Intelligent recommendation method, device and equipment of modular building component warehouse and medium

Abstract

The application discloses an intelligent recommendation method, device and equipment of a modular building component warehouse and a medium. First, current building component information is acquired, component shape features and numerical text attribute features are determined, and a current component vector is obtained based on a multi-modal feature fusion strategy. Then, based on the vector similarity between the current component vector and the component vectors of each entity node in a component data knowledge graph, candidate building component information that meets the vector similarity screening condition is determined. Finally, corresponding supplementary candidate building component information is acquired in the component data knowledge graph and is sent to a user terminal together with the building component information. The embodiment of the application can combine the relationship network specific to the building field, i.e. the component data knowledge graph, to screen candidate entity nodes for the current building component information to be searched, and can further acquire supplementary candidate building component information corresponding to the candidate building component information, so that more accurate and more comprehensive building component information is searched and intelligently recommended for the user.

Description

Technical Field

[0001] This invention relates to the field of building information technology, and in particular to intelligent recommendation methods, devices, equipment and media for modular building component libraries. Background Technology

[0002] With the popularization of prefabricated buildings and modular design, the size of building component libraries has been expanding year by year, and the types of components, from structural components and electromechanical components to interior decoration components, have all grown exponentially. Traditional component libraries mainly search for components through keywords or categories, which is not only inefficient but also difficult to automatically provide the optimal components based on the design scenario.

[0003] Existing technologies (such as the intelligent recommendation method, system and storage medium for industrial design products disclosed in patent publication number CN120952923A) have achieved recommendations based on voice data recognition of keywords and combined with price and material matching. However, these methods are mainly applicable to industrial product recommendations. They do not construct a unique relationship network between components in the construction field to assist in more accurate retrieval and recommendation of building components, nor are they optimized for the usage characteristics of prefabricated buildings. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for intelligent recommendation of modular building component libraries. It aims to solve the problem of recommending industrial products based on voice data recognition of keywords and price and material matching in the prior art, and to build a unique relationship network between components in the construction field to assist in more accurate building component retrieval and recommendation.

[0005] In a first aspect, embodiments of the present invention provide an intelligent recommendation method for a modular building components library, comprising:

[0006] In response to a building component retrieval command sent by a user terminal, obtain the current building component information to be retrieved corresponding to the building component retrieval command;

[0007] Obtain the component shape features and numerical text attribute features corresponding to the current building component information, and obtain the corresponding current component vector based on a preset multimodal feature fusion strategy;

[0008] Based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph, candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions are determined.

[0009] Obtain supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and combine it with the candidate building component information to form comprehensive building component recommendation information;

[0010] The comprehensive building component recommendation information is sent to the user terminal.

[0011] Secondly, embodiments of the present invention also provide an intelligent recommendation device for a modular building components library, comprising:

[0012] The current building component information acquisition unit is used to respond to the building component retrieval command sent by the user terminal and acquire the current building component information to be retrieved corresponding to the building component retrieval command.

[0013] The current component vector acquisition unit is used to acquire the component shape features and numerical text attribute features corresponding to the current building component information, and to acquire the corresponding current component vector based on a preset multimodal feature fusion strategy;

[0014] The candidate information acquisition unit is used to determine candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph.

[0015] The comprehensive recommendation information acquisition unit is used to acquire supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and to form comprehensive building component recommendation information with the candidate building component information.

[0016] The recommendation information sending unit is used to send the integrated building component recommendation information to the user terminal.

[0017] Thirdly, embodiments of the present invention also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect above.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the method described in the first aspect above.

[0019] This invention provides an intelligent recommendation method, apparatus, device, and medium for a modular building component library. The method includes: responding to a building component retrieval command sent by a user terminal, obtaining the current building component information to be retrieved corresponding to the building component retrieval command; obtaining the component shape features and numerical text attribute features corresponding to the current building component information, and obtaining the corresponding current component vector based on a preset multimodal feature fusion strategy; determining candidate entity nodes and corresponding candidate building component information that meet preset vector similarity screening conditions based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in a locally preset component data knowledge graph; obtaining supplementary candidate building component information corresponding to the candidate building component information in the component data knowledge graph, and combining it with the candidate building component information to form comprehensive building component recommendation information; and sending the comprehensive building component recommendation information to the user terminal. The embodiments of the present invention can combine the unique relationship network in the construction field, namely the component data knowledge graph, to filter candidate entity nodes for the current building component information to be retrieved, and can further supplement the candidate building component information with the supplementary candidate building component information, so as to realize more accurate and comprehensive retrieval and intelligent recommendation of building component information for users. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A schematic diagram illustrating an application scenario of the intelligent recommendation method for a modular building component library provided in this embodiment of the invention;

[0022] Figure 2 A flowchart illustrating the intelligent recommendation method for a modular building component library provided in an embodiment of the present invention;

[0023] Figure 3 A schematic diagram of a sub-process of the intelligent recommendation method for a modular building component library provided in an embodiment of the present invention;

[0024] Figure 4 Another sub-process diagram of the intelligent recommendation method for the modular building component library provided in the embodiments of the present invention;

[0025] Figure 5 This is another sub-process diagram of the intelligent recommendation method for the modular building component library provided in the embodiments of the present invention;

[0026] Figure 6A schematic block diagram of an intelligent recommendation device for a modular building component library provided in an embodiment of the present invention;

[0027] Figure 7 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0030] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0031] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0032] Please also refer to Figure 1 and Figure 2 ,in Figure 1 This is a schematic diagram illustrating a scenario of the intelligent recommendation method for the modular building component library according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the intelligent recommendation method for a modular building component library provided in an embodiment of the present invention. Figure 1 As shown, the intelligent recommendation method for the modular building component library provided in this embodiment of the invention is applied to server 10, and server 10 is communicatively connected to user terminal 20. Figure 2 As shown, the method includes the following steps S110-S150.

[0033] S110. In response to the building component retrieval command sent by the user terminal, obtain the current building component information to be retrieved corresponding to the building component retrieval command.

[0034] In this embodiment, the technical solution is described with the server as the execution entity. An intelligent recommendation platform for a modular building component library is deployed on the server. This intelligent recommendation platform can be embedded in the server's BIM system (Building Information Modeling) for users to perform online building model design. After a user logs into the intelligent recommendation platform using their login information (such as username and password), they can first select the current building component information to be retrieved through one of several methods on the corresponding user interface of the intelligent recommendation platform. For example, one approach is for a user to log into the online BIM system on the server and open a building model to be designed. If the user selects one of the building models as the current design model and chooses to perform intelligent recommendation processing on that model, triggering a building component retrieval command, then the user needs to obtain the current building component information (such as the building model component name, model number, model size, model setting location, model shape, model function, model style matching information, model usage scenario information, etc.). The current building component information of that building model can then be used as the initial retrieval information.

[0035] Of course, other ways to determine the current building component information to be retrieved include having the user upload a locally designed building model to the server, and selecting one of the models as the current design model. This allows the server to retrieve the current building component information as initial search information. Regardless of how the user determines the current building component information, once the initial determination is complete, it can be used as initial search information for subsequent intelligent retrieval and recommendation processing.

[0036] S120. Obtain the component shape features and numerical text attribute features corresponding to the current building component information, and obtain the corresponding current component vector based on a preset multimodal feature fusion strategy.

[0037] In this embodiment, multi-dimensional feature information can be extracted from the current building component information. This includes component numerical and textual information such as component name, model number, model function, model style matching information, and model usage scenario information; and component shape information such as model size, model location, and model shape (which can also be the specific 3D building model data corresponding to the current building component information). By performing multimodal feature fusion on the component shape features and numerical textual attribute features corresponding to the current building component information, the current component vector corresponding to the current building component information can be obtained.

[0038] In one embodiment, such as Figure 3 As shown, step S120 includes:

[0039] S121. Obtain a pre-trained geometric coding network, and obtain the component shape features corresponding to the component shape information in the current building component information through the geometric coding network;

[0040] S122. Obtain a pre-trained attribute embedding model, and obtain the numerical text attribute features corresponding to the component values ​​and text information in the current building component information through the attribute embedding model.

[0041] S123. Obtain the cross-modal attention mechanism corresponding to the multimodal feature fusion strategy, and perform multimodal feature fusion of the part shape feature and the numerical text attribute feature based on the cross-modal attention mechanism to obtain the current part vector.

[0042] In this embodiment, the geometric encoding network is a neural network that maps the architectural geometric data (point cloud, mesh, voxels, etc.) corresponding to the component shape information in the current building component information into fixed-dimensional feature vectors. It is specifically designed to capture and preserve the spatial structure, topological relationships, and geometric characteristics of buildings, laying the foundation for subsequent analysis and fusion. Specifically, it can convert unstructured / structured geometric data into computer-understandable vector representations; extract multi-level geometric features (from local components to global morphology); maintain geometric invariance such as rotation and translation, enhancing the model's generalization ability; and provide a geometric semantic basis for BIM model, architectural design retrieval, and generation. The component shape features corresponding to the component shape information in the current building component information are obtained through a pre-trained geometric encoding network on the server.

[0043] Attribute embedding models are neural network architectures used to map component values ​​and textual information from current building component information into low-dimensional continuous vectors (e.g., 512-dimensional vectors). Their core function is to capture semantic relationships between attributes and unify heterogeneous attributes into a common semantic space, laying the foundation for subsequent cross-modal fusion, similarity calculation, and knowledge reasoning. Attribute embedding models can convert textual attributes (e.g., high-level green building rating), numerical attributes (e.g., building area of ​​100 square meters), and categorical attributes (e.g., frame structure type) into unified vector representations; they can also capture semantic relationships between attributes (e.g., the similarity between environmentally friendly materials and sustainable design); and they can solve the problem of attribute data heterogeneity, providing a foundation for subsequent fusion operations.

[0044] After extracting the shape features and numerical textual attribute features of the component, a cross-modal attention mechanism is used to fuse multimodal features, resulting in the current component vector. The cross-modal attention mechanism is an attention mechanism that solves the problem of aligning and efficiently fusing heterogeneous modalities (architectural geometry, textual attributes, and numerical parameters). Essentially, it automatically learns the correlation strength between different modalities through dynamic query-key-value interactions, thereby achieving feature fusion.

[0045] In one embodiment, such as Figure 4 As shown, step S123 includes:

[0046] S1231. Map the shape features of the component to the corresponding first preset feature dimension based on the cross-modal attention mechanism and perform normalization processing to obtain the first fusion vector;

[0047] S1232. The numerical features in the numerical text attribute features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the second vector to be fused.

[0048] S1233. Map the text features in the numerical text attribute features to the corresponding first preset feature dimension based on the cross-modal attention mechanism and perform normalization processing to obtain the third vector to be fused.

[0049] S1234. The first vector to be fused, the second vector to be fused, and the third vector to be fused are weighted and summed based on the cross-modal attention mechanism to obtain the current component vector.

[0050] In this embodiment, the cross-modal attention mechanism includes core steps such as modal projection alignment, multi-head cross-modal attention processing, scaling and normalization, and weighted fusion. Referring to the above processing procedure, the component shape features can first be mapped to the corresponding first preset feature dimension (e.g., 512 dimensions) based on the cross-modal attention mechanism and normalized to obtain a first vector to be fused; the numerical features in the numerical text attribute features can be mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain a second vector to be fused; the text features in the numerical text attribute features can be mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain a third vector to be fused. For example, 1024-dimensional component shape features, 768-dimensional text features, and 32-dimensional numerical features are projected into a 512-dimensional unified feature space. Then, a multi-head cross-modal attention mechanism (using a standard Query-Key-Value attention mechanism as input) is determined. This multi-head cross-modal attention includes at least shape-text bidirectional attention and numerical feature-guided attention, with corresponding similarity matrices between the two attentions. Based on the shape-text bidirectional attention and numerical feature-guided attention, the component shape features, text features, and numerical features are scaled and normalized to obtain the first, second, and third vectors to be fused, respectively. These are then weighted and fused to obtain the current component vector. Thus, the multi-modal feature fusion of component shape features, text features, and numerical features is achieved through the aforementioned cross-modal attention mechanism.

[0051] S130. Based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph, determine the candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions.

[0052] In this embodiment, each entity node in the component data knowledge graph is pre-built locally on the server, and the component vector is also pre-calculated. When the current component vector of the current building component information to be retrieved is obtained, the vector similarity between it and the component vector corresponding to each entity node in the component data knowledge graph can be calculated, and the candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions can be quickly determined.

[0053] Specifically, when pre-building a component data knowledge graph locally on the server, a multi-dimensional component knowledge graph is constructed, comprising entity nodes of multiple component models. Each entity node includes component geometric features (shape, size), attribute features (specifications, material, price), functional attributes (use, compatible systems), and style tags (decoration style, design semantics), among other attribute information. This multi-dimensional component knowledge graph forms a network of relationships between components. The multi-dimensional component knowledge graph between entity nodes includes at least shape similarity relationships, functional substitution relationships, style matching relationships, and usage scenario dependencies. Through this component data knowledge graph, users can quickly obtain the recommended component information by using the current building component information as initial search information.

[0054] In one embodiment, such as Figure 5 As shown, step S130 includes:

[0055] S131. Obtain the cosine similarity or Euclidean distance between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph as the vector similarity.

[0056] S132. From each entity node in the component data knowledge graph, candidate entity nodes whose vector similarity ranking with the current component vector belongs to TopK or exceeds the preset similarity threshold are selected based on the vector similarity filtering conditions, and candidate building component information corresponding to each candidate entity node is obtained.

[0057] In TopK, K is a user-defined integer value, and K is greater than or equal to 3.

[0058] In this embodiment, the cosine similarity or Euclidean distance between the current component vector and the component vectors corresponding to each entity node in the locally preset component data knowledge graph can be calculated as the vector similarity. Then, based on the vector similarity filtering criteria, candidate entity nodes whose vector similarity ranking with the current component vector is either Top K or exceeds a preset similarity threshold are selected. The selected entity nodes are the building component information and building component models that are closest to the user's needs. Of course, this initial recommendation only recommends data similar to the current building component information to be retrieved; further analysis and processing can be performed to supplement the recommendations.

[0059] S140. Obtain supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and combine it with the candidate building component information to form comprehensive building component recommendation information.

[0060] In this embodiment, in order to recommend data other than approximate data for the current building component information to be retrieved, supplementary candidate building component information corresponding to the candidate building component information can be further filtered by combining the relationship network between the entity nodes in the component data knowledge graph, and combined with the building component information to form a comprehensive building component recommendation letter, thereby realizing richer information output and recommendation.

[0061] In one embodiment, step S140 includes:

[0062] Based on the shape similarity relationship, functional substitution relationship, style matching relationship and usage scenario relationship of each entity node in the component data knowledge graph, similar components, substitute components and complementary components related to the candidate building component information are determined, and supplementary candidate building component information corresponding to the candidate building component information is formed.

[0063] In this embodiment, when determining supplementary candidate building component information related to the candidate building component information, similar components, substitute components, and complementary components related to the candidate building component information can be determined one by one based on the shape similarity relationship, functional substitution relationship, style matching relationship, and usage scenario relationship of each entity node in the component data knowledge graph. These components are then used as supplementary candidate building components obtained by further filtering from the component data knowledge graph. Finally, the supplementary candidate building component information is composed of the candidate building component information and the component information corresponding to its similar, substitute, and complementary components. This achieves intelligent association retrieval capabilities based on shape similarity, functional substitution, and style matching for architectural design scenarios.

[0064] In one embodiment, the method further includes the following after step S140:

[0065] Obtain the historical usage frequency, historical scenario adaptability, and historical replacement rate of each entity node in the comprehensive building component recommendation information, and generate a thermal analysis view of each entity node accordingly.

[0066] In this embodiment, in addition to obtaining comprehensive building component recommendation information corresponding to the current building component information to be retrieved, the server can also combine the historical usage frequency, historical scene adaptability, and historical replacement rate of each entity node in the component data knowledge graph to determine the historical usage frequency, historical scene adaptability, and historical replacement rate of each entity node in the comprehensive building component recommendation information, and generate a time series heat map corresponding to the above three indicators within a week, which can also be sent to the user terminal along with the comprehensive building component recommendation information for the user to view.

[0067] S150. Send the integrated building component recommendation information to the user terminal.

[0068] In this embodiment, once the comprehensive building component recommendation information is obtained from the server, it can be pushed to the user terminal for local visualization viewing by the user.

[0069] In one embodiment, the method further includes the following after step S150:

[0070] If a knowledge graph update instruction is detected, the historical usage frequency, historical scenario adaptability, and historical replacement usage rate of each entity node in the component data knowledge graph are obtained.

[0071] The component data knowledge graph will be updated by merging entity nodes that have shape similarity and whose historical usage frequency, historical scenario adaptability, and historical replacement rate are all less than the corresponding preset parameter thresholds.

[0072] In this embodiment, to enable manual or periodic updates of the component data knowledge graph, maintenance personnel can manually generate knowledge graph update commands locally on the server, or generate them periodically according to a preset knowledge graph update cycle. When the server detects the knowledge graph update command, it obtains the historical usage frequency, historical scenario adaptability, and historical replacement rate of each entity node in the component data knowledge graph. Entity nodes that meet the criteria of shape similarity and historical usage frequency, historical scenario adaptability, and historical replacement rate all being less than the corresponding preset parameter thresholds are then merged. This allows for updating the component data knowledge graph based on user usage history.

[0073] As can be seen, the implementation of this method can combine the unique relationship network of the construction field, namely the component data knowledge graph, to filter candidate entity nodes of the current building component information to be retrieved, and can further supplement the acquisition of supplementary candidate building component information corresponding to the candidate building component information, so as to achieve more accurate and comprehensive retrieval and intelligent recommendation of building component information for users.

[0074] Figure 6 This is a schematic block diagram of an intelligent recommendation device for a modular building component library provided in an embodiment of the present invention. Figure 6 As shown, corresponding to the above-described intelligent recommendation method for modular building component libraries, the present invention also provides an intelligent recommendation device 100 for modular building component libraries. This intelligent recommendation device 100 includes a unit for executing the above-described intelligent recommendation method for modular building component libraries. Please refer to... Figure 6The intelligent recommendation device 100 of the modular building component library includes: a current building component information acquisition unit 110, a current component vector acquisition unit 120, a candidate information acquisition unit 130, a comprehensive recommendation information acquisition unit 140, and a recommendation information sending unit 150.

[0075] The current building component information acquisition unit 110 is used to respond to the building component retrieval command sent by the user terminal and acquire the current building component information to be retrieved corresponding to the building component retrieval command.

[0076] In this embodiment, the technical solution is described with the server as the execution entity. An intelligent recommendation platform for a modular building component library is deployed on the server. This intelligent recommendation platform can be embedded in the server's BIM system (Building Information Modeling) for users to perform online building model design. After a user logs into the intelligent recommendation platform using their login information (such as username and password), they can first select the current building component information to be retrieved through one of several methods on the corresponding user interface of the intelligent recommendation platform. For example, one approach is for a user to log into the online BIM system on the server and open a building model to be designed. If the user selects one of the building models as the current design model and chooses to perform intelligent recommendation processing on that model, triggering a building component retrieval command, then the user needs to obtain the current building component information (such as the building model component name, model number, model size, model setting location, model shape, model function, model style matching information, model usage scenario information, etc.). The current building component information of that building model can then be used as the initial retrieval information.

[0077] Of course, other ways to determine the current building component information to be retrieved include having the user upload a locally designed building model to the server, and selecting one of the models as the current design model. This allows the server to retrieve the current building component information as initial search information. Regardless of how the user determines the current building component information, once the initial determination is complete, it can be used as initial search information for subsequent intelligent retrieval and recommendation processing.

[0078] The current component vector acquisition unit 120 is used to acquire the component shape features and numerical text attribute features corresponding to the current building component information, and to acquire the corresponding current component vector based on a preset multimodal feature fusion strategy.

[0079] In this embodiment, multi-dimensional feature information can be extracted from the current building component information. This includes component numerical and textual information such as component name, model number, model function, model style matching information, and model usage scenario information; and component shape information such as model size, model location, and model shape (which can also be the specific 3D building model data corresponding to the current building component information). By performing multimodal feature fusion on the component shape features and numerical textual attribute features corresponding to the current building component information, the current component vector corresponding to the current building component information can be obtained.

[0080] In one embodiment, the current component vector acquisition unit 120 is specifically used for:

[0081] Obtain a pre-trained geometric coding network, and use the geometric coding network to obtain the component shape features corresponding to the component shape information in the current building component information;

[0082] Obtain a pre-trained attribute embedding model, and use the attribute embedding model to obtain the numerical text attribute features corresponding to the component values ​​and text information in the current building component information;

[0083] Obtain the cross-modal attention mechanism corresponding to the multimodal feature fusion strategy, and fuse the part shape feature and the numerical text attribute feature based on the cross-modal attention mechanism to obtain the current part vector.

[0084] In this embodiment, the geometric encoding network is a neural network that maps the architectural geometric data (point cloud, mesh, voxels, etc.) corresponding to the component shape information in the current building component information into fixed-dimensional feature vectors. It is specifically designed to capture and preserve the spatial structure, topological relationships, and geometric characteristics of a building, laying the foundation for subsequent analysis and fusion. Specifically, it can convert unstructured / structured geometric data into computer-understandable vector representations; extract multi-level geometric features (from local components to global morphology); maintain geometric invariance such as rotation and translation, enhancing the model's generalization ability; and provide a geometric semantic basis for BIM model, architectural design retrieval, and generation. The component shape features corresponding to the component shape information in the current building component information are obtained through a pre-trained geometric encoding network on the server.

[0085] Attribute embedding models are neural network architectures used to map component values ​​and textual information from current building component information into low-dimensional continuous vectors (e.g., 512-dimensional vectors). Their core function is to capture semantic relationships between attributes and unify heterogeneous attributes into a common semantic space, laying the foundation for subsequent cross-modal fusion, similarity calculation, and knowledge reasoning. Attribute embedding models can convert textual attributes (e.g., high-level green building rating), numerical attributes (e.g., building area of ​​100 square meters), and categorical attributes (e.g., frame structure type) into unified vector representations; they can also capture semantic relationships between attributes (e.g., the similarity between environmentally friendly materials and sustainable design); and they can solve the problem of attribute data heterogeneity, providing a foundation for subsequent fusion operations.

[0086] After extracting the shape features and numerical textual attribute features of the component, a cross-modal attention mechanism is used to fuse multimodal features, resulting in the current component vector. The cross-modal attention mechanism is an attention mechanism that solves the problem of aligning and efficiently fusing heterogeneous modalities (architectural geometry, textual attributes, and numerical parameters). Essentially, it automatically learns the correlation strength between different modalities through dynamic query-key-value interactions, thereby achieving feature fusion.

[0087] In one embodiment, the step of obtaining the cross-modal attention mechanism corresponding to the multimodal feature fusion strategy, and fusing the part shape features and the numerical text attribute features based on the cross-modal attention mechanism to obtain the current part vector, includes:

[0088] The component shape features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the first fusion vector;

[0089] The numerical features in the numerical text attribute features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the second vector to be fused.

[0090] The text features in the numerical text attribute features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the third vector to be fused.

[0091] The first vector to be fused, the second vector to be fused, and the third vector to be fused are weighted and summed based on the cross-modal attention mechanism to obtain the current component vector.

[0092] In this embodiment, the cross-modal attention mechanism includes core steps such as modal projection alignment, multi-head cross-modal attention processing, scaling and normalization, and weighted fusion. Referring to the above processing procedure, the component shape features can first be mapped to the corresponding first preset feature dimension (e.g., 512 dimensions) based on the cross-modal attention mechanism and normalized to obtain a first vector to be fused; the numerical features in the numerical text attribute features can be mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain a second vector to be fused; the text features in the numerical text attribute features can be mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain a third vector to be fused. For example, 1024-dimensional component shape features, 768-dimensional text features, and 32-dimensional numerical features are projected into a 512-dimensional unified feature space. Then, a multi-head cross-modal attention mechanism (using a standard Query-Key-Value attention mechanism as input) is determined. This multi-head cross-modal attention includes at least shape-text bidirectional attention and numerical feature-guided attention, with corresponding similarity matrices between the two attentions. Based on the shape-text bidirectional attention and numerical feature-guided attention, the component shape features, text features, and numerical features are scaled and normalized to obtain the first, second, and third vectors to be fused, respectively. These are then weighted and fused to obtain the current component vector. Thus, the multi-modal feature fusion of component shape features, text features, and numerical features is achieved through the aforementioned cross-modal attention mechanism.

[0093] The candidate information acquisition unit 130 is used to determine candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph.

[0094] In this embodiment, each entity node in the component data knowledge graph is pre-built locally on the server, and the component vector is also pre-calculated. When the current component vector of the current building component information to be retrieved is obtained, the vector similarity between it and the component vector corresponding to each entity node in the component data knowledge graph can be calculated, and the candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions can be quickly determined.

[0095] Specifically, when pre-building a component data knowledge graph locally on the server, a multi-dimensional component knowledge graph is constructed, comprising entity nodes of multiple component models. Each entity node includes component geometric features (shape, size), attribute features (specifications, material, price), functional attributes (use, compatible systems), and style tags (decoration style, design semantics), among other attribute information. This multi-dimensional component knowledge graph forms a network of relationships between components. The multi-dimensional component knowledge graph between entity nodes includes at least shape similarity relationships, functional substitution relationships, style matching relationships, and usage scenario dependencies. Through this component data knowledge graph, users can quickly obtain the recommended component information by using the current building component information as initial search information.

[0096] In one embodiment, the candidate information acquisition unit 130 is specifically used for:

[0097] Obtain the cosine similarity or Euclidean distance between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph as the vector similarity.

[0098] From the knowledge graph of the component data, candidate entity nodes whose vector similarity ranking with the current component vector is TopK or exceeds the preset similarity threshold are selected from each entity node based on the vector similarity filtering conditions, and candidate building component information corresponding to each candidate entity node is obtained.

[0099] In TopK, K is a user-defined integer value, and K is greater than or equal to 3.

[0100] In this embodiment, the cosine similarity or Euclidean distance between the current component vector and the component vectors corresponding to each entity node in the locally preset component data knowledge graph can be calculated as the vector similarity. Then, based on the vector similarity filtering criteria, candidate entity nodes whose vector similarity ranking with the current component vector is either Top K or exceeds a preset similarity threshold are selected. The selected entity nodes are the building component information and building component models that are closest to the user's needs. Of course, this initial recommendation only recommends data similar to the current building component information to be retrieved; further analysis and processing can be performed to supplement the recommendations.

[0101] The comprehensive recommendation information acquisition unit 140 is used to acquire supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and to form comprehensive building component recommendation information together with the candidate building component information.

[0102] In this embodiment, in order to recommend data other than approximate data for the current building component information to be retrieved, supplementary candidate building component information corresponding to the candidate building component information can be further filtered by combining the relationship network between the entity nodes in the component data knowledge graph, and combined with the building component information to form a comprehensive building component recommendation letter, thereby realizing richer information output and recommendation.

[0103] In one embodiment, the comprehensive recommendation information acquisition unit 140 is specifically used for:

[0104] Based on the shape similarity relationship, functional substitution relationship, style matching relationship and usage scenario relationship of each entity node in the component data knowledge graph, similar components, substitute components and complementary components related to the candidate building component information are determined, and supplementary candidate building component information corresponding to the candidate building component information is formed.

[0105] In this embodiment, when determining supplementary candidate building component information related to the candidate building component information, similar components, substitute components, and complementary components related to the candidate building component information can be determined one by one based on the shape similarity relationship, functional substitution relationship, style matching relationship, and usage scenario relationship of each entity node in the component data knowledge graph. These components are then used as supplementary candidate building components obtained by further filtering from the component data knowledge graph. Finally, the supplementary candidate building component information is composed of the candidate building component information and the component information corresponding to its similar, substitute, and complementary components. This achieves intelligent association retrieval capabilities based on shape similarity, functional substitution, and style matching for architectural design scenarios.

[0106] In one embodiment, the intelligent recommendation device 100 for the modular building components library further includes:

[0107] The thermal analysis visualization production unit is used to obtain the historical usage frequency, historical scenario adaptability and historical replacement rate of each entity node in the comprehensive building component recommendation information, and generate thermal analysis visualizations for each entity node accordingly.

[0108] In this embodiment, in addition to obtaining comprehensive building component recommendation information corresponding to the current building component information to be retrieved, the server can also combine the historical usage frequency, historical scene adaptability, and historical replacement rate of each entity node in the component data knowledge graph to determine the historical usage frequency, historical scene adaptability, and historical replacement rate of each entity node in the comprehensive building component recommendation information, and generate a time series heat map corresponding to the above three indicators within a week, which can also be sent to the user terminal along with the comprehensive building component recommendation information for the user to view.

[0109] The recommendation information sending unit 150 is used to send the integrated building component recommendation information to the user terminal.

[0110] In this embodiment, once the comprehensive building component recommendation information is obtained from the server, it can be pushed to the user terminal for local visualization viewing by the user.

[0111] In one embodiment, the intelligent recommendation device 100 for the modular building components library further includes:

[0112] The entity node historical usage data acquisition unit is used to acquire the historical usage frequency, historical scenario adaptability and historical replacement usage rate of each entity node in the component data knowledge graph if a knowledge graph update instruction is detected.

[0113] The knowledge graph update unit is used to merge entity nodes in the component data knowledge graph that have shape similarity relationships and whose historical usage frequency, historical scenario adaptability, and historical replacement usage rate are all less than the corresponding preset parameter thresholds, so as to update the component data knowledge graph.

[0114] In this embodiment, to enable manual or periodic updates of the component data knowledge graph, maintenance personnel can manually generate knowledge graph update commands locally on the server, or generate them periodically according to a preset knowledge graph update cycle. When the server detects the knowledge graph update command, it obtains the historical usage frequency, historical scenario adaptability, and historical replacement rate of each entity node in the component data knowledge graph. Entity nodes that meet the criteria of shape similarity and historical usage frequency, historical scenario adaptability, and historical replacement rate all being less than the corresponding preset parameter thresholds are then merged. This allows for updating the component data knowledge graph based on user usage history.

[0115] As can be seen, the embodiments implementing this device can combine the unique relationship network in the construction field, namely the component data knowledge graph, to filter candidate entity nodes for the current building component information to be retrieved, and can further supplement the acquisition of supplementary candidate building component information corresponding to the candidate building component information, so as to achieve more accurate and comprehensive retrieval and intelligent recommendation of building component information for users.

[0116] The aforementioned intelligent recommendation device for the modular building components library can be implemented as a computer program, which can, for example... Figure 7 It runs on the computer device shown.

[0117] Please see Figure 7 , Figure 7This is a schematic block diagram of a computer device provided in an embodiment of the present invention. This computer device integrates an intelligent recommendation device for any of the modular building component libraries provided in the embodiments of the present invention.

[0118] See Figure 7 The computer device 400 includes a processor 402, a memory, and a network interface 405 connected via a system bus 401. The memory may include a storage medium 403 and internal memory 404.

[0119] The storage medium 403 may store an operating system 4031 and a computer program 4032. The computer program 4032 includes program instructions that, when executed, cause the processor 402 to perform an intelligent recommendation method for a modular building components library.

[0120] The processor 402 provides computing and control capabilities to support the operation of the entire computer device.

[0121] The internal memory 404 provides an environment for the operation of the computer program 4032 in the storage medium 403. When the computer program 4032 is executed by the processor 402, the processor 402 can execute the above-mentioned intelligent recommendation method for the modular building components library.

[0122] This network interface 405 is used for network communication with other devices. Those skilled in the art will understand that... Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0123] The processor 402 is used to run a computer program 4032 stored in the memory to implement the above-mentioned intelligent recommendation method for the modular building component library.

[0124] It should be understood that, in this embodiment of the invention, the processor 402 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0125] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0126] Therefore, the present invention also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the aforementioned intelligent recommendation method for a modular building components library.

[0127] The storage medium can be any computer-readable storage medium that can store program code, such as a USB flash drive, external hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0128] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0129] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0130] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0131] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0132] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent recommendation method for a modular building component library, characterized in that, include: In response to a building component retrieval command sent by a user terminal, obtain the current building component information to be retrieved corresponding to the building component retrieval command; Obtain the component shape features and numerical text attribute features corresponding to the current building component information, and obtain the corresponding current component vector based on a preset multimodal feature fusion strategy; Based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph, candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions are determined. Obtain supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and combine it with the candidate building component information to form comprehensive building component recommendation information; The integrated building component recommendation information is sent to the user terminal; The step of obtaining supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph includes: Based on the shape similarity relationship, functional substitution relationship, style matching relationship and usage scenario relationship of each entity node in the component data knowledge graph, similar components, substitute components and complementary components related to the candidate building component information are determined, and supplementary candidate building component information corresponding to the candidate building component information is formed. After the step of sending the integrated building component recommendation information to the user terminal, the method further includes: If a knowledge graph update instruction is detected, the historical usage frequency, historical scenario adaptability, and historical replacement usage rate of each entity node in the component data knowledge graph are obtained. The component data knowledge graph will be updated by merging entity nodes that have shape similarity and whose historical usage frequency, historical scenario adaptability, and historical replacement rate are all less than the corresponding preset parameter thresholds.

2. The method according to claim 1, characterized in that, The step of acquiring the component shape features and numerical text attribute features corresponding to the current building component information, and acquiring the corresponding current component vector based on a preset multimodal feature fusion strategy, includes: Obtain a pre-trained geometric coding network, and use the geometric coding network to obtain the component shape features corresponding to the component shape information in the current building component information; Obtain a pre-trained attribute embedding model, and use the attribute embedding model to obtain the numerical text attribute features corresponding to the component values ​​and text information in the current building component information; Obtain the cross-modal attention mechanism corresponding to the multimodal feature fusion strategy, and fuse the part shape feature and the numerical text attribute feature based on the cross-modal attention mechanism to obtain the current part vector.

3. The method according to claim 2, characterized in that, The step of fusing the component shape features and the numerical text attribute features using the cross-modal attention mechanism to obtain the current component vector includes: The component shape features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the first fusion vector; The numerical features in the numerical text attribute features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the second vector to be fused. The text features in the numerical text attribute features are mapped to the corresponding first preset feature dimension based on the cross-modal attention mechanism and normalized to obtain the third vector to be fused. The first vector to be fused, the second vector to be fused, and the third vector to be fused are weighted and summed based on the cross-modal attention mechanism to obtain the current component vector.

4. The method according to claim 1, characterized in that, The process of determining candidate entity nodes and corresponding candidate building component information that meet preset vector similarity screening conditions based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the locally preset component data knowledge graph includes: Obtain the cosine similarity or Euclidean distance between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph as the vector similarity. From the entity nodes in the component data knowledge graph, candidate entity nodes whose vector similarity ranking with the current component vector belongs to TopK or exceeds the preset similarity threshold are selected based on the vector similarity filtering conditions, and candidate building component information corresponding to each candidate entity node is obtained; wherein, K in TopK is a user-defined integer value, and K is greater than or equal to 3.

5. The method according to claim 1, characterized in that, After the step of obtaining supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and combining it with the candidate building component information to form comprehensive building component recommendation information, the method further includes: Obtain the historical usage frequency, historical scenario adaptability, and historical replacement rate of each entity node in the comprehensive building component recommendation information, and generate a thermal analysis view of each entity node accordingly.

6. An intelligent recommendation device for a modular building component library, characterized in that, include: The current building component information acquisition unit is used to respond to the building component retrieval command sent by the user terminal and acquire the current building component information to be retrieved corresponding to the building component retrieval command. The current component vector acquisition unit is used to acquire the component shape features and numerical text attribute features corresponding to the current building component information, and to acquire the corresponding current component vector based on a preset multimodal feature fusion strategy; The candidate information acquisition unit is used to determine candidate entity nodes and corresponding candidate building component information that meet the preset vector similarity screening conditions based on the vector similarity between the current component vector and the component vectors corresponding to each entity node in the local preset component data knowledge graph. The comprehensive recommendation information acquisition unit is used to acquire supplementary candidate building component information corresponding to the candidate building component information from the component data knowledge graph, and to form comprehensive building component recommendation information with the candidate building component information. A recommendation information sending unit is used to send the integrated building component recommendation information to the user terminal; The comprehensive recommendation information acquisition unit is specifically used for: Based on the shape similarity relationship, functional substitution relationship, style matching relationship and usage scenario relationship of each entity node in the component data knowledge graph, similar components, substitute components and complementary components related to the candidate building component information are determined, and supplementary candidate building component information corresponding to the candidate building component information is formed. The intelligent recommendation device for the modular building components library also includes: The entity node historical usage data acquisition unit is used to acquire the historical usage frequency, historical scenario adaptability and historical replacement usage rate of each entity node in the component data knowledge graph if a knowledge graph update instruction is detected. The knowledge graph update unit is used to merge entity nodes in the component data knowledge graph that have shape similarity relationships and whose historical usage frequency, historical scenario adaptability, and historical replacement usage rate are all less than the corresponding preset parameter thresholds, so as to update the component data knowledge graph.

7. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the intelligent recommendation method for the modular building component library as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, can implement the intelligent recommendation method for the modular building component library as described in any one of claims 1-5.

Images