A decision system for land space planning physical examination supervision and monitoring and planning dynamic maintenance based on "AI + knowledge graph"

Abstract

This invention discloses a decision-making system for the monitoring, supervision, and dynamic maintenance of territorial spatial planning based on "AI + knowledge graph," comprising: a data acquisition and preprocessing module for collecting and preprocessing multi-source heterogeneous data related to the implementation of territorial spatial planning; a dynamic maintenance rule knowledge graph construction module for constructing a dynamic maintenance rule knowledge graph for territorial spatial master planning by integrating historical multi-source heterogeneous data with policy norms, planning knowledge, and expert experience based on artificial intelligence technology and urban health check indicator system; a three-dimensional indicator evaluation module for dynamically evaluating urban health check data through feature modeling and time series analysis; a dynamic maintenance rule intelligent diagnosis module for performing multi-dimensional feature modeling and dynamic diagnosis of evaluation results based on a knowledge graph and deep learning fusion mechanism; and a dynamic maintenance rule matching module for inputting diagnostic results into the constructed knowledge graph to generate intelligent decisions.

Description

Technical Field

[0001] This invention relates to a decision-making system for monitoring, supervising, and dynamically maintaining land and space planning based on "AI + knowledge graph". Background Technology

[0002] The existing territorial spatial planning health check system suffers from two major problems: first, the level of intelligent application of the "monitoring, supervision, and management" system for planning implementation is not high; second, a "data-assessment-maintenance" rule system for applying health check indicators to implementation and management has not yet been formed. These two problems directly restrict the effective functioning of the territorial spatial planning health check system, and further adversely affect the optimization of urban spatial structure, the improvement of functional quality, and the improvement of planning implementation efficiency. Summary of the Invention

[0003] To overcome the shortcomings of the existing technologies, this invention provides a decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph". By constructing an AI + knowledge graph-driven management decision-making system, the effective functioning of the physical examination of land and space planning is ensured.

[0004] According to one aspect of the present invention, a decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph" is provided, comprising: The data acquisition and preprocessing module is used to collect multi-source heterogeneous data related to the implementation of land and space planning and to preprocess the data to form a standardized input dataset for modeling and analysis. The dynamic maintenance rule knowledge graph construction module is used to construct a dynamic maintenance rule knowledge graph for the overall land space planning by integrating historical multi-source heterogeneous data with policy norms, planning knowledge and expert experience, based on artificial intelligence technology and combining two major categories of urban health check indicator systems: spatial support system and central urban area planning. The three-dimensional indicator evaluation module is used to dynamically evaluate urban physical examination data based on the three-dimensional model construction principle after acquiring the actual multi-source heterogeneous data. It uses feature modeling and time series analysis. The dynamic maintenance rule intelligent diagnosis module is used to perform multi-dimensional feature modeling and dynamic diagnosis on the evaluation results output by the three-dimensional index evaluation module. The dynamic maintenance rule matching module is used to match the diagnostic results output by the dynamic maintenance rule intelligent diagnosis module with the constructed dynamic maintenance rule knowledge graph to generate intelligent decisions.

[0005] As a further technical solution, the dynamic maintenance rule knowledge graph construction module includes: A multi-source data layer is used for data source integration and data processing. The knowledge extraction layer is used to identify key entities, extract semantic relationships, and construct structured knowledge from multi-source heterogeneous data. It adopts a dual technical path of entity recognition model based on BERT-BiLSTM-CRF and relationship extraction method with attention mechanism to realize the transformation from raw data to knowledge triples. The storage management layer is used for knowledge storage and management, and uses the Cypher query language to realize node creation, relationship mapping, index optimization and logical verification, providing data support for upper-layer intelligent applications. The service application layer is used to transform the underlying knowledge graph into technical tools that can directly serve planning practices, supporting the dynamic evaluation and decision-making process of planning assessment.

[0006] As a further technical solution, the multi-source data layer is also used to execute the following instructions: It performs structured extraction and semantic annotation, data mapping and standardization, multi-level data fusion, and quality verification and iterative optimization.

[0007] As a further technical solution, the knowledge extraction layer is also used to execute the following instructions: Entity recognition is performed using a hybrid BERT-BiLSTM-CRF model. First, the BERT model is used to preprocess and extract features from the text in policy documents, urban health indicators, and socioeconomic data. Then, BiLSTM is used to capture long-term dependencies in the sequence. Finally, the CRF layer is used to optimize the prediction of the label sequence. Relation extraction: An attention mechanism is introduced to dynamically focus on the parts of the text that are related to relation judgment. At the same time, it supports multi-category relation classification and stores it in association with the applicable conditions defined in previous research to identify complex semantic relationships between entities. The process employs a dual-technology approach: first, entities are extracted from multi-source data using an entity recognition model; then, a relation extraction model is used to construct semantic relationships between entities, forming a preliminary "target-indicator-strategy" network structure. This outputs knowledge triples containing applicable condition metadata, providing standardized input for the storage management layer.

[0008] As a further technical solution, the storage management layer adopts the Neo4j native graph database, which matches the network structure characteristics of knowledge graphs ("entity-relationship-attribute"), and the underlying storage mechanism directly maps to the graph structure. Specifically, during the data transformation and mapping stage, the storage management layer converts the structured knowledge triples output by the knowledge extraction layer into Cypher statements, dynamically creates nodes and relationships, and indexes high-frequency access node types. Simultaneously, it utilizes Neo4j's ACID transaction characteristics to ensure the atomicity and consistency of data writes. The storage management layer also constructs a complete logical verification mechanism, implementing logical verification through Cypher and combining it with visualization tools to assist manual review, outputting a high-quality knowledge graph.

[0009] As a further technical solution, the service application layer is also used to execute the following instructions: Deploy open API services, build visualization display functions, develop semantic retrieval and intelligent recommendation functions, build intelligent reasoning and decision support capabilities, and integrate with large language models.

[0010] As a further technical solution, the three-dimensional index evaluation module also includes: The feature modeling unit is used to construct a multi-dimensional indicator feature matrix after acquiring urban health check data, based on the classification of three main functional areas: urbanized areas, major agricultural production areas, and key ecological areas, and combined with the evaluation connotations of three indicators: health, progress, and completion. It also uses an expert judgment knowledge base formed by expert surveys and the Delphi method to extract features and assign weights to urban health check data, forming a feature set as feature input for subsequent evaluation. The time series analysis unit is used to train and fit the changing trends of urban physical examination data over the years based on machine learning algorithm ensemble models, and to learn the evolution patterns and interrelationships of three different indicators—health, progress, and completion—over time. The evaluation output unit is used for dynamic evaluation of the features provided by the feature modeling unit based on the fitting results of the time series analysis unit.

[0011] As a further technical solution, the dynamic maintenance rule intelligent diagnosis module is also used to execute the following instructions: By integrating multi-source urban health check data and combining it with policy documents, industry standards, historical cases, and expert experience stored in the land and space planning knowledge graph, multi-dimensional feature integration and semantic mapping can be achieved. With health, progress, and completion as the core indicators, a deep neural network model is used to quantitatively model and learn nonlinear features of the functional coordination of the land and space system, the achievement of planning goals, and the efficiency of implementation. This generates dynamic diagnostic results of the urban planning implementation status and outputs intermediate data, including the evaluation values ​​of the three indicators and the rule matching level, to provide input support for subsequent decision-making.

[0012] As a further technical solution, the dynamic maintenance rule matching module is also used to execute the following instructions: Feature learning and nonlinear pattern mining are performed on historical sample data of health, progress, and completion.

[0013] According to one aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to implement the decision-making system for the physical examination, supervision and monitoring of land and space planning and dynamic maintenance based on "AI + knowledge graph".

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention first constructs a knowledge graph module based on historical multi-source heterogeneous data. Then, for the actually collected multi-source heterogeneous data, it conducts expert surveys and simulations to generate a smart diagnostic module that evaluates three-dimensional indicators. The evaluation results are input into the knowledge graph, and a dynamic maintenance rule matching module outputs dynamic maintenance rules. Together, these three components constitute a "planning, implementation, and dynamic maintenance intelligent agent." The knowledge graph module constructs the dynamic maintenance rule system, the smart diagnostic module dynamically diagnoses the evaluation results of the input data, and the dynamic maintenance rule matching module inputs the evaluation results into the knowledge graph, thereby intelligently generating a report. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of a decision-making system for the physical examination, supervision and monitoring of land and space planning and dynamic maintenance based on "AI + knowledge graph" provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0018] This invention provides a decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph," such as... Figure 1 As shown, it includes: The data acquisition and preprocessing module is used to collect multi-source heterogeneous data related to the implementation of land and space planning and to preprocess the data to form a standardized input dataset for modeling and analysis.

[0019] Specifically, the data acquisition and preprocessing module collects multi-source heterogeneous data related to the implementation of territorial spatial planning from urban physical examination data sources, national and local policy documents, industry standards, academic papers, best practices, and local planning databases. It then performs formatting, feature extraction, semantic annotation, and data cleaning on this data to form a standardized input dataset suitable for modeling and analysis. Taking the indicator of "number of nursing home beds per thousand elderly people" as an example, the module first retrieves urban physical examination data sources for a specific region or city and uploads the value of the number of nursing home beds per thousand elderly people from the urban physical examination data to the system. Simultaneously, it extracts relevant information related to the number of nursing home beds per thousand elderly people from national and local policy documents, industry standards, academic papers, best practices, and relevant policies from the National Bureau of Statistics and local civil affairs departments, providing a knowledge foundation and reference for the subsequent construction of a knowledge graph of territorial spatial planning physical examinations.

[0020] The dynamic maintenance rule knowledge graph construction module is used to construct a dynamic maintenance rule knowledge graph for the overall land space planning based on artificial intelligence technology (including deep learning models, natural language processing and semantic reasoning algorithms) and combined with two major categories of urban health check indicator systems: "spatial support system" and "central urban area planning". It integrates collected data with policy norms, planning knowledge and expert experience.

[0021] The "Central Urban Area Planning" indicator system comprises five dimensions: industrial space, blue-green space, housing security, underground space, and urban renewal. It covers indicators such as water consumption per 10,000 yuan of GDP, forest coverage rate, per capita park green space area, 5-minute walking coverage rate of park green space and squares, per capita urban housing area, per capita underground space area, and the proportion of existing land. These indicators reflect the optimization of the central urban area's industrial structure, improvement of the ecological environment, adequacy of housing supply, and the multi-functional utilization of space. The "Spatial Support System" indicator system includes dimensions such as public service facilities. It covers the number of hospital beds per 1,000 people, the number of elderly care beds per 1,000 people, the 15-minute walking coverage rate of community health service facilities, community middle schools, community elderly care facilities, community sports facilities, community cultural activity facilities, community primary schools, and the 30-minute walking coverage rate of administrative villages for tertiary hospitals. These indicators reflect the supporting capacity and fairness of urban spatial operations. By extracting knowledge, identifying entities, and constructing relationships, semantic associations are established between the above indicators and planning objectives, policy requirements, evaluation standards, and dynamic maintenance rules. This forms a knowledge graph that supports dynamic evaluation and intelligent decision-making in territorial spatial planning, enabling the automatic generation of semantic associations, causal logical reasoning, and dynamic maintenance rules.

[0022] Taking the indicator of "number of nursing home beds per thousand elderly people" as an example, the system models the semantic structure, hierarchical logical relationships, and strategy adaptation rules of this indicator through a knowledge graph construction module. Specifically, this includes: First, in the planning objective hierarchy, the system categorizes the indicator of "number of elderly care beds per thousand elderly people" under the primary objective of "fair accessibility of public service facilities" within the "spatial support system," setting the secondary planning objective as "increasing the number of elderly care beds per thousand elderly people." This hierarchical structure achieves a semantic mapping from overall spatial functional objectives to quantitative indicators, providing a structural foundation for subsequent intelligent reasoning.

[0023] Secondly, at the planning strategy level, the system automatically identifies multiple strategy nodes associated with the indicator based on natural language processing and semantic reasoning algorithms. These include governance strategies (such as "guiding social forces to participate in elderly care services," "encouraging private medical institutions," and "promoting bed liability insurance") and technology strategies (such as "dynamic calculation of aging population and bed demand," "deployment of smart bed management systems," and "development and application of bed occupancy rate prediction algorithms"). Each strategy node includes attributes such as strategy name, type, strategy content, responsible entity, and implementation difficulty.

[0024] Furthermore, based on indicator characteristics and regional attributes, the system constructs multi-dimensional semantic attributes for the "planning objectives - planning strategies" relationship edge, including: Urban functional zones (such as urbanized areas, major agricultural production areas, and key ecological functional zones); Urban economic level (high, medium, low); City population size (small, medium, large, extra-large); The number of nursing home beds per thousand elderly people (low, medium, high); The system also includes "Reasons for Applicability" and "Matching Degree" fields to describe the applicability and priority of the strategy for different city types. For example, the system infers that the strategy of "guiding social forces to participate in elderly care services" has a high matching degree with the goal of "increasing the number of elderly care beds per thousand elderly people," making it suitable for major agricultural production areas and key ecological function areas with medium to high economic levels and a medium to high number of elderly care beds.

[0025] Finally, the system generates a complete knowledge structure network in the Neo4j graph database: Each "used to achieve" relation edge carries rich conditional attributes and logical descriptions. The system can perform multi-hop semantic reasoning based on Cypher queries, such as automatically retrieving "elderly care bed improvement strategies that should be prioritized in areas with medium economic levels and accelerating aging populations," and returning applicable governance and technical measures.

[0026] Through this process, the knowledge graph not only realizes the semantic association and causal logic reasoning of the indicator of "number of elderly care beds per thousand elderly people", but also supports the dynamic rule generation and regionally differentiated strategy recommendation based on artificial intelligence algorithms, providing technical support for intelligent assessment and accurate decision-making in territorial spatial planning.

[0027] Specifically, the knowledge graph construction module for the dynamic maintenance rules of territorial spatial planning includes a multi-source data layer, a knowledge extraction layer, a storage management layer, and a service application layer, wherein: The multi-source data layer includes data source integration and data processing: it integrates four major categories of key data: policy and regulatory data, urban health check data, socio-economic statistical data, and spatial and monitoring data, providing multi-dimensional data support for planning and evaluation work. The core data processing flow includes data extraction, mapping, standardization, and quality control.

[0028] The knowledge extraction layer, as the core component of the knowledge graph construction for the land and space planning system, undertakes the mission of identifying key entities, extracting semantic relationships, and constructing structured knowledge from multi-source heterogeneous data. It adopts a dual technical approach of entity recognition model based on BERT-BiLSTM-CRF and relationship extraction method with attention mechanism, deeply integrating deep learning and natural language processing technologies to achieve intelligent transformation from raw data to knowledge triples.

[0029] The storage management layer uses Neo4j native graph database as the knowledge storage and management platform. It uses Cypher query language to realize node creation, relationship mapping, index optimization and logical verification to ensure the integrity, consistency and scalability of the knowledge graph and provide data support for upper-layer intelligent applications.

[0030] The service application layer, by providing open API services, visualization interfaces and semantic retrieval functions, transforms the underlying knowledge graph into a technical tool that can directly serve planning practices, supports the dynamic evaluation and decision-making process of planning assessment, and promotes the formation of an integrated intelligent planning system that combines "knowledge-analysis-decision".

[0031] Furthermore, the processing of the multi-source data layer includes: Perform structured extraction and semantic annotation: Use natural language processing technology to perform preliminary analysis of policy texts and physical examination indicators, identify key entities such as planning goals, quantitative indicators, and strategy types, and ensure consistency through the construction of a terminology database (such as unifying "elderly care facility coverage rate" in different documents into a standardized indicator name), thereby achieving data semantic parsing and terminology unification; Perform data mapping and standardization: Map multi-source data to a common data model (e.g., define data format through XML or JSON schema), unify statistical data units (e.g., standardize population data to "person / square kilometer"), match spatiotemporal scales (e.g., align economic data by year and administrative division), and perform numerical standardization (e.g., normalization to avoid differences in units), thereby achieving uniformity in the format, scale, and values ​​of multi-source data. Multi-level data fusion is implemented: In the model design phase, based on the top-level framework of the "Regulations", elements such as policy objectives, quantitative indicators, and implementation strategies are linked into a semantic network of "category-indicator-strategy". Data on the applicable conditions of strategies in the previous research (such as applicable functional areas, economic level thresholds, and reasons for implementation) are integrated, and the embedding of conditional attributes is achieved through entity-relationship modeling to build a data structure with close semantic association. Perform quality verification and iterative optimization: Eliminate duplicate and conflicting data through data cleaning rules (such as using consistency check algorithms to verify the rationality of indicator values), use visualization tools to initially verify data correlation, and at the same time, respond to data updates and schema evolution through iterative optimization to ensure that the merged data supports efficient querying and subsequent knowledge extraction.

[0032] Furthermore, the processing of the knowledge extraction layer includes: Entity identification is performed using a hybrid BERT-BiLSTM-CRF model. First, the BERT model is used to preprocess and extract features from the text in policy documents, urban health indicators, and socioeconomic data (using pre-trained language representations to capture lexical polysemy and contextual dependencies, such as accurately identifying "number of beds per thousand people" as a quantitative indicator entity). Then, BiLSTM is used to capture long-term dependencies in the sequence (such as distinguishing "elderly care facility coverage rate" as a whole entity). Finally, the CRF layer is used to optimize the label sequence prediction (avoiding incorrect label combinations of "planning goals - indicators"). Relation extraction is performed by introducing an attention mechanism to dynamically focus on parts of the text related to relation judgment (e.g., when extracting the "indicator-strategy" relationship, attention is paid to condition descriptions such as "applicable economic level" and "applicable functional area" to determine the support relationship between the "increase investment in public service facilities" strategy and the "fair accessibility of public service facilities" indicator). It also supports multi-category relation classification (relationships are divided into spatial strategies, technical strategies, governance strategies, etc.) and stores them in association with the applicable conditions defined in previous research (such as economic thresholds and functional area attributes) to accurately identify complex semantic relationships between entities and endow knowledge triples with rich attribute information. The process employs a dual-technology approach: First, an entity recognition model extracts entities such as planning objectives, quantitative indicators, and planning strategies from multi-source data, including the "Regulations for Urban Physical Examination and Evaluation of Territorial Spatial Planning" and urban physical examination results. Then, a relation extraction model constructs semantic relationships between entities to form a preliminary network structure of "objective-indicator-strategy". Finally, a high-quality knowledge triple containing metadata of applicable conditions (such as "Public service facility configuration, including indicators, number of beds per thousand people") is output, providing standardized input for import into the storage management layer (Neo4j database).

[0033] Furthermore, the processing of the storage management layer includes: Database selection: Neo4j native graph database is adopted, which is in line with the network structure characteristics of knowledge graphs of "entity-relationship-attribute". The underlying storage mechanism directly maps the graph structure, avoiding the performance bottleneck of multi-table join queries in traditional relational databases (such as when storing complex relationships of "target-indicator-strategy", it can respond to multi-hop queries in milliseconds, such as traversing from the target of "public service facility configuration" to all related indicators and applicable strategies). Implementation technology: In the data transformation and mapping stage, the structured triples output by the knowledge extraction layer are transformed into Cypher statements, and nodes and relationships are dynamically created (such as creating a node for the "number of beds per thousand people" indicator, establishing a "contains indicator" relationship edge with the "public service facility configuration" target, and adding attributes such as applicable functional areas and economic level thresholds to the relationship edge). Indexes are built for high-frequency access node types such as planning targets and quantitative indicators to improve query performance. At the same time, Neo4j's ACID transaction characteristics are used to ensure the atomicity and consistency of data writing and avoid data redundancy or conflicts. Perform quality verification: Build a complete logical verification mechanism, use Cypher queries to detect issues such as isolated nodes (unrelated entities), circular dependencies, and missing attributes, combine visualization tools to assist manual review, strictly follow the requirements of the "quality verification" process, and output a high-quality knowledge graph.

[0034] Furthermore, the processing of the service application layer includes: Deploy open API services: Provide a complete graph query interface, supporting database access and invocation via Neo4j's local URL (localhost:7687) and authentication password; the interface encapsulates entity relationship query, conditional filtering query, time series analysis query and other functions, and adopts a RESTful architecture to provide standardized data interfaces to achieve seamless integration with business systems; The system enables visualization and display capabilities: it presents the knowledge graph structure and content through a graphical interface, providing view modes such as network topology diagrams, geospatial visualization (integrated with GIS systems), and multi-dimensional drill-down analysis; it not only displays the complex relationships between "goals, indicators, and strategies" but also supports multi-level drill-down analysis from macro goals to specific indicators. Develop semantic retrieval and intelligent recommendation functions: Based on the semantic understanding capabilities of knowledge graphs, provide natural language query and association reasoning services (e.g., when a user queries "how to improve the coverage rate of elderly care facilities in economically developed areas", the system identifies the semantic association of entities and returns relevant indicators, strategies and applicable conditions); the strategy recommendation engine automatically recommends appropriate planning strategies based on parameters such as regional characteristics and development level; Building intelligent reasoning and decision support capabilities: As the core value of knowledge graphs, it provides a rule-based reasoning engine (based on dynamically maintained rules generated by machine learning to achieve automatic reasoning and strategy generation, providing decision-making basis and reasoning path) and multi-scenario simulation and deduction functions (supporting simulation and comparative analysis of planning effects under different conditions), enhancing the transparency and credibility of AI decision-making; Integration of knowledge graphs with large language models: By combining knowledge graphs with large language models centered on DeepSeek, the application boundaries of knowledge graphs are expanded through natural language interaction interfaces, intelligent report generation, knowledge discovery and mining, etc., forming the underlying knowledge foundation of the intelligent decision-making agent for dynamic maintenance of land space physical examination, thus solving the problem of weak knowledge transformation capabilities.

[0035] The three-dimensional indicator evaluation module is used to dynamically evaluate urban physical examination data based on the principle of three-dimensional model construction after acquiring actual multi-source heterogeneous data. It uses feature modeling and time series analysis.

[0036] Feature modeling is based on three main functional areas: urbanized areas, agricultural production areas, and key ecological areas. It combines the evaluation criteria of health, progress, and completion, utilizing an expert judgment knowledge base formed by expert surveys and the Delphi method to extract features and assign weights to urban health checkup data. This establishes a feature matrix containing multi-dimensional factors such as economic level, population density, infrastructure level, and spatial structure. Time-series analysis, based on an integrated model of machine learning algorithms (including gradient boosting decision trees, random forests, and extreme gradient boosting algorithms), trains and fits the changing trends of urban health checkup data over the years. It learns the evolutionary patterns of different indicators over time and their interrelationships, thereby dynamically predicting and evaluating newly input urban health checkup data, and calculating and outputting the three quantitative results of health, progress, and completion.

[0037] Taking the indicator of "number of nursing home beds per thousand elderly people" as an example, the system uses a three-dimensional indicator evaluation module to assess the health, completion, and progress of this indicator. Specifically, this includes: First, in feature modeling, the main functional areas of county-level administrative regions are identified based on urban health check data.

[0038] Secondly, clarify the evaluation criteria for health, progress, and completion: Health status (number of nursing home beds per thousand elderly people in the current year / three-quarters of the number of nursing home beds per thousand elderly people in a certain province or county in the current year); Completion rate (the difference between the number of elderly care beds per thousand elderly people in the current year and the previous year, divided by the number of elderly care beds per thousand elderly people in the previous year). Progress rate (number of nursing home beds per thousand elderly people in the current year / standard value of number of nursing home beds per thousand elderly people in a certain province in the current year).

[0039] Furthermore, based on the evaluation criteria of health, progress, and completion, a multi-dimensional indicator feature matrix is ​​constructed, including: Urban functional zones (such as urbanized areas, major agricultural production areas, and key ecological functional zones); Urban economic level (GDP); Urban population size (urban resident population density). Urban infrastructure level (density of urban road network). Furthermore, based on the aforementioned expert survey using the Delphi method, experts assessed the health, completion, and progress of elderly care beds per thousand elderly people in various counties of a province, taking into account the influence of multi-dimensional indicator characteristic moments, using indicators of health, completion, and progress. Through multiple rounds of anonymous expert consultation, feedback collection, and iterative revision, consensus was ultimately reached by converging group opinions.

[0040] Finally, through the time-series analysis unit and the evaluation output unit, three quantitative results—health, progress, and completion—are generated. The comprehensive scores of each indicator are then normalized and graded, outputting the three-dimensional evaluation results of the city's physical examination. See Table 1.

[0041] Table 1. Evaluation Results of the Three-Dimensional Indicators of Urban Physical Examination

[0042]

[0043]

[0044] .

[0045] Specifically, the three-dimensional index evaluation module includes: The feature modeling unit, after acquiring urban health checkup indicator data, constructs a multi-dimensional indicator feature matrix based on three main functional areas: urbanized areas, major agricultural production areas, and key ecological areas, combined with the evaluation connotations of three indicators: health, progress, and completion. The feature modeling unit utilizes an expert judgment knowledge base formed by expert surveys and the Delphi method to extract features and assign weights to the urban health checkup data, forming a feature set that includes factors such as economic development level, population density, infrastructure level, spatial structure, and ecological constraints, providing feature input for subsequent evaluation. The time series analysis unit is used to perform time series modeling and trend learning on urban physical examination data over the years based on machine learning algorithm ensemble models (including gradient boosting decision tree, random forest and extreme gradient boosting algorithm). By training and fitting the evolution law and correlation coupling relationship of different indicators in the time dimension, it can dynamically predict and evaluate the newly input urban physical examination data and generate three quantitative results: health, progress and completion. The evaluation output unit is used to normalize and classify the comprehensive scores of each indicator based on the results of feature modeling and time series analysis, output the evaluation results of the three-dimensional indicators of urban health check, and use the evaluation results as input data for the subsequent dynamic maintenance rule knowledge graph construction module and the planning implementation intelligent diagnosis module, so as to realize the quantitative evaluation and intelligent analysis of the implementation status of territorial spatial planning.

[0046] The dynamic maintenance rule intelligent diagnosis module is used to perform multi-dimensional feature modeling and dynamic diagnosis on the evaluation results output by the three-dimensional index evaluation module.

[0047] The intelligent diagnostic module for dynamic maintenance rules is based on a fusion mechanism of knowledge graphs, expert surveys, and deep learning simulations. It performs structured language conversion of expert knowledge, multi-dimensional feature modeling, and dynamic diagnosis. Its construction process includes: classifying development scenarios into different economic levels, development scales, and city types; calculating three core indicators based on time-series urban health check data: health (compared to similar cities), progress (compared to the previous year), and completion (compared to the completion of planning goals); generating numerous virtual samples to participate in expert surveys; recording in detail the diagnostic process of experts judging the "good" or "bad" of the indicators and the process of generating planning and maintenance suggestions based on the diagnoses; and integrating the features of the expert survey results and mapping the computer language structure into an expert think tank using computer language. Furthermore, it uses deep learning to quantitatively model and learn the features of the expert think tank, employing multiple types of deep learning simulations and comparing and selecting the model with the best simulation effect. Thus, after inputting health check data, it outputs three-dimensional dynamic diagnostic conclusions (which can also be considered dynamic maintenance rules) and intermediate data matching the dynamic maintenance rules, providing input support for subsequent decision-making modules.

[0048] Specifically, for three types of regions—urbanized areas, major agricultural production areas, and key ecological areas—a questionnaire survey was conducted to evaluate the distribution range, excellent values, and warning values ​​of 11 public service facilities, such as the number of hospital beds per thousand people, in different regions. At the same time, experts conducted a status assessment of five virtual sample cities in each type of region based on four reference indicators: GDP, resident population, urban population density, and road network density, and determined the health, progress, and completion levels of each indicator.

[0049] Furthermore, the construction and operation of the AI ​​reasoning simulation agent are carried out: with "data-driven - regional reasoning - hierarchical judgment - linkage verification" as the core mechanism, and questionnaire data of three main functional areas (urbanized areas, major agricultural production areas, and key ecological areas) as the training basis, the deep learning model learns the distribution patterns, excellent / warning values ​​of 11 public service facility indicators (including the number of medical and health institution beds per thousand people) and the linkage characteristics of 4 reference indicators to form a reasoning knowledge base; during reasoning, the differentiated evaluation rules are automatically called according to the region, and after preprocessing the virtual sample city data, the health level, progress level, and quantified score are calculated and then mapped to six levels of status: "excellent", "good", "relatively good", "average", "poor" and "very poor".

[0050] When using deep learning to quantitatively model and learn features of expert think tanks, and employing multiple types of deep learning simulations and comparing them to select the model with the best simulation performance, the neural network model used is as follows: Input and output are defined as follows: Input layer: 3 independent variables; x1: Health Level; x2: Completion Rate; x3: Progress Level; Output layer: 1 dependent variable - Dynamic Maintenance Rule.

[0051] Forward propagation formula (where w are weight values): 1. Input layer to hidden layer 1: z1= x1·w11 + x2·w12 + x3·w13 + b1 a1 = LeakyReLU(z1) a1_drop = Dropout(a1, rate=0.25) 2. From hidden layer 1 to hidden layer 2: z2 = a1_drop·W2 + b2 a2 = LeakyReLU(z2) a2_drop = Dropout(a2, rate=0.2) 3. From hidden layer 2 to output layer z3 = a2_drop·W3 + b3 y_pred = Sigmoid(z3) (assuming the dynamically maintained rules are a binary classification problem) or Softmax(z3) (if it is a multi-class classification problem).

[0052] Loss function: Binary classification case: Binary cross-entropy loss function L = -[y_true·log(y_pred) + (1-y_true)·log(1-y_pred)] + λ·∑(||W||²); Multi-class classification: Multi-class cross-entropy loss function L = -∑(y_true·log(y_pred)) + λ·∑(||W||²) Where λ is the L2 regularization coefficient.

[0053] Explanation of variable meanings: Health (x1): The current health status indicator of the system or entity, ranging from [0,1]; Completion (x2): Percentage of task or project completed, ranging from [0,1]; Progress (x3): The rate of progress of the plan's execution, ranging from [0,1]; Dynamic maintenance rules (y_pred): Maintenance strategy rules output based on the above three indicators.

[0054] The dynamic maintenance rule matching module is used to match the diagnostic results output by the dynamic maintenance rule intelligent diagnosis module with the constructed dynamic maintenance rule knowledge graph to generate intelligent decisions. Based on a deep neural network (DNN) model, the dynamic maintenance rule matching module intelligently analyzes and recognizes the quantitative results of the three-dimensional indicators. According to a preset dynamic maintenance evaluation system, it automatically matches the urban planning implementation status to six categories of dynamic maintenance rules to generate urban planning dynamic maintenance levels and corresponding strategies.

[0055] The "intelligent analysis and pattern recognition" is achieved by performing feature learning and nonlinear pattern mining on historical sample data of three core indicators: health, progress, and completion. It adopts a DNN model with a multilayer perceptron structure, uses the three indicators as input layer features, performs high-order feature mapping through a nonlinear activation function of two fully connected hidden layers, and finally generates a six-dimensional probability distribution in the output layer through the Softmax function to achieve automatic classification of the urban planning implementation status.

[0056] The "pre-defined dynamic maintenance evaluation system" comprises three dimensions: health, progress, and completion. By comparing the deviations of these indicators from policy standards, historical progress, and planning goals, it comprehensively assesses the overall state of urban planning implementation, resulting in six levels: excellent, good, relatively good, average, poor, and very poor. The "urban planning implementation status" is derived from the evaluation results calculated based on the "three-dimensional" model, reflecting the city's health level, goal completion rate, and progress rate during planning execution. The "six types of dynamic maintenance rules" correspond to six core maintenance strategies extracted from knowledge graphs and refined through artificial intelligence clustering: excellent (promoted as a model); good (incentive mechanisms encouraged for continuous improvement); relatively good (requiring targeted maintenance improvement plans); average (phased assessment and regular evaluation); poor (included in key research and rectification plans); and very poor (identifying responsible parties and establishing a special rectification team).

[0057] The six types of dynamic maintenance rules are interconnected with the dynamic maintenance rule knowledge graph. The knowledge graph, as a semantic foundation layer, stores policy norms, indicator thresholds, and expert knowledge, providing training samples and decision semantic constraints for the DNN model. This enables automatic matching and intelligent decision output of urban planning status based on three-dimensional indicators to the six types of maintenance rules. According to the preset dynamic maintenance evaluation system, the urban planning implementation status is automatically matched to the six types of dynamic maintenance rules to generate urban planning dynamic maintenance levels and corresponding strategies.

[0058] Taking the number of elderly care beds per thousand elderly people as an example, the results of intelligent diagnosis based on dynamic maintenance rules are generated. See Table 2.

[0059] Table 2 Results of Intelligent Diagnosis of Dynamic Maintenance Rules

[0060]

[0061]

[0062]

[0063] .

[0064] Specifically, based on the above three-dimensional model, a systematic evaluation of 11 key public service indicators was conducted to obtain quantitative scores for the "health," "completion," and "progress" of 11 public service indicators in various cities of a certain province. Based on DeepSeek's large language model technology, through the LLM-driven "knowledge distillation" process, 12 general maintenance rules are condensed into 6 core rule systems, forming the basic data system for model input and output; Construct a dynamic maintenance rule automatic classification model based on a deep neural network (DNN). Adopt a multi-layer perceptron architecture, and determine the final topological structure through system hyperparameter optimization. There are 2 fully connected layers in total. The input layer is the quantitative scores of three core indicators, namely "health degree, progress degree, and completion degree". The hidden layer performs high-order feature transformation and abstraction through a network structure with a non-linear activation function. The output layer generates a six-dimensional probability distribution corresponding to 6 major core rules through the Softmax function to achieve classification decision-making; Use the backpropagation algorithm to carry out model training. The learning rate of the optimizer is set to 0.001. The training process goes through 1000 iteration cycles to ensure the full convergence of model parameters. After performance verification, the optimized DNN model has an accuracy rate of 90% in the dynamic maintenance rule classification task. It can automatically learn the non-linear mapping relationship between input indicators and complex rule patterns, and achieve accurate and automatic classification into 6 predetermined categories based on subtle differences in rule features.

[0065] Preferably, the embodiment of the present invention further provides a report generation module: which is used to automatically generate a dynamic maintenance suggestion report for the municipal and county-level territorial space planning according to the output result of the dynamic maintenance rule matching module. The report includes local development priorities, index threshold adjustment, and planning management optimization suggestions.

[0066] Taking the number of elderly care beds per thousand elderly people in County A as an example, generate a dynamic maintenance suggestion report for the territorial space planning of County A from four aspects: problem diagnosis, provincial regulation suggestions, dynamic maintenance of the implementation of the municipal and county-level planning, and special suggestions. For example: I. Problem diagnosis As a key ecological area, the number of elderly care beds per thousand elderly people in County A is at a "medium" level, and the dynamic maintenance rules are good. However, considering the characteristics of the elderly service needs in the key ecological area and the goal of "continuous improvement", there are still three core optimization shortboards. First, the adaptability of the elderly care bed structure to the care needs of the elderly is insufficient; second, the spatial layout of the elderly care beds does not match well with the terrain characteristics of the ecological area; third, the professionalism of the supporting medical services for the elderly care beds is insufficient.

[0067] II. Provincial regulation suggestions Combined with the positioning of County A as a key ecological area and the dynamic maintenance rules (good, need to strengthen incentives), the provincial level provides support from three aspects: policy coordination, resource tilt, and mechanism establishment. First, set up a special incentive fund for elderly care and medical care in key ecological areas and clarify the incentive standards; second, introduce a targeted support policy for elderly care and medical care talents in key ecological areas; third, build a provincial-level regional elderly care and medical care cooperation platform, etc.

[0068] III. Municipal and county implementation and maintenance Building upon County A's existing sound dynamic maintenance foundation, and focusing on the specific needs of the elderly in key ecological areas, the county aims to transform its elderly care bed development from simply increasing quantity to prioritizing both quality and coverage. This involves: 1) optimizing the structure and function of elderly care beds; 2) improving the spatial layout of elderly care and medical resources; and 3) deepening regional cooperation in elderly care and medical services.

[0069] IV. Specific Recommendations The A County Elderly Care and Medical Care Talent Targeted Training and Incentive Program, the Key Ecological Area Elderly Care and Medical Care Resource Targeted Downward Transfer Project, and the Cross-Regional Elderly Care and Medical Care Collaborative Service Mechanism, etc.

[0070] The intelligent decision-making system for planning management provided by this invention integrates the output of the intelligent diagnosis module with a dynamic maintenance rule knowledge graph to form a closed-loop logic of "monitoring-evaluation-decision". Under the constraints of the pre-set early warning rules in the dynamic maintenance rule knowledge graph, it calls the local Large Language Model (LLM) to integrate the urban health check result data from the intelligent diagnosis module with the structured rules and unstructured text information (including indicator interpretation, planning outline and expert opinions) in the knowledge graph. Through semantic reasoning and strategy generation mechanisms, it achieves deep coupling of "rules-data-strategy" and automatically generates an urban health check maintenance rule report. The report includes four dimensions: problem diagnosis, provincial control suggestions, city and county implementation maintenance plans and special suggestions. It has data support, strategy feasibility and rule guidance, thereby realizing dynamic prediction of the implementation status of land and space planning, strategy output and self-learning optimization.

[0071] Specifically, the system of this invention integrates facility maintenance rules (including operation and maintenance thresholds, response time, and maintenance frequency) identified by deep learning models with existing maintenance strategies (aggregating historical cases, regional standards, and industry norms) in the Neo4j knowledge graph to form a basic data set to be matched; through three-dimensional matching of "facility type - problem scenario - management objective", a logical mapping between rules and strategies is established, and the consistency between problem scenarios and maintenance standards is verified and ensured; after the integration and mapping are completed, the structured knowledge is imported into the local large language model to provide unified knowledge input for subsequent intelligent applications.

[0072] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to describe the relevant electronic device. For this purpose, an embodiment of the invention provides an electronic device comprising: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to implement all or part of the modules of the system provided in the foregoing system embodiments.

[0073] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0074] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0075] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph", characterized in that, include: The data acquisition and preprocessing module is used to collect multi-source heterogeneous data related to the implementation of land and space planning and to preprocess the data to form a standardized input dataset for modeling and analysis. The dynamic maintenance rule knowledge graph construction module is used to construct a dynamic maintenance rule knowledge graph for the overall land space planning by integrating historical multi-source heterogeneous data with policy norms, planning knowledge and expert experience, based on artificial intelligence technology and combining two major categories of urban health check indicator systems: spatial support system and central urban area planning. The dynamic maintenance rule knowledge graph construction module includes: a multi-source data layer for data source integration and processing; a knowledge extraction layer for identifying key entities, extracting semantic relationships, and constructing structured knowledge from multi-source heterogeneous data, employing a dual technical path of an entity recognition model based on BERT-BiLSTM-CRF and a relationship extraction method incorporating an attention mechanism to achieve the transformation from raw data to knowledge triples; a storage management layer for knowledge storage and management, and using the Cypher query language to achieve node creation, relationship mapping, index optimization, and logical verification, providing data support for upper-layer intelligent applications; and a service application layer for transforming the underlying knowledge graph into technical tools that can directly serve planning practices, supporting the dynamic evaluation and decision-making process of planning assessment. The three-dimensional indicator evaluation module is used to dynamically evaluate urban health checkup data based on the three-dimensional model construction principle after acquiring actual multi-source heterogeneous data. This module utilizes feature modeling and time-series analysis. The module includes: a feature modeling unit, which, after acquiring urban health checkup data, constructs a multi-dimensional indicator feature matrix based on three main functional areas: urbanized areas, major agricultural production areas, and key ecological areas, combined with the evaluation connotations of three indicators: health, progress, and completion. It also uses an expert judgment knowledge base formed by expert surveys and the Delphi method to extract features and assign weights to the urban health checkup data, forming a feature set as the feature input for subsequent evaluation. A time-series analysis unit is used to train and fit the changing trends of urban health checkup data over the years based on a machine learning algorithm ensemble model, learning the evolutionary patterns and interrelationships of the three different indicators (health, progress, and completion) over time. Finally, an evaluation output unit is used to dynamically evaluate the features provided by the feature modeling unit based on the fitting results of the time-series analysis unit. The dynamic maintenance rule intelligent diagnosis module is used to perform multi-dimensional feature modeling and dynamic diagnosis on the evaluation results output by the three-dimensional indicator evaluation module. This module also executes the following instructions: integrating multi-source urban health check data and combining it with policy documents, industry standards, historical cases, and expert experience stored in the land and space planning knowledge graph to achieve multi-dimensional feature integration and semantic mapping; using health, progress, and completion as the core indicators, it quantitatively models and learns nonlinear features of the functional coordination of the land and space system, the achievement of planning goals, and the efficiency of progress through a deep neural network model, generating dynamic diagnostic results of the urban planning implementation status, and outputting intermediate data including the three-dimensional indicator evaluation values ​​and rule matching levels to provide input support for subsequent decision-making. The dynamic maintenance rule matching module is used to match the diagnostic results output by the dynamic maintenance rule intelligent diagnosis module with the constructed dynamic maintenance rule knowledge graph to generate intelligent decisions.

2. The decision-making system for land spatial planning health check-up, supervision and monitoring, and dynamic maintenance based on "AI + knowledge graph" as described in claim 1, characterized in that, The multi-source data layer is also used to execute the following instructions: It performs structured extraction and semantic annotation, data mapping and standardization, multi-level data fusion, and quality verification and iterative optimization.

3. The decision-making system for land spatial planning health check-up, supervision and monitoring, and dynamic maintenance based on "AI + knowledge graph" as described in claim 1, characterized in that, The knowledge extraction layer is also used to execute the following instructions: Entity recognition is performed using a hybrid BERT-BiLSTM-CRF model. First, the BERT model is used to preprocess and extract features from the text in policy documents, urban health indicators, and socioeconomic data. Then, BiLSTM is used to capture long-term dependencies in the sequence. Finally, the CRF layer is used to optimize the prediction of the label sequence. Relation extraction: An attention mechanism is introduced to dynamically focus on the parts of the text that are related to relation judgment. At the same time, it supports multi-category relation classification and stores it in association with the applicable conditions defined in previous research to identify complex semantic relationships between entities. The process employs a dual-technology approach: first, entities are extracted from multi-source data using an entity recognition model; then, a relation extraction model is used to construct semantic relationships between entities, forming a preliminary "target-indicator-strategy" network structure. This outputs knowledge triples containing applicable condition metadata, providing standardized input for the storage management layer.

4. The decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph" as described in claim 1, characterized in that, The storage management layer employs Neo4j, a native graph database, which aligns with the mesh structure of knowledge graphs ("entity-relationship-attribute"). The underlying storage mechanism directly maps to the graph structure. During the data transformation and mapping phase, the storage management layer converts the structured knowledge triples output from the knowledge extraction layer into Cypher statements, dynamically creating nodes and relationships, indexing frequently accessed node types, and leveraging Neo4j's ACID transaction characteristics to ensure the atomicity and consistency of data writes. Furthermore, the storage management layer constructs a complete logical verification mechanism, implementing logical verification through Cypher and combining it with visualization tools to assist manual review, outputting a high-quality knowledge graph.

5. The decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph" as described in claim 1, characterized in that, The service application layer is also used to execute the following instructions: Deploy open API services, build visualization display functions, develop semantic retrieval and intelligent recommendation functions, build intelligent reasoning and decision support capabilities, and integrate with large language models.

6. The decision-making system for the physical examination, supervision, monitoring, and dynamic maintenance of land and space planning based on "AI + knowledge graph" as described in claim 1, characterized in that, The dynamic maintenance rule matching module is also used to execute the following instructions: Feature learning and nonlinear pattern mining are performed on historical sample data of health, progress, and completion.

7. An electronic device, characterized in that, The system includes a memory and a processor. The memory stores program instructions that are executed by the processor. The processor invokes the program instructions to implement the decision-making system for the physical examination, supervision and monitoring of land and space planning and dynamic maintenance based on "AI + knowledge graph" as described in any one of claims 1 to 6.

Images