Knowledge base intelligent retrieval and knowledge reasoning method and system oriented to urban styles and features

Abstract

The invention discloses a knowledge base intelligent retrieval and knowledge reasoning method and system oriented to city styles and features, and the method comprises the steps: collecting multi-source city data of a preset city, and carrying out the preprocessing of the multi-source city data; constructing a three-dimensional matrix based on style stability, style freshness and development kinetic energy through the multi-source city data, dividing city modes according to the three-dimensional matrix to obtain mode data, and constructing a style topology network according to the multi-source city data and the mode data; a graph neural network is adopted to carry out community detection on the text topology network to identify a style and appearance cluster, the style and appearance cluster is evaluated according to a four-quadrant matrix to obtain a text value degree, and style and appearance causes are deduced according to the text topology network; and constructing an intelligent retrieval and knowledge reasoning model of the urban style and appearance knowledge base according to the value degree of the veins, optimizing the intelligent retrieval and knowledge reasoning model of the urban style and appearance knowledge base according to the causes of the style and appearance, and outputting a retrieval reasoning result.

Description

Technical Field

[0001] This invention relates to the field of intelligent knowledge base retrieval and knowledge reasoning, and in particular to a method and system for intelligent knowledge base retrieval and knowledge reasoning oriented towards urban landscape. Background Technology

[0002] Against the backdrop of rapid urbanization and accelerated urban renewal, the protection, inheritance, and rational development of urban landscape, as a carrier of regional culture and the core of urban characteristics, have become important issues. Current urban planning and landscape management face challenges such as difficulty in integrating multi-source data, ambiguity in identifying landscape features, and a lack of systematic value assessment. Traditional urban landscape research relies heavily on manual surveys and qualitative analysis, which is inefficient and highly subjective, making it difficult to handle the processing needs of massive amounts of multi-dimensional urban data (such as architectural form, historical culture, spatial layout, and human activities).

[0003] Meanwhile, existing retrieval and reasoning methods lack consideration for the relevance and dynamic development of urban context, failing to accurately match the characteristics and causes of urban landscape clusters. This results in insufficient grasp of the continuity of historical context and the adaptability to innovative development in planning decisions. Furthermore, the fragmented and unstructured nature of multi-source heterogeneous data further exacerbates the difficulty of urban landscape knowledge mining and intelligent application, hindering the formation of a scientific urban landscape assessment system and an efficient retrieval and reasoning mechanism, thus restricting the precise implementation of urban landscape protection and sustainable development. Therefore, there is an urgent need to construct a set of urban landscape knowledge base retrieval and reasoning methods that integrate multi-source data and incorporate intelligent algorithms to overcome current technical bottlenecks. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent retrieval and knowledge reasoning method for knowledge bases oriented towards urban landscape.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Collect multi-source city data for a preset city, and preprocess the multi-source city data; the multi-source city data includes image metadata, building morphology parameters, historical and cultural data, spatial data, dynamic development data, and human activity data; A three-dimensional matrix based on contextual stability, contextual freshness, and development momentum is constructed using the multi-source city data. City patterns are divided according to the three-dimensional matrix to obtain pattern data. A contextual topology network is constructed based on the multi-source city data and the pattern data. A graph neural network is used to perform community detection and identify landscape clusters on the context topology network. The context value is obtained by evaluating the landscape clusters based on the four-quadrant matrix. The causes of landscape are deduced based on the context topology network. Based on the stated contextual value, an intelligent retrieval and knowledge reasoning model for urban landscape knowledge base is constructed. Based on the stated causes of landscape, the intelligent retrieval and knowledge reasoning model for urban landscape knowledge base is optimized. The data to be retrieved and reasoned is input into the intelligent retrieval and knowledge reasoning model for urban landscape knowledge base, and the retrieval and reasoning results are output.

[0006] Furthermore, the method for constructing a three-dimensional matrix based on contextual stability, contextual freshness, and development momentum using the aforementioned multi-source urban data includes: Historical and cultural data, spatial data, dynamic development data, and human activities data are preprocessed to obtain a structured format, and the number of historical buildings under protection, the total number of buildings in the region, the retention rate of cultural symbols, the total number of cultural facilities, the proportion of the youth population in the region, and the infrastructure renewal rate are obtained. Based on GIS topology analysis, the degree of conformity between the existing traditional street and alley network and the historical map is calculated to obtain the integrity of the historical street and alley texture; the number of emerging cultural facilities is obtained based on the number of newly added cultural and creative spaces, art districts, and trendy commercial facilities in the past 5 years; the social media mention rate is obtained through the standardized value of the annual check-in volume on social media platforms in the region; and the industrial upgrading coefficient is obtained based on the change rate of the proportion of cultural creativity and digital economy in the tertiary industry. Calculate contextual stability, contextual freshness, and development momentum:

[0007]

[0008]

[0009] in The number of historical buildings protected in the area. This represents the total number of buildings in the area. To preserve the integrity of the historical street layout, For the retention rate of cultural symbols, The number of emerging cultural facilities, The total number of cultural facilities, For social media mentions, The proportion of the region's youth population, For the stability of the cultural context, For the sake of the freshness of the cultural context, To develop driving forces, This represents the annual growth rate of fixed asset investment. As the industrial upgrading coefficient, For infrastructure renewal rate, Weighting historical facilities As a key element of the historical streetscape, As cultural symbols weight, Weighting of emerging cultural facilities For social media weight, Weighting of the youth population As the weighting of fixed asset investment, For industrial upgrading weight, Weighting for facility upgrades; Construct a three-dimensional matrix with stability as the X-axis, freshness as the Y-axis, and development momentum as the Z-axis.

[0010] Furthermore, the method for obtaining pattern data by dividing the city pattern according to the three-dimensional matrix includes: Based on the city's overall development strategy and the requirements for the protection of historical features, the k-means clustering algorithm is used to cluster the cultural context stability value, cultural context freshness value, and development momentum value within the sample area, and the threshold values ​​of the three-dimensional indicators in the three-dimensional matrix are automatically generated. Urban development models are categorized using a combination of three-dimensional indicators, including continuation model, transformation model, derivative model, and mutation model. When stability is greater than 70, freshness is less than 30, and development momentum is less than 40, the urban development model is a continuation model; when stability is greater than or equal to 30 and less than or equal to 70, freshness is greater than or equal to 30 and less than or equal to 60, and development momentum is greater than or equal to 40 and less than 60, the urban development model is a transformation model; when stability is less than 50, freshness is greater than or equal to 60, and development momentum is greater than or equal to 60, the urban development model is a derivative model; and when stability is less than 30, freshness is greater than or equal to 60, and development momentum is less than 40, the urban development model is a mutation model. If indicators overlap, a weighted voting method is used to determine the dominant model: calculate the matching degree of each city's development model, select the model with the highest matching degree, and if there is a tie, determine the model according to the context stability weight; output the determination result of the city's development model as model data.

[0011] Furthermore, the method for constructing a contextual topology network based on the multi-source city data and the pattern data includes: Core nodes are defined based on the characteristics of urban landscape elements, and each node is assigned three types of features: morphological attributes, contextual attributes, and pattern attributes. Core nodes include architectural nodes, street and alley nodes, cultural symbol nodes, and regional pattern nodes. Morphological attributes include height, density, textural complexity, and material similarity. Contextual attributes include historical period, protection level, and cultural type. Pattern attributes include urban development patterns and corresponding three-dimensional matrix parameter values. Edges are defined based on spatial and contextual relationships between nodes. Edge types include spatially adjacent edges, functionally dependent edges, contextually inherited edges, pattern-dependent edges, and semantically similar edges. A weighting calculation rule is given by integrating multiple dimensions of spatial, morphological, and contextual indicators, expressed as:

[0012]

[0013]

[0014]

[0015] in The weights of adjacent edges in space, The weights are the morphological similarity between nodes. Weights representing the contextual relationships between nodes. This represents the actual distance between nodes. This represents the maximum threshold distance between nodes. For the contextual stability between nodes, To ensure the freshness of the context between nodes, To provide impetus for development between nodes. As a stability weight, Weighted by freshness, To develop kinetic weight, For height, For density, Let i be the height of node i. Let the density of node i be... Let j be the height of node j. Let j be the density of node j. The ratio of adjacent sides in space. The ratio of morphological similarity. For the proportion of contextual relevance, For comprehensive weighting; Remove weakly related edges with weights less than 0.2, merge duplicate edges, automatically add isolated nodes without related edges as pattern dependent edges of the pattern nodes in their respective regions, calculate the average clustering coefficient of the network, calculate the network diameter, and if the network diameter does not meet the standard, backtrack to adjust the weight calculation rules.

[0016] Furthermore, the method for community detection and feature cluster identification using graph neural networks on the context topology network includes: The connection strength between nodes is calculated based on a graph similarity algorithm. Weights are fused from spatial topological similarity and contextual semantic similarity. A graph attention network is used as the core framework, learning the association weights between nodes through an attention mechanism. Combining connection loss and density loss, the compactness of node clusters and the discriminative power between clusters are optimized. The expression is:

[0017]

[0018] in For the total loss function, For the connection loss function, Let density loss function be used. These are the weighting coefficients. Let be the true label of the edge connecting node i and node j. An edge exists connecting node i and node j. There is no edge connecting node i and node j; The predicted connection probability of the edge connecting node i and node j; Construct a dual-view network of spatial and contextual perspectives. The spatial view inputs node coordinates and morphological parameters, while the contextual view inputs historical labels and cultural relevance. The view features are fused through an attention mechanism perceived by intercellular communication. An iterative aggregation strategy using a hierarchical graph neural network algorithm is adopted to gradually merge connected subgraphs starting from the bottom-level nodes until the convergence condition is met; the convergence condition is that the number of newly added edges is less than or equal to 1 / average size of the cluster in the previous level. By optimizing the mask matrix using GNNExplainer, the subgraph structure that contributes most to cluster partitioning is identified. Node feature masks are learned, and the most critical contextual attributes for cluster identification are selected. The expression is:

[0019] in For the masked feature subset, As a key subgraph, For the target variable, For node feature masks, For label entropy, For conditional entropy, For mutual information; output feature clusters based on contextual attributes.

[0020] Furthermore, the method for evaluating the contextual value of the landscape clusters based on the four-quadrant matrix includes: The X-axis represents the freshness of the context, the Y-axis represents the stability of the context, and the Z-axis represents the development momentum, resulting in a four-quadrant matrix; where the first quadrant represents high stability and high freshness; the second quadrant represents low stability and high freshness; the third quadrant represents low stability and low freshness; and the fourth quadrant represents high stability and low freshness. Based on the fusion of multi-source urban data, a cultural heritage value assessment system is constructed, which includes primary dimensions and secondary indicators. The primary dimensions include historical value, cultural value, and social value; the secondary indicators include historical period index, protection level index, cultural symbol abundance, relevance to historical events, public recognition, functional vitality, and continuity of inheritance. The Analytic Hierarchy Process (AHP) combined with expert scoring is used to calculate the weights of secondary indicators through eigenvalue decomposition. These secondary indicators are then standardized and normalized. A weighted sum of the secondary indicators yields the basic contextual value. Correction coefficients are determined based on whether the landscape cluster coordinates fall into the corresponding quadrant, and a Z-axis adjustment factor is introduced. Finally, the contextual value is calculated based on the basic contextual value.

[0021] in To assess the value of cultural context, Based on the value of the basic cultural context, For correction factor, This is the Z-axis adjustment factor.

[0022] Furthermore, the method for constructing the intelligent retrieval and knowledge reasoning model of the urban landscape knowledge base includes: The intelligent retrieval and knowledge reasoning model for the urban landscape knowledge base includes a data layer, a feature layer, and a reasoning layer. The user inputs target landscape features, and keywords are extracted through natural language processing, which are then mapped to visual feature vectors and contextual labels. A pre-trained CLIP model is used to calculate the cosine similarity between the input text and visual features in the image database to screen an initial candidate set. Relevant contextual entities are retrieved through knowledge graph triples. A weighted edge matrix is ​​constructed, and the topological similarity is calculated based on the height difference, density difference, and cluster difference. The height difference is the standard deviation of the average height of buildings within the cluster; the density difference is the KL divergence of the spatial distribution of building density; and the cluster difference is the Jaccard similarity coefficient of the node degree distribution. Cross-graph attention interaction is achieved through a graph matching network, and the topological similarity score between the target feature and the candidate cluster is calculated:

[0023] in Candidate reference street map, Map of the target historic district. For the image The set of nodes, For the image The set of nodes, For the image The number of nodes, For the image The number of nodes, For the image The c-th node, For the image The b-th node, For cross-attention weights, For the image and picture Topological similarity score; A comprehensive score is obtained by weighting visual similarity, contextual tag matching degree, and topological structure similarity. The candidate cluster with the highest comprehensive score is returned, along with the contextual attributes of the candidate cluster.

[0024] Furthermore, the method for optimizing the intelligent retrieval and knowledge reasoning model of the urban landscape knowledge base based on the aforementioned causes of landscape formation includes: The factors contributing to urban landscape formation are extracted from multi-source urban data. A directed weighted graph is constructed with these factors as core nodes and the types of formation as edge relationships. The factors contributing to urban landscape formation include cultural connotation, spatial form, functional needs, and environmental adaptation. Visual features of the image are extracted using the CLIP model, and basic vectors are generated by combining contextual tags. The topological similarity between the target and the candidate clusters is calculated based on the graph similarity algorithm. The causal chain features of the target landscape are extracted through the causal association network, and the cosine similarity between the causal chain of the target landscape and the causal chain of the candidate clusters is calculated. Adjust the weights of the three-dimensional vectors according to the application scenario, and output the objective weighted sum of the basic vector, topological similarity, and causal chain cosine similarity as the comprehensive similarity. Output the top 3 results in descending order of comprehensive similarity, and attach the causal matching degree.

[0025] Secondly, a knowledge base intelligent retrieval and knowledge reasoning system for urban landscape design includes: Data acquisition and preprocessing module: used to acquire multi-source urban data of a preset city and preprocess the multi-source urban data; the multi-source urban data includes image metadata, building morphology parameters, historical and cultural data, spatial data, dynamic development data, and human activity data; The urban pattern segmentation and topology network construction module is used to construct a three-dimensional matrix based on contextual stability, contextual freshness, and development momentum using the multi-source urban data, segment urban patterns according to the three-dimensional matrix to obtain pattern data, and construct a contextual topology network based on the multi-source urban data and the pattern data. The module for calculating the value of cultural features and context and deriving their causes is used to perform community detection and identification of cultural features clusters on the context topology network using a graph neural network, evaluate the cultural features clusters based on a four-quadrant matrix to obtain the cultural features value, and derive the causes of cultural features based on the context topology network. Model building and optimization module: This module is used to build an intelligent retrieval and knowledge reasoning model for the urban landscape knowledge base based on the contextual value, optimize the model based on the causes of the landscape, input the data to be retrieved and reasoned into the model, and output the retrieval and reasoning results.

[0026] The beneficial effects of this invention are: This invention relates to an intelligent knowledge retrieval and knowledge reasoning method and system for urban landscape, which has the following technical advantages compared with existing technologies: This invention offers more comprehensive data integration, fusing multi-source urban data to overcome data fragmentation limitations and achieve a multi-dimensional depiction of urban landscape. It innovates the construction methods of three-dimensional matrices and topological networks to accurately classify urban development patterns and clearly identify landscape clusters and contextual connections. Combining graph neural networks and four-quadrant matrices, it scientifically assesses the value of context and derives its causes, improving the accuracy of assessment and reasoning. The model is optimized for landscape causes and comprehensively ranks results based on multi-dimensional similarity, making the retrieved and reasoned results more aligned with actual needs. The entire process is intelligently processed, replacing traditional manual qualitative analysis, improving efficiency while reducing subjectivity, and providing reliable decision support for urban landscape protection and development. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the steps of the intelligent knowledge retrieval and knowledge reasoning method for urban landscape based on the present invention. Detailed Implementation

[0028] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0029] The present invention relates to an intelligent knowledge retrieval and knowledge reasoning method and system for urban landscape, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Collect multi-source city data for a preset city, and preprocess the multi-source city data; the multi-source city data includes image metadata, building morphology parameters, historical and cultural data, spatial data, dynamic development data, and human activity data; In the actual assessment, the core area of ​​the ancient city B was taken as the research object. The core area covers the Pingjiang Road historical and cultural block, the Guanqian Street business district, and the typical landscape units of the area surrounding the Humble Administrator's Garden. It combines the inheritance of historical and cultural heritage with the vitality of modern development, which is in line with the characteristics of multi-mode urban landscape research. The data collection period is from 2018 to 2023, covering multi-source heterogeneous data types. The public satisfaction survey adopted a stratified sampling method, with samples covering different age and occupation groups, a sample size of no less than 1,000, a Cronbach's alpha coefficient of greater than or equal to 0.8 for reliability, and validity was verified by factor analysis. Image metadata includes photos of building facades, panoramic street views, and satellite imagery within the area; sourced from drone aerial photography, street view maps, and planning department archives; preprocessing is performed at a uniform resolution of 1920×1080, using OpenCV to remove blur and extract visual features (color, texture, and outline). Building form parameters: including building height, building area, plot ratio, building density, and structural type; sourced from the property registration system, planning approval archives, and on-site measurements; preprocessing method: standardized units, outlier removal; Historical and cultural data includes: a list of historical buildings under protection, the level of cultural relics protection units, cultural symbols, and records of historical events; sourced from the archives of the Bureau of Culture and Tourism, the Ancient City B Protection Plan, and local chronicles; the preprocessing method is structured coding (protection level: national level = 3, provincial level = 2, municipal level = 1, district level = 0), and extracting keyword tags for cultural symbols; Spatial data includes street and alley network layout, plot boundaries, traffic nodes, and public space distribution; it is sourced from GIS maps and urban planning databases; the preprocessing method is a unified coordinate system (WGS-84), and the street and alley texture consistency is calculated through GIS topology analysis. Dynamic development data includes fixed asset investment amount, the proportion of tertiary industry output value, and a list of infrastructure renovation projects; sourced from the National Bureau of Statistics yearbook and the National Development and Reform Commission project database; preprocessing method is to calculate the annual growth rate (investment, output value) and the standardized infrastructure renewal rate (number of renovated facilities / total number of facilities). Humanities and social activities data include social media check-in data (Douyin, Xiaohongshu), youth population household registration statistics, and public satisfaction surveys; the data are sourced from social media platform API interfaces, public security household registration systems, and questionnaires (sample size = 1200); the preprocessing method is to standardize the check-in data (normalize the maximum value) and test the reliability of the questionnaire data; A three-dimensional matrix based on contextual stability, contextual freshness, and development momentum is constructed using the multi-source city data. City patterns are divided according to the three-dimensional matrix to obtain pattern data. A contextual topology network is constructed based on the multi-source city data and the pattern data. A graph neural network is used to perform community detection and identify landscape clusters on the context topology network. The context value is obtained by evaluating the landscape clusters based on the four-quadrant matrix. The causes of landscape are deduced based on the context topology network. In practical assessments, the formation logic of urban landscape is inverted through the node relationships and quantitative indicators in the graph structure: First, landscape clusters are identified using graph neural networks, and the synergy of elements within the clusters is judged by the strength of weighted edges between nodes; Second, combined with development pattern data in the three-dimensional matrix, when a certain area is classified as a transformation mode, the landscape renewal mechanism driven by industrial transformation can be deduced by the change in edge weight from node 1 to node 2 in the topological network; Finally, by calculating the morphological entropy of clusters and the degree of fusion of inter-cluster boundaries, the role of contextual continuity in maintaining the overall landscape is revealed. Based on the stated contextual value, an intelligent retrieval and knowledge reasoning model for urban landscape knowledge base is constructed. Based on the stated causes of landscape, the intelligent retrieval and knowledge reasoning model for urban landscape knowledge base is optimized. The data to be retrieved and reasoned is input into the intelligent retrieval and knowledge reasoning model for urban landscape knowledge base, and the retrieval and reasoning results are output.

[0030] In this embodiment, the method for constructing a three-dimensional matrix based on contextual stability, contextual freshness, and development momentum using the multi-source urban data includes: Historical and cultural data, spatial data, dynamic development data, and human activities data are preprocessed to obtain a structured format, and the number of historical buildings under protection, the total number of buildings in the region, the retention rate of cultural symbols, the total number of cultural facilities, the proportion of the youth population in the region, and the infrastructure renewal rate are obtained. Based on GIS topology analysis, the degree of conformity between the existing traditional street and alley network and the historical map is calculated to obtain the integrity of the historical street and alley texture; the number of emerging cultural facilities is obtained based on the number of newly added cultural and creative spaces, art districts, and trendy commercial facilities in the past 5 years; the social media mention rate is obtained through the standardized value of the annual check-in volume on social media platforms in the region; and the industrial upgrading coefficient is obtained based on the change rate of the proportion of cultural creativity and digital economy in the tertiary industry. Calculate contextual stability, contextual freshness, and development momentum:

[0031]

[0032]

[0033] in The number of historical buildings protected in the area. This represents the total number of buildings in the area. To preserve the integrity of the historical street layout, For the retention rate of cultural symbols, The number of emerging cultural facilities, The total number of cultural facilities, For social media mentions, The proportion of the region's youth population, For the stability of the cultural context, For the sake of the freshness of the cultural context, To develop driving forces, This represents the annual growth rate of fixed asset investment. As the industrial upgrading coefficient, For infrastructure renewal rate, Weighting historical facilities As a key element of the historical streetscape, As cultural symbols weight, Weighting of emerging cultural facilities For social media weight, Weighting of the youth population As the weighting of fixed asset investment, For industrial upgrading weight, Weighting for facility upgrades; Construct a three-dimensional matrix with stability as the X-axis, freshness as the Y-axis, and development momentum as the Z-axis; In actual assessments, the retention rate of cultural symbols is the proportion of time-honored shops and traditional handicraft sites; the infrastructure renewal rate is the proportion of newly added / renovated transportation, greening, and public service facilities in the past 5 years; the annual fixed asset investment growth rate is the average growth rate in the past 3 years; and the proportion of young population is the proportion of the resident population aged 15-35. It is 0.4. It is 0.3. It is 0.3. It is 0.3. It is 0.4. It is 0.3. It is 0.3. It is 0.4. It is 0.3; The area has 186 historical buildings under protection, 42 newly built cultural facilities, a total of 1243 buildings, 156 cultural facilities, a historical street and alley integrity rate of 0.78, a social media mention rate of 0.65, a cultural symbol preservation rate of 0.82, a youth population ratio of 0.32%, an annual fixed asset investment growth rate of 0.086, an industrial upgrading coefficient of 0.15, and an infrastructure renewal rate of 0.23. The cultural heritage stability of the core area B of the ancient city is 0.54, the cultural heritage freshness is 0.437, and the development momentum is 0.1548.

[0034] In this embodiment, the method for obtaining pattern data by dividing the city pattern according to the three-dimensional matrix includes: Based on the city's overall development strategy and the requirements for the protection of historical features, the k-means clustering algorithm is used to cluster the cultural context stability value, cultural context freshness value, and development momentum value within the sample area, and the threshold values ​​of the three-dimensional indicators in the three-dimensional matrix are automatically generated. Urban development models are categorized using a combination of three-dimensional indicators, including continuation model, transformation model, derivative model, and mutation model. When stability is greater than 70, freshness is less than 30, and development momentum is less than 40, the urban development model is a continuation model; when stability is greater than or equal to 30 and less than or equal to 70, freshness is greater than or equal to 30 and less than or equal to 60, and development momentum is greater than or equal to 40 and less than 60, the urban development model is a transformation model; when stability is less than 50, freshness is greater than or equal to 60, and development momentum is greater than or equal to 60, the urban development model is a derivative model; and when stability is less than 30, freshness is greater than or equal to 60, and development momentum is less than 40, the urban development model is a mutation model. If indicators overlap, the dominant model is determined by a weighted voting method: calculate the matching degree of each city's development model, select the model with the highest matching degree, and if there is a tie, determine the model according to the context stability weight; output the determination result of the city's development model as model data. In actual evaluation, the stability of cultural context, the freshness of cultural context, and the momentum of development are all converted into a 0-100 point system through standardized processing; The three-dimensional indicators of the 12 sub-areas of the core area of ​​the ancient city B were clustered using the k-means clustering algorithm (k=4) to generate threshold ranges. Based on the indicator combination rules, the overall three-dimensional indicators of the core area of ​​the ancient city B are (S=54, F=44, Kr=52), which meet the criteria for transformation mode (stability 30-70, freshness 30-60, development momentum 40-60). The subdivided area distribution patterns are as follows: Continuation pattern: Core section of A1 Road historical and cultural block (S=76, F=28, Kr=35); Transformation pattern: A2 Street commercial district and surrounding area of ​​Humble Administrator's Garden (S=54-62, F=40-50, Kr=48-55); Derivative pattern: A3 commercial district renewal area (S=42, F=65, Kr=63); Mutation pattern: None (there are no areas in this region with a stability of less than 30 and a freshness of greater than or equal to 60).

[0035] In this embodiment, the method for constructing a contextual topology network based on the multi-source city data and the pattern data includes: Core nodes are defined based on the characteristics of urban landscape elements, and each node is assigned three types of features: morphological attributes, contextual attributes, and pattern attributes. Core nodes include architectural nodes, street and alley nodes, cultural symbol nodes, and regional pattern nodes. Morphological attributes include height, density, textural complexity, and material similarity. Contextual attributes include historical period, protection level, and cultural type. Pattern attributes include urban development patterns and corresponding three-dimensional matrix parameter values. Edges are defined based on spatial and contextual relationships between nodes. Edge types include spatially adjacent edges, functionally dependent edges, contextually inherited edges, pattern-dependent edges, and semantically similar edges. A weighting calculation rule is given by integrating multiple dimensions of spatial, morphological, and contextual indicators, expressed as:

[0036]

[0037]

[0038]

[0039] in The weights of adjacent edges in space, The weights are the morphological similarity between nodes. Weights representing the contextual relationships between nodes. This represents the actual distance between nodes. This represents the maximum threshold distance between nodes. For the contextual stability between nodes, To ensure the freshness of the context between nodes, To provide impetus for development between nodes. As a stability weight, Weighted by freshness, To develop kinetic weight, For height, For density, Let i be the height of node i. Let the density of node i be... Let j be the height of node j. Let j be the density of node j. The ratio of adjacent sides in space. The ratio of morphological similarity. For the proportion of contextual relevance, For comprehensive weighting; Remove weakly related edges with weights less than 0.2, merge duplicate edges, automatically add isolated nodes without related edges as pattern dependent edges of the pattern nodes in their respective regions, calculate the average clustering coefficient of the network, calculate the network diameter, and if the network diameter does not meet the standard, backtrack and adjust the weight calculation rules. In actual assessment, the model attributes include the three-dimensional parameter values ​​of urban development model and corresponding contextual stability, contextual freshness, and development momentum; the maximum threshold distance is determined according to the scale of urban landscape unit, and is set at twice the average distance between core nodes in the region, which is automatically calculated through GIS spatial analysis. Building nodes: 1243, C1 garden entrance building: height 8m, density 0.8 buildings / hectare, texture complexity 0.7, material similarity 0.85; historical period 18th century, protection level national level (3), cultural type garden auxiliary building; transformation mode (S=58, F=45, Kr=50). Street and alley nodes: 326; A1 Road: width 4m, density 1.2 streets / hectare, texture complexity 0.9, material similarity 0.92; historical period 16th century, protection level municipal, cultural type traditional commercial street and alley; continuity pattern (S=76, F=28, Kr=35). Cultural symbol nodes: 48; D1 Museum: height 12m, density 0.3 buildings / hectare, texture complexity 0.6, material similarity 0.78; historical period 20th century, protection level district level (1), cultural type is intangible cultural heritage display carrier; transformation mode (S=52, F=58, Kr=53). Regional pattern nodes: 12; A1 Road area: average height 6m, density 1.5 buildings / hectare, texture complexity 0.85, material similarity 0.88; average historical age 17th century, average protection level 2.1, cultural type is historical and cultural block; continuous pattern (S=76, F=28, Kr=35). The actual distance is 50m, the maximum threshold distance is 200m, and the weight of adjacent edges in space is 0.75; building height. For 8m, It is 4m long and has a density of 4m. 0.8 The weight is 1.2, the morphological similarity weight is 0.725; the stability weight is 0.4, the freshness weight is 0.3, the development momentum weight is 0.3, and the contextual relevance weight is 42.5; the spatial proportion is 0.3, the morphological proportion is 0.3, the contextual proportion is 0.4, and the overall weight is 0.6445. This edge is retained. Remove weakly related edges with a weight less than 0.2 (a total of 124 edges were removed, accounting for 3.2% of the total number of edges); merge duplicate edges; handle isolated nodes: add 3 newly added temporary building nodes as pattern subordinate edges of the pattern nodes of the Guanqian Street area; network verification: the average clustering coefficient is 0.68 and the network diameter is 8 (meets the standard, the preset threshold is ≤10), and there is no need to backtrack to adjust the weight rules.

[0040] In this embodiment, the method for community detection and feature cluster identification using a graph neural network on the context topology network includes: The connection strength between nodes is calculated based on a graph similarity algorithm. Weights are fused from spatial topological similarity and contextual semantic similarity. A graph attention network is used as the core framework, learning the association weights between nodes through an attention mechanism. Combining connection loss and density loss, the compactness of node clusters and the discriminative power between clusters are optimized. The expression is:

[0041]

[0042] in For the total loss function, For the connection loss function, Let be the density loss function, representing the probability that nodes i and j belong to the same cluster. The number of nodes; These are the weighting coefficients. Let be the true label of the edge connecting node i and node j. An edge exists connecting node i and node j. There is no edge connecting node i and node j; The predicted connection probability of the edge connecting node i and node j; Construct a dual-view network of spatial and contextual perspectives. The spatial view inputs node coordinates and morphological parameters, while the contextual view inputs historical labels and cultural relevance. The view features are fused through an attention mechanism perceived by intercellular communication. An iterative aggregation strategy using a hierarchical graph neural network algorithm is adopted to gradually merge connected subgraphs starting from the bottom-level nodes until the convergence condition is met; the convergence condition is that the number of newly added edges is less than or equal to 1 / average size of the cluster in the previous level. By optimizing the mask matrix using GNNExplainer, the subgraph structure that contributes most to cluster partitioning is identified. Node feature masks are learned, and the most critical contextual attributes for cluster identification are selected. The expression is:

[0043] in For the masked feature subset, As a key subgraph, For the target variable, For node feature masks, For label entropy, For conditional entropy, For mutual information; output feature clusters based on contextual attributes; In the actual assessment, the weighting coefficient is 0.5. The dual-view network input includes a spatial view and a contextual view. The spatial view includes node coordinates, height, and density. The contextual view includes historical age, protection level, and cultural type labels.

[0044] In this embodiment, the method for evaluating the contextual value of the landscape clusters based on the four-quadrant matrix includes: The X-axis represents the freshness of the context, the Y-axis represents the stability of the context, and the Z-axis represents the development momentum, resulting in a four-quadrant matrix; where the first quadrant represents high stability and high freshness; the second quadrant represents low stability and high freshness; the third quadrant represents low stability and low freshness; and the fourth quadrant represents high stability and low freshness. Based on the fusion of multi-source urban data, a cultural context value assessment system is constructed, comprising primary dimensions and secondary indicators. The primary dimensions include historical value, cultural value, and social value. The secondary indicators include a historical age index, a protection level index, cultural symbol abundance, historical event relevance, public recognition, functional vitality, and continuity of inheritance. The historical age index represents the average construction date of historical buildings within a cluster; the protection level index represents the proportion of protected buildings within the cluster and their level-weighted values; cultural symbol abundance represents the quantity and uniqueness of material carriers embodying local cultural characteristics within the cluster; historical event relevance represents the strength of the cluster's association with major historical events and activities of prominent figures; public recognition represents residents' / tourists' subjective evaluation of the cluster's cultural representativeness; functional vitality represents the frequency and intensity of cultural activities within the cluster; and continuity of inheritance represents the contemporary state of inheritance of the cluster's cultural context elements. The basic context value is obtained by combining the analytic hierarchy process (AHP) with expert scoring, calculating the weights of secondary indicators through eigenvalue decomposition, standardizing and normalizing the secondary indicators, and then weighting and summing the secondary indicators. Based on the coordinates of the landscape clusters falling into the corresponding quadrants, correction coefficients are determined, and a Z-axis adjustment factor is introduced. The context value is then calculated based on the basic context value.

[0045] in To assess the value of cultural context, Based on the value of the basic cultural context, For correction factor, Z-axis adjustment factor; In actual assessments, the correction factor for the first quadrant is 1.2, for the second quadrant it is 0.8, for the third quadrant it is 0.5, and for the fourth quadrant it is 1. The weights for historical value, cultural value, and social value are 0.3 and 0.3 respectively; the weight for historical period index is 0.25, with an index value of 0.82; the weights for protection level index are 0.15 and 0.76; the weight for cultural symbol abundance is 0.15, with an index value of 0.91; the weight for historical event relevance is 0.18, with an index value of 0.85; the weight for public recognition is 0.12, with an index value of 0.93; the weight for functional vitality is 0.1, with an index value of 0.78; and the weight for inheritance continuity is 0.08, with an index value of 0.88. , This is an adjustment coefficient (ranging from 0.1 to 0.3, adaptively adjusted according to the city's development stage). This represents the standardized value of development momentum. The basic cultural context value is 0.284; the coordinates of the traditional garden-street cluster (S=67, F=36.5) fall into the fourth quadrant (high stability, low novelty), with a correction coefficient of 1.0; the development momentum is 42.5, the standardization is 0.425, and the Z-axis adjustment factor is 1+0.425×0.2=1.085; The cultural context value score is 0.308, among which the cultural context value scores of other clusters are: Modern Business-Culture Integration Cluster (78.5 points), Emerging Cultural and Creative-Trend Cluster (72.3 points), and Infrastructure-Supporting Services Cluster (65.8 points).

[0046] In this embodiment, the method for constructing the intelligent retrieval and knowledge reasoning model of the urban landscape knowledge base includes: The intelligent retrieval and knowledge reasoning model of the urban landscape knowledge base includes a data layer, a feature layer, and a reasoning layer. The data layer stores image metadata, building morphology parameters, and historical and cultural archives, constructing a multi-source heterogeneous database. The feature layer transforms urban elements into topological nodes through graph neural networks. The node attributes include visual features, contextual tags, and topological relationships. The reasoning layer combines graph similarity algorithms and rule-based reasoning to achieve knowledge reasoning from target features to cluster matching, dynamically outputting retrieval results. The user inputs target landscape features, and keywords are extracted through natural language processing, which are then mapped to visual feature vectors and contextual labels. A pre-trained CLIP model is used to calculate the cosine similarity between the input text and visual features in the image database to screen an initial candidate set. Relevant contextual entities are retrieved through knowledge graph triples. A weighted edge matrix is ​​constructed, and the topological similarity is calculated based on the height difference, density difference, and cluster difference. The height difference is the standard deviation of the average height of buildings within the cluster; the density difference is the KL divergence of the spatial distribution of building density; and the cluster difference is the Jaccard similarity coefficient of the node degree distribution. Cross-graph attention interaction is achieved through a graph matching network, and the topological similarity score between the target feature and the candidate cluster is calculated:

[0047] in Candidate reference street map, Map of the target historic district. For the image The set of nodes, For the image The set of nodes, For the image The number of nodes, For the image The number of nodes, For the image The c-th node, For the image The b-th node, For cross-attention weights, For the image and picture Topological similarity score; A comprehensive score is obtained by weighting visual similarity, contextual tag matching degree, and topological structure similarity. The candidate cluster with the highest comprehensive score and the contextual attributes of the candidate cluster are returned. In the actual evaluation, the data layer stores preprocessed data such as 1243 building nodes and 326 street and alley nodes, constructing a multi-source heterogeneous database (MySQL + Neo4j knowledge graph); the feature layer extracts node feature vectors (256 dimensions) through a graph neural network, fusing visual features (CLIP model output) with contextual tags; the inference layer combines graph matching networks and knowledge graph triple inference, setting weights for: visual similarity (0.3), contextual tag matching degree (0.3), and topological similarity (0.4). Visual similarity: The cosine similarity between the target text and the cluster image features was calculated using the CLIP model. The similarity between the traditional garden-street cluster was 0.82. Contextual label matching: The matching degree between the target label (Ming and Qing style, high historical value, moderate commercial activity) and the cluster label was 0.78. Topological similarity: The height difference was 0.12 (standard deviation of the average height of cluster buildings = 1.2m), the density difference was 0.15 (KL divergence), and the clustering difference was 0.82 (Jaccard similarity coefficient). The topological similarity score calculated using a graph matching network was 0.85. The overall score was 0.3×0.82+0.3×0.78+0.4×0.85=0.82.

[0048] In this embodiment, the method for optimizing the intelligent retrieval and knowledge reasoning model of the urban landscape knowledge base based on the causes of landscape features includes: The factors contributing to urban landscape formation are extracted from multi-source urban data. A directed weighted graph is constructed with these factors as core nodes and the types of formation as edge relationships. The factors contributing to urban landscape formation include cultural connotation, spatial form, functional needs, and environmental adaptation. Visual features of the image are extracted using the CLIP model, and basic vectors are generated by combining contextual tags. The topological similarity between the target and the candidate clusters is calculated based on the graph similarity algorithm. The causal chain features of the target landscape are extracted through the causal association network, and the cosine similarity between the causal chain of the target landscape and the causal chain of the candidate clusters is calculated. Adjust the weights of the three-dimensional vectors according to the application scenario, and output the objective weighted sum of the basic vectors, topological similarity, and causal chain cosine similarity as the comprehensive similarity. Output the top 3 results in descending order of comprehensive similarity, and attach the causal matching degree. In practical evaluation, the application scenarios are divided into three categories: landscape protection scenarios, urban renewal scenarios, and new area planning scenarios. Landscape protection scenarios: basic vector 0.1, topological similarity 0.5, causal chain cosine similarity 0.4; Urban renewal scenarios: basic vector 0.2, topological similarity 0.3, causal chain cosine similarity 0.5; New area planning scenarios: basic vector 0.3, topological similarity 0.2, causal chain cosine similarity 0.5. Causes of Cultural Connotation: Core elements include intangible cultural heritage, historical events, and literati activities; the causal chain is: gathering of literati → garden construction → development of traditional crafts → contemporary integration of culture and tourism. Causes of Spatial Form: Core elements include street and alley texture, building scale, and material craftsmanship; the causal chain is: dense river network → street and alley layout along waterways → low and dense buildings → continuation of traditional brick and wood structures. Causes of Functional Needs: Core elements include commercial services, residential needs, and public activities; the causal chain is: residents' living needs → layout of shops along the street → increased number of tourists → integration of commercial and cultural tourism functions. Causes of Environmental Adaptation: Core elements include climate conditions and geographical pattern; the causal chain is: rainy in the ancient city B → sloping roof design of buildings → street and alley drainage system → creation of landscape water system. Adjust the weights of the three-dimensional vectors: the basic vector is 0.2, the topological similarity is 0.3, and the causal chain cosine similarity is 0.5; the causal chain cosine similarity between the target landscape and the traditional garden-street cluster is 0.91, and the optimized comprehensive score is 0.874.

[0049] Secondly, a knowledge base intelligent retrieval and knowledge reasoning system for urban landscape design includes: Data acquisition and preprocessing module: used to acquire multi-source urban data of a preset city and preprocess the multi-source urban data; the multi-source urban data includes image metadata, building morphology parameters, historical and cultural data, spatial data, dynamic development data, and human activity data; The urban pattern segmentation and topology network construction module is used to construct a three-dimensional matrix based on contextual stability, contextual freshness, and development momentum using the multi-source urban data, segment urban patterns according to the three-dimensional matrix to obtain pattern data, and construct a contextual topology network based on the multi-source urban data and the pattern data. The module for calculating the value of cultural features and context and deriving their causes is used to perform community detection and identification of cultural features clusters on the context topology network using a graph neural network, evaluate the cultural features clusters based on a four-quadrant matrix to obtain the cultural features value, and derive the causes of cultural features based on the context topology network. Model building and optimization module: This module is used to build an intelligent retrieval and knowledge reasoning model for the urban landscape knowledge base based on the contextual value, optimize the model based on the causes of the landscape, input the data to be retrieved and reasoned into the model, and output the retrieval and reasoning results.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning, characterized in that, The method comprises the following steps: Collecting multi-source city data of a preset city, and preprocessing the multi-source city data; the multi-source city data comprises image metadata, building shape parameters, historical and cultural data, spatial data, dynamic development data, and human activity data; Constructing a three-dimensional matrix based on context stability, context freshness, and development energy through the multi-source city data, dividing city patterns according to the three-dimensional matrix to obtain pattern data, and constructing a context topology network according to the multi-source city data and the pattern data; Using a graph neural network to detect and identify style clusters in the context topology network, evaluating the context value degree of the style clusters according to a four-quadrant matrix, and deducing the style causes according to the context topology network; Constructing an intelligent retrieval and knowledge reasoning model of a city style knowledge base according to the context value degree, optimizing the intelligent retrieval and knowledge reasoning model of the city style knowledge base according to the style causes, inputting to-be-retrieved and reasoned data into the intelligent retrieval and knowledge reasoning model of the city style knowledge base, and outputting a retrieval and reasoning result.

2. The method according to claim 1, wherein the cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method is characterized in that, The method for constructing a three-dimensional matrix based on context stability, context freshness, and development energy through the multi-source city data comprises: Preprocessing historical and cultural data, spatial data, dynamic development data, and human activity data to obtain a structured format, and obtaining the number of historical protected buildings in a region, the total number of buildings in the region, the cultural symbol retention rate, the total number of cultural facilities, the proportion of young population in the region, and the infrastructure update rate; Calculating the degree of coincidence between the existing traditional street network and the historical map based on GIS topology analysis to obtain the historical street texture integrity, obtaining the number of emerging cultural facilities according to the number of newly added cultural and creative spaces, art districts, and trendy commercial facilities in the past five years, obtaining the social media mention degree through the annual check-in quantity standardized value of the social media platform in the region, and obtaining the industrial upgrading coefficient according to the change rate of the proportion of cultural creativity and digital economy in the tertiary industry; Calculating the context stability, context freshness, and development energy: ; ; ; wherein is the number of historical protected buildings in the area, is the total number of buildings in the area, is the completeness of the historical street texture, is the cultural symbol retention rate, is the number of emerging cultural facilities, is the total number of cultural facilities, is the social media mention degree, is the proportion of young population in the area, is the cultural context stability, is the cultural context freshness, is the development momentum, is the annual fixed asset investment growth rate, is the industrial upgrading coefficient, is the infrastructure updating rate, is the historical facility weight, is the historical street texture weight, is the cultural symbol weight, is the emerging cultural facility weight, is the social media weight, is the young population weight, is the fixed asset investment weight, is the industrial upgrading weight, is the facility updating weight; Constructing a three-dimensional matrix with stability as the X-axis, freshness as the Y-axis, and development energy as the Z-axis.

3. The method of claim 1, wherein the cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method is characterized by, The method for dividing city patterns according to the three-dimensional matrix to obtain pattern data comprises: According to the overall development strategy of the city and the historical style protection requirements, clustering the context stability value, context freshness value, and development energy value in the sample region through a k-means clustering algorithm, and automatically generating the threshold values of the three-dimensional indexes in the three-dimensional matrix; Dividing city development patterns through three-dimensional index combinations; the city development patterns include a continuation mode, a transformation mode, a derivation mode, and a mutation mode; when the stability is greater than 70, the freshness is less than 30, and the development energy is less than 40, the city development pattern is the continuation mode; when the stability is greater than or equal to 30 and less than or equal to 70, the freshness is greater than or equal to 30 and less than or equal to 60, and the development energy is greater than or equal to 40 and less than 60, the city development pattern is the transformation mode; when the stability is less than 50, the freshness is greater than or equal to 60, and the development energy is greater than or equal to 60, the city development pattern is the derivation mode; and when the stability is less than 30, the freshness is greater than or equal to 60, and the development energy is less than 40, the city development pattern is the mutation mode. If the index crosses, the dominant mode is determined by weighted voting method: calculate the matching degree of each city development mode, select the mode with the highest matching degree, if there is a tie, determine according to the weight of context stability; The determination result of the city development mode is output as mode data.

4. The method of claim 1, wherein the cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method is characterized by, The method for constructing the context topology network according to the multi-source city data and the mode data comprises: According to the characteristics of cityscape elements, define core nodes, each node is given three types of features: form attribute, context attribute and mode attribute; The core nodes include building nodes, street nodes, cultural symbol nodes and regional mode nodes; The form attribute includes height, density, texture complexity and material similarity; The context attribute includes historical age, protection level and cultural type; The mode attribute includes city development mode and corresponding three-dimensional matrix parameter value; Define edges based on the space and context association between nodes, edge types include spatial adjacent edge, functional dependent edge, context inheritance edge, mode subordinate edge and semantic similar edge; Give weight calculation rules based on spatial, form and context multidimensional indexes, the expression is: ; ; ; ; wherein is a spatial adjacent edge weight, is a node inter-form similarity weight, is a node inter-context correlation weight, is a node inter-actual distance, is a node inter-maximum threshold distance, is a node inter-context stability, is a node inter-context freshness, is a node inter-development momentum, is a stability weight, is a freshness weight, is a development momentum weight, is a height, is a density, is a height of node i, is a density of node i, is a height of node j, is a density of node j, is a spatial adjacent edge proportion, is a form similarity proportion, is a context correlation proportion, is a comprehensive weight; Remove weakly connected edges with edge weight less than 0.2, merge duplicate edges, automatically add mode subordinate edges for isolated nodes that have no association, calculate the average clustering coefficient of the network, calculate the network diameter, and if the network diameter is not up to standard, backtrack and adjust the weight calculation rules.

5. The method of claim 1, wherein the cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method is characterized by, The method for community detection and identification of landscape clusters by using graph neural network on the context topology network comprises: Calculate the connection strength between nodes based on graph similarity algorithm, weight fusion spatial topology similarity and context semantic similarity, use graph attention network as the core framework, learn the association weight between nodes through attention mechanism, combine connection loss and density loss to optimize the compactness of node clustering and the distinction between clusters, the expression is: ; ; wherein is a total loss function, is a connection loss function, is a density loss function, is a weight coefficient, is a true label of the edge connecting node i and node j, is an existence of the edge connecting node i and node j, is a non-existence of the edge connecting node i and node j; is a predicted connection probability of the edge connecting node i and node j; Build a spatial and context dual-view network, input node coordinates and form parameters in spatial view, input historical labels and cultural correlation in context view, and fuse view features through intercellular communication perception attention mechanism; Use the iterative aggregation strategy of hierarchical graph neural network algorithm to merge connected subgraphs step by step from the bottom nodes until the convergence condition is met; The convergence condition is that the number of new edges is less than or equal to 1 / cluster average size of the previous level; Optimize the mask matrix through GNNExplainer to identify the subgraph structure that contributes most to cluster division, learn node feature mask, and filter the most critical context attributes for cluster identification, the expression is: ; wherein is a masked feature subset, is a key subgraph, is a target variable, is a node feature mask, is a label entropy, is a conditional entropy, is a mutual information; output style clusters according to contextual attributes.

6. The cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method according to claim 1, characterized in that, The method for evaluating the context value of the landscape cluster according to the four-quadrant matrix comprises: X-axis represents context freshness, Y-axis represents context stability, and Z-axis represents development energy, and four-quadrant matrix is divided; The first quadrant is high stability and high freshness; The second quadrant is low stability and high freshness; The third quadrant is low stability and low freshness; The fourth quadrant is high stability and low freshness; A context value evaluation system is constructed based on multi-source city data fusion, including a first dimension and a second index. The first dimension includes historical value, cultural value, and social value. The second index includes historical age index, protection level index, cultural symbol abundance, historical event correlation, public recognition, functional vitality, and inheritance continuity. The weights of the second index are calculated by eigenvalue decomposition using the analytic hierarchy process combined with expert scoring. The second index is standardized and normalized, and the basic context value is obtained by weighted summation of the second index. The context value is calculated according to the context value degree, the coordinate of the style cluster falls into the corresponding quadrant, the correction coefficient is determined, and the Z-axis adjustment factor is introduced. ; wherein is the contextual value, is the base contextual value, is the correction factor, is the Z-axis adjustment factor.

7. The cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method according to claim 1, characterized in that, The method for constructing the intelligent retrieval and knowledge reasoning model of the city style knowledge base includes: The intelligent retrieval and knowledge reasoning model of the city style knowledge base includes data layer, feature layer, and reasoning layer. The user inputs the target style feature, extracts the keywords through natural language processing, and maps them to visual feature vectors and context labels. The pre-trained CLIP model is used to calculate the cosine similarity between the input text and the visual features in the image library to filter the preliminary candidate set. The relevant context entities are retrieved through the knowledge graph triple. A weighted edge matrix is constructed, and the topological structure similarity is calculated based on height difference, density difference, and clustering difference. The height difference is the average height standard deviation of the buildings in the cluster. The density difference is the KL divergence of the spatial distribution of building density. The clustering difference is the Jaccard similarity coefficient of node degree distribution. Cross-graph attention interaction is realized through a graph matching network to calculate the topological similarity score of the target feature and the candidate cluster. ; wherein is a candidate reference block graph, is a target historical block graph, is a graph of node sets, is a graph of node sets, is a number of nodes of graph , is a number of nodes of graph , is a c-th node of graph , is a b-th node of graph , is a cross-attention weight, is a topological similarity score of graph and graph ; The comprehensive score is obtained by weighting the visual similarity, context label matching degree, and topological structure similarity. The candidate cluster with the highest comprehensive score and the context attributes of the candidate cluster are returned.

8. The cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning method according to claim 1, characterized in that, The method for optimizing the intelligent retrieval and knowledge reasoning model of the city style knowledge base based on the style cause includes: The style cause elements are extracted from the multi-source city data, and a directed weighted graph is constructed with the style elements as core nodes and the cause types as edge relationships. The style cause elements include cultural connotation cause, spatial form cause, functional demand cause, and environmental adaptation cause. The image visual features are extracted through the CLIP model, and the basic vector is generated combined with the context label. The topological structure similarity of the target and the candidate cluster is calculated based on the graph similarity algorithm. The cause chain features of the target style are extracted through the cause association network, and the cosine similarity of the cause chain of the target style and the candidate cluster is calculated. The three-dimensional vector weight is adjusted according to the application scenario. The objective weighted sum of the basic vector, topological structure similarity, and cause chain cosine similarity is output as the comprehensive similarity. The top 3 results are output in descending order of comprehensive similarity, and the cause matching degree is attached.

9. A cityscape-oriented knowledge base intelligent retrieval and knowledge reasoning system for performing the method of any one of claims 1-8, characterized in that, It includes: A data acquisition and preprocessing module is used to collect and preprocess multi-source city data of a predetermined city. The multi-source city data includes image metadata, building form parameters, historical and cultural data, spatial data, dynamic development data, and human activity data. The urban pattern division and topology network construction module is configured to construct a three-dimensional matrix based on context stability, context freshness and development momentum by using the multi-source urban data, divide urban patterns according to the three-dimensional matrix to obtain pattern data, and construct a context topology network according to the multi-source urban data and the pattern data; The style context value degree calculation and cause derivation module is configured to identify style clusters by performing community detection on the context topology network by using a graph neural network, obtain a context value degree according to a four-quadrant matrix evaluation of the style clusters, and derive style causes according to the context topology network; The model construction and optimization module is configured to construct an urban style knowledge base intelligent retrieval and knowledge reasoning model according to the context value degree, optimize the urban style knowledge base intelligent retrieval and knowledge reasoning model according to the style causes, input to-be-retrieved and reasoned data into the urban style knowledge base intelligent retrieval and knowledge reasoning model, and output a retrieval and reasoning result.

Images