Cultural constraint rule coding and conflict graph solving method, system and device for community intelligent protection of human settlement heritage and storage medium

Abstract

The application discloses a cultural constraint rule coding and conflict graph solving method, system and device for intelligent protection of human settlement heritage community and a storage medium, and relates to the technical field of digital heritage protection and rule reasoning. The method comprises the following steps: obtaining multi-source heterogeneous data, performing standardization pretreatment, constructing an environmental state vector to uniformly represent the community protection state, extracting entities and relationships from a regulation text through natural language processing to construct a cultural knowledge graph, converting constraint clauses into machine executable condition-action constraint expressions coded in an abstract syntax tree to form a rule base, constructing a rule driven decision model for multiple subjects, performing legality clipping on candidate actions to generate a legal action space, generating a joint action combination and constructing a conflict graph, finally forming a target protection scheme in the form of a space unit level control instruction set, and performing incremental updating according to execution feedback to form a closed loop decision. The application realizes automatic coding and execution of cultural constraints and computable collaborative solving of multiple subjects.

Description

Technical Field

[0001] This invention belongs to the fields of digital heritage protection, knowledge graph and rule reasoning technology, and specifically relates to a method, system, device and storage medium for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities. Background Technology

[0002] Human settlement heritage communities refer to living communities that combine heritage protection with the life and livelihood of residents, including historic districts, traditional villages, ancient towns, and traditional residential settlements. The protection and management of these communities typically involves multiple objectives simultaneously, including the preservation of historical buildings, ensuring residents' quality of life, maintaining commercial order, regulating visitor capacity, and ensuring the continuation of traditional culture. With the widespread application of digital technology in the field of cultural heritage protection, technologies such as knowledge graphs, multi-agent systems, and rule-based reasoning are gradually being introduced into protection and management practices.

[0003] However, existing technologies still have significant shortcomings in this type of scenario. These shortcomings are mainly reflected in the following aspects:

[0004] In terms of data perception and status representation, human settlement heritage communities involve multi-source heterogeneous data, including images, videos, environmental monitoring, structural monitoring, community feedback, operational activities, management rules, and expert annotations. These data sources are complex, with varying granularities and update frequencies. Currently, there is a lack of a unified way to represent the environmental status, making it difficult to incorporate the status of historical buildings, environmental changes, visitor flow, and community feedback into a unified decision-making framework.

[0005] In terms of multi-stakeholder collaboration, residents, merchants, managers, and experts have different objectives: residents focus on living environment and convenience, merchants focus on operating profits and customer traffic, managers focus on protection effectiveness and budget control, and experts focus on authenticity, integrity, and security. Existing solutions typically rely on manual negotiation and meeting discussions, lacking a computationally achievable multi-stakeholder joint solution mechanism.

[0006] In terms of the implementation of cultural rules, the requirements for cultural heritage protection are mostly in the form of legal texts, guidelines, and expert experience, such as the "Regulations on the Protection of Famous Historical and Cultural Cities, Towns and Villages", local protection and management measures, and style control guidelines. These textual requirements are difficult to directly translate into machine-executable constraints, which often requires manual review of the compliance of candidate protection plans afterward.

[0007] Regarding the dynamic updating of the scheme, most existing protection schemes are fixed schemes based on static evaluation results, making it difficult to dynamically adjust and update them in a closed loop according to real-time monitoring data, scheme implementation feedback, and newly added protection rules. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide a method, system, device and storage medium for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities, which solves the problems of inconsistent environmental state expression, difficulty in multi-subject collaboration, difficulty in machine execution of cultural rules and difficulty in dynamic updating of protection schemes in the prior art.

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0010] A method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities, including:

[0011] Multi-source heterogeneous data of human settlement heritage communities were acquired, and standardized preprocessing was performed on the acquired multi-source heterogeneous data to obtain a standardized dataset including numerical data, image and video data;

[0012] Based on a standardized dataset, numerical data is normalized and mapped to a preset range, and target detection and feature extraction are performed on image and video data to construct an environmental state vector with spatial units as the granularity, which is used to characterize the community protection status.

[0013] Based on the texts of cultural heritage protection regulations, local management rules, expert experience rules, and community consultation rules, a cultural knowledge graph is constructed by extracting entities and relationships through natural language processing. The constraint clauses are then transformed into machine-executable condition-action constraint expressions to generate a cultural constraint rule library.

[0014] To construct rule-driven decision-making models for multiple stakeholders involved in human settlement heritage communities, each decision-making model uses the current environmental state vector as the observation input, a predefined action parameter template as the candidate action, and a utility evaluation function as the basis for quantifying action preferences.

[0015] The candidate action parameters of each decision model are logically compared with the condition-action constraint expressions one by one, and legality pruning is performed to generate the legal action space of each decision model.

[0016] Based on the legal action space of each decision model, a joint action combination is generated. A conflict graph is constructed with actions as nodes and resource occupation relationship and spatiotemporal interference relationship as edges. Connected conflict components in the conflict graph are detected to identify and eliminate joint action combinations with conflicts. The target protection scheme is determined after sorting the remaining joint action combinations.

[0017] The environmental state vector is recalculated based on the monitoring feedback data after the implementation of the target protection scheme. When the change exceeds the preset threshold, a new round of decision-making process is triggered, and the rule parameters in the cultural constraint rule base are incrementally updated based on the statistics of violation events and the deviation of scheme implementation.

[0018] The multi-source heterogeneous data includes image and video data, environmental monitoring data, structural monitoring data, community feedback data, business activity data, management rule data, and expert-annotated data. Image and video data originates from monitoring equipment; environmental monitoring data includes noise, temperature, humidity, air quality, illuminance, and pedestrian density; structural monitoring data includes crack width, settlement, tilt value, and vibration anomalies; community feedback data includes resident satisfaction ratings, written opinions, complaint records, and voting records; business activity data includes merchant customer flow statistics, operating hours, business types, and records of high-noise business activities; management rule data includes protected area boundaries, cultural relic protection levels, landscape control requirements, business restrictions, and repair requirements; and expert-annotated data includes risk area annotations, value element annotations, protection recommendations, and rule revision recommendations.

[0019] The environmental state vector comprises seven dimensions: heritage health index, pedestrian carrying capacity index, noise impact index, environmental quality index, community satisfaction index, cultural activity index, and compliance risk index. The value range of each dimension is mapped to a preset interval. Among them, the heritage health index reflects the health of the building structure, the pedestrian carrying capacity index reflects the load of population density relative to the carrying capacity threshold, the noise impact index reflects the degree of noise interference to the community, the environmental quality index reflects the overall air quality and temperature and humidity, the community satisfaction index reflects residents' feedback, the cultural activity index reflects the degree of participation in traditional activities and exhibitions, and the compliance risk index reflects the degree of occurrence of violations.

[0020] The cultural knowledge graph includes a set of entity nodes, a set of relation edges, and a set of relation types. The set of entity nodes includes heritage building entities, street and alley space entities, traditional activity entities, main entities, protection measure entities, risk event entities, and rule and clause entities.

[0021] The rule priority in the condition-action constraint expression is determined by the source of the rule, from high to low: rules of national laws and regulations, rules of local management measures, rules of expert experience, and rules of community consultation. Within the same priority, the rule with more entity attribute dimensions involved in the triggering condition has a higher priority.

[0022] The specific process for performing legality-based cropping is as follows:

[0023] Candidate actions from each decision model are matched one by one with the condition-action constraint expressions in the cultural constraint rule base. Based on the matching results, candidate actions are divided into a set of allowed actions, a set of actions to be corrected, and a set of prohibited actions. Specifically, when a candidate action satisfies any prohibited condition-action constraint expression, it is included in the prohibited action set. When a candidate action does not satisfy a prohibited condition but satisfies a corrected condition-action constraint expression, it is included in the set of actions to be corrected. When a candidate action does not trigger any prohibited or corrected condition-action constraint expressions, it is included in the allowed action set. When multiple rules apply to the same candidate action, the prohibited result takes precedence over the corrected result, and the corrected result takes precedence over the allowed result.

[0024] Generating the legal action space for each decision model also includes performing at least one of the following correction methods on the actions in the action set to be corrected: adjusting the action execution area, adjusting the action execution time period, adjusting the action resource quota, replacing the action implementation object, and adding supplementary protection measures; performing legality pruning again on the corrected actions, merging the corrected actions through the second pruning with the actions in the allowed action set, and generating the legal action space for each decision model.

[0025] Detecting connected conflict components in a conflict graph includes:

[0026] Resource conflict detection is used to detect conflicts where multiple actions simultaneously occupy the same budget, equipment, or construction resources.

[0027] Spatial conflict detection is used to detect conflicts where multiple actions are performed in the same spatial unit and cannot be performed in parallel.

[0028] Time conflict detection is used to detect conflicts where multiple actions occur within the same time window and interfere with each other.

[0029] Rule conflict detection is used to detect conflicts where a combination of actions as a whole violates a rule base of cultural constraints.

[0030] Target conflict detection is used to detect conflicts where a combination of actions causes the utility of a decision model to fall below a preset lower limit.

[0031] The target protection scheme is output in the form of a spatial unit-level control instruction set, which includes at least one of the following: spatial unit-level equipment scheduling instructions, regional pedestrian flow threshold setting parameters, monitoring equipment alarm threshold configuration parameters, repair project construction constraint parameter table, budget resource allocation parameters, and monitoring task issuance parameters.

[0032] Candidate actions for each decision model are generated in the following way:

[0033] The values ​​of each dimension of the current environmental state vector, the subject type identifier, the subject's historical action records, the attribute information of the spatial unit to which it belongs, and the cultural constraint prompt information are input into the large language model as structured prompt words. The large language model outputs a structured list of candidate action parameters according to a predefined action parameter template format. After format verification, it is used as the candidate action set of the decision model.

[0034] The incremental update includes at least one of the following operations:

[0035] The environmental state vector is recalculated based on the latest collected monitoring data. When the change in any dimension exceeds the preset change threshold, a new round of decision-making process is triggered.

[0036] Based on the frequency of newly added violations, the priority of the corresponding rules in the cultural constraint rule base will be increased by a preset amount;

[0037] Based on the newly added expert-annotated data, entity nodes and relationship edges are added to the cultural knowledge graph, and entity attribute values ​​are updated.

[0038] Based on the deviation between the actual results of the scheme and the expected results, the ranking weight parameters are updated through iterative optimization and normalized projection is performed after the update.

[0039] Based on the failure rate of the re-correction action, the scope of application of the correction instructions for the corresponding rules in the cultural constraint rule base is expanded.

[0040] The newly added regulatory clauses and new negotiation rules are encoded according to the condition-action constraint expression format and then written into the cultural constraint rule library.

[0041] A system for coding and resolving conflict graphs of cultural constraint rules for the protection of human settlement heritage communities, including:

[0042] The data preprocessing module is used to acquire multi-source heterogeneous data of human settlement heritage communities and perform standardized preprocessing to obtain standardized datasets;

[0043] The environmental state modeling module is used to perform normalization mapping and feature extraction based on the standardized dataset to construct an environmental state vector with spatial units as the granularity.

[0044] The text rule extraction and constraint encoding module is used to extract entities and relationships from cultural protection regulations texts through artificial intelligence natural language processing to construct a cultural knowledge graph, and to transform constraint clauses into machine-executable condition-action constraint expressions encoded with abstract syntax trees to generate a cultural constraint rule library.

[0045] The multi-agent decision modeling module is used to build rule-driven decision models for multiple agents, and to define observation inputs, action parameter templates and utility evaluation functions for each decision model.

[0046] The legality pruning module is used to logically compare the candidate action parameters of each decision model with the condition-action constraint expression, and perform legality pruning to generate the legal action space of each decision model.

[0047] The conflict graph construction and solution module is used to generate joint action combinations based on the legal action space of each decision model, construct the conflict graph and detect connected conflict components, and remove joint action combinations that have conflicts.

[0048] The control instruction set generation module is used to determine the target protection scheme after sorting the remaining joint action combinations, and outputs it in the form of a space unit-level control instruction set and a resource scheduling parameter table.

[0049] The incremental update module is used to recalculate the environmental state vector based on the monitoring feedback data after execution, and to perform incremental updates on the rule parameters based on the violation event statistics and execution deviations.

[0050] An electronic device includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program, when executed by the processor, implements the method.

[0051] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1. Multi-source heterogeneous data are unified into a 7-dimensional environmental state vector through spatiotemporal alignment, normalization mapping, and feature extraction. The value range of each dimension is normalized to [0,1], which facilitates the unified expression and quantitative assessment of multi-dimensional protection status under the same decision-making framework. This solves the problem that it is difficult to uniformly incorporate the status of historical buildings, environmental changes, tourist flow, and community feedback into decision-making.

[0054] 2. By using natural language processing, entities and relationships are extracted from cultural protection regulations, local guidelines, expert experience, and community consultation outcomes, and structured into six-tuple condition-action constraint expressions. The expressions are encoded using abstract syntax trees to achieve machine-executable judgment, realizing the automatic transformation from text description to computable constraint rules, and improving the compliance screening capability before the solution is generated.

[0055] 3. By generating a legal action space through legality pruning based on condition-action constraint expressions (directly eliminating prohibited classes, re-judging modified classes after adjustment instructions based on parameters, and directly retaining classes), and then generating joint action combinations through Cartesian product, the collaborative decision-making ability and solution consistency among different subjects are improved.

[0056] 4. By constructing a conflict graph and using depth-first search to detect connected conflict components, five types of conflicts are identified: resource conflicts, spatial conflicts, temporal conflicts, rule conflicts, and target conflicts. Combined with a multi-index weighted scoring function order, the feasibility and protection adaptability of the scheme are improved, and a target protection scheme in the form of a spatial unit-level control instruction set is output.

[0057] 5. Through six closed-loop feedback update operations—enhancing the system's dynamic adaptability to state changes and new rules—including recalculating the environment state vector, adjusting rule priorities, updating the knowledge graph, iteratively optimizing ranking weights, expanding the scope of application of correction instructions, and updating the rule base content, the system's ability to dynamically adapt to state changes and new rules is improved. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the process for encoding cultural constraint rules and solving conflict diagrams for the protection of human settlement heritage communities, as described in this invention.

[0059] Figure 2 This is a schematic diagram of the multi-source heterogeneous data acquisition and preprocessing process of the present invention.

[0060] Figure 3 This is a schematic diagram of the cultural knowledge graph structure of this invention.

[0061] Figure 4 This is a schematic diagram of the multi-subject candidate action screening and legality trimming process of the present invention.

[0062] Figure 5 This is a schematic diagram of the process for constructing and solving the joint action conflict graph of the present invention.

[0063] Figure 6 This is a schematic diagram of the closed-loop feedback update process of the present invention.

[0064] Figure 7 This is the initial interface diagram of the system of the present invention.

[0065] Figure 8 This is a diagram of the main operation interface of the system of the present invention.

[0066] Figure 9 This is a diagram of the data analysis interface of the system of the present invention.

[0067] Figure 10 This is a simulation interface diagram of the system of the present invention. Detailed Implementation

[0068] The structure and working process of the present invention will be further described below with reference to the accompanying drawings.

[0069] A method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities, including:

[0070] Multi-source heterogeneous data of human settlement heritage communities were acquired, and standardized preprocessing was performed on the acquired multi-source heterogeneous data to obtain a standardized dataset including numerical data, image and video data;

[0071] Based on a standardized dataset, numerical data is normalized and mapped to a preset range, and target detection and feature extraction are performed on image and video data to construct an environmental state vector with spatial units as the granularity, which is used to characterize the community protection status.

[0072] Based on the texts of cultural heritage protection regulations, local management rules, expert experience rules, and community consultation rules, a cultural knowledge graph is constructed by extracting entities and relationships through natural language processing. The constraint clauses are then transformed into machine-executable condition-action constraint expressions to generate a cultural constraint rule library.

[0073] To construct rule-driven decision-making models for multiple stakeholders involved in human settlement heritage communities, each decision-making model uses the current environmental state vector as the observation input, a predefined action parameter template as the candidate action, and a utility evaluation function as the basis for quantifying action preferences.

[0074] The candidate action parameters of each decision model are logically compared with the condition-action constraint expressions one by one, and legality pruning is performed to generate the legal action space of each decision model.

[0075] Based on the legal action space of each decision model, a joint action combination is generated. A conflict graph is constructed with actions as nodes and resource occupation relationship and spatiotemporal interference relationship as edges. Connected conflict components in the conflict graph are detected to identify and eliminate joint action combinations with conflicts. The target protection scheme is determined after sorting the remaining joint action combinations.

[0076] The environmental state vector is recalculated based on the monitoring feedback data after the implementation of the target protection scheme. When the change exceeds the preset threshold, a new round of decision-making process is triggered, and the rule parameters in the cultural constraint rule base are incrementally updated based on the statistics of violation events and the deviation of scheme implementation.

[0077] Specific embodiments, such as Figures 1 to 10 As shown:

[0078] The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities mainly includes the following seven parts:

[0079] I. Acquire multi-source heterogeneous data of human settlement heritage communities, and perform standardized preprocessing on the acquired multi-source heterogeneous data to obtain a standardized dataset including numerical data, image and video data;

[0080] In this embodiment, to address the data perception needs of human settlement heritage communities, the following seven types of multi-source heterogeneous data are collected:

[0081] 1. Image and Video Data: Sourced from monitoring equipment deployed at street intersections, building facades, and public activity spaces. Image data resolution is no less than 1920 x 1080 pixels, and video frame rate is no less than 25 frames per second, used to identify pedestrian density, activity occupancy, and facade changes. Pedestrian count and movement trajectories are extracted from the video stream using object detection algorithms, and new damage or unauthorized alterations to the facade are detected using image comparison algorithms.

[0082] 2. Environmental Monitoring Data: Data is collected through environmental sensors deployed at key points in the community, including noise levels (30-120 dB), temperature (-40°C to 60°C), humidity (0-100%RH), air quality index (AQI), illuminance (0-100,000 Lux), and population density (number of people per square meter). Environmental monitoring data is collected every 5 minutes.

[0083] 3. Structural monitoring data: Data is collected through sensors deployed at key structural parts of the historical building, including crack width (accuracy 0.01 mm), settlement (accuracy 0.1 mm), tilt (accuracy 0.001 degrees), and abnormal vibration (peak acceleration, accuracy 0.001 m / s²). The structural monitoring data is collected every 30 minutes, automatically increasing to every 5 minutes in abnormal conditions.

[0084] 4. Community Feedback Data: Collected through the community management platform, mobile applications, and offline feedback channels, including resident satisfaction ratings (using a 1-5 Likert scale), written opinions (free text format), complaint records (including time, location, complaint type, and processing status), and voting records (including voting topics, options, number of participants, and voting results). Community feedback data is summarized daily.

[0085] 5. Business activity data: Collected through the merchant management platform and customer flow statistics equipment, including merchant customer flow statistics (average daily customer flow), business hours (start time and end time), business type (catering, retail, cultural experience, etc. classification codes), and high-noise business activity records (including activity type, time, noise level and duration).

[0086] 6. Management rules data: These are derived from regulations and documents issued by cultural heritage protection management departments and local governments, including protection boundaries (polygonal areas represented by GIS coordinates), cultural heritage protection levels (national, provincial, municipal, and district / county level classifications), style control requirements (building height restrictions, facade color specifications, signboard size restrictions, etc.), business restrictions (business hours restrictions, noise emission standards, etc.), and repair requirements (repair approval procedures, material and process standards, etc.).

[0087] 7. Expert-annotated data: This data is entered by experts in the field of cultural heritage protection through an annotation platform. It includes risk area annotation (risk area polygons marked on the community floor plan and their risk levels), value element annotation (annotation of architectural components, spatial patterns and cultural elements with conservation value), protection recommendations (structured protection recommendation texts), and rule revision recommendations (revision opinions on existing rules and proposals for new rules).

[0088] A unified preprocessing workflow is performed on the above-mentioned multi-source heterogeneous data, specifically including:

[0089] Standardization: For numerical data, the Min-Max normalization method is used to map each indicator to the [0,1] interval. For indicators with clear physical thresholds (such as noise decibels, AQI, etc.), a piecewise linear mapping is performed based on the threshold.

[0090] Missing value handling: Data with a missing value rate of less than 20% is filled using linear interpolation or mean imputation methods. Data with a missing value rate of more than 20% is marked as unusable and removed from the current decision-making cycle.

[0091] Time alignment: All data is aligned to the same timestamp according to a preset time window. Environmental monitoring data and structural monitoring data are aggregated in hourly windows, community feedback data and business activity data are aggregated in daily windows, and keyframe features are extracted from image and video data in 15-minute windows. At the decision-making moment, the value of the most recent time window from each data source is taken as the current input.

[0092] Spatial Alignment: Data is uniformly mapped to standard spatial units according to community spatial hierarchy. The spatial hierarchy, from largest to smallest, is: street block, alley segment, building unit, and point of interest. Each spatial unit is assigned a unique identifier, and all data is mapped to the corresponding spatial unit based on GIS coordinates.

[0093] Outlier Handling: Data exceeding a preset reasonable range is marked as anomaly. For example, a sudden increase in crack width exceeding 0.5 mm / day is marked as anomaly, and pedestrian traffic exceeding three times the area's carrying capacity threshold is marked as anomaly. Anomaly data is corrected according to anomaly handling rules: for anomalies caused by sensor malfunctions, the previous normal value is used as a replacement; for anomalies caused by real events, the original value is retained and an anomaly label is added for subsequent decision-making reference.

[0094] After the above preprocessing, a standardized dataset is obtained, which is organized according to a uniform timestamp and spatial unit, and is used for subsequent environmental state vector construction and rule matching.

[0095] Second, based on a standardized dataset, the numerical data is normalized and mapped to a preset interval, and the image and video data are subjected to target detection and feature extraction to construct an environmental state vector with spatial units as the granularity, which is used to characterize the community protection status.

[0096] In this embodiment, a 7-dimensional environmental state vector is constructed based on a standardized dataset to uniformly represent the multidimensional protection status of human settlement heritage communities. The environmental state vector is represented as follows:

[0097]

[0098] The definitions, data sources, and calculation methods for each dimension are as follows:

[0099] 1. Heritage Health Index Used to reflect the health status of a building structure. The sub-indicators are calculated by weighting at least two of the following: structural damage level, leakage risk, settlement, and vibration anomaly. They can be normalized to the [0,1] interval by linear weighted summation and complementation. Each sub-indicator is calculated based on the standardized results of structural monitoring data, and each weight is determined by the analytic hierarchy process (AHP) or the entropy weight method. The value range is [0,1]. The closer the value is to 1, the healthier the building structure is, and the closer the value is to 0, the worse the building structure is.

[0100] 2. Passenger carrying capacity index : Used to reflect the load level of tourist or personnel density relative to the carrying capacity threshold. The ratio of measured pedestrian flow to the regional carrying capacity threshold is obtained by normalizing using a saturation mapping function. The mapping method can be a combination of linear mapping and a saturation function, making... The value range falls within the [0,1] interval. The closer it is to 1, the more severe the overload.

[0101] 3. Noise Impact Index Used to reflect the degree of noise disturbance to a community. Noise decibel data from environmental monitoring is normalized to the [0,1] interval through linear or piecewise mapping. The closer the value is to 1, the more severe the noise interference.

[0102] 4. Environmental Quality Index Used to reflect the overall situation of air quality, temperature, humidity, and comfort. The result is calculated by weighting the inversely normalized value of the Air Quality Index (AQI), temperature comfort, and humidity comfort. The AQI is normalized to a preset range through inverse mapping (the lower the AQI, the closer the value is to 1). Temperature comfort and humidity comfort are calculated using mapping functions of preset comfort ranges. The sum of all weights is 1.

[0103] 5. Community Satisfaction Index Used to reflect residents' feedback. The weighted aggregation of resident satisfaction scores is used to calculate the average value. The weights are determined by normalization based on factors such as the length of residence and distance from the core protected area, ensuring that the results are normalized to [0,1].

[0104] 6. Cultural Activity Index Used to reflect traditional activities, exhibitions, and levels of participation. The results are calculated by combining the frequency of traditional events, the number of participants in exhibitions, and the coverage of cultural activity spaces. Each sub-indicator is standardized and weighted, with the sum of all weights being 1.

[0105] 7. Compliance Risk Index Used to reflect the extent of a violation. It is calculated by the ratio of the number of violations to the total number of monitored events within a preset period, with a value range of [0,1]. The closer the value is to 1, the higher the risk of violation.

[0106] The environmental state vector The update frequency is consistent with the decision-making cycle. The time step index, representing the decision-making cycle, is updated daily by default. After the target protection plan is implemented, the state vector for the next moment is recalculated and generated based on the latest monitoring feedback data, thus forming a closed-loop dynamic evolution mechanism. In emergency situations (such as sudden structural anomalies, extreme overcrowding, or violations), immediate updates can be triggered. The environmental state vector serves as part of the observation information of various agents and is input into the subsequent decision-making process, providing a data foundation for the generation of candidate actions and utility evaluation for each agent.

[0107] Third, based on the texts of cultural heritage protection laws and regulations, local management rules, expert experience rules, and community consultation rules, we extract entities and relationships through natural language processing to construct a cultural knowledge graph, and transform the constraint clauses into machine-executable condition-action constraint expressions to generate a cultural constraint rule library.

[0108] In this embodiment, a cultural knowledge graph and a cultural constraint rule base are constructed to transform cultural protection requirements from textual descriptions into structured, machine-executable rules.

[0109] First, construct a cultural knowledge graph. ,in For a set of entity nodes, Let be the set of relation edges. This is a collection of relation types. The knowledge graph construction process is as follows:

[0110] (1) Perform natural language processing on the text of cultural protection regulations and local management rules, use the Named Entity Recognition (NER) model to extract entity nodes such as heritage building names, protection levels, and spatial locations, and use the relation extraction model to extract relation edges such as location, belonging, and constraints between entities;

[0111] (2) For expert experience rules and community consultation rules, semi-structured mapping is performed through predefined entity relationship templates, and the rules are entered into the knowledge graph after being reviewed and confirmed by experts;

[0112] (3) Standardize the encoding of entity attribute values. For example, the protection level adopts enumeration encoding (national level = 4, provincial level = 3, municipal level = 2, district / county level = 1), and the spatial location adopts GIS coordinate encoding. The knowledge graph is stored using an attribute graph model and supports entity relationship retrieval and path reasoning based on SPARQL or Cypher query language.

[0113] Entity node set It includes at least the following ten types of entities:

[0114] 1. Heritage Building Entities: Each heritage building is an independent node, with attributes including building number, construction date, structural type (timber structure, brick-timber structure, brick-stone structure, etc.), protection level (national, provincial, municipal, district / county level), protected area (core protection zone, construction control zone, environmental coordination zone), and current health status.

[0115] 2. Street and alley space entities: Each street or alley or public space is an independent node, with attributes including space number, space type (pedestrian street, square, courtyard, alley, etc.), area, functional positioning, and carrying capacity threshold.

[0116] 3. Traditional Activity Entities: Each traditional activity is an independent node, with attributes including activity number, activity type (festival activities, folk performances, traditional handicraft displays, etc.), event duration, cultural value level, and required space conditions.

[0117] 4. Resident Subject Entities: Resident groups are clustered according to their residential area and participation characteristics to form representative resident subject nodes. Attributes include subject number, representative area, population size, and degree of cultural participation.

[0118] 5. Merchant Entity: Merchants are clustered according to their operating area and business type to form representative merchant entity nodes. The attributes include entity number, operating area, business type, and business scale.

[0119] 6. Management Entity: The management organization and functional departments are used as nodes, and the attributes include the entity number, management scope and responsibility type (planning approval, daily inspection, emergency response, etc.).

[0120] 7. Expert Entities: The experts participating in the protection assessment are designated as nodes, with attributes including entity number, professional field (architecture, planning, folklore, structural engineering, etc.) and assessment authority.

[0121] 8. Protection Measure Entities: Each type of protection measure is an independent node, with attributes including measure number, measure type (routine maintenance, preventive protection, repair projects, environmental remediation, etc.), applicable objects, and implementation conditions.

[0122] 9. Risk Event Entities: Each type of risk event is an independent node, with attributes including event number, event type (structural damage, illegal modification, overload operation, environmental pollution, etc.), scope of impact, and urgency level.

[0123] 10. Rule Clause Entities: Each protection rule is an independent node, with attributes including clause number, source of law (such as the "Cultural Relics Protection Law", "Regulations on the Protection of Famous Historical and Cultural Cities, Towns and Villages", local protection and management measures, etc.), scope of application, and level of validity.

[0124] Relation type set This includes, but is not limited to, the following relationship types: located (the heritage building is located in a street or alley space), belonging to (the heritage building belongs to a protection level), held in (traditional activities are held in a street or alley space), resided in (residents mainly reside in a street or alley space), operated in (businesses mainly operate in a street or alley space), management scope (the management scope of the management entity covers the space or building), assessment object (the assessment object of the expert entity is the heritage building or protection measures), applicable object (the applicable object of the protection measures is the heritage building or space), triggering rules (the risk event triggers the rule clauses), and binding relationship (the rule clauses bind the protection measures or the behavior of the entity), etc.

[0125] Based on a cultural knowledge graph, cultural protection regulations, local management rules, expert experience rules, and community consultation outcomes are transformed into structured rules to construct a cultural constraint rule base. Each rule is represented using a six-tuple structure:

[0126]

[0127] in: This represents the index of a rule in the cultural constraint rule base, used to uniquely identify each rule. The rule base is stored using an ordered linked list. The rule triggering condition describes the preconditions for the rule to take effect, encoded using logical expressions based on the environment state vector. Boolean judgment is performed on the dimension values ​​and entity attributes in the cultural knowledge graph; For the constraint object, specify which type of intelligent agent or which type of action the rule applies to; For the constrained action, specify the specific action type that the rule constrains; The constraint result type can be one of three types: "prohibited", "corrected", or "allowed". To correct the instruction, only when If "correction" is not empty, describe the specific parameter adjustment content (including the target parameter to be adjusted, the adjustment direction, and the adjustment range). This represents the rule priority, and its value is a positive integer, arranged from highest to lowest. =1,2,…,K (where K is the total number of rules), a larger value indicates a higher priority. Rule triggering condition expression. The following data structure is used for encoding: represented by an Abstract Syntax Tree (AST), with leaf nodes serving as dimension identifiers for the environment state vector (e.g., ...). , ) or knowledge graph entity attribute query results, where non-leaf nodes are logical operators (AND, OR, NOT) and comparison operators. During rule matching, the attribute values ​​of candidate actions are substituted into the AST of the conditional expression for recursive evaluation, and a Boolean result is returned.

[0128] When matching rules, the system starts from... Starting with rule =1 (highest priority), the algorithm iterates through each rule. If a prohibited result is found, the matching process immediately terminates, and the candidate action is added to the prohibited action set. If a corrective result is found, it is recorded... It then continues to match subsequent rules to ensure that there are no higher-priority prohibition rules.

[0129] Here are five specific example rules:

[0130] Rule 1 (derived from national laws and regulations): =(Target building is located in a core protected area AND protection level is national level, facade modification action of the target building, modification action, prohibited, modification instruction is empty, 10). This rule means: when the target building is located in a core protected area and the protection level is national level, any modification action that damages the historical appearance of the facade is prohibited, with a priority of 10.

[0131] Rule 2 (derived from local management regulations): =(Merchant located in a sensitive residential area AND applied closing time > 23:00, merchant's business hours adjustment action, extending nighttime business hours action, correction, adjustment target: closing time, adjustment direction: down to, adjustment range: 22:30, 7). This rule indicates that when a merchant is located in a sensitive residential area and applied for a closing time exceeding 23:00, the closing time must be adjusted to 22:30 before implementation, with a priority of 7.

[0132] Rule 3 (derived from community consultation rules): = (Traditional event space has already scheduled cultural activities within the preset date AND application to schedule high-noise commercial activities, commercial activity scheduling action, high-noise commercial activities, prohibited, correction instruction is empty, 4). This rule means: when a traditional event space has already scheduled cultural activities within the preset date, scheduling high-noise commercial activities within the same time window is prohibited, with a priority of 4.

[0133] Rule 4 (derived from expert experience rules): = (The object to be repaired is at the provincial level or above, the repair action is to use materials that are not in the original style, the correction and adjustment target is the repair materials, the adjustment direction is to replace them with traditional materials that are consistent with the original style, and the adjustment range is 6). This rule means that when the object to be repaired is at the provincial level or above, it is required to use repair materials and techniques that are consistent with the original style, and the priority is 6.

[0134] Rule 5 (derived from local management regulations): =(Population carrying capacity index of a certain area) If the pedestrian flow carrying capacity index exceeds 0.5 for three consecutive monitoring periods, the manager's crowd control actions, or actions that are left unchecked, should be corrected. The adjustment target is: control strategy; the adjustment direction is: priority selection; the adjustment range is: flow restriction or diversion actions, 8). This rule means that when the pedestrian flow carrying capacity index of a certain area continuously exceeds the carrying capacity threshold, the manager is required to prioritize flow restriction or diversion actions, with a priority of 8.

[0135] Rule priority The priority is determined according to the following principles: rules derived from national laws and regulations have the highest priority (level 9-10), followed by rules derived from local management regulations (level 7-8), then rules derived from expert experience (level 5-6), and rules derived from community consultation have a relatively low priority (level 3-4). Within the same priority level, rules with more entity attribute dimensions in their triggering conditions are judged first. The above priority range is set during system initialization and can be dynamically adjusted based on violation statistics during operation: if a rule from a certain source is frequently triggered within a preset period but still results in violations, the priority of the relevant rules under that source will be increased by 1-2 levels, with the increase not exceeding 10.

[0136] The above rules are written into the cultural constraint rule base, which is stored in an ordered linked list, arranged from highest to lowest priority. During rule matching, each candidate action is matched one by one starting from the head of the linked list. Once a prohibited action is matched, the matching immediately terminates and the action is added to the prohibited action set. If a corrective action is matched, the corrective instruction is recorded and the matching of subsequent rules continues to ensure that there are no higher-priority prohibited actions.

[0137] Fourth, rule-driven decision-making models are constructed for multiple stakeholders involved in the human settlement heritage community. Each decision-making model uses the current environmental state vector as the observation input, a predefined action parameter template as the candidate action, and a utility evaluation function as the basis for quantifying action preferences.

[0138] In this embodiment, rule-driven decision-making models are constructed for four types of social actors in the scenario of human settlement heritage community protection. Each decision-making model consists of triplets. Definition, where In order to observe information, For a set of candidate actions represented by a parameterized template, For utility evaluation function, The index represents the subject type corresponding to the decision-making model, used to distinguish four types of social subjects: residents ( =1), Merchants ( =2), Managers ( =3), Experts ( =4). The operating mechanism of each decision model is as follows: within each decision cycle, it receives the current environmental state vector. Using relevant dimensions as observation input, candidate actions are generated from a predefined action parameter template. The utility value of each candidate action is calculated using a utility evaluation function as the basis for preference ranking. Then, all candidate actions are submitted to the legality trimming module. The strategy generation of each decision model adopts a deterministic utility maximization strategy: the action with the highest utility value in the legal action space is selected as the preferred action to participate in the generation of joint action combinations.

[0139] Candidate actions for each decision model are represented by parameterized templates. Each action template includes an action type identifier, an execution space unit ID, an execution time period parameter, a resource quota parameter, and constraint condition parameters. The action parameter templates for four types of subjects are given below:

[0140] 1. Resident-led decision-making model Observational information Includes: environment state vector Data related to dimensions of the living environment (noise impact index, environmental quality index) and community satisfaction.

[0141] Candidate action set The following parameterized action templates are included: (1) Environmental improvement request parameters, including target noise threshold, target environmental quality threshold, applicable spatial unit ID and request time period parameters; (2) Area occupancy preference parameters, including preference area ID, preference time period and population density preference value; (3) Activity participation parameters, including target activity type, participation time period and participation spatial unit ID; (4) Keep unchanged, indicating that no parameter change request is submitted.

[0142] Utility evaluation function The calculation is based on a weighted combination of factors such as the degree of improvement in the living environment, the convenience of living, and the accessibility of cultural activities. The initial values ​​of each weight are set by experts according to the stage of community protection. During operation, the weights are fine-tuned through iterative optimization based on the statistical correlation between historical actions and implementation effects.

[0143] 2. Merchant Decision-Making Model Observational information Includes: environment state vector The data includes operational dimensions (passenger flow carrying capacity index, compliance risk index) and the company's own operational data.

[0144] Candidate action set The following parameterized action templates are included: (1) Business hours application parameters, including application start time, application end time, applicable space unit ID and noise emission estimate; (2) Space renovation application parameters, including renovation type, renovation scope coordinates, involved facade sign positions and proposed material codes; (3) Business demand parameters, including demand type code, expected passenger flow threshold and resource demand; (4) Maintain the status quo and do not submit any parameter changes.

[0145] Utility evaluation function It is calculated by weighted combination of expected operating revenue, customer flow satisfaction, and operating constraint satisfaction.

[0146] 3. Managerial Decision-Making Model Observational information Includes: complete environment state vector Budget balance and usage, management objectives, and historical data on program implementation effectiveness.

[0147] Candidate action set The following parameterized action templates are included: (1) Pedestrian flow control parameters, including control space unit ID, control time period, upper limit of pedestrian flow threshold and diversion target space unit ID; (2) Repair and construction parameters, including target building ID, construction type code, construction time period, upper limit of construction personnel, vibration threshold and material constraint parameters; (3) Area control parameters, including control area ID, control level and applicable time period; (4) Resource scheduling parameters, including resource type code, allocation quantity, target space unit ID and scheduling time period; (5) Path guidance parameters, including guidance start point ID, guidance end point ID, guidance time period and guidance method code; (6) Business constraint parameters, including constraint space unit ID, constraint time period, noise upper limit and business type restriction code.

[0148] Utility evaluation function The calculation is based on a weighted combination of protection effectiveness, budget control, order maintenance, and the degree to which management objectives are achieved.

[0149] 4. Expert-led decision-making model Observational information Includes: complete environment state vector Professional assessment data and risk warning information.

[0150] Candidate action set The following parameterized action templates are included: (1) Scheme review parameters, including review result code (pass / reject / conditionally pass), rejection reason code and condition parameters; (2) Protection constraint supplementary parameters, including constraint type code, applicable object ID, constraint condition parameters and priority suggestion value; (3) Evaluation parameter adjustment parameters, including target indicator ID, adjustment direction and adjustment range.

[0151] Utility evaluation function The calculation is based on a weighted combination of authenticity evaluation, integrity evaluation, and safety evaluation. The observed information in each decision model includes an environmental state vector. The relevant dimensions enable each decision-making model to generate reasonable parameterized candidate actions based on the current community protection status. The weight parameters of each utility evaluation function are determined as follows: initial weights are set by domain experts according to the community protection stage; during operation, the weights are fine-tuned through iterative optimization based on the statistical correlation between the historical action selections and actual execution effects of each decision-making model, and are updated once per decision cycle. After the update, a normalized projection is performed on the weight vector to ensure that each weight is non-negative and the sum is 1.

[0152] Implementation of Candidate Action Parameter Generation by Large Language Model: In this embodiment, the candidate action parameters for each decision model can be generated by a Large Language Model (LLM). Specifically, the values ​​of each dimension of the current environmental state vector, the subject type identifier, the subject's historical action records, the attribute information of the spatial unit to which it belongs, and cultural constraint prompts are input into the Large Language Model as structured prompt words. The Large Language Model is required to output a structured list of candidate action parameters according to a predefined action parameter template format. After format validation, the output of the Large Language Model serves as the candidate action set for the decision model, which is then entered into the legality trimming module for constraint expression matching. This implementation utilizes the semantic understanding capability of the Large Language Model to generate more adaptive candidate action parameters based on the dynamic changes in the environmental state, while ensuring that all candidate actions conform to the constraint requirements of the cultural constraint rule base through the legality trimming module.

[0153] 5. The candidate action parameters of each decision model are logically compared with the condition-action constraint expressions one by one, and legality pruning is performed to generate the legal action space of each decision model.

[0154] For the sake of brevity, the term "intelligent agent" will be used consistently throughout this document. "Obsi, Actsi, Utilityi" refers to each decision model. Each agent corresponds to a decision model of a type of social subject (e.g., residents, merchants, managers, or experts).

[0155] In this embodiment, the candidate actions of each agent are matched with the rules in the cultural constraint rule base, and the legal action space of each agent is generated by rule determination.

[0156] The detailed process for rule determination is as follows:

[0157] Step a: For each agent Each candidate action This involves matching the candidate action against all rules in the cultural constraint rule base. The matching process is based on attributes such as the action type, execution area, execution time period, and action target of the candidate action, along with the triggering conditions of the rules. and constraint objects Perform a logical comparison.

[0158] Step b: Divide the candidate actions into three sets based on the matching results:

[0159] Forbidden Actions When a candidate action satisfies any prohibited class condition - action constraint expression (i.e., a rule exists). Make For true and If the action is marked as "prohibited", it will be added to the prohibited action set. Actions in the prohibited action set may not be executed during the current decision-making cycle.

[0160] Action set to be corrected When a candidate action satisfies the modification class condition - action constraint expression (i.e., a rule exists). Make For true and If the action is designated as "corrected," it is added to the action set to be corrected. Actions in the action set to be corrected need to be corrected according to the rules. The procedure can only be executed after the parameters have been adjusted.

[0161] Allowed action set When a candidate action does not trigger any prohibited or corrective conditional-action constraint expressions, the candidate action is included in the allowed action set. Actions in the allowed action set can directly enter the legal action space.

[0162] Step c: Priority Rules. When multiple rules apply to the same candidate action simultaneously, the final result is determined according to the following priority: prohibition takes precedence over correction, and correction takes precedence over permission. That is, if the same action is simultaneously determined as prohibited by one rule and permitted by another, the action will ultimately be included in the prohibited action set. Rule Priority Used to determine the execution order of rule decisions: rules with higher priority are executed first.

[0163] Step d: The set of actions to be corrected Each action in the process performs at least one of the following correction methods:

[0164] (1) Adjust the execution area of ​​the action: Adjust the spatial scope of the action from the original area to the compliant area. For example, adjust a merchant's outdoor stall application from the core protection zone to the construction control zone.

[0165] (2) Adjust the execution time of the action: Adjust the time window for the action from the original time period to the compliant time period. For example, adjust the merchant's business closing time from 24:00 to 22:30.

[0166] (3) Adjust the allocation of resources for the operation: Adjust the allocation of human, material or financial resources involved in the operation to the compliant range. For example, adjust the upper limit of the number of workers in the repair project from 50 to 30 to reduce the risk of vibration.

[0167] (4) Replace the target of the action: Replace the target of the action with a compliant object. For example, replace the repair materials from modern paint with traditional materials that are consistent with the original style.

[0168] (5) Add supplementary protection measures: Add additional protection measures to meet compliance requirements while the action is being performed. For example, add vibration monitoring and temporary support measures in the repair work.

[0169] The corrected action re-enters the rule-determination process, meaning it is matched against all rules in the cultural constraint rule base again. If the corrected action passes all rule determinations (is not marked as prohibited or corrected by any rule), it is added to the allowed action set; if the corrected action still triggers prohibited rules, it is added to the prohibited action set. This iterative determination process executes a maximum of 3 rounds. Actions that fail to pass determination after more than 3 rounds are directly added to the prohibited action set.

[0170] Step e: Finally, all actions in the action set will be allowed as the legal action space for the corresponding agent. Legal action space Represents intelligent agents The set of all actions permitted to be performed under the current environmental conditions and cultural constraints.

[0171] VI. Generate joint action combinations based on the legal action space of each decision model, construct a conflict graph with actions as nodes and resource occupation relationship and spatiotemporal interference relationship as edges, detect connected conflict components in the conflict graph to identify and eliminate joint action combinations with conflicts, and determine the target protection scheme after sorting the remaining joint action combinations.

[0172] In this embodiment, a joint action combination is generated based on the legal action space of each intelligent agent, conflict detection and comprehensive scoring are performed, and the final target protection scheme is determined.

[0173] Joint action combination generation: Let the legal action spaces of resident intelligent agent, merchant intelligent agent, manager intelligent agent and expert intelligent agent be respectively... , , , The combined action is generated through a Cartesian product:

[0174]

[0175] Each combination of actions This involves a joint scheme consisting of four agents, each selecting one legal action. When the legal action space is large, a pruning strategy can be used to reduce the number of combinations: low-quality actions with utility evaluation function values ​​below their respective preset thresholds are removed, and the top N actions in terms of utility for each agent are retained for combination.

[0176] Combination of actions Perform five types of collision detection. The collision detection algorithm is implemented by constructing a collision graph. ,in It is the set of nodes representing the actions of each agent in a combined action sequence. This is the set of edges where there are conflicts between actions. For each pair of action nodes... Check the following five conflict types in sequence. If any type is matched, add a conflicting edge between the two nodes:

[0177] The specific judgment logic for the five types of conflict detection is as follows:

[0178] 1. Resource Conflict Detection: This function checks whether multiple actions in a combined action sequence simultaneously occupy the same budget, equipment, or construction resources, and whether the total demand exceeds the available resources. For example, if the manager agent's repair scheduling and resource allocation actions simultaneously request the same repair budget, and the total amount exceeds the budget balance, this is considered a resource conflict.

[0179] 2. Spatial Conflict Detection: This function checks whether multiple actions in a combined action sequence are performed within the same spatial unit and cannot be performed in parallel. For example, if a merchant agent's space renovation application and a manager agent's repair arrangement both target the same historical building, it is considered a spatial conflict.

[0180] 3. Time Conflict Detection: This function checks whether multiple actions in a combined action sequence occur within the same time window and interfere with each other. For example, if a merchant agent requests to hold a promotional event during a traditional holiday, but a cultural event is already scheduled for the same time slot, this is considered a time conflict.

[0181] 4. Rule Conflict Detection: This checks whether a combination of actions, as a whole, violates rules in the cultural constraint rule base. Individual actions may pass rule judgment, but combinations of actions may create new violation scenarios. For example, the overlap of a merchant's legal operating hours and the manager's flow restriction measures may prevent the merchant from obtaining a reasonable customer flow, triggering protective constraint rules.

[0182] 5. Target Conflict Detection: Check whether a combination of joint actions causes the utility value of a certain agent to fall below a preset lower limit. For each combination of joint actions, calculate the utility evaluation function value of each agent. If any Below the agent's lower utility limit If so, it is determined to be a target conflict.

[0183] After the conflict graph is constructed, a depth-first search (DFS) algorithm is used to identify conflicts: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Perform a Depth-First Search (DFS) traversal starting from each node, marking all connected components containing conflicting edges. If a joint action combination contains a connected component with a conflicting edge (i.e., has at least one conflicting edge), the joint action combination is marked as conflicting and immediately removed, not proceeding to the next sorting step. Only when the graph is conflicting... middle When there are no conflicting edges, the combined action combination passes the conflict detection. The time complexity of this DFS detection process is O(log n). It can efficiently handle large-scale combined actions.

[0184] In the extreme case where the number of joint action combinations that pass the conflict detection is zero, the system adopts the following degradation strategy: relax the lower limit threshold of utility in the target conflict detection, lower the threshold by a preset ratio, and then re-execute the conflict detection.

[0185] For the conflict-free joint action combinations that pass the above five types of conflict detection, a multi-index weighted ranking is used to determine the target protection scheme. The ranking can be performed using the following multi-index weighting function:

[0186]

[0187] The calculation methods for each scoring indicator are as follows:

[0188] Cultural compatibility This measure assesses the degree of consistency between the proposed solution and cultural preservation regulations and local guidelines. It calculates the degree of consistency based on the matching of each action in the combined action sequence with the cultural constraint rule base, identifying solutions that fully comply with all rules. For each non-prohibited rule violated, a certain number of points will be deducted.

[0189] Multi-subject satisfaction This measures the weighted satisfaction of the proposed solution with the utility evaluation functions of each agent. The calculation method is as follows:

[0190]

[0191] in, The weight parameters for each agent reflect the importance of different agents in the decision-making process.

[0192] .

[0193] Feasibility of implementation This measure assesses the feasibility of a solution in terms of resources, time, and space. It is calculated based on the current availability of resources, the sufficiency of the execution time window, and the accessibility of the implementation space.

[0194] Protect income This measures the expected improvement of the proposed solutions on dimensions of the environmental state vector, such as heritage health index and cultural activity. The estimated impact of each action on the environmental state vector is based on this estimation.

[0195] Implementation costs Measure the budget cost, labor cost, and time cost of the proposed solution. Standardize the three types of costs and then weight and sum them.

[0196] Weight parameters to This reflects the relative importance of each scoring indicator in the overall score, with the sum of all weights equal to 1. (These are deduction items). Initial weight values ​​can be determined using the Analytic Hierarchy Process (AHP) or the entropy weight method, or a combination of both: subjective weights are obtained using AHP, objective weights are obtained using the entropy weight method, and the final weights are obtained by combining them according to a preset ratio. In the default settings, cultural fit has the highest weight, reflecting the primary goal of cultural heritage protection. Each weight parameter is iteratively optimized based on the implementation effect of the scheme through a closed-loop feedback mechanism. After updating, a normalized projection is performed on the weight vector to ensure that each weight is non-negative and that the sum is 1.

[0197] The optimal combination of joint actions is output as the target protection scheme. The target protection scheme is output in the form of a spatial unit-level control instruction set, which includes at least two of the following: spatial unit-level equipment scheduling instructions, regional pedestrian flow threshold setting parameters, monitoring equipment alarm threshold configuration parameters, repair project construction constraint parameter table, budget resource allocation parameters, and monitoring task issuance parameters.

[0198] 7. Recalculate the environmental state vector based on the monitoring feedback data after the implementation of the target protection plan. When the change exceeds the preset threshold, trigger a new round of decision-making process and perform incremental updates on the rule parameters in the cultural constraint rule base based on the statistics of violation events and the deviation of plan implementation.

[0199] In this embodiment, after the target protection scheme is executed, the system dynamically adjusts the environmental state, rule parameters, and decision parameters through a continuous monitoring and feedback update mechanism to form a closed-loop collaborative decision.

[0200] Data collection after implementation of the plan: After the target protection plan is implemented, the system will re-collect multi-source heterogeneous data according to the preset monitoring cycle (default is daily), including environmental monitoring data, structural monitoring data, community feedback data and business activity data, etc., to evaluate the effectiveness of the plan implementation.

[0201] Closed-loop feedback updates include at least one of the following six operations:

[0202] 1. Recalculate the environmental state vector: Based on the latest collected monitoring data, recalculate the environmental state vector according to the calculation method in Implementation Method 2. The values ​​of each dimension of the environment state vector. If the change in the environment state vector exceeds a preset change threshold (e.g., any dimension changes by more than 20%), a new round of decision-making process is triggered.

[0203] 2. Adjust rule priority based on new violations: If new violations are detected during the implementation of the plan (such as unauthorized modifications, exceeding operating hours, etc.), the priority of the relevant rules will be adjusted. Upgrade. For example, if illegal outdoor parking is detected three times consecutively within the core protected area, the priority of the relevant prohibition rule will be increased from 8 to 10.

[0204] 3. Update the cultural knowledge graph based on new expert opinions: When experts propose new conservation recommendations or discover new risk factors, update the cultural knowledge graph accordingly. New entity nodes and relationship edges are added to the knowledge graph. For example, if an expert discovers that a certain architectural component has unrecognized conservation value, a new entity node for that component is added to the knowledge graph, and its relationship with the heritage building entity is established.

[0205] 4. Adjust the comprehensive scoring parameters based on implementation results: Based on the deviation between the actual and expected results of the scheme, update the weight parameters in the comprehensive scoring formula using an iterative optimization method. During updates, adjustments are made along the gradient direction of the scoring function based on the actual effect deviation, and a normalized projection is performed on the weight vector after each update to ensure that each weight is non-negative and their sum is 1. For example, if a scheme with high cultural compatibility is found to have poor actual protection effects, the weight of cultural compatibility is automatically reduced while the weight of protection benefits is increased.

[0206] 5. Adjust action correction rules based on execution deviations: If the correction scheme for a certain type of action repeatedly fails (i.e., it is still judged as prohibited by the rules after correction), then update the applicable conditions of the correction rules for that type of action, such as expanding the correction scope or lowering the correction threshold.

[0207] 6. Update the rule base content based on new rules or events: When new local protection regulations are issued, new management rules are generated through community consultation, or new constraint requirements arise from sudden risk events, the new content is converted into a six-tuple rule format and written into the cultural constraint rule base. For example, when a local government issues new nighttime noise management regulations, the relevant clauses are converted into rules. And add it to the rule base.

[0208] The triggering conditions for closed-loop decision-making include three types: time cycle triggering (automatically triggered according to the preset decision cycle, the default is once a week), event triggering (triggered immediately when any dimension in the environmental state vector exceeds the preset emergency threshold), and threshold triggering (triggered when the state change trend of multiple consecutive monitoring cycles exceeds the preset trend threshold).

[0209] Through the above feedback and update process, after each decision cycle, the system inputs the updated environmental state vector, rule base and scoring parameters into the next round of decision-making process, and starts again from the second part, forming a continuously optimized closed-loop collaborative decision-making mechanism.

[0210] To further illustrate the feasibility and effectiveness of the above scheme, the following example of a historical district will be used to illustrate the complete execution process of the method through specific numerical data.

[0211] This historical district is a national-level historical and cultural district, covering an area of ​​approximately 15 hectares, and includes three protection levels: a core protection zone, a construction control zone, and an environmental coordination zone. The district contains 3 national-level protected buildings, 8 provincial-level protected buildings, and 15 municipal-level protected buildings, with approximately 1,200 permanent residents and about 80 registered businesses.

[0212] During the current decision-making cycle, the environmental state vector collected by the system is as follows:

[0213] Heritage Health Index =0.62 (one of the provincial-level cultural heritage buildings was found to have cracks expanding, resulting in a lower index).

[0214] Passenger carrying capacity index =0.84 (The real-time pedestrian flow in the core streets and alleys is 3500 people / hour, the carrying capacity threshold is 3000 people / hour, and the normalized value after saturation mapping exceeds the carrying capacity range, indicating overload).

[0215] Noise Impact Index =0.81 (the measured average noise level is 72dB, which is close to the severe interference threshold).

[0216] Environmental Quality Index =0.73 (Air quality is good, but temperature and humidity comfort is average).

[0217] Community Satisfaction Index =0.54 (Residents expressed dissatisfaction with noise and excessive tourists);

[0218] Cultural Activity Index =0.66 (One traditional festival is being prepared this period);

[0219] Compliance Risk Index =0.29 (7 violations were detected in the past 30 days, 4 of which were for operating beyond the permitted time limit).

[0220] In this state, some of the candidate actions for each agent are as follows: The resident agent proposes two actions: "demand for noise reduction at night" (hoping to control nighttime noise below 50dB) and "oppose the influx of tourists".

[0221] The merchant intelligence agent proposed to "extend business hours from 22:00 to 24:00" (a joint application from 12 catering merchants).

[0222] The management AI proposed two actions: "limiting the flow of people in core streets and alleys from 18:00 to 21:00" and "arranging structural repairs for Building No. 3, a provincial-level cultural relic protection building".

[0223] The expert agent proposed two actions: "requiring noise restrictions to be added to applications for nighttime business operations" and "requiring the use of traditional materials and techniques for repairs."

[0224] Rule determination process:

[0225] The resident's smart agent's "nighttime noise reduction request" action did not trigger any prohibited rules and was consistent with the community's protection goals, so it was deemed permissible.

[0226] Merchant Smart Agent's "Business Hours Extended to 24:00" Action Triggering Rules (Restrictions on nighttime operation for businesses located in sensitive residential areas) have been added to the set of actions to be corrected. The correction method involves adjusting the action's execution time, changing the closing time from 24:00 to 22:30. The corrected action will be re-evaluated: the rule was not triggered at 22:30. The condition (the preset time limit is 23:00) is deemed allowed.

[0227] The administrator agent's action of "time-based flow control in core streets and alleys" did not trigger any prohibition rules and was therefore deemed permitted. The administrator agent's action of "renovation of Provincial-level Cultural Relics Protection Building No. 3" triggered certain rules. (Renovation of provincial-level or higher protected buildings requires the use of materials that reflect the original style), and this is included in the set of actions to be revised. The revision method is to add supplementary protection measures: to add a requirement to the renovation plan to use traditional brick and tile materials and lime mortar techniques consistent with the original style. After the revision, it will be re-evaluated as permissible.

[0228] The expert agent's actions of "adding constraints on repair materials" and "adding constraints on nighttime noise" did not trigger prohibited rules and were therefore deemed permissible.

[0229] After rule-based determination, the legal action space for each agent is as follows: Resident Agent ={Demands for nighttime noise reduction, opposition to tourist influx}, merchant smart system ={Open until 22:30, maintain the status quo}, Managerial agent ={Time-based flow control, repair arrangements (including material requirements)}, expert intelligent agent ={Add noise constraints, add repair material constraints}.

[0230] Combined action combination generation: A total of 2×2×2×2=16 combined action combinations are generated.

[0231] Conflict Detection: Sixteen combined action combinations were examined. Among them, the combination (opposing tourist influx, operating until 22:30, time-limited flow control, increased noise constraints) seemingly contradicts the "opposing tourist influx" and "operating until 22:30" principles, but in essence, it represents a balance between residents' demands for limiting tourist flow and the reasonable operational needs of businesses, and does not constitute a target conflict (business utility is not below the lower limit). Another combination (nighttime noise reduction demands, maintaining the status quo, repair arrangements, increased constraints on repair materials) has a time conflict between the construction noise generated by the repair arrangements and residents' noise reduction demands during the 18:00-21:00 period; this combination was eliminated. After conflict detection, the remaining 12 non-conflicting combined action combinations were included in the comprehensive evaluation.

[0232] Ranking (using cultural fit, multi-stakeholder satisfaction, feasibility of implementation, protection benefits, and implementation costs as ranking indicators, with the weights of each indicator determined using the analytic hierarchy process as an example):

[0233] Combinations =(Nighttime noise reduction, business hours until 22:30, time-limited flow control, increased noise constraints) Multi-indicator evaluation: High cultural compatibility (all actions comply with cultural protection rules), high satisfaction among multiple stakeholders (high utility for residents, managers and experts, moderate utility for merchants), high feasibility of implementation (resources and space can meet the requirements), high protection benefits (expected to improve noise and pedestrian flow), and moderate implementation costs (flow control and supervision have certain human resource costs).

[0234] After evaluating and ranking all 12 combinations, this combination was ranked as the best overall and was determined as the target protection scheme.

[0235] The target protection scheme is output in the form of a space unit-level control command set, the specific contents of which are as follows:

[0236] 1. Spatial unit level pedestrian flow scheduling instruction: Implement time-sharing flow restriction for core streets and alleys (spatial unit ID: SX-001) from 18:00 to 21:00, and set the peak period pedestrian flow threshold parameter on the equipment side to 2500 people / hour;

[0237] 2. Equipment-side noise alarm threshold configuration: The alarm threshold for noise monitoring equipment in the core protection zone is set to 55dB, and the triggering time is after 22:30;

[0238] 3. Construction constraint parameters table for the renovation project: The constraint parameters for the renovation project of Provincial Cultural Relics Protection Building No. 3 include the material type being limited to traditional bricks and tiles and lime mortar, the maximum number of construction workers being 30, and the vibration acceleration threshold being 0.005m / s².

[0239] 4. Monitoring task distribution parameters: The monitoring task plan is distributed to the community management platform for public announcement, and the feedback channel is open for a period of 7 days.

[0240] Feedback update example: One week after the solution was implemented, the system re-collected data. Noise impact index. The community satisfaction index dropped from 0.81 to 0.63. The pedestrian carrying capacity index increased from 0.54 to 0.68. The score decreased from 0.84 to 0.48 (pedestrian flow returned to within the carrying capacity range). The system determined that the flow control and business hours adjustment rules were effective, and the priority of the relevant rules remained unchanged. However, the repair cost assessment showed that the actual cost exceeded the budget by 15%. Based on this, the system used a gradient descent method to lower the weight of similar repair actions in subsequent scoring. The corresponding adjustments will be made, and the renovation plan will be adjusted or an additional budget will be requested in the next decision-making cycle.

[0241] Based on the above method, the present invention also provides a system for encoding cultural constraint rules and solving conflict graphs for the protection of human settlement heritage communities, comprising the following eight modules:

[0242] 1. The data preprocessing module is used to acquire multi-source heterogeneous data from human settlement heritage communities and perform spatiotemporal alignment, missing value imputation, and outlier removal to obtain a standardized dataset. This module receives raw data from monitoring equipment, environmental sensors, structural monitoring sensors, community management platforms, merchant management platforms, and expert annotation platforms, and performs spatiotemporal alignment, missing value imputation, and outlier removal operations.

[0243] 2. An environmental state modeling module is used to perform normalization mapping and feature extraction based on the standardized dataset, constructing an environmental state vector with spatial unit granularity. This module converts the standardized dataset into a multidimensional state vector according to a preset mapping method and parameters.

[0244] 3. The text rule extraction and constraint encoding module is used to extract entities and relationships from cultural protection regulations texts using natural language processing to construct a cultural knowledge graph. It then transforms constraint clauses into machine-executable condition-action constraint expressions encoded with an abstract syntax tree, generating a cultural constraint rule base. This module converts textual protection requirements into structured six-tuple condition-action constraint expressions.

[0245] 4. Multi-agent decision modeling module, used to build rule-driven decision models for multiple agents, and define observation inputs, parameterized action templates and utility evaluation functions for each decision model.

[0246] 5. A legality pruning module is used to logically compare the candidate action parameters of each decision model with the condition-action constraint expression, and perform legality pruning to generate the legal action space of each decision model.

[0247] 6. Conflict graph construction and solution module, used to generate joint action combinations based on the legal action space of each decision model, construct conflict graphs and detect connected conflicting components through graph traversal algorithm, and eliminate joint action combinations with conflicts.

[0248] 7. Control instruction set generation module, used to determine the target protection scheme after sorting the remaining joint action combinations, and output it in the form of space unit level control instruction set and resource scheduling parameter table.

[0249] 8. Incremental update module, used to recalculate the environmental state vector based on the monitoring feedback data after execution, and to perform incremental updates on the rule parameters based on the violation event statistics and execution deviation.

[0250] The modules described above can be deployed on the same server, or distributed across multiple servers, edge nodes, or cloud platforms. The system deployment is illustrated using a project in an ancient town as an example. Figures 7 to 10 As shown.

[0251] Figure 7 This is the initial system interface. Users access the system login page via a mobile application or web link, enter their username and password to complete authentication, and then enter the main system interface. The system automatically loads the corresponding permission configuration and operation interface based on the logged-in user's role (administrator, expert, merchant, or resident), thereby enabling differentiated access for different entities on the same platform. This login verification process is automatically completed by the system server, handling user identification and role mapping without manual intervention.

[0252] like Figure 8 The image shown is a screenshot of the system's main user interface. After logging in, users enter the main user interface, which integrates the following core function entry points:

[0253] (1) Project overview viewing function, which is used to display the real-time values ​​and historical trend curves of the environmental status vector of the current community. The system automatically pulls multi-source heterogeneous data from various sensors and platform interfaces through the data preprocessing module, and automatically calculates and refreshes the status vector values ​​through the environmental status modeling module.

[0254] (2) Data upload function, which is used to upload pre-collected historical data, expert-annotated data and management rule data to the system database. After the uploaded data is automatically standardized, spatiotemporally aligned and outlier detected by the data preprocessing module, it enters the standardized dataset.

[0255] (3) Online discussion and evaluation function, which is used to initiate multi-party online consultation. Each party submits opinions and suggestions through this function. The system automatically converts the consultation results into community consultation rules through the text rule extraction and constraint coding module and writes them into the cultural constraint rule library.

[0256] (4) Analysis function, used to call the environmental status modeling module and the multi-subject decision modeling module to perform a comprehensive analysis of the current protection status.

[0257] (5) Simulation function, used to call the conflict graph construction and solution module to simulate and deduce the solution. After the operation of each of the above function entry points is triggered, the corresponding module in the system background will automatically perform data processing and calculation tasks. The user only needs to select and confirm on the front-end interface.

[0258] like Figure 9 The image shows the system's data analysis interface. This interface displays the specific analysis results for each dimension of the environmental state vector. The interface presents the current values ​​and trends of the Heritage Health Index, Population Carrying Capacity Index, Noise Impact Index, Environmental Quality Index, Community Satisfaction Index, Cultural Activity Index, and Compliance Risk Index in chart form. When a user clicks on a dimension, the system automatically invokes the environmental state modeling module to display the calculation process for that dimension, including data sources, normalization mapping parameters, weighted calculation weights, and spatial unit-level subdivision data. Simultaneously, the system automatically invokes the legality trimming module to display the matching results between the current candidate actions of each subject and the condition-action constraint expressions in the cultural constraint rule base, including the classification of allowed action sets, action sets to be corrected, and prohibited action sets, as well as the triggering criteria for each rule. This allows users to intuitively understand the process and results of the system's automatic rule judgment.

[0259] like Figure 10 The diagram shows the system's simulation interface. This interface is used to simulate and extrapolate target protection schemes. Users can select different simulation scenarios (such as peak passenger flow scenarios, extreme weather scenarios, sudden structural risk scenarios, festival event scenarios, etc.) and set parameters for each scenario (such as passenger flow ratio, environmental parameter offset values, sudden risk levels, etc.). After the parameters are set, the system automatically executes the following calculation process: First, the environmental state modeling module generates a virtual environmental state vector based on the set parameters; then, the multi-agent decision modeling module generates candidate actions for each agent based on the virtual state vector; next, the legality pruning module performs rule matching to filter the legal action space; then, the conflict graph construction and solution module generates joint action combinations and performs five types of conflict detection, using a depth-first search algorithm to eliminate joint action combinations with resource conflicts, spatial conflicts, temporal conflicts, rule conflicts, or target conflicts; finally, the control instruction set generation module performs multi-index weighted sorting on the conflict-free joint action combinations, determines the target protection scheme, and outputs the simulation results in the form of a spatial unit-level control instruction set. Users can compare the simulation results under different scenarios to evaluate the robustness and adaptability of the scheme, providing a reference for the final decision.

[0260] The above embodiments are merely illustrative examples of the present invention. In actual application and deployment, different projects can make adaptive modifications and supplements based on the characteristics of project scale, data access conditions, main body composition, and protection objectives.

[0261] The data preprocessing module and environmental status modeling module can be deployed on edge computing nodes in the community to achieve low-latency data acquisition and status updates; the text rule extraction and constraint coding module, multi-agent decision modeling module, and legality trimming module can be deployed on cloud servers to utilize sufficient computing resources; the conflict graph construction and solving module, control instruction set generation module, and incremental update module can be deployed on a dedicated server in the management center to ensure decision security and traceability.

[0262] The present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and the computer program, when executed by the processor, implements the steps of the method. The electronic device may be a server, an industrial control host, an edge computing gateway, a cloud computing node, or other device with data processing capabilities. The processor may be a general-purpose processor (such as a CPU) or a dedicated processor (such as a GPU, FPGA, ASIC, etc.). The memory may be volatile memory (such as DRAM, SRAM) or non-volatile memory (such as a hard disk, SSD, flash memory, etc.), or a combination thereof.

[0263] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method. The computer-readable storage medium may include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination thereof. Specific examples include: portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any combination thereof.

[0264] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities, characterized by: include: Multi-source heterogeneous data of human settlement heritage communities were acquired, and standardized preprocessing was performed on the acquired multi-source heterogeneous data to obtain a standardized dataset including numerical data, image and video data; Based on a standardized dataset, numerical data is normalized and mapped to a preset range, and target detection and feature extraction are performed on image and video data to construct an environmental state vector with spatial units as the granularity, which is used to characterize the community protection status. Based on the texts of cultural heritage protection regulations, local management rules, expert experience rules, and community consultation rules, a cultural knowledge graph is constructed by extracting entities and relationships through natural language processing. The constraint clauses are then transformed into machine-executable condition-action constraint expressions to generate a cultural constraint rule library. To construct rule-driven decision-making models for multiple stakeholders involved in human settlement heritage communities, each decision-making model uses the current environmental state vector as the observation input, a predefined action parameter template as the candidate action, and a utility evaluation function as the basis for quantifying action preferences. The candidate action parameters of each decision model are logically compared with the condition-action constraint expressions one by one, and legality pruning is performed to generate the legal action space of each decision model. Based on the legal action space of each decision model, a joint action combination is generated. A conflict graph is constructed with actions as nodes and resource occupation relationship and spatiotemporal interference relationship as edges. Connected conflict components in the conflict graph are detected to identify and eliminate joint action combinations with conflicts. The target protection scheme is determined after sorting the remaining joint action combinations. The environmental state vector is recalculated based on the monitoring feedback data after the implementation of the target protection scheme. When the change exceeds the preset threshold, a new round of decision-making process is triggered, and the rule parameters in the cultural constraint rule base are incrementally updated based on the statistics of violation events and the deviation of scheme implementation.

2. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The multi-source heterogeneous data includes image and video data, environmental monitoring data, structural monitoring data, community feedback data, business activity data, management rule data, and expert-annotated data. Image and video data originates from monitoring equipment; environmental monitoring data includes noise, temperature, humidity, air quality, illuminance, and pedestrian density; structural monitoring data includes crack width, settlement, tilt value, and vibration anomalies; community feedback data includes resident satisfaction ratings, written opinions, complaint records, and voting records; business activity data includes merchant customer flow statistics, operating hours, business types, and records of high-noise business activities; management rule data includes protected area boundaries, cultural relic protection levels, landscape control requirements, business restrictions, and repair requirements; and expert-annotated data includes risk area annotations, value element annotations, protection recommendations, and rule revision recommendations.

3. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The environmental state vector comprises seven dimensions: heritage health index, pedestrian carrying capacity index, noise impact index, environmental quality index, community satisfaction index, cultural activity index, and compliance risk index. The value range of each dimension is mapped to a preset interval. Among them, the heritage health index reflects the health of the building structure, the pedestrian carrying capacity index reflects the load of population density relative to the carrying capacity threshold, the noise impact index reflects the degree of noise interference to the community, the environmental quality index reflects the overall air quality and temperature and humidity, the community satisfaction index reflects residents' feedback, the cultural activity index reflects the degree of participation in traditional activities and exhibitions, and the compliance risk index reflects the degree of occurrence of violations.

4. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The cultural knowledge graph includes a set of entity nodes, a set of relation edges, and a set of relation types. The set of entity nodes includes heritage building entities, street and alley space entities, traditional activity entities, main entities, protection measure entities, risk event entities, and rule and clause entities.

5. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The rule priority in the condition-action constraint expression is determined by the source of the rule, from high to low: rules of national laws and regulations, rules of local management measures, rules of expert experience, and rules of community consultation. Within the same priority, the rule with more entity attribute dimensions involved in the triggering condition has a higher priority.

6. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The specific process for performing legality-based cropping is as follows: Candidate actions from each decision model are matched one by one with the condition-action constraint expressions in the cultural constraint rule base. Based on the matching results, candidate actions are divided into a set of allowed actions, a set of actions to be corrected, and a set of prohibited actions. Specifically, when a candidate action satisfies any prohibited condition-action constraint expression, it is included in the prohibited action set. When a candidate action does not satisfy a prohibited condition but satisfies a corrected condition-action constraint expression, it is included in the set of actions to be corrected. When a candidate action does not trigger any prohibited or corrected condition-action constraint expressions, it is included in the allowed action set. When multiple rules apply to the same candidate action, the prohibited result takes precedence over the corrected result, and the corrected result takes precedence over the allowed result.

7. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 6, characterized in that: Generating the legal action space for each decision model also includes performing at least one of the following correction methods on the actions in the action set to be corrected: adjusting the action execution area, adjusting the action execution time period, adjusting the action resource quota, replacing the action implementation object, and adding supplementary protection measures; performing legality pruning again on the corrected actions, merging the corrected actions through the second pruning with the actions in the allowed action set, and generating the legal action space for each decision model.

8. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: Detecting connected conflict components in a conflict graph includes: Resource conflict detection is used to detect conflicts where multiple actions simultaneously occupy the same budget, equipment, or construction resources. Spatial conflict detection is used to detect conflicts where multiple actions are performed in the same spatial unit and cannot be performed in parallel. Time conflict detection is used to detect conflicts where multiple actions occur within the same time window and interfere with each other. Rule conflict detection is used to detect conflicts where a combination of actions as a whole violates a rule base of cultural constraints. Target conflict detection is used to detect conflicts where a combination of actions causes the utility of a decision model to fall below a preset lower limit.

9. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The target protection scheme is output in the form of a spatial unit-level control instruction set, which includes at least one of the following: spatial unit-level equipment scheduling instructions, regional pedestrian flow threshold setting parameters, monitoring equipment alarm threshold configuration parameters, repair project construction constraint parameter table, budget resource allocation parameters, and monitoring task issuance parameters.

10. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: Candidate actions for each decision model are generated in the following way: The values ​​of each dimension of the current environmental state vector, the subject type identifier, the subject's historical action records, the attribute information of the spatial unit to which it belongs, and the cultural constraint prompt information are input into the large language model as structured prompt words. The large language model outputs a structured list of candidate action parameters according to a predefined action parameter template format. After format verification, it is used as the candidate action set of the decision model.

11. The method for encoding cultural constraint rules and solving conflict graphs for intelligent protection of human settlement heritage communities according to claim 1, characterized in that: The incremental update includes at least one of the following operations: The environmental state vector is recalculated based on the latest collected monitoring data. When the change in any dimension exceeds the preset change threshold, a new round of decision-making process is triggered. Based on the frequency of newly added violations, the priority of the corresponding rules in the cultural constraint rule base will be increased by a preset amount; Based on the newly added expert-annotated data, entity nodes and relationship edges are added to the cultural knowledge graph, and entity attribute values ​​are updated. Based on the deviation between the actual results of the scheme and the expected results, the ranking weight parameters are updated through iterative optimization and normalized projection is performed after the update. Based on the failure rate of the re-correction action, the scope of application of the correction instructions for the corresponding rules in the cultural constraint rule base is expanded. The newly added regulatory clauses and new negotiation rules are encoded according to the condition-action constraint expression format and then written into the cultural constraint rule library.

12. A system for coding cultural constraint rules and solving conflict graphs for the protection of human settlement heritage communities, characterized in that, include: The data preprocessing module is used to acquire multi-source heterogeneous data of human settlement heritage communities and perform standardized preprocessing to obtain standardized datasets; The environmental state modeling module is used to perform normalization mapping and feature extraction based on the standardized dataset to construct an environmental state vector with spatial units as the granularity. The text rule extraction and constraint encoding module is used to extract entities and relationships from cultural protection regulations texts through artificial intelligence natural language processing to construct a cultural knowledge graph, and to transform constraint clauses into machine-executable condition-action constraint expressions encoded with abstract syntax trees to generate a cultural constraint rule library. The multi-agent decision modeling module is used to build rule-driven decision models for multiple agents, and to define observation inputs, action parameter templates and utility evaluation functions for each decision model. The legality pruning module is used to logically compare the candidate action parameters of each decision model with the condition-action constraint expression, and perform legality pruning to generate the legal action space of each decision model. The conflict graph construction and solution module is used to generate joint action combinations based on the legal action space of each decision model, construct the conflict graph and detect connected conflict components, and remove joint action combinations that have conflicts. The control instruction set generation module is used to determine the target protection scheme after sorting the remaining joint action combinations, and outputs it in the form of a space unit-level control instruction set and a resource scheduling parameter table. The incremental update module is used to recalculate the environmental state vector based on the monitoring feedback data after execution, and to perform incremental updates on the rule parameters based on the violation event statistics and execution deviations.

13. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, implements the method as described in any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 11.

Images