Method for analyzing, studying and judging value of demolition land parcels based on GIS and one-household one-file data

Abstract

The invention relates to a demolition land parcel value analysis, research and judgment method based on GIS and one-household-one-file data, and the method comprises the steps: carrying out the collection and standardization processing of one-household-one-file attribute data and GIS spatial data, and achieving the integrity and spatial consistency of a structured data set; establishing accurate mapping between attributes and spatial entities by utilizing a spatial analysis technology, generating a composite analysis unit, and verifying a spatial affiliation relationship of the composite analysis unit; constructing a multi-dimensional plot evaluation model, calculating a development value and implementation cost comprehensive index by adopting feature normalization, dimension reduction and machine learning methods, and forming a semantic rule knowledge base in combination with historical text mining; generating an optimal development suggestion based on semantic reasoning and similarity matching, and outputting a policy combination and policy adaptation scheme by integrating location, ownership and corollary factors through a decision tree; according to the method, the precision of land parcel evaluation, the decision transparency and the strategy implementation are improved, and optimization of the city updating process is facilitated.

Description

Technical Field

[0001] This invention relates to data intelligent analysis and decision support technology in the field of urban renewal, and in particular to a method for value analysis and judgment of demolition plots based on GIS and household-specific data. Background Technology

[0002] In the current field of urban renewal and demolition site development value assessment, the mainstream technical solutions generally adopt the fusion analysis method based on GIS spatial data and multi-source attribute data of one household, and present the development potential and implementation cost of the proposed demolition site in an index-based manner by constructing a quantitative assessment model.

[0003] The prominent problems with existing technologies include: a "semantic gap" between model output and business decisions, making it difficult for business departments to directly adopt the evaluation results and affecting the efficiency of decision-making; in addition, in terms of model interpretability, there is a lack of systematic presentation of the supporting logic and implementation path of the evaluation results, and the causal network between evaluation data and actual business actions is lacking, affecting the transparency and credibility of the decision-making process. Summary of the Invention

[0004] This application provides a method for analyzing and judging the value of demolished land parcels based on GIS and individual household data, aiming to solve one of the problems or issues of the existing technology mentioned in the background.

[0005] This application provides a method for value analysis and assessment of demolition sites based on GIS and household-specific data, specifically including: S1: Collect "one household, one file" attribute data and GIS spatial data from urban renewal projects. The attribute data includes property type, population structure, and historical compensation records. The spatial data includes plot boundaries, plot ratio, and surrounding supporting information. Standardize and clean the "one household, one file" attribute data and perform coordinate system unification processing on the GIS spatial data.

[0006] S2: Based on GIS spatial analysis technology, a mapping relationship between household-specific data and GIS spatial data elements is established through spatial connection tools. Unstructured archive data is converted into structured analysis units with geographic coordinate association by using inclusion and intersection topology rules.

[0007] S3: Construct a multi-dimensional evaluation model for demolition sites, with separate modules for calculating development value index and implementation cost. Use a weighted summation method to generate a comprehensive evaluation index for the sites. The weight parameters of the development value module are dynamically adjusted according to regional characteristics.

[0008] S4: Utilize natural language processing technology to perform text mining on historical urban renewal project archives, extract decision-making rules for compensation standards, plot ratio adjustments, and resettlement plans, and construct a semantic rule knowledge base containing triples of plot characteristics, business actions, and implementation effects.

[0009] S5: Construct a semantic reasoning engine including a feature mapping module, a rule matching module, and a logical reasoning module. Input the comprehensive evaluation index of the land parcel into the semantic reasoning engine, calculate the optimal decision path for retrieving the similarity between the current land parcel features and the historical land parcel features in the semantic rule knowledge base, and have the logical reasoning module reason based on the mapping relationship between business actions and implementation effects in the triplet to generate a set of candidate suggestions including applications to increase the plot ratio, adjustments to compensation standards, and optimization of resettlement plans.

[0010] S6: Based on the spatial location, ownership complexity, and surrounding infrastructure maturity of the land parcel, perform multi-dimensional decision tree analysis on the candidate suggestion set, and output a combination of development strategies with implementation priority and corresponding policy adaptation schemes.

[0011] S7: Generate a visual decision map, showing the causal relationship network of 'assessment index - land parcel characteristics - recommended strategy', marking the supporting evidence sources for each recommended strategy, and automatically generating an assessment report summary containing value basis and implementation path.

[0012] S8: Establish a closed-loop mechanism for human feedback. When the actual decision deviates from the system's recommended solution, record the reasons for the deviation and update the decision weight parameters in the semantic rule knowledge base. Optimize the case matching accuracy of the semantic reasoning engine through online learning.

[0013] This application provides a method for value analysis and assessment of demolition sites based on GIS and individual household data, which has the following beneficial effects: (1) To address the problem that existing methods for assessing the development value of demolished land parcels are disconnected from business decisions and lack operability, this invention introduces a semantic mapping engine to simultaneously construct a "assessment index-business action" mapping knowledge base during the model training phase. It utilizes natural language processing technology to extract key decision-making rules from unstructured texts such as compensation standards, floor area ratio adjustment records, and resettlement plans in historical project archives, and transforms them into structured semantic rule triples. This mechanism achieves the systematic accumulation of implicit experiential knowledge in multi-source heterogeneous data, enabling the model output to move beyond numerical scoring and directly relate to specific planning applications, policy applications, or implementation processes. This significantly improves the business interpretability and action orientation of the assessment results, effectively overcoming the problems of low decision conversion rates and high departmental collaboration costs caused by the abstract nature of traditional methods.

[0014] (2) During the evaluation and implementation phase, this invention employs a semantic reasoning mechanism based on historical case matching to compare the development value index of the current land parcel with similar cases in the knowledge base from multiple dimensions. It dynamically generates a candidate suggestion set by combining contextual features such as spatial location, ownership complexity, and supporting infrastructure maturity. Furthermore, it optimizes strategies by weightedly integrating decision-making paths from multiple reference cases. Compared to the traditional model of triggering recommendations with fixed thresholds, this solution possesses stronger contextual adaptability, automatically selecting the most feasible strategy combinations under different regional policy environments and implementation conditions, significantly improving the suitability of the recommended solutions. Simultaneously, the system automatically generates an evaluation report summary containing a logical chain of "value basis—influencing factors—recommended actions," and displays the supporting evidence sources for each suggestion in a visual graph format, significantly enhancing the transparency and credibility of the decision-making process and providing strong support for cross-departmental consultation and public communication.

[0015] (3) To further enhance the system's continuous learning ability and long-term availability, this invention designs a closed-loop human feedback mechanism. When the actual decision deviates from the system's recommendation, users are allowed to input the reasons for the deviation and use them to update the semantic rule base, thereby realizing the dynamic evolution of the knowledge system. This mechanism avoids the drawbacks of traditional models that require frequent retraining or manual parameter tuning to adapt to policy changes, enabling the system to have self-optimization capabilities and maintain the relevance and timeliness of recommendations in the context of dynamic adjustments to urban renewal policies and the emergence of new development models. Overall, this solution realizes a technological leap from "data-driven assessment" to "knowledge-guided decision-making," which not only improves the efficiency of transforming assessment results into actual operations but also constructs a closed-loop intelligent decision support system that integrates historical experience, current constraints, and human feedback, significantly enhancing the scientific, standardized, and collaborative level of early-stage assessment of urban renewal projects. Attached Figure Description

[0016] Figure 1 This is the main flowchart of a method for analyzing and judging the value of demolished land parcels based on GIS and individual household data.

[0017] Figure 2 This is a sub-flowchart of a method for analyzing and assessing the value of demolished land parcels based on GIS and individual household data.

[0018] Figure 3 This is another sub-flowchart of the method for analyzing and judging the value of demolished land parcels based on GIS and individual household data. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0021] like Figure 1 As shown, this application provides a method for value analysis and assessment of demolition land parcels based on GIS and household-specific data, specifically including: S1: Collect household-specific attribute data and GIS spatial data from urban renewal projects. The attribute data includes property type, population structure, and historical compensation records. The spatial data includes plot boundaries, floor area ratio, and surrounding supporting information. Standardize and clean the household-specific attribute data and perform coordinate system unification processing on the GIS spatial data.

[0022] S2: Based on GIS spatial analysis technology, a mapping relationship between household-specific data and GIS spatial data elements is established through spatial connection tools. Unstructured archive data is converted into structured analysis units with geographic coordinate association by using inclusion and intersection topology rules.

[0023] S3: Construct a multi-dimensional evaluation model for demolition sites, with separate modules for calculating development value index and implementation cost. Use a weighted summation method to generate a comprehensive evaluation index for the sites. The weight parameters of the development value module are dynamically adjusted according to regional characteristics.

[0024] S4: Utilize natural language processing technology to perform text mining on historical urban renewal project archives, extract decision-making rules for compensation standards, plot ratio adjustments, and resettlement plans, and construct a semantic rule knowledge base containing triples of plot characteristics, business actions, and implementation effects.

[0025] S5: Construct a semantic reasoning engine including a feature mapping module, a rule matching module, and a logical reasoning module. Input the comprehensive evaluation index of the land parcel into the semantic reasoning engine, calculate the optimal decision path for retrieving the similarity between the current land parcel features and the historical land parcel features in the semantic rule knowledge base, and have the logical reasoning module reason based on the mapping relationship between business actions and implementation effects in the triplet to generate a set of candidate suggestions including applications to increase the plot ratio, adjustments to compensation standards, and optimization of resettlement plans.

[0026] S6: Based on the spatial location, ownership complexity, and surrounding infrastructure maturity of the land parcel, perform multi-dimensional decision tree analysis on the candidate suggestion set, and output a combination of development strategies with implementation priority and corresponding policy adaptation schemes.

[0027] S7: Generate a visual decision map, showing the causal relationship network of 'assessment index - land parcel characteristics - recommended strategy', marking the supporting evidence sources for each recommended strategy, and automatically generating an assessment report summary containing value basis and implementation path.

[0028] S8: Establish a closed-loop mechanism for human feedback. When the actual decision deviates from the system's recommended solution, record the reasons for the deviation and update the decision weight parameters in the semantic rule knowledge base. Optimize the case matching accuracy of the semantic reasoning engine through online learning.

[0029] Step S1 involves collecting "one household, one file" attribute data and GIS spatial data from urban renewal projects. The attribute data includes property type, population structure, and historical compensation records. The spatial data includes plot boundaries, floor area ratio, and surrounding amenities. The "one household, one file" attribute data is then standardized and cleaned, and the GIS spatial data undergoes coordinate system unification processing. Specifically, this includes: S1.1: Obtain the attribute data of each household in the urban renewal project. The attribute data includes property type, population structure, and historical compensation records. Based on the data dictionary, the field content is semantically standardized to generate a structured attribute dataset.

[0030] The input data includes original attribute data files for each household from urban renewal projects. Fields cover categories such as property type, population structure, and historical compensation records. The format is CSV or Excel, and encoding may vary. A data dictionary-based matching method (parameters: predefined standard field set, regular expression matching rules) is used to map field names to standardized terms, eliminating differences in field naming across different data sources.

[0031] Furthermore, through a data type conversion algorithm (parameters: field data type mapping table, exception type detection rules), the conversion of each attribute field from the original data type to a unified data type set is achieved, and an attribute data set with type consistency is obtained.

[0032] Furthermore, through a semantic normalization processing algorithm (parameters: synonym thesaurus, word segmentation rules, word form reduction model), the unified expression of field content in the semantic space is achieved, and a set of attribute values with consistent semantic labels is generated.

[0033] Furthermore, by adopting a coding mapping table generation method (parameters: standard coding systems such as GB / T 2260, GB / T 17710, field value - coding comparison table), the unified coding processing of attribute values is achieved, and coded attribute data that can be docked with other systems is generated.

[0034] Through a field validity detection algorithm (parameters: null value detection rules, illegal value range), the results of the previous step are converted into a structured and semantically consistent attribute data set, achieving the expected technical effect of high-quality data input.

[0035] Exemplarily, in an urban renewal project in a certain city, the original data file of one household - one file containing 500 households of residents is read. Among them, the property right type field includes non - unified names such as "private house", "state - owned property", "unit - allocated housing", etc. The population number in the population structure field is expressed in a mixed form of Chinese numerals and Arabic numerals, and the historical compensation record field contains compensation amount records in different time formats. Based on data dictionary matching, both "private house" and "private property house" are mapped to the standard field value "private property right". Using regular rules, Chinese numerals such as "three" are converted to and the type conversion algorithm is used to unify the compensation amount field into currency type. Combining with the synonym thesaurus, "state - owned property" and "public housing" are unified into "state - owned property right", and the administrative division codes corresponding to each record are generated using the GB / T 2260 standard. After validity detection, 3 records with null values in the property right type are excluded, and finally a structured attribute data set is output. Its property right type, population structure, and historical compensation record fields all meet the requirements of unified coding and semantic normalization, providing stable and high - quality attribute input for subsequent spatial data matching and model analysis.

[0036] S1.2: Collect GIS spatial data. The spatial data includes vector layers of plot boundaries, plot ratios, and surrounding supporting information. Based on the EPSG standard coordinate system, the projection conversion of the layer data is performed to obtain a spatial data set under a unified coordinate system.

[0037] The input condition is the original set of GIS spatial data from urban renewal projects. The spatial data includes plot boundary vector layers, plot ratio vector layers, and surrounding supporting information vector layers. The original layers may be based on different projection coordinate systems and spatial reference standards.

[0038] The spatial data acquisition interface module (parameters: supports WFS, WMS and local Shapefile formats) is used to realize the data capture function of the vector layer of plot boundaries, plot ratio and surrounding supporting information.

[0039] Furthermore, through a spatial data format parsing algorithm (parameter: OGC standard schema mapping rules), the structured parsing of vector layer geometric data and attribute tables is achieved, and the original spatial dataset containing coordinate point columns and attribute fields is obtained.

[0040] Furthermore, by using a coordinate system identification and matching algorithm (parameters: EPSG encoding library, CRS parsing rules), the current coordinate system of each layer in the original spatial dataset is automatically identified, and an EPSG coordinate system identifier code for each layer is generated.

[0041] Furthermore, through a projection transformation algorithm (parameters: target EPSG coordinate system, transformation accuracy threshold of 0.001 meters), vector layers under different coordinate systems are uniformly transformed to the target coordinate system. During the transformation process, the centroid translation method is used to correct latitude and longitude errors of geographic coordinates, and a spatial dataset under a unified coordinate system is generated.

[0042] Furthermore, by using a spatial geometric accuracy optimization algorithm (parameters: minimum node spacing 0.0001 degrees, simplification tolerance ratio 0.01), the geometric accuracy of polygons, polylines, and point features in the spatial dataset under a unified coordinate system is adjusted, and a standardized spatial dataset after spatial geometric error correction is generated.

[0043] Through the above projection transformation and geometric accuracy optimization methods, the original multi-source, heterogeneous coordinate system GIS layer is transformed into a high-precision spatial dataset under a unified coordinate system, achieving the technical effect of coordinate benchmark consistency for subsequent spatial association and evaluation analysis.

[0044] For example, in an urban renewal project involving demolition sites, the original coordinate system of the collected site boundary layer is EPSG:4326, the plot ratio layer is EPSG:3857, and the surrounding supporting information layer is EPSG:4490. The spatial data acquisition interface module reads three Shapefile files and parses out the geometric location and attribute fields for each. The coordinate system identification and matching algorithm calls the EPSG encoding library to identify the three different coordinate systems mentioned above, and sets the target unified coordinate system as EPSG:4526. The projection transformation algorithm uses a 7-parameter Helmert transformation for the 4326 and 4490 coordinate system layers and a Mercator back projection for the 3857 coordinate system layer, ensuring that the planar accuracy error does not exceed 0.001 meters when all layers are transformed to the EPSG:4526 coordinate system. The spatial geometric accuracy optimization algorithm sets a minimum node spacing threshold of 0.0001 degrees and a simplification tolerance ratio of 0.01. During the optimization process, redundant nodes of the plot ratio polygon layer are removed, reducing storage space and eliminating morphological distortion of the graphic boundaries. The final output spatial dataset under a unified coordinate system has geometric precision in all three types of layers that meet the requirements of subsequent spatial overlay analysis, maintains consistent coordinate references, and supports accurate mapping with 'one household, one file' attribute data.

[0045] S1.3: Perform missing value detection and completion processing on the structured attribute dataset, and fill in the missing fields based on interpolation algorithms or historical data averages to generate an attribute dataset with enhanced integrity.

[0046] S1.4: Perform topology checking and geometric repair operations on the spatial dataset under the unified coordinate system. Based on the spatial index optimization algorithm, perform self-intersection correction and redundant node removal on the polygon boundaries to generate a spatial dataset with enhanced topology consistency.

[0047] S1.5: Perform field mapping and encoding alignment between the integrity-enhanced attribute dataset and the topology-consistency-enhanced spatial dataset, and establish a mapping table between attributes and spatial data based on unique identifiers to generate structured analysis units with a unified data format.

[0048] Step S2: Based on GIS spatial analysis technology, a mapping relationship between household-specific data and GIS spatial data elements is established using spatial connectivity tools. Unstructured archival data is converted into structured analysis units with geographic coordinate associations using inclusion and intersection topology rules. Specifically, this includes: S2.1: Standardize the property address and registered population distribution information in the attribute data of each household and file, and convert them into point spatial elements based on geocoding service to obtain geographic coordinate data that matches the GIS spatial data.

[0049] S2.2: Based on the GIS platform, a spatial index database is constructed. The standardized point-based archive data and the polygon boundary data of the demolition plots are spatially overlaid and analyzed to identify the topological relationship between the archive points and the plot boundaries.

[0050] S2.3: The spatial relationship determination algorithm of inclusion and intersection is used to classify and mark the spatial relationship between archive points and plot boundaries. Spatial ownership identifiers are established for archive points that fall inside the plot or intersect with the plot boundary to achieve the initial mapping between archive data and spatial entities.

[0051] S2.4: Based on the spatial attribution identifier, perform spatial connection operation on the data of each household and the land parcel boundary data to generate a composite analysis unit with a unified spatial coordinate system. The composite analysis unit includes a joint data structure of property rights type, population structure, historical compensation records and land parcel boundaries, plot ratio, and surrounding supporting information.

[0052] S2.5: Perform spatial consistency verification on the composite analysis unit, and use buffer analysis and nearest neighbor query algorithm to detect and correct spatial mapping errors to ensure that the spatial relationship between each household file and the corresponding demolition plot is accurate.

[0053] like Figure 2 As shown, step S3 involves constructing a multi-dimensional assessment model for demolition sites, including a development value index calculation module and an implementation cost calculation module. A weighted summation method is used to generate a comprehensive assessment index for the sites. The weight parameters of the development value module are dynamically adjusted based on regional characteristics. Specifically, this includes: S3.1: Based on the attribute data and GIS spatial data collected in S1, the property rights type, population structure, historical compensation records, plot boundaries, plot ratio and surrounding supporting information of the land are normalized to eliminate the influence of different dimensions on the evaluation model and obtain a standardized feature vector set.

[0054] Based on the integrity enhancement attribute dataset and topology consistency enhancement spatial dataset obtained in step S1, the Z-score standardization method is used to achieve unified dimensionalization of continuous and discrete features such as property type, population structure, historical compensation records, land boundary coordinates, plot ratio, and surrounding facilities scores.

[0055] Furthermore, by using the Min-Max normalization method (parameters: minimum value min, maximum value max), interval mapping is achieved for numerical features such as latitude and longitude coordinates, plot ratio, and surrounding amenities ratings, compressing all feature values ​​to a minimum. Within the standard interval, so as to facilitate unified calculation of subsequent feature vectors.

[0056] Furthermore, for categorical features such as property type and historical compensation records, a One-Hot encoding method (parameter: total number of categories K) is adopted to convert each category into a K-dimensional binary vector, and the encoded vector set is optimized through a sparse matrix storage structure to reduce storage and computation pressure.

[0057] Furthermore, by using a weighted missing value imputation algorithm (parameter: feature correlation coefficient matrix R), the values ​​of other highly correlated features are introduced as weighting benchmarks during the imputation value calculation process, thereby optimizing and completing the missing values ​​of features such as population structure and surrounding supporting facilities ratings, so as to improve the completeness of features before normalization.

[0058] By using the feature splicing method, the standardized property type coding vector, population structure normalized value, historical compensation record coding vector, land parcel boundary coordinate normalized value, plot ratio normalized value, and surrounding supporting facilities score normalized value are merged to form a multi-dimensional standardized feature vector set, thereby realizing data fusion between different feature domains.

[0059] Through the above algorithms or processing methods, the results of the previous step are transformed into a unified and standardized feature vector that is adapted to subsequent dimensionality reduction and dynamic weight adjustment operations, thereby achieving the expected technical effects of structured consistency and feature dimension elimination of multi-source heterogeneous data at the input of the evaluation model.

[0060] S3.2: Based on the standardized feature vector set, the principal component analysis (PCA) method is used to reduce the dimensionality of the feature space and extract the feature subset with the highest relevance to the development value assessment, so as to reduce the computational complexity of the model and improve the robustness of the assessment, and obtain the key assessment feature factor matrix.

[0061] S3.3: Based on the key evaluation feature factor matrix, the feature weights are spatially localized using the regional geographical unit division results. A regional division algorithm based on K-means clustering is adopted to divide geographical sub-regions with similar development potential, providing a spatial partitioning basis for dynamic weight allocation and generating a regional feature weight configuration template.

[0062] S3.4: Based on the regional feature weight configuration template, combined with the historical data training set of development value influencing factors, the gradient boosting decision tree (GBDT) algorithm is used to model the development value index calculation module, construct the nonlinear mapping relationship between feature factors and development value index, and generate a development value prediction model.

[0063] Based on the regional feature weight configuration template and the historical data training set of development value influencing factors, the gradient boosting decision tree (GBDT) algorithm (parameter settings: learning rate, maximum tree depth, subsampling ratio, feature sampling ratio) is used to realize the nonlinear mapping relationship between feature factors and development value index.

[0064] Furthermore, by using a feature importance assessment method (parameters: based on information gain and split count statistics), the contribution of each development value influencing factor is ranked, and the feature importance weight vector is obtained, providing a basis for the iterative optimization of the GBDT model.

[0065] S3.5: Based on the development value index output by the development value prediction model, and combined with the demolition compensation cost, resettlement cost, infrastructure renovation cost and other dimensions calculated in the implementation cost calculation module, a weighted summation method is used to calculate the comprehensive evaluation index. The weight parameters are dynamically adjusted according to the regional characteristics to generate a comprehensive evaluation index matrix for the land parcel.

[0066] Based on the development value index output by S3.4 and the data such as demolition compensation costs, resettlement costs, and infrastructure renovation costs from the implementation cost calculation module, a multi-dimensional weighted summation method (parameters: regional feature weight, cost category weight) is adopted to achieve joint quantitative integration of development value and cost data.

[0067] Furthermore, by using a normalization method (parameter: minimum-maximum normalization interval [0,1]), the proportional standard conversion of cost data under different dimensions is achieved, and a dimensionless cost index set is obtained.

[0068] Furthermore, by using a regional feature weight dynamic adjustment algorithm (parameters: spatial partition template, feature importance coefficient), the evaluation weights are adaptively adjusted according to the changes in the features of the geographic sub-regions, and a weight adjustment matrix is ​​generated.

[0069] Furthermore, the comprehensive evaluation index matrix I is calculated using a weighted comprehensive formula: in, To develop value weight, To develop the value index, As a comprehensive cost weight, This is the total cost index.

[0070] Furthermore, a matrix iterative optimization method is used (parameters: iteration step size 0.01, convergence threshold). This enables the convergence calculation of the comprehensive index matrix under the constraints of historical cases, and outputs a stable set of comprehensive evaluation results.

[0071] By dynamically adjusting the weights and using weighted summation, the development value index and various cost data are transformed into a comprehensive land parcel evaluation index matrix, achieving accurate value and cost balance assessment under different regional conditions.

[0072] For example, in a certain urban renewal area, the development value prediction model outputs a land parcel value index of 0.72, a demolition compensation cost index of 0.65, a resettlement cost index of 0.58, and an infrastructure renovation cost index of 0.40. Regional clustering results show that the land parcel is located in a high-potential area, and its regional characteristic weights are configured as development value weights. Cost weight The above cost index is first subjected to min-max normalization to obtain the dimensionless total cost index. Substituting into the weighted summation formula: The comprehensive evaluation index was calculated. Under the constraints of historical project cases, matrix iterative optimization was performed, eventually converging to... The output heat map color grading shows that the plot is in the range of medium to high development potential and medium cost pressure, providing a reliable basis for subsequent strategy recommendations.

[0073] S3.6: Based on the comprehensive evaluation index matrix of land parcels, the evaluation results are modeled in a continuous spatial distribution using a spatial interpolation algorithm to generate a heat map of land parcel development value and a cost distribution map, providing a spatial evaluation basis for subsequent semantic reasoning and visualization expression modules.

[0074] like Figure 3 As shown, step S4 involves using natural language processing technology to perform text mining on historical urban renewal project archives, extracting decision rules for compensation standards, plot ratio adjustments, and resettlement plans, and constructing a semantic rule knowledge base containing triples of land parcel characteristics, business actions, and implementation effects. Specifically, this includes: S4.1: Preprocess the unstructured text data in the archives of historical urban renewal projects, including word segmentation, stop word removal, stemming, and named entity recognition, to obtain structured text feature vectors.

[0075] S4.2: Based on the BERT pre-trained language model, semantic embedding encoding is performed on the feature vectors of structured text to generate a text semantic representation matrix, thereby improving the accuracy of text semantic understanding.

[0076] Based on the structured text feature vectors generated by S4.1, the BERT pre-trained language model is used to perform embedding encoding on each feature vector, thereby mapping the original lexical-level features into context-dependent dense vector representations.

[0077] Furthermore, by using a multi-head self-attention mechanism to calculate the relevance weights between the input feature vector and the semantic context, global dependency information modeling of decision rule paragraphs in long texts is achieved, and a coding tensor that preserves the semantic context is obtained.

[0078] Furthermore, a positional encoding algorithm is used to incorporate positional information into the encoding tensor, thereby achieving the distinguishability of text features in the sequence order dimension and generating a unified sequence representation that adapts to a mixture of long and short sentences.

[0079] Furthermore, layer normalization and residual connection processing are performed on the encoded tensor, using a normalization parameter ε=1e-12 to ensure the stability of the deep network during gradient propagation and to generate a deep semantic embedding matrix resistant to gradient vanishing.

[0080] S4.3: A method combining rule matching and dependency parsing is adopted to identify and extract triplet information fragments related to land parcel characteristics, business actions and implementation effects from the text semantic representation matrix to form a preliminary candidate set of decision rules.

[0081] Based on the text semantic representation matrix generated in step S4.2, a rule-based matching algorithm (parameters: regularized matching mode for compensation standards, floor area ratio adjustment, and resettlement schemes) is used to initially identify target keywords and their semantic dependencies in the matrix content and establish a candidate matching index.

[0082] Furthermore, through a dependency parsing algorithm (parameter: constructing rule templates using dependency relation tag sets and syntactic trees), the dependency path of the syntactic structure containing the matched keywords is parsed, and a dependency path set containing land parcel feature description nodes, business action nodes, and implementation effect nodes is obtained.

[0083] Furthermore, by utilizing the keywords output by rule matching and the node set extracted by dependency parsing, a node relationship assembly algorithm (parameters: based on dependency path length and direction weight factor) is adopted to combine relevant nodes into triplet information fragments conforming to the format of <land parcel characteristics, business actions, implementation effects>, and generate a preliminary decision rule candidate set data structure.

[0084] Furthermore, for each triple in the candidate set, a semantic consistency scoring algorithm (parameters: based on word vector cosine similarity and context dependency distance weights) is used to evaluate the semantic coherence between elements within the triple and generate a semantic consistency score matrix.

[0085] By using a triplet filtering method, triples with scores below a set threshold from the previous step are removed and transformed into a highly consistent candidate set of triples, thereby enabling the output of high-quality decision rule data required for subsequent knowledge graph construction.

[0086] For example, in the archives of a historical urban renewal project, the text semantic representation matrix contains the field "This plot has a low plot ratio; it is recommended to apply for an increase in the plot ratio to improve its development value." A rule-matching algorithm uses regular expressions to match plot parameters and business action keywords (plot ratio, application, increase) to identify the target combination. A dependency parsing algorithm, based on the sentence's syntactic tree, obtains a dependency path set {(plot characteristics: low plot ratio), (business action: apply for an increase in plot ratio), (implementation effect: improved development value)}. A node relationship assembly algorithm combines these dependency paths into a triple <low plot ratio, apply for an increase in plot ratio, improve development value> after processing with a direction weight of 0.8 and a path length correction coefficient of 1.2. A semantic consistency scoring algorithm calculates a weighted value based on the word vector cosine similarity of 0.87 and the context dependency distance weight of 0.92, resulting in a score. ,Right now If the triplet value exceeds the threshold of 0.85, it is retained in the candidate set. After performing the above processing, the candidate set contains multiple highly consistent triplets, providing accurate input for subsequent relation classification and knowledge graph embedding. The application effect is a significant improvement in the accuracy and logical completeness of triplet extraction.

[0087] S4.4: Perform relation classification and entity alignment on the preliminary candidate set of decision rules, and perform consistency verification on the relations in the triples based on the knowledge graph embedding model (TransE) to eliminate semantic ambiguity and improve the logical consistency of the triples.

[0088] Based on the initial candidate set of decision rules as input, a relation classification algorithm (parameters: relation label set, feature vector encoding rule) is used to systematically identify the relation types of land parcel features, business actions, and implementation effects in triples, and to group relations of the same type into a unified classification label. The classification results establish a mapping structure between relation categories and semantic sets, ensuring that the input relation types for subsequent consistency checks are clear and structurally stable.

[0089] Furthermore, through an entity alignment algorithm (parameters: entity standard lexicon, similarity threshold), the alignment processing of land parcel feature entities in candidate triples with a unified land parcel feature lexicon is achieved, and synonym merging of business action and implementation effect entities based on context window vector is performed, thereby eliminating entity name differences caused by different text sources and generating a set of normalized entity triples.

[0090] Furthermore, by employing a vectorization method based on the knowledge graph embedding model TransE (parameters: embedding dimension d, number of iterations), a continuous vector space mapping of triple relationships is achieved. Vector translation operations are then used to characterize the logical chain of <land parcel features, business actions, implementation effects> in a low-dimensional space. The basic mapping formula for TransE is as follows: in, For the feature vector of the land parcel, For relation vectors, For a business action or implementation effect vector, consistency is determined by minimizing the following squared distance loss function: Where i represents the i-th candidate triple in the knowledge base, n is the total number of triples, and triples with a loss value L lower than the preset consistency threshold are judged to be logically consistent and retained to enter the knowledge base construction process.

[0091] Furthermore, illogical triples are filtered out by consistency verification results, and the weight coefficients of triples near the edge values ​​are readjusted based on the distance distribution of relation embedding to ensure that logical consistency and semantic coherence are satisfied at the same time, ultimately forming a set of triples that have been classified, aligned and logically verified.

[0092] By combining relation classification, entity alignment, and TransE embedding verification, the initial candidate triples are transformed into logically consistent, entity-standardized, and relationally clear structured decision rule data, thereby achieving high-precision construction of triple semantic rules.

[0093] For example, in a processing scenario involving archives of nine historical urban renewal projects, the initial number of candidate triples was 720, the relation classification parameters were set to five relation labels (compensation adjustment, floor area ratio application, resettlement optimization, infrastructure construction, policy adaptation), and the feature encoding dimension was 128. In the entity alignment stage, a unified land parcel feature lexicon (containing 72 standard terms such as property type, population structure, and historical compensation records) was used, with a similarity threshold set to 0.85. Finally, approximately 12% of synonymous entities in the candidate set were aligned and merged. The TransE embedding model dimension was set to 64, the training iterations were 800, and the loss function parameter was minimized using Euclidean distance as a metric. The system checks for logical consistency, filtering out approximately 7% of logically inconsistent triples. After this processing, the output set of triples contains 655 triples, all of which meet the requirements of logical consistency and semantic standardization. This significantly improves the reliability and interpretability of decision path matching in subsequent knowledge base construction tests.

[0094] S4.5: Store the triple information that has passed consistency verification into the graph database according to a unified semantic expression format, build a semantic rule knowledge base that includes land parcel features, business actions, and implementation effects, and attach weight parameters to each triple to reflect its decision-making influence in historical cases.

[0095] Step S5: Construct a semantic reasoning engine including a feature mapping module, a rule matching module, and a logical reasoning module. Input the comprehensive land parcel evaluation index into the semantic reasoning engine to calculate the optimal decision path for retrieving the similarity between the current land parcel features and historical land parcel features in the semantic rule knowledge base. The logical reasoning module then infers based on the mapping relationship between business actions and implementation effects in the triplet, generating a set of candidate suggestions including applications to increase floor area ratio, adjustments to compensation standards, and optimization of resettlement plans. Specifically, this includes: S5.1: Based on the comprehensive evaluation index of land parcels, feature space mapping is performed on historical cases in the semantic rule knowledge base to construct an initial retrieval set of candidate suggestions.

[0096] Based on the input condition of the land parcel comprehensive evaluation index, the index matrix is ​​used as a feature carrier and imported into the data preprocessing module of the semantic reasoning engine to ensure the structural consistency and dimensional uniformity of the input data during the algorithm execution process.

[0097] The feature vector normalization method is adopted (parameter: the scaling interval between the maximum and minimum values ​​is set to [0,1]) to unify the amplitude ratio of different evaluation indicators in the same feature space and generate a set of normalized land parcel feature vectors.

[0098] Furthermore, by using a feature space mapping algorithm (parameter: the mapping dimension is consistent with the feature dimension of the knowledge base), the normalized set of land parcel feature vectors is projected onto the unified feature coordinate system adopted by the semantic rule knowledge base, thereby achieving spatial alignment of features across data sources and obtaining the mapped feature representation matrix.

[0099] Furthermore, a feature weighted fusion method (parameter: weights are derived from historical case triples with added weight parameters) is used to perform weighted operations on each dimension of the mapped feature representation matrix to improve the retrieval contribution of key features in the similarity matching process, and output a weighted fusion feature matrix.

[0100] Furthermore, by constructing a candidate retrieval set (parameter: the initial set size is set to 30% of the total number of cases in the knowledge base), feature distance calculation is performed between the weighted fusion feature matrix and the knowledge base triplet feature matrix to generate an initial candidate set containing the most likely matching historical cases, providing input data for subsequent similarity screening.

[0101] Through the above mapping and fusion processing, the comprehensive evaluation index of land parcels is transformed into a feature space representation compatible with the semantic rule knowledge base, thereby optimizing the retrieval and computational efficiency of the decision reasoning engine.

[0102] For example, in an urban renewal project, the comprehensive evaluation index matrix of the land parcel contains five feature dimensions: property rights stability index (0.72), population structure index (0.65), historical compensation matching degree index (0.58), plot ratio potential index (0.83), and surrounding infrastructure maturity index (0.77). Using maximum-minimum normalization, all indicators are mapped to the [0,1] interval, with normalization results of 0.72, 0.65, 0.58, 0.83, and 0.77. Using a feature space mapping algorithm, this 5-dimensional vector is transformed to the 6-dimensional feature coordinate system used by the knowledge base, where the newly added 6th dimension represents regional geoeconomic activity, with an initial value set to 0.61. Based on the triplet-added weight parameters (such as property rights stability weight 0.25, population structure weight 0.20, historical compensation matching degree weight 0.15, plot ratio potential weight 0.20, surrounding infrastructure maturity weight 0.15, and geoeconomic activity weight 0.05), a new feature vector is generated using weighted fusion. In constructing the candidate retrieval set, the initial set size was set to 30% of the total 200 cases in the knowledge base, i.e., 60 cases. The Euclidean distance between the weighted fused feature vector and the feature vectors of each historical case was calculated, and the 60 historical cases with the smallest distance were selected from the knowledge base as the initial candidate set. This candidate set retained the historical decision samples most similar to the current land parcel's characteristics, significantly improving the accuracy and matching speed of subsequent similarity calculations.

[0103] S5.2: Use the cosine similarity algorithm to calculate the similarity between the current land parcel feature vector and the triple <land parcel features, business actions, implementation effects> in the semantic rule knowledge base, in order to filter out historical cases with similarity higher than a preset threshold.

[0104] Using the current land parcel feature vector obtained from S5.1 mapping and the triplet data in the semantic rule knowledge base as input conditions, the cosine similarity algorithm (parameter: feature vector dimension d = number of key evaluation factors) is used to measure the similarity between the current land parcel and the feature space of historical cases.

[0105] By using cosine similarity calculation and threshold filtering, the mapping results from the previous step are transformed into a set of highly relevant historical cases, thereby improving the efficiency of knowledge base retrieval and ensuring the accuracy of decision path matching.

[0106] For example, in an urban renewal project, the current plot's feature vector dimension is set to 8, including the property type code value (0.85), population structure index (0.65), historical compensation record average (0.72), plot boundary shape coefficient (0.54), plot ratio index (0.78), supporting infrastructure maturity (0.81), surrounding traffic index (0.74), and ownership complexity location signal (0.69). The corresponding feature vector for historical case A is [0.88, 0.66, 0.70, 0.55, 0.80, 0.83, 0.72, 0.68], and for case B it is [0.50, 0.40, 0.35, 0.42, 0.55, 0.65, 0.58, 0.60]. Using the above formula, the similarity between case A and the current plot is 0.994, and for case B it is 0.765. Setting a similarity threshold of 0.85, case A is selected into the high-relevance set, and case B is removed. The final output of the highly relevant set enters the semantic clustering stage. The verification results show that the decision path of the highly relevant set has a significantly improved policy fit with the current land parcel, which improves the accuracy of the candidate suggestion generation stage.

[0107] S5.3: Semantically cluster the business actions corresponding to the selected historical cases, and use the K-means algorithm to merge similar actions to generate a structured set of candidate suggestion categories.

[0108] S5.4: Based on the logical reasoning module in the semantic reasoning engine, a knowledge graph-based path matching algorithm is executed on the candidate suggestion category set to identify the optimal decision path that best matches the current plot characteristics.

[0109] S5.5: Based on the output of the optimal decision path, generate a set of candidate suggestions that include business actions such as applying for increased floor area ratio, adjusting compensation standards, and optimizing resettlement plans, and attach a corresponding predicted value for the implementation effect to each suggestion.

[0110] The input conditions are the optimal decision path set and its matching weight parameters obtained through step S5.4. The path set includes the business action type and association rules for the current land parcel characteristics.

[0111] A rule parsing algorithm (parameters: optimal decision path set, matching weight threshold) is used to identify the type of business action and extract attributes for each node in the path set.

[0112] Furthermore, through the action template generation algorithm (parameters: business action type, attribute field mapping table), the identified action type is transformed into a structured candidate suggestion record and encoded according to the unified strategy metadata format.

[0113] Furthermore, through the effect prediction model (parameters: historical implementation effect dataset, current land feature vector), the quantitative relationship between business actions and implementation effects is modeled, and the expected effect value of each candidate suggestion is obtained.

[0114] Furthermore, by using a normalized scoring algorithm (parameters: predicted implementation effect value, upper and lower limits of the effect domain), effect values ​​of different dimensions are mapped to a comparable unified scoring space, and a candidate suggestion scoring matrix is ​​generated.

[0115] The candidate suggestion set generation algorithm integrates the structured suggestion records from the previous step with the scoring matrix to construct a final candidate suggestion set that includes business actions such as applications to increase floor area ratio, adjustments to compensation standards, and optimization of resettlement plans, thereby achieving a direct link between suggestion content and quantitative effects.

[0116] For example, the feature vector of a certain plot consists of the property type field "commercial land", the population structure field "aging ratio 0.35", the historical average compensation of 1.2 million yuan, the plot boundary area of ​​5400 square meters, the planned plot ratio of 2.5, and the surrounding supporting facilities score of 0.82. The optimal decision path set includes three business actions: <plot characteristics, apply for increased plot ratio, increase development profit>, <plot characteristics, adjust compensation standard, optimize capital expenditure>, and <plot characteristics, optimize resettlement plan, improve resident satisfaction>. The rule parsing algorithm identifies the three types of actions and converts them into structured records through the action template generation algorithm. The effect prediction model adopts a gradient boosting regression tree. Based on historical data, it predicts that the profit increase value of the increased plot ratio application is 3.7 million yuan, the capital optimization value of the compensation standard adjustment is 2.1 million yuan, and the satisfaction index of the resettlement plan optimization increases by 0.18. The profit increase value and the capital optimization value are normalized to the economic benefit domain [0,500] million yuan, and the satisfaction index is normalized to the satisfaction domain [0,1]. The scoring matrix is ​​calculated as follows: Profit increase =0.74, Funds Optimization =0.42, satisfaction increased =0.18. The final candidate suggestion set is obtained by integrating the score and suggestion records, which includes three suggestions and their corresponding scores. This realizes the quantitative binding between the recommendation strategy and the expected effect, and significantly improves the scientificity and interpretability of the solution adoption.

[0117] Step S6: Based on the spatial location, ownership complexity, and surrounding infrastructure maturity characteristics of the land parcel, perform multi-dimensional decision tree analysis on the candidate suggestion set, and output a combination of development strategies with implementation priority ranking and corresponding policy adaptation schemes. Specifically, this includes: S6.1: Based on the spatial location characteristics data of the land parcel, perform spatial heat map overlay analysis to obtain the strategic value level of the current land parcel in the urban functional zoning.

[0118] S6.2: Perform fuzzy comprehensive evaluation on the ownership complexity index, and use the analytic hierarchy process (AHP) to calculate the weight coefficients of sub-items such as property type, co-ownership status, and historical disputes, so as to generate an ownership risk index as one of the bases for splitting the decision tree.

[0119] S6.3: Based on the maturity data of surrounding facilities, perform a 15-minute living circle coverage analysis, and use buffer analysis and POI density calculation to generate a supporting capability score, which serves as the input feature vector for the decision tree model.

[0120] S6.4: Construct a multi-dimensional decision tree model, inputting spatial location strategic value level, ownership risk index, and supporting capability score, and perform recursive splitting training based on the CART algorithm to generate the optimal development strategy path.

[0121] S6.5: Perform policy adaptability assessment on the development strategy path output by the decision tree, and perform rule matching based on the policy text knowledge graph to generate a policy support list and implementation constraints that are suitable for the characteristics of the current land parcel.

[0122] The input conditions for the development strategy path selection based on the output of a multi-dimensional decision tree include the set of strategy nodes, path dependency attribute data, and the spatial location, ownership complexity, and surrounding infrastructure maturity characteristics of the current land parcel.

[0123] A policy text parsing algorithm (parameters: word segmentation dictionary = urban renewal policy corpus, named entity recognition model = BERT-NER) is used to achieve structured decomposition of policy text content and extract standardized semantic vectors of policy clauses involving plot ratio, compensation standards, resettlement conditions, etc.

[0124] Furthermore, by using a policy knowledge graph construction algorithm (parameters: relation type set = support, restriction, condition; entity category set = policy clause, land parcel feature, implementation requirements), structured semantic vectors are instantiated into policy graph nodes and relation edges, resulting in a knowledge network data structure containing policy constraints.

[0125] Furthermore, a rule-based matching algorithm (parameters: matching threshold = 0.85, similarity calculation method = cosine similarity) is adopted to compare and associate the strategy nodes in the optimal development path with the policy support nodes in the policy knowledge graph, and generate a strategy-policy matching score matrix.

[0126] Furthermore, by using a constraint extraction algorithm (parameters: dependency parsing mode = constraint-oriented semantics, feature weight mode = historical case weights), differentiated implementation requirements are extracted from the matched policy nodes, and a policy implementation constraint vector is generated.

[0127] By using the above rule matching and constraint extraction processing methods, the strategy path of the previous step is transformed into a policy support list and corresponding implementation constraints that are adapted to the characteristics of the current land parcel, thereby realizing the feasibility assessment of the strategy in the policy environment.

[0128] S6.6: Based on the priority ranking of strategy implementation, combined with the policy support list and constraints, generate a feasible combination of development strategies and output a structured strategy package for subsequent visualization and feedback mechanisms.

[0129] Based on the strategy implementation priority ranking results and the generated policy support list and constraint data, a multi-dimensional strategy combination method (parameters: priority sequence, policy support weight, implementation constraint level) is used to screen the feasibility of candidate suggestion sets. Furthermore, a constraint conflict detection algorithm (parameters: policy clause keywords, local regulatory restriction set) is used to verify the constraint consistency of strategy combinations and obtain a candidate strategy comparison table that meets the policy and implementation conditions. Further, a weighted combination calculation of multiple strategies is performed based on a weighted matching optimization algorithm (parameters: strategy implementation priority, policy support degree, constraint resolution degree), generating an optimal strategy combination score matrix. Further, a combination path optimization model (parameters: strategy dependency graph, implementation step sequence length limit) is used to globally optimize the strategy paths in the score matrix and generate a set of implementation paths that satisfy dependencies and execution sequence. Through structured serialization processing, the optimized strategy combinations and implementation path sets are transformed into structured strategy packages, achieving organic integration of strategy content, policy basis, and implementation constraints, for subsequent visualization and manual feedback mechanisms.

[0130] For example, in an urban renewal project, the input strategy implementation priority sequence is [application to increase plot ratio (priority value 85), compensation standard adjustment (priority value 75), resettlement plan optimization (priority value 80)]. The policy support list includes the "Local Plot Ratio Management Measures" (support weight 0.9), the "Demolition Compensation Guidance Standards" (support weight 0.85), and the "Resettlement Housing Construction Specifications" (support weight 0.8). Implementation constraints include 5 hard legal restrictions and 3 soft enforcement conditions. A constraint conflict detection algorithm identifies a clause conflict between the resettlement plan optimization and a local planning regulation. After clause resolution, the conflict level is reduced to an executable level. Based on the adjusted data, a weighted matching optimization algorithm is used for combined calculations, and its objective function is defined as: in, This represents the priority value for strategy implementation. As a weight for policy support, The coefficients for eliminating constraints are... The total number of strategies is calculated. The scores for increasing floor area ratio (FAR) application are 76.5, compensation standard adjustment is 63.75, and resettlement plan optimization is 64.0. Using a combined path optimization model, the implementation sequence in the strategy dependency graph is determined as FAR application → resettlement plan optimization → compensation standard adjustment. The generated structured strategy package includes the implementation steps of each strategy, the reference codes of the underlying policy clauses, and explanations of constraints. Application results show that the combined approach can significantly reduce policy review time and improve the stability of plan implementation in actual practice.

[0131] Step S7: Generate a visual decision map, displaying the causal relationship network of 'assessment index - land parcel characteristics - recommended strategy', marking the supporting evidence sources for each recommended strategy, and automatically generating an assessment report summary containing value justification and implementation path. Specifically, this includes: S7.1: Based on the comprehensive evaluation index of land parcels and land parcel characteristic data, construct a set of nodes for a causal relationship network, where nodes include evaluation index segments, land parcel characteristic items, and recommendation strategy items, in order to achieve preliminary modeling of the graph structure.

[0132] S7.2: Utilize graph database technology to model the edge connection relationship of the causal relationship network, and establish the edges in the graph based on the statistical correlation between the characteristics of land parcels and the recommendation strategy in historical decision-making cases, so as to express the basis chain of strategy recommendation.

[0133] S7.3: The causal relationship network is processed by a graph visualization rendering engine, and a node spatial distribution map is generated based on the force-directed graph algorithm to form an interactive and visual graph structure.

[0134] S7.4: Perform evidence annotation processing on the recommendation strategy nodes in the visualization map, and match the supporting evidence sources based on the triple <land parcel characteristics, business actions, implementation effects> in the semantic rule knowledge base to enhance the credibility and traceability of the recommendation strategy.

[0135] S7.5: Based on natural language generation technology, text summaries are generated for the comprehensive evaluation index of land parcels, recommendation strategies and supporting evidence. A combination of template-based and semantic filling is used to generate evaluation report summaries that include value basis and implementation path, so as to improve the business readability of the system output.

[0136] Based on the input land parcel comprehensive evaluation index matrix, recommendation strategy set and supporting evidence vector, a templated natural language generation method (parameters: predefined structure template set, syntax rule base) is used to build the basic framework of the evaluation report summary.

[0137] Furthermore, through a semantic fill algorithm (parameters: semantic map table, context dependency parser), the numerical segment labels, recommendation strategy items and corresponding supporting evidence sources in the comprehensive evaluation index matrix are embedded into placeholders in the templated framework, and a complete structured semantic fill result is obtained.

[0138] Furthermore, a rule-driven sentence restructuring algorithm (parameters: dependency syntax tree matching rules, part-of-speech constraints) is used to optimize the language sequence of the filling results and generate a preliminary report text with logical coherence and readability.

[0139] Furthermore, by combining the technical explanatory generation module (parameters: causal chain graph node sequence, influencing factor weight table), a logical chain description of "value basis - influencing factors - recommendation action" is added to each recommendation strategy, and a detailed explanation paragraph containing the reasoning path is generated.

[0140] Furthermore, a text consistency verification algorithm (parameters: vocabulary coverage threshold, fact consistency rule set) is used to perform semantic and factual verification on the report text, correct inconsistencies, and generate the final version of the evaluation report summary.

[0141] By combining template-based and semantic fill-in natural language generation technology, the comprehensive evaluation index and recommendation strategy from the previous step are transformed into structured and highly readable report text, achieving a precise alignment between evaluation results and business decision-making intentions.

[0142] For example, for the input land parcel comprehensive evaluation index matrix, let the strategic value level node label be "high" and the ownership risk index value be... The supporting capabilities were rated as follows: The corresponding recommended strategy is "apply to increase the plot ratio," with supporting evidence sourced from "historical case A." Within the templated framework, the following placeholders are predefined: {strategic value}, {ownership risk}, {supporting facilities score}, {strategy}, {evidence}. The semantic filling algorithm maps the above values ​​and labels to the placeholders, generating: "This plot has a high strategic value level, an ownership risk index of 0.35, and a supporting facilities score of 82. It is recommended to adopt the strategy of applying to increase the plot ratio, with supporting evidence from historical case A." The rule-driven sentence restructuring algorithm, while retaining the original information, optimizes the sentence order to: "Based on the reasoning results of historical case A, this plot has a high strategic value level, a low ownership risk index, and strong supporting facilities capabilities; it is recommended to apply to increase the plot ratio." The technical explanatory generation module adds a logical chain to this: "Value basis: high strategic value; Influencing factors: low ownership risk index, strong supporting facilities capabilities; Recommended action: apply to increase the plot ratio," forming a complete paragraph. The text consistency verification algorithm detected that the coverage of all words exceeded the threshold and that the facts and data were consistent, generating a summary of the final evaluation report. The application effect is that the report has a rigorous structure, traceable content, and can be directly used as a decision-making reference document for urban renewal projects.

[0143] Step S8: Establish a closed-loop mechanism for human feedback. When the actual decision deviates from the system's recommended solution, record the reasons for the deviation and update the decision weight parameters in the semantic rule knowledge base. Optimize the case matching accuracy of the semantic reasoning engine through online learning. Specifically, this includes: S8.1: Identify and record the deviation between actual decisions and system recommended strategies in urban renewal projects, and classify the causes of deviation based on a deviation type classification model. The causes of deviation include policy restrictions, ownership complexity, public opinion, etc., in order to generate a structured deviation feedback sample set.

[0144] S8.2: Based on the land parcel feature description information in the deviation feedback sample set, the semantic similarity calculation model is used to calculate the matching degree between it and the existing <land parcel features, business actions, implementation effects> triple in the semantic rule knowledge base, so as to locate the most similar historical decision cases and their corresponding recommendation paths.

[0145] S8.3: Perform reverse evaluation of the implementation effect of the matched historical decision cases, dynamically adjust the causal relationship strength between business actions and implementation effects in the semantic rule triples based on the actual deviation results, and output the updated semantic rule weight parameter set.

[0146] S8.4: Input the updated set of semantic rule weight parameters into the semantic reasoning engine, and use an incremental knowledge fusion algorithm to update the decision path weight matrix in the semantic reasoning model online, so as to optimize the generation quality of candidate suggestion sets in subsequent land parcel evaluation.

[0147] S8.5: Based on the updated semantic reasoning engine, it performs case retrieval and recommendation strategy generation, continuously monitors the consistency between the recommendation strategy and the actual decision, and outputs a model optimization effect evaluation report through a feedback closed-loop log recording mechanism to support the continuous iterative optimization of the system's recommendation capabilities.

[0148] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0149] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0150] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for value analysis and assessment of demolition sites based on GIS and household-by-household data, characterized in that, Specifically, it includes: S1: Collect household-specific attribute data and GIS spatial data from urban renewal projects, and perform standardized cleaning of the household-specific attribute data and coordinate system unification processing on the GIS spatial data. S2: Establish a mapping relationship between household-specific data and GIS spatial data elements through spatial connection tools; S3: Construct a multi-dimensional evaluation model for demolition sites, with separate modules for calculating development value index and implementation cost, and use a weighted summation method to generate a comprehensive evaluation index for the sites; S4: Conduct text mining on historical urban renewal project archives to extract decision-making rules for compensation standards, plot ratio adjustments, and resettlement plans, and construct a semantic rule knowledge base containing triplets of plot characteristics, business actions, and implementation effects; S5: Construct a semantic reasoning engine including a feature mapping module, a rule matching module, and a logical reasoning module. Input the comprehensive evaluation index of the land parcel into the semantic reasoning engine, calculate the similarity between the current land parcel features and the historical land parcel features in the semantic rule knowledge base to retrieve the optimal decision path, and generate a set of candidate suggestions including applications to increase the plot ratio, adjustments to compensation standards, and optimization of resettlement plans based on the mapping relationship between business actions and implementation effects in the triplet. S6: Based on the spatial location, ownership complexity, and surrounding infrastructure maturity of the land parcel, perform multi-dimensional decision tree analysis on the candidate suggestion set, and output a combination of development strategies with implementation priority and corresponding policy adaptation schemes.

2. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 1, characterized in that, Step S6 is followed by: S7: Generate a visual decision map, showing the causal relationship network of "assessment index - land parcel characteristics - recommended strategy", marking the supporting evidence sources for each recommended strategy, and automatically generating an assessment report summary containing value basis and implementation path; S8: Establish a closed-loop mechanism for human feedback. When the actual decision deviates from the system's recommended solution, record the reasons for the deviation and update the decision weight parameters in the semantic rule knowledge base. Optimize the case matching accuracy of the semantic reasoning engine through online learning.

3. The method for analyzing and judging the value of demolition sites based on GIS and household-by-household data as described in claim 1, characterized in that, The attribute data for each household includes property type, population structure, and historical compensation records.

4. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 1, characterized in that, The GIS spatial data includes plot boundaries, plot ratio, and surrounding supporting information.

5. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 4, characterized in that, The spatial data acquisition interface module is used to capture data from the land parcel boundary, plot ratio, and surrounding supporting information vector layers to obtain the land parcel boundary vector layer, plot ratio vector layer, and surrounding supporting information vector layer.

6. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 5, is characterized in that... The geometric data and attribute tables of the land parcel boundary vector layer, plot ratio vector layer, and surrounding supporting information vector layer are structured and analyzed to obtain the original spatial dataset containing coordinate point columns and attribute fields. Based on the EPSG standard coordinate system, the vector layers in the original spatial dataset are uniformly transformed to the target coordinate system to obtain a spatial dataset under a unified coordinate system.

7. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 2, characterized in that, The aforementioned closed-loop mechanism for human feedback categorizes and attributes deviations from system recommendations, with deviation types covering policy restrictions, ownership complexity, and public opinion.

8. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 1, characterized in that, Step S4 specifically includes: Unstructured text data in historical urban renewal project archives are preprocessed to obtain structured text feature vectors. Based on the BERT pre-trained language model, semantic embedding encoding is performed on the feature vectors of structured text to generate a text semantic representation matrix; A method combining rule matching and dependency parsing is adopted to identify and extract triplet information fragments related to land parcel characteristics, business actions and implementation effects from the text semantic representation matrix, so as to form a preliminary candidate set of decision rules; The preliminary candidate set of decision rules is subjected to relation classification and entity alignment processing, and the consistency of relations in triples is verified based on the knowledge graph embedding model TransE. The triple information that has passed consistency verification is stored in a graph database according to a unified semantic expression format, and a semantic rule knowledge base containing land parcel features, business actions, and implementation effects is constructed.

9. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 8, characterized in that, Preprocessing of unstructured text data in historical urban renewal project archives includes word segmentation, stop word removal, stemming, and named entity recognition.

10. The method for value analysis and assessment of demolition land parcels based on GIS and household-by-household data as described in claim 1, characterized in that, Step S5 specifically includes: Based on the comprehensive land parcel evaluation index, feature space mapping is performed on historical cases in the semantic rule knowledge base to construct an initial retrieval set of candidate suggestions; The cosine similarity algorithm is used to calculate the similarity between the current land parcel feature vector and the triplet land parcel features, business actions, and implementation effects in the semantic rule knowledge base, so as to filter out historical cases with similarity higher than a preset threshold. Semantic clustering is performed on the business actions corresponding to the selected historical cases, and the K-means algorithm is used to merge similar actions to generate a structured set of candidate suggestion categories. A knowledge graph-based path matching algorithm is performed on the set of candidate suggestion categories to identify the optimal decision path that best matches the characteristics of the current parcel; Based on the output of the optimal decision path, a set of candidate suggestions is generated, including business actions such as applying for increased floor area ratio, adjusting compensation standards, and optimizing resettlement plans. Each suggestion is then accompanied by a corresponding predicted value for its implementation effect.

Images