on the overall compliance status. The normal deviation rate of the dynamic compliance state vector at the constraint boundary is calculated; this rate represents the change in compliance satisfaction when the dynamic compliance state vector moves a unit distance along the normal direction of the constraint boundary. The normal deviation rates of each compliance dimension are then mapped using a gradient according to the risk transmission direction, generating a compliance risk gradient that characterizes the compliance deviation trend and the urgency of regulation. Each element of the compliance risk gradient corresponds to a specific compliance dimension; a larger element value indicates a higher compliance risk for that dimension, requiring priority regulation.

[0081] In this embodiment, the core function of step S3 is to drive a multi-dimensional operation monitoring model based on the dynamic compliance state vector and compliance risk gradient. It employs a cross-modal feature alignment mechanism to map the compliance constraint vector to real-time operation monitoring data, calculates the time-series deviation of each operation dimension, and generates an operation anomaly pattern identifier through a time-series anomaly detection algorithm. The detailed steps are as follows:

[0082] S3-1: Cross-modal feature alignment and mapping relationship generation:

[0083] The system receives dynamic compliance status vectors and compliance risk gradients, and simultaneously accesses real-time operational monitoring data streams generated during the land operation phase. These real-time operational monitoring data streams include multi-source sensor monitoring values ​​and operational business record values. Using the compliance constraint vector as a cross-modal alignment benchmark anchor, and leveraging a pre-built compliance operational semantic mapping dictionary, the system performs semantic-level dimensional association and weight allocation between the status representations of each compliance dimension in the dynamic compliance status vector and the monitoring values ​​of each operational dimension in the real-time operational monitoring data stream. The semantic-level dimensional association process is achieved by calculating the cosine similarity between the compliance dimension semantic vector and the operational dimension semantic vector, and the weight allocation is dynamically adjusted based on the dimensional sensitivity indicated by the compliance risk gradient. A cross-modal alignment mapping relationship is generated, which defines the semantic association strength and numerical conversion rules between each compliance dimension and its corresponding operational dimension.

[0084] S3-2: Construction of the expected baseline surface and calculation of instantaneous deviation in the operational dimension:

[0085] Based on the cross-modal alignment mapping relationship, an expected baseline surface for the operational dimension is constructed. This surface is jointly defined by a rigid compliance boundary constrained by a compliance constraint vector and a flexible compliance execution tolerance represented by a dynamic compliance state vector. The rigid compliance boundary represents an insurmountable absolute constraint, while the flexible compliance execution tolerance represents a permissible range of small fluctuations. For each operational monitoring moment, the actual monitoring value of the corresponding operational dimension in the real-time operational monitoring data stream is projected onto the expected baseline surface. The multidimensional spatial deviation distance between the actual monitoring value projection point and the baseline point on the surface is calculated, forming the instantaneous deviation value of each operational dimension at the monitoring moment. The formula for calculating the instantaneous deviation value is as follows: ,in, For the first The instantaneous deviation value at each monitoring moment. For the first The monitoring time of the first monitoring moment Actual monitoring values ​​for each operational dimension For the first The monitoring time of the first monitoring moment The baseline surface values ​​for each operational dimension The total number of operational dimensions;

[0086] S3-3: Deviation-based temporal sequence construction and noise removal:

[0087] The instantaneous deviation values ​​of each operational dimension are serialized and accumulated along the time axis for continuous monitoring moments to form an initial deviation time series. A sliding window smoothing method is used to denoise the initial deviation time series. The size of the sliding window is set according to the monitoring frequency and business needs, generally ranging from 5 to 15 consecutive monitoring moments. The sliding window smoothing is achieved by calculating the arithmetic mean of all instantaneous deviation values ​​within the window. High-frequency jitter components caused by occasional sensor noise are removed, and low-frequency trend components that characterize the long-term evolution trend of operational behavior are extracted to construct the deviation time series corresponding to each operational dimension.

[0088] S3-4: Adaptive Threshold Temporal Anomaly Detection and Anomaly Pattern Recognition

[0089] The deviation time series of each operational dimension is input in parallel into an anomaly detection engine based on time series morphology analysis. The anomaly detection engine adaptively adjusts the detection threshold sensitivity according to the dimension sensitivity indicated by the compliance risk gradient. The higher the compliance risk gradient value of the operational dimension, the higher the detection threshold sensitivity, and the earlier it can detect small abnormal changes. The anomaly detection engine performs time series shape feature extraction on the deviation time series. The extracted features include trend slope, fluctuation amplitude, period length, and boundary proximity. It identifies a variety of abnormal behavior patterns, including continuous upward trend pattern, periodic violent oscillation pattern, and boundary critical adhesion pattern. The continuous upward trend pattern indicates that the operational indicators are continuously developing in the direction of non-compliance, the periodic violent oscillation pattern indicates that the operational status is unstable, and the boundary critical adhesion pattern indicates that the operational indicators are close to the compliance boundary for a long time.

[0090] S3-5: Anomaly Pattern Clustering and Rating, and Generation of Operational Anomaly Pattern Identifiers:

[0091] Based on the duration and cumulative amplitude of the identified abnormal behavior patterns in the corresponding operational dimensions, the abnormal patterns are clustered and their severity is classified. The K-means clustering algorithm is used to group similar abnormal behavior patterns into the same category. The severity classification is divided into four levels: low risk, medium risk, high risk, and extremely high risk. The longer the duration and the greater the cumulative amplitude of the abnormal pattern, the higher its risk level. An operational abnormal pattern identifier with a pattern classification label and risk intensity level is generated, and the operational abnormal pattern identifier is transmitted to step S4 as a pre-judgment basis for triggering bidirectional coupling optimization.

[0092] In this embodiment, the core function of step S4 is to construct an optimization mechanism that couples compliance constraints and operational control in two directions, establish a closed-loop channel for positive mapping and negative correction, and transform operational deviation into constraint penalty terms when operational anomalies meet preset trigger conditions. A joint optimization model is established with dynamic compliance status as the hard constraint boundary and multi-dimensional operational comprehensive efficiency as the objective. The optimal control scheme is solved iteratively through dynamic penalty coefficients to generate an executable adaptive control instruction set, achieving a dynamic balance between compliance baseline and operational efficiency. The detailed steps are as follows:

[0093] S4-1: Compliance-Operations Two-Way Coupled Topology Architecture Setup

[0094] Establish a positive mapping channel from compliance constraints to operational control and a reverse correction channel from operational feedback to compliance boundary correction, forming a two-way coupled topology; the two channels operate independently and achieve data communication through a central data bus;

[0095] The positive mapping channel is responsible for converting compliance constraints into adjustable parameters in the operational dimension, enabling precise transmission of compliance requirements to operational actions. The channel is internally set up with a parameter conversion submodule and an instruction generation submodule. The parameter conversion submodule completes the spatial mapping from compliance status to operational parameters. The instruction generation submodule converts the optimal parameters into executable control instructions.

[0096] The reverse correction channel is responsible for feeding back deviation information generated during operation to the compliance rules system, enabling dynamic adjustment of the compliance constraint strength. The channel is equipped with a deviation analysis submodule and a rule adjustment submodule. The deviation analysis submodule quantifies the impact of operational anomalies on the compliance status. The rule adjustment submodule dynamically corrects the execution parameters of the compliance rules based on the deviation analysis results.

[0097] A data isolation and verification module is set up between the two channels; all data transmitted across channels must undergo integrity verification and logical consistency verification; data that fails verification will be intercepted and an anomaly alarm will be triggered.

[0098] S4-2: Configuration of bidirectional channel functional units and data interface reservation:

[0099] A compliance-operation parameter conversion unit is configured in the positive mapping channel. This unit receives cross-modal alignment mapping relationships and dynamic compliance status vectors, and establishes a bidirectional reversible mapping relationship library between compliance dimensions and operational control parameters. The mapping relationship library is stored in a hierarchical structure. The first layer is the compliance dimension index; the second layer is the corresponding operational parameter list; and the third layer is the quantitative relationship table of parameter adjustment magnitude and compliance status change. The quantitative relationship table is generated through statistical analysis of historical control data and is continuously updated as the system runs.

[0100] An operational deviation backtracking unit is configured in the reverse correction channel. This unit receives the deviation time series and compliance risk gradient, and constructs a dynamic compensation mechanism for the constraint strength level of compliance rule nodes based on operational anomaly characteristics. The compensation mechanism adopts a tiered adjustment strategy: low-risk anomalies correspond to a 5% reduction in constraint strength; medium-risk anomalies correspond to a 15% reduction in constraint strength; high-risk anomalies correspond to a 30% reduction in constraint strength; and extremely high-risk anomalies trigger a manual review process. The unit is equipped with a historical deviation database to store the characteristic information of all operational anomalies and their corresponding adjustment records for subsequent optimization of the compensation mechanism.

[0101] A standardized data interaction interface is reserved at the central node of the bidirectional coupled topology architecture; the interface adopts the RESTful protocol and supports data transmission in JSON format; the interface reserves input ports for receiving operational anomaly mode identifiers, dynamic compliance status vectors, and deviation time series sequences; and reserves output ports for sending adaptive control instruction sets and optimization process logs.

[0102] S4-3: Integration and Module Encapsulation of Joint Optimization Solver Engines

[0103] A joint optimization solution engine is deployed at the central node of the bidirectional coupled topology architecture; the joint optimization solution engine integrates objective function construction sub-units, hard constraint boundary solidification sub-units, and dynamic penalty coefficient adjustment sub-units.

[0104] The objective function construction sub-unit is responsible for generating a quantitative expression of multi-dimensional operational efficiency based on operational needs; the sub-unit has multiple preset indicator weight configuration schemes built-in, which support automatic switching according to different land parcel types and operational stages; at the same time, it provides a custom weight interface, allowing users to adjust the relative importance of each indicator according to actual needs;

[0105] The hard constraint boundary solidification sub-unit is responsible for transforming compliance requirements into insurmountable parameter boundaries during the optimization process; the sub-unit resolves the projection trajectory of the dynamic compliance state vector in the multidimensional control space into the boundary surface of the parameter feasible region; during the optimization solution process, any solution that exceeds the boundary surface will be automatically projected back into the feasible region to ensure that all control schemes meet the basic compliance requirements;

[0106] The dynamic penalty coefficient adjustment subunit is responsible for adaptively adjusting the penalty coefficient value based on feedback information during the optimization process. The subunit sets the upper and lower limits of the penalty coefficient; the lower limit is 0.1; the upper limit is 10.0; the initial penalty coefficient is set to 1.0. During the iteration process, the subunit adjusts the penalty coefficient in real time based on the compliance boundary approach residual and the slope of operational efficiency improvement, balancing the strictness of compliance constraints and the optimization space of operational efficiency.

[0107] The objective function construction sub-unit, hard constraint boundary solidification sub-unit, dynamic penalty coefficient adjustment sub-unit, compliance-operation parameter conversion unit, and operation deviation backtracking unit are logically encapsulated and interconnected with a data bus to complete the structured construction of the compliance-operation bidirectional coupling optimization module; the module provides a unified start and stop interface to the outside world, supporting remote calls and status monitoring;

[0108] S4-4: Operational Anomaly Trigger Condition Determination and Module Wake-up:

[0109] Real-time analysis of the pattern classification label and risk intensity level in the operational anomaly pattern identifier; determination of whether the triggering conditions are met according to preset judgment rules;

[0110] The risk intensity level is divided into four levels: low risk indicates that the deviation is small and the duration is short, with no significant impact on the compliance status; medium risk indicates that the deviation is moderate and the duration exceeds 24 hours, with potential compliance risks; high risk indicates that the deviation is large and the duration exceeds 72 hours, and it is about to break through the compliance boundary; and extremely high risk indicates that the deviation has already broken through the compliance boundary, resulting in actual violations.

[0111] The preset intervention types include three types: continuous upward trend pattern, which refers to the deviation showing a monotonically increasing trend for 10 consecutive monitoring times, and the trend slope is greater than 0.05; periodic violent oscillation pattern, which refers to the fluctuation amplitude of the deviation exceeding 3 times the normal range, and the oscillation period being less than 24 hours; and boundary critical adhesion pattern, which refers to the deviation being within 10% of the compliance boundary for 20 consecutive monitoring times.

[0112] When the risk intensity level reaches medium risk or above, or the pattern classification label matches any preset intervention type, the preset trigger condition is determined to be met; after the trigger condition is met, the compliance-operation two-way coupling optimization module is automatically activated; after the module is activated, the initialization operation is first performed, and the latest dynamic compliance status vector, compliance risk gradient and deviation time series are loaded.

[0113] If the operation anomaly mode identifier does not meet the preset trigger conditions, the module remains in a dormant state; during the module dormant period, a new operation anomaly mode identifier is received every 5 minutes for real-time monitoring.

[0114] S4-5: Construction of the joint optimization model and iterative solution of the dynamic penalty coefficient:

[0115] Extract the deviation time series for the 30 days prior to the trigger point; based on the sensitivity weights of each operational dimension indicated by the compliance risk gradient, perform nonlinear amplitude mapping and time decay weighting on the deviation time series to generate dynamic constraint penalty terms that correspond one-to-one with the operational dimension space; the nonlinear amplitude mapping adopts an exponential function form; the mapping formula is: ,in, For the first Non-linear mapping values ​​for each operational dimension These are nonlinear coefficients. For the first Average deviation of each operational dimension;

[0116] The time decay weighting process uses an exponential decay function; the weighting formula is: ,in, For the first Time weight of each moment The time decay coefficient, The trigger time;

[0117] The dynamic constraint penalty term for each operational dimension is obtained by multiplying the nonlinear mapping value by the time weight and taking the average.

[0118] Construct an objective function with multi-dimensional operational comprehensive efficiency as the core optimization objective; multi-dimensional operational comprehensive efficiency is calculated by weighted fusion of land resource utilization efficiency, environmental load control level, and economic benefit output indicators; the objective function expression is: ,in, To improve the overall efficiency of multi-dimensional operations, To improve the efficiency of land resource utilization, To control environmental load levels, As an economic output indicator, , , The weighting coefficients are and satisfy the following conditions: ;

[0119] The dynamic constraint penalty term is embedded into the objective function in an adaptive augmented form, and hard constraint boundaries are superimposed to construct a complete joint optimization model.

[0120] Initiate the iterative solution process for the dynamic penalty coefficient; based on the compliance boundary approximation residual and operational efficiency improvement slope of the current iteration cycle, adaptively adjust the convergence step size and update frequency of the dynamic penalty coefficient; the dynamic penalty coefficient update formula is: ,in, For the first The dynamic penalty coefficient for the next iteration. For the first The dynamic penalty coefficient for the next iteration. For adjustment coefficients, For the first The compliance boundary of the next iteration approaches the residual. For the first The slope of operational efficiency improvement in each iteration;

[0121] An iterative strategy combining gradient optimization and boundary projection truncation is employed to solve the joint optimization model. Gradient optimization is used to find the optimal solution within the feasible region, while boundary projection truncation projects solutions that exceed the hard constraint boundaries back into the feasible region. The iterative process continues until a preset convergence tolerance is met or the maximum number of iterations is reached. The preset convergence tolerance is defined as the change in the objective function value between two consecutive iterations being less than [a certain value]. The maximum number of iterations is set to 100. If convergence is not achieved after the maximum number of iterations, the suboptimal solution is used as the final result, and an alarm for non-convergence is generated.

[0122] S4-6: Optimal Control Parameter Inversion and Adaptive Control Instruction Set Generation:

[0123] The optimal control parameter sequence output after convergence is input into the compliance-operation parameter conversion unit; parameter space inversion is performed through a bidirectional invertible mapping relation library; for parameters with exact matching in the mapping relation library, the corresponding conversion value is directly adopted;

[0124] For parameters that do not have an exact match, an approximate conversion value is calculated using linear interpolation; the converted executable control parameters are encapsulated with an instruction protocol; and an adaptive control instruction set is generated that includes the control object code, control action type, execution time window, expected compliance recovery threshold, instruction priority, execution responsibility person, and feedback requirements.

[0125] The control object code uses a globally unique identifier to uniquely identify the operational equipment or business process that needs to be controlled; the control action types include three categories: parameter adjustment, process optimization, and resource scheduling; the execution time window specifies the start and end time of the control action; the expected compliance recovery threshold is the target value that each operational dimension should achieve after the control is completed; the instruction priority is divided into three levels: general, urgent, and extremely urgent; the execution responsibility person specifies the personnel or department responsible for executing the instruction; the feedback requirements specify the content and frequency of feedback during and after the execution process.

[0126] The generated adaptive control instruction set is sent to the instruction verification module for verification. The verification includes the correctness of the instruction format, the rationality and compliance of the parameters. Instructions that pass the verification are sent to the execution control terminal. Instructions that fail the verification are returned to the joint optimization solution engine for re-solution.

[0127] After receiving the instruction, the execution control terminal sends a confirmation receipt to the system; the system tracks the execution status of the instruction and receives execution feedback data in real time; if the instruction does not receive a confirmation receipt within the specified time, it is automatically resent; if no receipt is received after three resentments, a manual intervention process is triggered.

[0128] In this embodiment, the core function of step S5 is to receive the execution feedback data of the adaptive control instruction set, perform residual calculation between the actual control effect and the expected compliance target, integrate compliance risk gradient information to generate a state update tensor, trigger the node feature reset and edge weight dynamic reconstruction of the initial land parcel state graph, generate an iterative evolution graph and feed it back to the data processing link, and realize continuous closed-loop optimization of land parcel full life cycle compliance assessment and operation monitoring until the life cycle ends or the system converges; the detailed steps are as follows:

[0129] S5-1: Execution Feedback Data Reception and Compliance - Operational Status Update Residual Tensor Generation:

[0130] Establish a real-time data acquisition channel for execution feedback, connecting the execution control terminal and each monitoring sensor node; acquire full execution feedback data after the issuance of adaptive control command sets according to a preset acquisition frequency; the acquisition frequency is dynamically adjusted according to the urgency of the control command; the acquisition frequency for general control commands is once per hour;

[0131] Emergency control commands are collected once per minute; core parameters are extracted from the execution feedback data; core parameters include real-time monitoring values ​​of various operational dimensions, completion rate of control actions, resource consumption data, equipment operating status data, and on-site inspection record data; the expected compliance recovery threshold and expected execution time window preset in the adaptive control command set are extracted simultaneously.

[0132] Standardized preprocessing operations are performed on the extracted execution feedback data; outlier removal and missing value imputation are performed on numerical data; outlier removal adopts the 3σ principle; missing value imputation adopts the linear interpolation method of adjacent monitoring points in the same time period; semantic standardization processing is performed on text data to unify terminology.

[0133] The preprocessed actual control effectiveness parameters are aligned with the compliance risk gradient in both time and space dimensions. Time dimension alignment is achieved by unifying the timestamp to the UTC+8 time zone; spatial dimension alignment is achieved by unifying the spatial coordinates to the WGS84 coordinate system; ensuring that the two are completely matched in both time and space dimensions.

[0134] Calculate the residual sequence of state deviation between the expected compliance status and the actual execution status; the method for calculating the residual sequence of state deviation is the difference between the actual value and the expected value of the corresponding dimension; the calculation formula is: ,in, For the first The residuals of state deviation in each operational dimension For the first Actual monitoring values ​​for each operational dimension For the first Expected compliance recovery thresholds for each operational dimension;

[0135] A compliance risk gradient decay factor is introduced to weight and fuse the state deviation residual sequences, generating a compliance-operational state update residual tensor; the fusion formula is: ,in, To ensure compliance and operational status, the residual tensor is updated. As a compliance risk gradient decay factor, For the state to deviate from the residual sequence, Compliance risk gradient;

[0136] The compliance risk gradient attenuation factor ranges from 0.2 to 0.8; it is dynamically adjusted according to the execution cycle of the regulatory instructions; the longer the execution cycle, the smaller the attenuation factor and the lower the weight of the compliance risk gradient; when the execution cycle is less than 7 days, the attenuation factor is 0.8; when the execution cycle is greater than 30 days, the attenuation factor is 0.2.

[0137] S5-2: Resetting the node features of the initial land parcel state map:

[0138] Based on the compliance-operation status update residual tensor, a node-by-node feature reset operation is performed on the four types of core nodes in the initial land parcel status diagram; the four types of core nodes include land parcel entity nodes, life cycle stage nodes, compliance rule nodes, and operation indicator nodes.

[0139] For land parcel entity nodes, the spatial coordinate range attribute of the land parcel entity node is corrected based on the spatial offset compensation amount in the compliance-operation status update residual tensor; the spatial offset compensation amount is calculated from the difference between the latest measured land parcel boundary data and the initial registration data; the land parcel area attribute and plot ratio attribute of the land parcel entity node are corrected simultaneously; the current ownership status attribute of the land parcel entity node is updated based on the latest ownership change registration information; the responsible entity attribute of the land parcel entity node is updated based on the latest responsible entity change information.

[0140] For each lifecycle stage node, the start and end time attributes of the stage are recalibrated based on the temporal evolution lag in the residual tensor updated according to the compliance-operation status. The temporal evolution lag is the difference between the actual stage completion time and the planned stage completion time. The stage completion progress attribute and stage leader attribute of the lifecycle stage node are updated synchronously. The stage status identifier attribute of the lifecycle stage node is automatically switched according to the actual project progress and operation status. The stage status identifier includes five states: not started, in progress, completed, delayed, and suspended.

[0141] For compliance rule nodes, the constraint strength level attribute of the compliance rule node is dynamically adjusted based on the constraint violation residual value in the constraint-operation status update residual tensor; the larger the constraint violation residual value, the greater the reduction in constraint strength level; the regulatory clause applicability confidence attribute and constraint violation count attribute of the compliance rule node are updated simultaneously; when the number of constraint violations exceeds the preset threshold, the compliance rule node is automatically marked as a high-risk rule node and the rule applicability review process is triggered.

[0142] For operational indicator nodes, the indicator current value attribute of the operational indicator node is updated by combining the indicator deviation correction amount in the compliance-operation status update residual tensor; the indicator threshold range attribute of the operational indicator node is dynamically fine-tuned based on the statistical analysis results of long-term operational data; the monitoring frequency attribute and monitoring location attribute of the operational indicator node are updated synchronously; when the indicator deviation exceeds the preset time continuously, the monitoring frequency of the operational indicator node is automatically increased.

[0143] After resetting the features of all nodes, perform consistency verification on the attribute values ​​of all nodes. The verification includes the numerical range of attribute values, data types, and the rationality of business logic. Store the set of node attributes that pass the verification in the graph database to complete the node feature reset process.

[0144] S5-3: Dynamic update of edge weights in the initial plot state graph:

[0145] Based on the node attribute set after feature reset, dynamic update of all edge weights is performed along the topology of the initial land parcel state graph; the update scope includes temporally dependent edges and all types of business logic edges;

[0146] Calculate the state transition probabilities of the lifecycle stage nodes at both ends of the temporally dependent edge under the compliance-operational state update residual tensor mapping; the formula for the state transition probability is: ,in, For nodes To the node The probability of state transitions, For nodes With nodes The overall similarity score, For nodes The set of all adjacent nodes;

[0147] The temporal transit decay weights of temporal dependent edges are reset based on the calculated state transition probabilities. The temporal transit decay weights are proportional to the state transition probabilities. The higher the state transition probability, the greater the temporal transit decay weights. The weights of all temporal dependent edges are normalized to ensure that the sum of the weights of all temporal dependent edges in the same plot is 1.

[0148] Perform weight update operations on different types of business logic edges; business logic edges include constraint logic edges, monitoring logic edges, and main responsibility edges.

[0149] For each constraint logic edge, calculate the correlation coefficient between the change in the constraint strength level of the compliance rule node and the change in the deviation of the corresponding operational indicator node; update the business coupling association weight of the constraint logic edge based on the correlation coefficient; the higher the correlation coefficient, the greater the business coupling association weight.

[0150] For the monitoring logic edge, calculate the correlation coefficient between the change in the monitoring frequency of the operation indicator node and the change in the risk level of the corresponding land parcel entity node; update the business coupling association weight of the monitoring logic edge based on the correlation coefficient; the higher the correlation coefficient, the greater the business coupling association weight.

[0151] For the main responsibility edge, calculate the correlation coefficient between the change information of the responsible entity and the change in the operational status of the corresponding land parcel entity node; update the business coupling association weight of the main responsibility edge based on the correlation coefficient; the higher the correlation coefficient, the greater the business coupling association weight;

[0152] Normalize the weights of all business logic edges; ensure that the sum of the weights of all outgoing edges of the same node is 1; store the updated edge weight information in the graph database to complete the adaptive reconstruction of edge weights;

[0153] S5-4: Iterative Evolution Graph Generation and State Loop Iteration Termination Determination

[0154] The initial parcel state graph that has completed node feature reset and edge weight dynamic update is marked as the iterative evolution graph; a unique version number is assigned to the iterative evolution graph; the version number adopts the format of major version number plus minor version number; the major version number increments with the life cycle stage; the minor version number increments with each closed loop iteration;

[0155] The iterative evolution graph is injected into the data processing chain of step S1 as a contextual prior graph feedback; prior knowledge includes historical compliance status records, historical operational anomaly mode records, and historical control instruction execution effect records; prior knowledge is used to guide the priority setting of the next round of data collection and improve the accuracy of entity relationship extraction;

[0156] Define a closed-loop iteration termination criterion; the termination criterion includes two independent conditions; the iteration will terminate if either condition is met.

[0157] The first condition is that the magnitude of the change in the compliance risk gradient generated in two consecutive iterations is lower than the preset convergence threshold; the preset convergence threshold is set to 10. -4 This condition indicates that the system has entered a stable operating state and requires no further adjustment.

[0158] The second condition is that the stage status identifier attribute of the life cycle stage node flows to the final operation termination identifier; the final operation termination identifier includes three states: land transfer termination, land demolition, and land use change; this condition indicates that the entire life cycle of the land has ended;

[0159] When the closed-loop iteration of the judgment state is completed, the final state result of the dynamic assessment of the compliance of the land parcel throughout its entire life cycle and the archived data of multi-dimensional operation monitoring are output.

[0160] The final results of the dynamic compliance assessment include compliance status records for each lifecycle stage, risk event statistics, rectification completion status, and the final compliance rating; the final compliance rating is divided into four levels: excellent, good, qualified, and unqualified.

[0161] The multi-dimensional operational monitoring archived data includes time-series data of operational indicators throughout the entire lifecycle, abnormal event records, control command execution logs, equipment operation logs, and on-site inspection records; all archived data is standardized in accordance with national standards and stored in a permanent archiving system.

[0162] If the termination condition is not met, the next round of closed-loop iteration process is started; steps S1 to S5 are repeated; the next iteration will use the iteration evolution diagram generated in the previous round as the initial state diagram to realize the continuous self-optimization and update of the system.

[0163] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, characterized by: Includes the following steps: S1: Obtain multi-source heterogeneous data of the land parcel at each stage of its entire life cycle, perform spatiotemporal benchmark alignment and entity relationship extraction on the multi-source heterogeneous data, and construct an initial land parcel state diagram. The initial land parcel state diagram includes land parcel entity nodes, life cycle stage nodes, compliance rule nodes and operation indicator nodes. The nodes are connected through temporal dependency edges and business logic edges. S2: Input the initial land parcel state map into the compliance dynamic assessment model. A differentiable rule embedding algorithm maps discrete regulatory policy texts into continuous compliance constraint vectors. Differentiable logic approximation operators use the Sigmoid function to approximate the logical AND operation and the Softplus function to approximate the logical OR operation. Compliance features are aggregated along temporal dependency edges based on a spatiotemporal graph attention network. The updated initial land parcel state map is used as the graph computation input, activating the temporal information propagation path of the spatiotemporal graph attention network. The spatiotemporal graph attention network is specifically designed to process graph-structured data with temporal dependencies, capable of simultaneously capturing spatial topological relationships and temporal evolution relationships. Dynamic attention weights are calculated between adjacent lifecycle stage nodes along temporal dependency edges. These dynamic attention weights are jointly determined by node feature similarity, stage time span decay coefficient, and semantic matching degree between the compliance constraint vector and the stage state. The output is the dynamic compliance state vector and compliance risk gradient of the land parcel in the current stage. S3: Based on dynamic compliance state vector and compliance risk gradient, drive multi-dimensional operation monitoring model, adopt cross-modal feature alignment mechanism to map compliance constraint vector to real-time operation monitoring data, calculate the deviation time series of each operation dimension, and generate operation anomaly pattern identifier through time series anomaly detection algorithm; S4: Construct a compliance-operation bidirectional coupling optimization module. When the operational anomaly mode identifier meets the preset trigger conditions, the deviation time series is transformed into a constraint penalty term. Establish a joint optimization model with the dynamic compliance state vector as the hard constraint boundary and the multi-dimensional operational comprehensive efficiency as the objective function. Generate an adaptive control instruction set by iteratively solving the dynamic penalty coefficient. S5: The execution feedback data of the adaptive control instruction set is residually fused with the compliance risk gradient to trigger the node feature reset and edge weight dynamic update of the initial land parcel state graph, thus completing the closed-loop iteration of the state for the full life cycle compliance assessment and operation monitoring of the land parcel.

2. The method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, as described in claim 1, is characterized in that: Acquire multi-source heterogeneous data of land parcels at each stage of their entire lifecycle, perform spatiotemporal benchmark alignment and entity relationship extraction on the multi-source heterogeneous data, and construct an initial land parcel state diagram. This initial land parcel state diagram includes land parcel entity nodes, lifecycle stage nodes, compliance rule nodes, and operational indicator nodes. Nodes are connected through temporal dependency edges and business logic edges, specifically including: Acquire raw data of the land parcel at each stage of its entire life cycle. The raw data includes planning and land use permit data, land transfer contract data, construction project planning permit data, completion acceptance and filing data, as well as real-time IoT sensor data and inspection record data during the operation phase. The raw data is classified according to data structure type to obtain a multi-source heterogeneous dataset including structured data records, semi-structured document files and unstructured text descriptions. The timestamp information and spatial coordinate reference information carried by various types of data in the multi-source heterogeneous dataset are uniformly parsed and transformed. The timestamps of all data are unified to the same standard time format, and the spatial coordinate reference system of all data is aligned to the same target coordinate system. This achieves spatiotemporal benchmark alignment of multi-source heterogeneous data, resulting in a spatiotemporally aligned dataset. Semantic parsing is performed on each data record in the spatiotemporal aligned dataset. Entity recognition algorithms are used to extract the land parcel spatial code, project unique identifier, approval item name, regulatory clause number, and monitoring indicator name from each data record. The extracted entities are labeled as land parcel entities, stage entities, rule entities, and indicator entities, respectively. The relationships between these entities are extracted simultaneously. These relationships include the temporal relationship between land parcel entities and stage entities, the constraint relationship between stage entities and rule entities, and the monitoring relationship between land parcel entities and indicator entities, forming the entity relationship extraction results. Based on the entity relationship extraction results, land parcel entity nodes, lifecycle stage nodes, compliance rule nodes, and operational indicator nodes are generated respectively, and each node is assigned a corresponding set of attributes. Among them, the attribute set of the land parcel entity node includes the unique identifier of the land parcel, the spatial coordinate range, the land use type, and the current ownership status; the attribute set of the lifecycle stage node includes the stage name, the start and end time of the stage, and the stage status identifier; the attribute set of the compliance rule node includes the content of the legal clause, the legal level identifier, and the constraint strength level; the attribute set of the operational indicator node includes the indicator name, the current value of the indicator, and the threshold range of the indicator. Based on the relationships extracted from the entity relationships, connection edges are constructed between nodes. Specifically, land parcel entity nodes are connected to lifecycle stage nodes in chronological order to generate temporal dependency edges; lifecycle stage nodes are connected to compliance rule nodes according to the corresponding rule application stages to generate constraint logic edges; land parcel entity nodes are connected to operational indicator nodes according to the corresponding indicator monitoring objects to generate monitoring logic edges; and temporal dependency edges, constraint logic edges, and monitoring logic edges are uniformly merged into business logic edges. The initial land parcel state diagram is constructed by integrating the land parcel entity nodes, lifecycle stage nodes, compliance rule nodes, and operational indicator nodes, as well as the temporal dependency edges and business logic edges. This initial land parcel state diagram is then used as the input data for the compliance dynamic assessment model in step S2.

3. The method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, as described in claim 1, is characterized in that: The initial land parcel state diagram is input into the compliance dynamic assessment model. A differentiable rule embedding algorithm maps discrete regulatory and policy texts into continuous compliance constraint vectors. Compliance features are aggregated along temporal dependency edges using a spatiotemporal graph attention network. The model outputs the dynamic compliance state vector and compliance risk gradient of the land parcel at the current stage, specifically including: Extract the content attributes of the legal clauses of each compliance rule node in the initial land parcel status diagram, use a pre-trained semantic coding model to perform structured parsing of discrete legal policy texts, extract rule condition features and threshold constraint features, and generate an initial semantic feature sequence; The initial semantic feature sequence is input into the differentiable rule embedding algorithm module. The constraint relaxation mapping mechanism and differentiable logic approximation operator are used to project the rule condition features into a continuous space, transforming the discrete regulatory logic constraints into a smooth continuous compliance constraint vector. The continuous compliance constraint vector is then used as a dynamic attribute to update the corresponding compliance rule node. Using the initial parcel state map after attribute update as the graph computation input, the temporal information propagation path of the spatiotemporal graph attention network is activated, and the dynamic attention weights between adjacent life cycle stage nodes are calculated along the temporal dependency edges. The dynamic attention weights are jointly determined by the node feature similarity, the stage time span decay coefficient, and the semantic matching degree between the compliance constraint vector and the stage state. Based on dynamic attention weights, multi-channel feature aggregation operations are performed in the spatiotemporal graph attention network to weight and fuse the compliance status representation of historical life cycle stages, the compliance rule representation activated in the current stage, and the operational benchmark representation of land parcel entity nodes to generate a temporal aggregated compliance feature representation of the compliance evolution trend in the current stage. The time-series aggregated compliance feature representation is input into the compliance status decoding layer. The state satisfaction of each compliance dimension of the land parcel in the current life cycle stage is calculated through a multi-dimensional compliance discrimination function. The state satisfaction of each dimension is vectorized and assembled according to a preset compliance dimension sequence, and the dynamic compliance status vector of the land parcel in the current stage is output. Based on the dynamic compliance state vector and the corresponding continuous compliance constraint vector, a compliance constraint boundary sensitivity assessment model is constructed. The normal deviation rate of the dynamic compliance state vector at the constraint boundary is calculated, and the normal deviation rate of each compliance dimension is mapped in a gradient according to the risk transmission direction to generate a compliance risk gradient that represents the compliance deviation trend and the urgency of regulation.

4. The method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, as described in claim 1, is characterized in that: Based on dynamic compliance state vectors and compliance risk gradients, a multi-dimensional operational monitoring model is driven. A cross-modal feature alignment mechanism is used to map compliance constraint vectors to real-time operational monitoring data, calculating the time-series deviation of each operational dimension. An operational anomaly pattern identifier is generated through a time-series anomaly detection algorithm, specifically including: It receives dynamic compliance status vectors and compliance risk gradients, and simultaneously accesses real-time operation monitoring data streams generated during the land operation phase. The real-time operation monitoring data streams include multi-source sensor monitoring values ​​and operation business record values. The compliance constraint vector is used as the cross-modal alignment benchmark anchor point. Using a pre-built compliance-operation semantic mapping dictionary, the state representation of each compliance dimension in the dynamic compliance status vector and the monitoring values ​​of each operation dimension in the real-time operation monitoring data stream are semantically correlated and weighted to generate cross-modal alignment mapping relationships. Based on the cross-modal alignment mapping relationship, an expected baseline surface for the operational dimension is constructed. The expected baseline surface for the operational dimension is jointly defined by the rigid compliance boundary limited by the compliance constraint vector and the flexible compliance execution tolerance represented by the dynamic compliance state vector. For each operational monitoring moment, the actual monitoring value of the corresponding operational dimension in the real-time operational monitoring data stream is projected onto the expected baseline surface for the operational dimension. The multi-dimensional spatial deviation distance between the actual monitoring value projection point and the baseline point of the surface is calculated to form the instantaneous deviation value of each operational dimension at the monitoring moment. The instantaneous deviation values ​​of each operational dimension at continuous monitoring time along the time axis are serialized, accumulated, and smoothed by a sliding window. High-frequency jitter components caused by occasional sensor noise are removed, and low-frequency trend components that characterize the long-term evolution trend of operational behavior are extracted to construct the deviation time series corresponding to each operational dimension. The deviation time series of each operational dimension is input in parallel into the anomaly detection engine based on time series morphology analysis. The anomaly detection engine adaptively adjusts the detection threshold sensitivity according to the dimension sensitivity indicated by the compliance risk gradient. The anomaly detection engine performs time series shape feature extraction on the deviation time series to identify a variety of abnormal behavior patterns, including continuous upward deviation trend, periodic violent oscillation, and boundary critical adhesion. Based on the duration and cumulative intensity of the identified abnormal behavior patterns in the corresponding operational dimensions, the abnormal patterns are clustered and their severity is classified. An operational abnormal pattern identifier with a pattern classification label and risk intensity level is generated, and the operational abnormal pattern identifier is transmitted to step S4 as a preliminary judgment basis for triggering bidirectional coupling optimization.

5. The method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, as described in claim 4, is characterized in that: The construction of the compliance-operations two-way coupling optimization module specifically includes: Establish a positive mapping channel from compliance constraints to operational control and a reverse correction channel from operational feedback to compliance boundary correction, forming a two-way coupled topology architecture; Configure a compliance-operation parameter conversion unit in the positive mapping channel to receive cross-modal alignment mapping relationships and dynamic compliance state vectors, and establish a bidirectional reversible mapping relationship library between compliance dimensions and operational control parameters; Configure an operational deviation backtracking unit in the reverse correction channel to receive the deviation time series and compliance risk gradient, and build a dynamic compensation mechanism for the constraint strength level of compliance rule nodes based on operational anomaly characteristics. A joint optimization solution engine is deployed at the central node of the bidirectional coupled topology architecture. The joint optimization solution engine integrates objective function construction sub-units, hard constraint boundary solidification sub-units, and dynamic penalty coefficient adjustment sub-units, and reserves data interaction interfaces with compliance-operation parameter conversion units and operation deviation backtracking units. The objective function construction sub-unit, hard constraint boundary solidification sub-unit, dynamic penalty coefficient adjustment sub-unit, compliance-operation parameter conversion unit, and operation deviation backtracking unit are logically encapsulated and interconnected with a data bus to complete the structured construction of the compliance-operation bidirectional coupling optimization module.

6. The method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, as described in claim 5, is characterized in that: When the operational anomaly pattern indicator meets the preset trigger conditions, the deviation time series is transformed into constraint penalty terms. A joint optimization model is established with the dynamic compliance state vector as the hard constraint boundary and multi-dimensional operational comprehensive efficiency as the objective function. An adaptive control instruction set is generated by iteratively solving the dynamic penalty coefficient, specifically including: The system analyzes the pattern classification label and risk intensity level in the abnormal operation pattern identifier in real time. When the risk intensity level crosses the preset safety threshold or the pattern classification label matches the preset intervention type, it determines that the preset triggering condition is met and activates the compliance-operation bidirectional coupling optimization module. Extract the deviation time series corresponding to the trigger time, and perform nonlinear amplitude mapping and time decay weighting on the deviation time series according to the sensitivity weight of each operational dimension indicated by the compliance risk gradient, to generate dynamic constraint penalty terms that correspond one-to-one with the operational dimension space. The joint optimization solution engine is invoked to resolve the projection trajectory of the dynamic compliance state vector in the multidimensional control space as a parameter insurmountable boundary, which is then fixed as the hard constraint boundary of the joint optimization model. A target function is constructed with multi-dimensional operational comprehensive efficiency as the core optimization objective. The multi-dimensional operational comprehensive efficiency is calculated by weighted fusion of land resource utilization efficiency, environmental load control level and economic benefit output indicators. The dynamic constraint penalty term is embedded into the objective function in an adaptive augmented form, and hard constraint boundaries are superimposed to construct a complete joint optimization model. Initiate the dynamic penalty coefficient iterative solution process, approximate the residual and operational efficiency improvement slope based on the compliance boundary of the current iteration cycle, adaptively adjust the convergence step size and update frequency of the dynamic penalty coefficient, and use an iterative strategy combining gradient optimization and boundary projection truncation to solve the joint optimization model until the preset convergence tolerance is met or the maximum iteration round is reached. The optimal control parameter sequence output after convergence is input into the compliance-operation parameter conversion unit. The parameter space is inverted and the instruction protocol is encapsulated through a bidirectional reversible mapping relation library to generate an adaptive control instruction set containing the control object code, control action type, execution time window and expected compliance recovery threshold, and then sent to the execution control terminal.

7. The method for dynamic assessment and multi-dimensional operational monitoring of land parcel compliance throughout its entire lifecycle, as described in claim 6, is characterized in that: The execution feedback data of the adaptive control instruction set is residually fused with the compliance risk gradient to trigger the reset of node features and dynamic update of edge weights in the initial land parcel state graph. This completes the closed-loop iteration of the land parcel's full lifecycle compliance assessment and operational monitoring, specifically including: After receiving the execution feedback data collected after the adaptive control instruction set is issued, the actual control effect parameters and expected compliance recovery thresholds are extracted from the execution feedback data. The actual control effect parameters and compliance risk gradients are aligned in the spatiotemporal dimensions, the state deviation residual sequence between the expected compliance state and the actual execution state is calculated, and the compliance risk gradient decay factor is introduced to perform weighted fusion on the state deviation residual sequence to generate the compliance-operation state update residual tensor. Based on the compliance-operation status update residual tensor, node feature reset operations are performed on the land parcel entity nodes, lifecycle stage nodes, compliance rule nodes, and operation indicator nodes in the initial land parcel status diagram: For land parcel entity nodes, the spatial coordinate range attribute and current ownership status attribute of the land parcel entity node are corrected according to the spatial offset compensation amount in the compliance-operation status update residual tensor; for lifecycle stage nodes, the stage start and end time attribute and stage status identifier attribute of the lifecycle stage node are recalibrated according to the temporal evolution lag amount in the compliance-operation status update residual tensor; for compliance rule nodes, the constraint strength level attribute of the compliance rule node is dynamically downgraded and the application confidence attribute of the regulatory clause is updated according to the constraint violation residual value in the compliance-operation status update residual tensor; for operation indicator nodes, the current value attribute and threshold range attribute of the operation indicator node are updated by combining the indicator deviation correction amount in the compliance-operation status update residual tensor, resulting in the node attribute set after feature reset. Based on the node attribute set after feature reset, the edge weights are dynamically updated along the topology of the initial land parcel state graph: the state transition probabilities of the lifecycle stage nodes at both ends of the temporally dependent edge under the compliance-operation state update residual tensor mapping are calculated, and the temporal transit decay weights of the temporally dependent edge are reset accordingly; the residual co-evolution relationship between the compliance rule nodes and operation indicator nodes connected by the business logic edge is analyzed, and the business coupling association weights of the business logic edge are updated through the attention redistribution mechanism; the reset temporally transit decay weights and the updated business coupling association weights are mapped to the temporally dependent edge and the business logic edge respectively, completing the adaptive reconstruction of the edge weights; The initial land parcel state graph, after completing node feature reset and edge weight dynamic update, is marked as an iterative evolution graph. The iterative evolution graph is then injected into the data processing link of step S1 as a contextual prior graph. At the same time, a closed-loop iteration termination criterion is set. When the magnitude of the change in compliance risk gradient generated by two consecutive iterations is lower than the preset convergence threshold or the stage status identifier attribute of the life cycle stage node flows to the final operation termination identifier, the state closed-loop iteration is determined to be completed, and the final state result of the dynamic evaluation of the land parcel's full life cycle compliance and multi-dimensional operation monitoring archive data are output.

Images