An elevator three-dimensional model recommendation method, system and electronic device

Abstract

The application discloses an elevator three-dimensional model recommendation method, system and electronic equipment, and the method comprises the following steps: S1, performing semantic analysis on the supplementary features of the elevator to be designed, obtaining the shape parameters and structure parameters of the components of the elevator to be designed, and inputting the shape parameter weight value and the structure parameter weight value by the user; S2, performing normalization processing on the shape parameter weight value, the structure parameter weight value and the default weight value to generate a normalized weight vector; S3, calculating the initial matching score of the candidate three-dimensional model; S4, optimizing and correcting the initial matching score based on the preference data of the user to generate an optimized matching score; and S5, sorting the candidate three-dimensional model according to the optimized matching score to generate a three-dimensional model recommendation list. The application can recommend different three-dimensional models based on the operation habits and preference features of different users, provide personalized recommendation services, reduce the screening time of the user, improve the elevator design efficiency, and greatly shorten the design time.

Description

Technical Field

[0001] This application relates to the field of elevator technology, and in particular to a method, system and electronic device for recommending three-dimensional elevator models. Background Technology

[0002] Elevator design is a core part of elevator product development. In recent years, digital design methods based on 3D models have gradually become an important development direction in the field of elevator design. This method realizes the digital modeling of elevator components through 3D visualization technology, which can effectively improve the accuracy of design information expression and the verification efficiency of design schemes, and provide technical support for the rapid iterative development of elevator products.

[0003] In traditional 3D design methods, technicians need to manually create a large number of 3D models based on user orders, resulting in a heavy workload and long design cycles. 3D model recommendation can quickly filter 3D models that match design requirements from an existing model library, thereby reducing the manual modeling workload of technicians and improving design efficiency. However, existing technologies have revealed many shortcomings in implementing elevator 3D model recommendation, severely restricting the improvement of design efficiency.

[0004] Although enterprises have accumulated a wealth of historical design data and 3D model resources, the lack of effective intelligent retrieval and recommendation methods means that existing technologies cannot fully utilize this historical data to assist in the design and development of new products. This forces engineers to continue conceptualizing and modeling from scratch when designing non-standard or customized elevators, resulting in persistently low design efficiency. Existing model recommendation methods based on model libraries rely solely on simple keyword matching or limited parameter conditions for model selection and ranking. These methods fail to deeply understand the design intent and actual needs of engineers, often leading to significant discrepancies between recommended results and actual application scenarios. This makes it difficult to meet the diverse needs of non-standard elevator design, requiring engineers to spend considerable time sifting through recommended lists to select the necessary models, severely impacting the practicality of 3D model recommendations and the user experience. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, and electronic device for recommending elevator 3D models to address the aforementioned technical problems.

[0006] In a first aspect, the present invention discloses a method for recommending three-dimensional elevator models, comprising the following steps: S1. Obtain the supplementary features of the elevator to be designed. Based on semantic recognition, perform semantic parsing on the supplementary features of the elevator to be designed to obtain the component names, shape parameters and structural parameters of the components, as well as the weight values ​​of the shape parameters and structural parameters input by the user. The supplementary features are the natural language text input by the user. S2. Normalize the shape parameter weight values, structure parameter weight values ​​and default weight values ​​to generate a normalized weight vector. S3. Based on the component name, select several candidate 3D models from the elevator 3D model library, extract the matching factor of each candidate 3D model, and calculate the initial matching score of the candidate 3D model by combining the component's shape and structural parameters, matching factor, preset weighting factor and normalized weight vector. S4. Optimize and correct the initial matching score based on user preference data to generate an optimized matching score; S5. Sort the candidate 3D models according to the optimization matching score and generate a recommended list of 3D models.

[0007] Furthermore, it also includes: Obtain the key features of the elevator to be designed, and determine the basic configuration and shape structure of the elevator to be designed based on the key features; the key features include the elevator type, rated load capacity, and number of floors and doors of the elevator to be designed; Based on key features, generate elevator design constraint rules; If a user's design or selection operation is detected to violate the elevator design constraints, a corresponding alarm operation will be triggered.

[0008] Furthermore, step S1 includes: Perform word segmentation, part-of-speech tagging, and syntactic analysis on natural language text, and output the language processing results; Input the language processing results into the design intent classification model, and output the intent label and confidence score; Key entity information is identified and extracted from part-of-speech-tagged text using named entity recognition, and the entity extraction results are output. Based on intent tags, the extracted entities are semantically mapped to the elevator component parameter library to generate shape parameters and structural parameters. The system identifies natural language text and obtains the weight values ​​of shape parameters and structure parameters input by the user.

[0009] Furthermore, step S4 includes: Acquire user preference data, which includes historical 3D model recommendation lists, historical model selections, and historical list browsing behavior. Based on user preference data and the attribute characteristics of candidate 3D models, calculate the probability that a user will select a particular candidate 3D model. The probability of a certain preference feature occurring And the probability of this preference feature occurring under the conditions of this model. Obtain the preference correction coefficient : ; Based on initial matching score and preference adjustment coefficient This will result in an optimized matching score.

[0010] Furthermore, it also includes: If there are multiple elevators to be designed, obtain the key features and supplementary features of the first elevator, execute steps S1-S5, and generate a recommended list of three-dimensional models of the first elevator. Based on the user's selection behavior of the recommended list of 3D models of the first elevator, the 3D model configuration of the first elevator is updated to the target number of other elevators, and the key features of the other elevators are analyzed to identify the differences between the key features of the first elevator and the key features of the other elevators. Based on semantic recognition, semantic parsing of the differences is performed to generate component shape parameters and structural parameters for other elevators; Perform steps S3-S5 to generate a recommended list of 3D models for other elevators within the target number.

[0011] Furthermore, it also includes: Based on the number of elevators to be designed, a consistency weight is generated, where the consistency weight is positively correlated with the number of elevators; The initial matching score is optimized and corrected based on user preference data and consistency weights to generate an optimized matching score.

[0012] Furthermore, it also includes: If multiple users participate in the elevator design, obtain the key characteristics of the elevator, and divide the elevator system into several subsystems, the subsystems include the elevator car subsystem, the elevator shaft subsystem, and the elevator electrical subsystem; Obtain the supplementary features of each subsystem, execute steps S2-S5, and generate a 3D model recommendation list for each subsystem. Based on the selection behavior of each user on the recommended list of 3D models of each subsystem, the coordination relationship between each subsystem is verified to meet the elevator design constraints. If there is a conflict between parameters of the subsystems, a conflict report is generated and fed back to the relevant users. The conflict report includes the name of the conflicting component, the name of the parameter, the parameter value of each subsystem and the conflict level, and marks the conflict location and conflict type.

[0013] Furthermore, the step of obtaining supplementary features for each subsystem and executing steps S2-S5 to generate a 3D model recommendation list for each subsystem includes: Obtain the current user's preference data and generate the current user's preference correction coefficient; Obtain the team's preference data, which includes multiple users, and generate a team preference correction coefficient. The current user preference correction coefficient and the team preference correction coefficient are processed by combining weights. Based on the initial matching score and the weighted preference correction coefficient, the optimized matching score is obtained. Based on the optimized matching score, the candidate 3D models are sorted to generate a recommended list of 3D models.

[0014] Secondly, the present invention discloses an elevator 3D model recommendation system for implementing the above-mentioned elevator 3D model recommendation method, including an input layer, a data processing layer, an output layer and a database. The input layer includes a key feature input module and a supplementary feature input module for obtaining the key features and supplementary features of the elevator to be designed. The data processing layer is used to determine the basic configuration and shape structure of the elevator to be designed based on key features; based on semantic recognition, it performs semantic parsing on the supplementary features of the elevator to be designed to obtain the component names, shape parameters, and structural parameters of the components, as well as the user-input weight values ​​of the shape parameters and structural parameters; it normalizes the shape parameter weight values ​​and structural parameter weight values ​​with the default weight values ​​to generate a normalized weight vector; based on the basic configuration and shape structure of the elevator to be designed, it selects several candidate 3D models from the elevator 3D model library, extracts the matching factors of each candidate 3D model, and calculates the initial matching score of the candidate 3D model by combining the component shape parameters and structural parameters, the matching factors, the preset weighting factors, and the normalized weight vector; based on the user's preference data, it optimizes and corrects the initial matching score to generate an optimized matching score. The key features include the elevator type, rated load capacity, rated speed, and number of floors and doors of the elevator to be designed; the supplementary features are natural language text; and the component parameters include shape parameters and structural parameters. The output layer is used to sort the candidate 3D models based on the optimization matching score and generate a recommended list of 3D models; The database includes a historical design data repository, a 3D model repository, and a user preference database, which are used to store historical design data, 3D model data, and user preference data, and enable data reading and writing.

[0015] Thirdly, the present invention discloses an electronic device, comprising: a chip, a processor, and a memory, wherein the memory is used to store computer program code, the computer program code including computer instructions, and when the chip executes the computer instructions, the electronic device executes an elevator 3D model recommendation method as described above.

[0016] Compared with existing technologies, this invention provides a method for recommending 3D elevator models. Through semantic recognition and parsing, it converts user-input natural language text into precise shape and structural parameters of elevator components, solving the problem of inaccurate recommendations due to imprecise feature information and improving the matching degree between recommendation results and actual needs. Furthermore, it optimizes and corrects the initial matching score based on user preference data, adapting to different users' operating habits and preferences, making the recommendation results more consistent with user preferences, providing personalized recommendation services, reducing user filtering time, and improving design efficiency.

[0017] Intelligent recommendations reduce the time users spend manually searching and selecting the required models, significantly improving the design efficiency of non-standard elevators. Through the synergistic effect of semantic enhancement and preference learning, users can obtain candidate 3D models that meet their needs through automatic recommendations without having to perform tedious keyword searches and manual filtering, greatly shortening the design time. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the elevator 3D model recommendation method described in this invention; Figure 2 This is a flowchart illustrating step S1 of the elevator 3D model recommendation method described in this invention. Figure 3 This is a flowchart illustrating step S12b of the elevator 3D model recommendation method described in this invention. Figure 4 This is a schematic diagram of the elevator 3D model recommendation system described in this invention; Figure 5 This is a schematic diagram of the structure of the electronic device described in this invention; Marker explanation: 110. Key Feature Input Module; 120. Supplementary Feature Input Module; 210. Natural Language Parsing Module; 220. Intent Classification and Recognition Module; 230. Entity Extraction Module; 240. Relationship Mapping Module; 250. Weight Normalization Processing Module; 260. Matching Factor Extraction Module; 270. Matching Scoring Optimization Module; 280. Preference Learning Optimization Module; 290. Preference Update Module; 200. Validation Module; 310. Model ranking module; 320. Recommendation list generation module; 410. Historical design database; 420. 3D model library; 430. User preference database; 510. Processor; 520. Memory; 530. Input device; 540. Output device. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0021] In one embodiment, such as Figure 1 This invention provides a method for recommending 3D elevator models, comprising the following steps: S1. Obtain the key features and supplementary features of the elevator to be designed, and determine the basic configuration and shape structure of the elevator to be designed based on the key features; based on semantic recognition, perform semantic parsing on the supplementary features of the elevator to be designed to obtain the component names, component shape parameters and structural parameters of the elevator to be designed, as well as the user-input shape parameter weight values ​​and structural parameter weight values.

[0022] Specifically, such as Figure 2 Step S1 includes: S11. Obtain the key features and supplementary features of the elevator to be designed. The key features are the elevator type, rated load capacity, number of floors and doors, shaft size, lifting height and other core parameters of the elevator to be designed. The supplementary features are the natural language text input by the user.

[0023] Key features are input via structured data forms, with each parameter having a clearly defined value range and data type validation rules. Elevator type parameters are enumerated using a categorized approach, including machine-room passenger elevators, machine-room-less passenger elevators, medical elevators, freight elevators, residential elevators, and sightseeing elevators. Rated load capacity parameters are in kilograms, with a range varying depending on the elevator type; freight elevators typically range from 1000kg to 5000kg, while passenger elevators typically range from 450kg to 2500kg. The number of floors and doors is expressed in the format of "x floors x stops," such as 3 floors x 3 stops, 5 floors x 5 stops, etc. Shaft dimensions include shaft width and depth, in millimeters, with typical value ranges determined by referring to tables based on the rated load capacity and elevator type.

[0024] S12a. Analyze and process key features to determine the basic configuration and shape structure of the elevator to be designed. Key features are those not directly related to the elevator, such as the elevator type, number of floors and doors, and shaft dimensions. Based on the key features, the basic configuration of the elevator to be designed, such as the elevator car dimensions and rated operating speed, is automatically calculated.

[0025] For example, the input key features of the elevator to be designed include: the elevator type is a machine room-less freight elevator, the rated load capacity is 2000kg, the lifting height is 25 meters, the number of floor landing doors is 4 floors and 4 stops, the hoistway width is 2400mm, and the hoistway depth is 2300mm. The basic configuration of the elevator is initially determined according to these key features, for example, the car size is automatically set to approximately 1800mm by 2000mm, and the rated speed is 1.0m / s.

[0026] S12b: Perform semantic parsing on the supplementary features, obtain the shape parameters and structural parameters of the elevator component to be designed from the elevator component parameter library, and obtain the shape parameter weight value and the structural parameter weight value input by the user.

[0027] Specifically, for example Figure 3 , step S12b includes: S12b-1: Perform word segmentation, part-of-speech tagging and syntactic analysis on the natural language text, and output a language processing result.

[0028] Call a Chinese word segmentation algorithm to segment the input natural language text into a word sequence. The Chinese word segmentation algorithm adopts a conditional random field-based sequence labeling model, which is trained in combination with a large-scale Chinese corpus, and can accurately process professional terms and quantifier expressions. For example, if the input natural language text is "The car door needs to be widened by 20cm, adopting a center-opening door, and equipped with a safety light curtain", the word segmentation result is "car door / need / widen / 20 / cm / , / adopt / center-opening / door, / be equipped with / safety / light curtain". After the word segmentation is completed, part-of-speech tagging is performed on each word, and the tagging system includes part-of-speech categories such as nouns, verbs, quantifiers, numerals, localizers, and auxiliary words. The part-of-speech tagging result of the above word segmentation result is "car door / noun / need / verb / widen / verb / 20 / numeral / cm / quantifier / , / punctuation / adopt / verb / center-opening / noun / door / noun / , / punctuation / be equipped with / verb / safety / noun / light curtain / noun". Syntactic analysis is performed on the part-of-speech tagging result to identify syntactic relationships such as subject-predicate, verb-object, attributive-modified, and coordinate relationships between words. For example, "widen" forms a verb-object relationship with "car door", and "20cm" acts as a complement component to modify "widen". The results of word segmentation, part-of-speech tagging and syntactic analysis are output in a structured data format.

[0029] S12b-2: Input the language processing result into a design intention classification model, and output an intention label and a confidence level.

[0030] The language processing output is input into the design intent classification model to identify the design intent types contained in the natural language text, and output intent labels and confidence scores. The pre-set intent type label library covers various dimensions of elevator design, including door width adjustment, door type selection, material requirements, configuration selection, safety devices, surface treatment, and color schemes. The design intent classification model is fine-tuned based on a pre-trained language model, and its parameters are optimized using a large amount of labeled data. When the natural language text is "The car door needs to be widened by 20cm, adopt a center-split opening, and have a safety light curtain," the model outputs the intent labels "door width adjustment" and "door opening method selection," with confidence scores of 0.92 and 0.88, respectively. The intent classification results are represented in the form of intent vectors, with each dimension corresponding to an intent type label, and the value representing the confidence score of that intent.

[0031] S12b-3. Identify and extract key entity information from part-of-speech tagged text using named entity recognition, and output the entity extraction results.

[0032] Named entity recognition (NAME) technology is employed, building an entity recognition model based on a BERT pre-trained model combined with a conditional random field decoding layer. This model identifies and extracts key entity information from part-of-speech tagged text, outputting the entity extraction results. Entity types include component names, size parameters, structural types, material types, and accessory names. For the above part-of-speech tagging results, the entity extraction results are: component name entity "car door", size parameter entity "20cm", structural type entity "center-split", and accessory entity "safety light curtain". Each extracted entity is accompanied by location information, identifying its start and end positions in the original text for traceability and verification. The entity extraction results are organized as an entity list, with each entity containing four fields: entity text, entity type, entity confidence score, and location offset.

[0033] S12b-4. Based on intent tags, semantically map the extracted entities to the elevator component parameter library to generate shape parameters and structural parameters.

[0034] The extracted entities are semantically mapped to an elevator component parameter library to generate precise shape and structural parameters for each component. The elevator component parameter library is a parameter mapping dictionary based on elevator design standards, organized by component type. Each component type defines its own set of shape and structural parameters. Shape parameters include dimensional parameters such as length, width, height, thickness, and diameter, all using international standard units. Structural parameters include type selection parameters such as door opening method, support method, and connection method, expressed using enumerated values. Taking the "car door" as an example, its shape parameters include door opening width, door panel height, door panel thickness, and guide rail spacing, while its structural parameters include door opening method (center-opening, side-opening, straight-opening), safety device type, and door operator type. The following mapping process is performed for each extracted entity: the target parameter set is determined based on the entity type; then, the most similar parameter items in the parameter set are matched based on the entity's semantics; finally, the natural language values ​​are converted into standard numerical values ​​or enumerated values. For example, the entity "20cm" is mapped to the shape parameter "door opening width" with a value of 200mm, and the entity "safety light curtain" is mapped to the structure parameter "safety device" with an enumeration value of "safety light curtain type".

[0035] In this embodiment, after semantic parsing is completed, a set of shape parameters and a set of structure parameters are generated. The shape parameters are stored in key-value pairs, where the key is the parameter name and the value is a numerical value with units. The structure parameters are also stored in key-value pairs, where the key is the parameter name and the value is an enumeration type or a Boolean type.

[0036] In this embodiment, step S12b further includes: S12b-5. Recognize natural language text and obtain the weight values ​​of shape parameters and structure parameters input by the user. The user can specify a weight value between 0 and 1 for each shape parameter and structure parameter in the parameter setting interface, indicating the relative importance of the parameter in the matching score calculation.

[0037] S2. Normalize the shape parameter weight value, structural parameter weight value and default weight value to generate a normalized weight vector. The default weight value is obtained based on statistical analysis of historical design data and reflects the average contribution of each parameter to the matching quality.

[0038] The normalization process uses the following method to convert the original weight vector into a normalized weight vector: Let the original weight vector be... equal , to The set of, where, For the first The original weight values ​​of each parameter. The normalized weight vector. equal , to The set of components is calculated using the following formulas:

[0039] in, For the first A normalized weight vector; This represents the method for representing each normalized weight vector. =1~n; The normalized weight vector satisfies the condition that the sum of all its components equals 1. Normalizing the weight vector effectively balances weight values ​​from different sources, allowing user-inputted custom shape parameter weight values, structural parameter weight values, and default weight values ​​to be compared and calculated on the same numerical scale, avoiding calculation errors caused by inconsistent weight dimensions. After normalization, a normalized weight vector is generated. .

[0040] S3. Select several candidate 3D models from the elevator 3D model library based on the component name and / or component type, extract the matching factor of each candidate 3D model, and calculate the initial matching score of the candidate 3D model by combining the component's shape and structural parameters, matching factor, preset weighting factor and normalized weight vector.

[0041] The 3D model library is organized according to part type. Each part type includes several 3D model instances. Each model instance stores a model identifier, model file path, model metadata, and model feature vector. The model feature vector includes multi-dimensional features, such as part type identifier, shape parameter set, structural parameter combination, and associated parameter set.

[0042] Four types of matching factors are extracted from the feature vector of the 3D model: component type matching factor f1, shape parameter matching factor f2, structural parameter matching factor f3, and association parameter matching factor f4. The component type matching factor f1 is used to measure the semantic similarity between the core component type of the candidate 3D model and the target component type. It is obtained by calculating the semantic distance between the two component type identifiers and takes a value from 0 to 1, with a larger value indicating a higher degree of matching. The target component type is the elevator component type obtained from the semantic analysis in step S1.

[0043] The shape parameter matching factor f2 is used to measure the degree of matching between the shape parameters of the candidate 3D model and the target shape parameters. For each shape parameter, the relative deviation between the candidate 3D model parameter value and the target parameter value is calculated, and the relative deviation is mapped to a matching score between 0 and 1. The weighted average of the matching scores of each parameter is used as the value of the shape parameter matching factor. The target shape parameter is the shape parameter of the elevator to be designed obtained in step S1.

[0044] The structural parameter matching factor f3 is used to measure the matching degree between the structural parameters of the candidate 3D model and the target structural parameters. For enumerated structural parameters, if the candidate value and the target value are exactly the same, the matching degree is 1; otherwise, a partial matching score is calculated based on the semantic distance of the enumerated values, and the matching degree scores of each structural parameter are summarized. The target structural parameters are the structural parameters of the elevator to be designed obtained in step S1.

[0045] The association parameter matching factor f4 is used to measure the fit constraints between the candidate 3D model and adjacent parts. It compares whether the candidate model meets the fit constraints and calculates the fit satisfaction rate as the value of the association parameter matching factor.

[0046] The preset weighting factors are weighting factors pre-set by the user, reflecting the relative importance of each matching factor in the scoring system, and can be adjusted according to the elevator type and design scenario.

[0047] The initial matching score is calculated using a weighted summation method, taking into account the contribution of each matching factor. Let there be a total of... There are matching factors, and the k-th matching factor is... The corresponding normalized weights are Initial matching score The calculation formula is:

[0048] The initial matching score is obtained by multiplying each matching factor by its normalized weight and then summing the results. After the initial matching score is calculated, all candidate 3D models are sorted in descending order of score to generate an initial candidate list of 3D models.

[0049] S4. Optimize and correct the initial matching score based on user preference data to generate an optimized matching score.

[0050] Specifically, step S4 includes: S41. Obtain user preference data from the user preference database, including historical 3D model recommendation lists, historical selection models, historical list browsing behavior, and other data.

[0051] S42. Based on user preference data and the attribute characteristics of each candidate 3D model, calculate the probability that the user selects a particular candidate 3D model. The probability of a certain preference feature occurring And the probability of this preference feature occurring under the conditions of this model. Obtain the preference correction coefficient : .

[0052] Specifically, the system retrieves users' historical selection records from user preference data, including historical 3D model recommendation lists, historical model selections, unselected models, and historical browsing behavior. This historical selection record analysis includes frequency analysis, feature preference analysis, time pattern analysis, and contextual relationship analysis. Frequency analysis statistically analyzes the distribution of users' first choice in the 3D model recommendation list, choices made after page browsing, and choices made after browsing multiple pages. Feature preference analysis statistically analyzes the distribution of user preferences for specific feature dimensions of the model, such as the frequency with which a user selects different door opening methods, different materials, and different security configurations. The analysis results are stored in the form of preference feature vectors, with each dimension corresponding to a feature item, the value of which represents the user's preference strength for that feature, ranging from 0 to 1.

[0053] The preference correction coefficient is calculated using Bayesian inference: Based on historical selection records and the attribute features of candidate 3D models, the probability of a user selecting a candidate 3D model given the preference features is calculated. Obtain the probability that a user selects a particular candidate 3D model. The probability of a certain preference feature occurring And the probability of this preference feature occurring under the conditions of this model. Calculate the preference correction coefficient : The preference correction coefficient is typically between 0.8 and 1.2. If the preference correction coefficient is greater than 1, it can be considered that the user has a positive preference for the attribute features of the candidate 3D model; if the preference correction coefficient is less than 1, it can be considered that the user has a negative preference for the attribute features of the candidate 3D model.

[0054] S43, Based on initial matching score And preference correction coefficient To obtain an optimized matching score: .

[0055] By using the above product correction method, the recommendation results can be dynamically adjusted according to the user's personal preferences while maintaining the basic matching logic, making the recommendation results more in line with the user's operating habits and selection preferences.

[0056] S5. Sort each candidate 3D model according to the optimization matching score and generate a 3D model recommendation list.

[0057] Specifically, the top N 3D models from the sorted candidate 3D models are selected to generate a 3D model recommendation list, where N is typically 10. The 3D model recommendation list also includes a model thumbnail corresponding to each candidate 3D model, the model name, the matching percentage, and the model's main parameters. The model thumbnail displays the model's appearance as a pre-rendered thumbnail; the model name follows standard naming conventions and includes a combined description of component type, theme specifications, and configuration features; the matching percentage is displayed as a percentage of the optimized matching score, such as "95.5% match"; and the main parameters are presented in a table format, showing the model's key shape and structural parameters. In this embodiment, the candidate 3D models are sorted in descending order of their optimized matching scores; for candidate 3D models with the same optimized matching score, a secondary sorting criterion is used, ranking them in ascending order of component complexity. The quantitative indicator of component complexity is a comprehensive score based on the number of components and parameters contained in the candidate 3D model.

[0058] Specifically, it also includes: S6. Record user selection behavior data in the 3D model recommendation list, and update user preference data and preference model.

[0059] Selection behavior data includes the following fields: a selection timestamp recording the precise time of the selection operation, in the format "year-month-day hour:minute:second"; a selected model identifier recording the unique identifier of the selected model, encoded using an alphanumeric combination; the original sorting position in the recommendation list recording the ranking number of the selected model in the recommendation list; and browsing history before selection recording the list of model identifiers viewed by the user before making the final selection and their browsing duration. The recording of selection behavior data is triggered by the event listener module of the front-end interface, sending the data to the back-end data processing module asynchronously. After parsing and formatting the selection behavior data, the back-end data processing module updates the historical selection record table in the user preference data. The preference model update adopts an incremental update strategy; each time a new selection record is appended to the historical record queue, the retraining process of the preference model is triggered. During retraining, the updated historical selection records are re-counted, generating updated preference feature vectors and preference correction coefficients. The updated preference model takes effect immediately and participates in subsequent recommendation scoring calculations.

[0060] Specifically, it also includes: During the 3D model recommendation and selection process, elevator design constraint rules are automatically generated based on key features. If the user's current operation is detected to violate the elevator design constraint rules, the corresponding alarm operation is triggered.

[0061] The system automatically generates elevator design constraint rules based on key features and detects whether current operations violate these rules during the 3D model recommendation and selection process. The rule generation engine, based on parameters such as rated load capacity, elevator type, and lifting height from key features, queries a pre-defined elevator design specification knowledge base to automatically generate a set of constraint rules relevant to the current design task. The elevator design specification knowledge base contains structured clauses from national elevator design standards, organized according to safety requirements, performance requirements, and dimensional requirements.

[0062] The alarm triggering mechanism adopts a three-level progressive mode: Level 1 alarm is the prompt level. When it detects that the current operation may violate design constraints, a warning icon and prompt text are displayed on the interface, describing the specific violation, such as "door width is lower than standard requirements." Level 1 alarms do not interrupt the operation process, and users can choose to ignore them and continue. Level 2 alarm is the warning level. Based on Level 1 alarms, a confirmation dialog box pops up, requiring the user to select "Confirm Modification" or "Cancel Modification." When "Confirm Modification" is selected, the system records the operation as an exception and executes the modification. When "Cancel Modification" is selected, the system undoes the current modification and restores the state before the modification. Level 3 alarm is the prohibition level. When it detects that the current operation seriously violates safety regulations, the system directly blocks the operation, does not pop up a confirmation dialog box, and displays a prohibition icon and prohibition instructions on the interface. Level 3 alarms are typically used for mandatory clauses involving personal safety, such as the minimum car area requirement corresponding to the rated load capacity.

[0063] In another embodiment, if there are multiple elevators to be designed, the batch design mode is entered: Obtain the key and supplementary features of the first elevator, and execute steps S1-S5 to generate a recommended list of 3D models of the first elevator.

[0064] Based on the user's selection behavior of the recommended list of 3D models of the first elevator, the 3D model configuration of the first elevator is updated to the target number of other elevators, and the key features of the other elevators are analyzed to identify the differences between the key features of the first elevator and the key features of the other elevators.

[0065] Based on semantic recognition, semantic parsing is performed on the differences to generate component shape parameters and structural parameters for other elevators.

[0066] Perform steps S3-S5 to generate a recommended list of 3D models for other elevators within the target number.

[0067] Specifically, it also includes: Based on the number of elevators to be designed, a consistency weight is generated, where the consistency weight is positively correlated with the number of elevators; The initial matching score is optimized and corrected based on user preference data and consistency weights to generate an optimized matching score.

[0068] The consistency of component selection among elevators to be designed in batch projects is analyzed, encouraging the use of 3D models with the same or similar configurations within the same batch to improve manufacturing efficiency and maintenance convenience. The consistency weight is dynamically adjusted based on the number of batch projects; the more projects, the higher the consistency weight. In this embodiment, the consistency weight typically ranges from 0.95 to 1.05.

[0069] In batch design mode, batch rule verification is supported, allowing users to check the rule compliance of all elevators to be designed at once. The rule violations of each elevator to be designed are summarized and displayed in a list, which facilitates centralized processing and correction.

[0070] In another embodiment, multiple users participate in the elevator design, entering a collaborative design mode: The key characteristics of the elevator are obtained, and the elevator system is divided into several subsystems, including the elevator car subsystem, the elevator shaft subsystem, and the elevator electrical subsystem.

[0071] Obtain supplementary features for each subsystem, execute steps S2-S5, and generate a recommended list of 3D models for each subsystem. Based on the selection behavior of each user on the recommended list of 3D models of each subsystem, the coordination relationship between each subsystem is verified to meet the elevator design constraints. If there is a conflict between parameters of the subsystems, a conflict report is generated and fed back to the relevant users. The conflict report includes the name of the conflicting component, the name of the parameter, the parameter value of each subsystem and the conflict level, and marks the conflict location and conflict type.

[0072] Specifically, the step of obtaining supplementary features for each subsystem, executing steps S2-S5, and generating a 3D model recommendation list for each subsystem includes: Obtain the current user's preference data and generate the current user's preference correction coefficient; Obtain the team's preference data, which includes multiple users, and generate a team preference correction coefficient. The current user preference correction coefficient and the team preference correction coefficient are processed by combining weights. Based on the initial matching score and the weighted preference correction coefficient, the optimized matching score is obtained. Based on the optimized matching score, the candidate 3D models are sorted to generate a recommended list of 3D models.

[0073] In another embodiment, such as Figure 4This invention provides an elevator 3D model recommendation system, comprising an input layer, a data processing layer, an output layer, and a database. The input layer includes a key feature input module 110 and a supplementary feature input module 120, used to acquire key features and supplementary features of the elevator to be designed. The data processing layer generates an optimized matching score based on the key features, supplementary features, and user preference data of the elevator to be designed, wherein the supplementary features are natural language text. The output layer sorts candidate 3D models according to the optimized matching score and generates a 3D model recommendation list. The database includes a historical design data database, a 3D model database, and a user preference database, used to store historical design data, 3D model data, and user preference data, and to enable data reading and writing.

[0074] Specifically, the key feature input module 110 acquires the basic design information of the elevator to be designed, input by the user, including elevator type, rated load capacity, rated speed, number of floors and doors, shaft dimensions, and lifting height. Key features are input via a structured data form, with each parameter having a value range and data type validation rules. The elevator type parameter uses a categorized enumeration method, including machine-room passenger elevators, machine-room-less passenger elevators, medical elevators, freight elevators, machine-room-less freight elevators, residential elevators, and sightseeing elevators. The rated load capacity parameter is in kilograms, with a value range set according to the specific elevator type; freight elevators typically range from 1000kg to 5000kg, while passenger elevators typically range from 450kg to 2500kg. The number of floors and doors is expressed in the format "x floors x stops," where x is a positive integer, such as 3 floors x 3 stops, or 5 floors x 5 stops. The shaft dimensions include shaft width and shaft depth, in millimeters, with specific value ranges determined by referring to a table based on the rated load capacity and elevator type. The key feature input module also encapsulates the acquired key features into key feature vectors and stores them in the feature cache area in the system memory.

[0075] Specifically, the data processing layer includes a natural language parsing module 210, an intent classification and recognition module 220, an entity extraction module 230, a relation mapping module 240, a weight normalization processing module 250, a matching factor extraction module 260, a matching score optimization module 270, and a preference learning optimization module 280.

[0076] The Natural Language Processing (NLP) module 210 performs word segmentation, part-of-speech tagging, and syntactic analysis on natural language text, outputting the language processing results. It calls a Chinese word segmentation algorithm to divide the input natural language text into word sequences. This algorithm employs a sequence labeling model based on Conditional Random Fields (CRF), trained using a large-scale Chinese corpus, and is capable of accurately handling technical terms and quantifier expressions. For example, if the input natural language text is "The car door needs to be widened by 20cm, adopts a center-part opening, and has a safety light curtain," the segmentation result would be "Car door / needs / widening / 20 / cm / , / adopts / center-part / opening / , / with / safety / light curtain." After segmentation, each word is tagged with its part of speech, including nouns, verbs, quantifiers, numerals, locative words, and auxiliary words. The part-of-speech tagging results for the above word segmentation are: "car door / noun / need / verb / widen / verb / 20 / numeral / cm / quantifier / , / punctuation / adopt / verb / partial / noun / open door / noun / , / punctuation / with / verb / safety / noun / light curtain / noun". Syntactic analysis is performed on the part-of-speech tagging results to identify syntactic relationships between words, such as subject-predicate, verb-object, modifier-head, and coordinate relationships. For example, "widen" and "car door" form a verb-object relationship, and "20cm" acts as a complement modifying "widen". The results of word segmentation, part-of-speech tagging, and syntactic analysis are output in a structured data format.

[0077] The intent classification and recognition module 220 is used to input the language processing results output by the natural language parsing module into the design intent classification model, identify the types of design intent contained in the user's natural language text, and output intent labels and confidence scores. The design intent classification model adopts a multi-label classification architecture, capable of identifying one or more design intents. The preset intent type label library covers various dimensions of elevator design, including door width adjustment, door type selection, material requirements, configuration selection, safety devices, surface treatment, and color schemes. The design intent classification model is fine-tuned based on a pre-trained language model, and its parameters are optimized using a large amount of labeled data. When the input text is "The car door needs to be widened by 20cm, adopt a center-split opening, and have a safety light curtain," the model outputs the intent labels "door width adjustment" and "door opening method selection," with confidence scores of 0.92 and 0.88, respectively. The intent classification results are represented in the form of intent vectors, with each dimension corresponding to an intent type label, and the value representing the confidence score of that intent.

[0078] The entity extraction module 230 employs named entity recognition technology, building an entity recognition model based on a BERT pre-trained model combined with a conditional random field decoding layer. It identifies and extracts key entity information from part-of-speech tagged text, outputting the entity extraction results. Entity types include component names, size parameters, structural types, material types, and accessory names. For the above part-of-speech tagging results, the entity extraction results are: component name entity "car door", size parameter entity "20cm", structural type entity "center-split", and accessory entity "safety light curtain". Each extracted entity is accompanied by location information, identifying the start and end positions of the entity in the original text for traceability and verification. The entity extraction results are organized in the form of an entity list, with each entity containing four fields: entity text, entity type, entity confidence, and position offset.

[0079] The relation mapping module 240 is used to semantically map extracted entities to an elevator component parameter library based on intent tags, generating accurate component shape and structural parameters. The elevator component parameter library 110 is a parameter mapping dictionary established based on elevator design standards, organized by component type. Each component type defines its shape parameter set and structural parameter set. Shape parameters include dimensional parameters such as length, width, height, thickness, and diameter. Structural parameters include type selection parameters such as door opening method, support method, and connection method, expressed using enumerated values. Taking the "car door" as an example, its shape parameters include door opening width, door panel height, door panel thickness, and guide rail spacing, while its structural parameters include door opening method (center-opening, side-opening, straight-opening), safety device type, and door operator type. The relation mapping module performs the following mapping process for each extracted entity: determining the target parameter set based on the entity type, then matching the most similar parameter items in the parameter set based on entity semantics, and finally converting natural language values ​​into standard numerical values ​​or enumerated values. For example, the entity "20cm" is mapped to the shape parameter "door opening width" with a value of 200mm, and the entity "safety light curtain" is mapped to the structural parameter "safety device" with an enumeration value of "safety light curtain type". In this embodiment, shape parameters are stored in key-value pairs, where the key is the parameter name and the value is a numerical value with units; structural parameters are also stored in key-value pairs, where the key is the parameter name and the value is an enumeration type or a Boolean type.

[0080] The weight normalization module 250 normalizes the parameter weight values ​​with preset weight values ​​to generate a normalized weight vector. The parameter weight values ​​include component shape parameter weight values ​​and structural parameter weight values. The weight normalization module obtains the parameter weight values ​​input by the user through the supplementary feature input module and loads the preset default weight values. Users can specify weight values ​​between 0 and 1 for each shape and structural parameter to represent the relative importance of that parameter in the matching score calculation. The system's preset default weight values ​​are derived from statistical analysis of historical design data, reflecting the average contribution of each parameter to the matching quality.

[0081] Let the original weight vector be... equal , to The set of, where For the first The original weight values ​​of each parameter. The normalized weight vector. equal , to The set of components is calculated using the following formulas:

[0082] in, For the first A normalized weight vector; This represents the method for representing each normalized weight vector. =1~n; The normalized weight vector satisfies the condition that the sum of all its components equals 1. Normalizing the weight vector effectively balances weight values ​​from different sources, allowing user-defined weights and system-preset weights to be compared and calculated on the same numerical scale, avoiding calculation errors caused by inconsistent weight dimensions. After normalization, a normalized weight vector is generated. .

[0083] The matching factor extraction module 260 is used to select several candidate 3D models based on the basic configuration and shape structure of the elevator to be designed, extract the matching factors of the candidate 3D models, and calculate the initial matching score of the candidate 3D models by combining the matching factors, preset weighting factors and normalized weight vectors.

[0084] Based on the basic configuration and shape structure of the elevator to be designed, several candidate 3D models are selected from a 3D model library. The 3D model library is organized according to component type, and each component type includes several 3D model instances. Each model instance stores the model identifier, model file path, model metadata, and model feature vector. The model feature vector includes multi-dimensional features, such as component type identifier, shape parameter set, structural parameter combination, and associated parameter set.

[0085] Four types of matching factors are extracted from the feature vector of the 3D model: component type matching factor f1, shape parameter matching factor f2, structural parameter matching factor f3, and association parameter matching factor f4. Component type matching factor f1 measures the semantic similarity between the core component type of the candidate 3D model and the target component type. It is obtained by calculating the semantic distance between the two component type identifiers and ranges from 0 to 1, with a higher value indicating a higher degree of matching. Shape parameter matching factor f2 measures the matching degree between the shape parameters of the candidate 3D model and the target shape parameters. For each shape parameter, the relative deviation between the candidate 3D model parameter value and the target parameter value is calculated, and the relative deviation is mapped to a matching score between 0 and 1. The weighted average of the matching scores of each parameter is used as the value of the shape parameter matching factor. Structural parameter matching factor f3 measures the matching degree between the structural parameters of the candidate 3D model and the target structural parameters. For enumerated structural parameters, if the candidate value and the target value are exactly the same, the matching degree is 1; otherwise, a partial matching score is calculated based on the semantic distance of the enumerated values. The matching scores of each structural parameter are then summarized. The association parameter matching factor f4 is used to measure the fit constraints between the candidate 3D model and adjacent parts. It compares whether the candidate model meets the fit constraints and calculates the fit satisfaction rate as the value of the association parameter matching factor.

[0086] The initial matching score is calculated using a weighted summation method, taking into account the contribution of each matching factor. This can be adjusted based on elevator type and design scenario. Assume there are a total of... There are matching factors, and the k-th matching factor is... The corresponding normalized weights are Initial matching score The calculation formula is:

[0087] The initial matching score is obtained by multiplying each matching factor by its normalized weight and then summing the results. After the initial matching score is calculated, all candidate 3D models are sorted in descending order of score to generate an initial candidate list of 3D models.

[0088] The matching score optimization module 270 is used to acquire and calculate the probability of a user selecting a certain candidate 3D model based on user preference data and the attribute features of the candidate 3D models. The probability of a certain preference feature occurring And the probability of this preference feature occurring under the conditions of this model. Obtain the preference correction coefficient : Based on the initial matching score and preference correction coefficient, an optimized matching score is obtained.

[0089] Specifically, the preference learning and optimization module 280 obtains the user's historical selection records from user preference data, including the model recommendation list history, model selection history, unselected models, and browsing behavior history. It analyzes these historical selection records, including frequency analysis, feature preference analysis, time pattern analysis, and contextual correlation analysis. Frequency analysis statistically analyzes the distribution ratio of users' first choice in the 3D model recommendation list, choices made after page browsing, and choices made after browsing multiple pages. Feature preference analysis statistically analyzes the distribution of user preferences for specific feature dimensions of the model, such as the frequency with which the user selects different door opening methods, different materials, and different security configurations. The analysis results are stored in the form of preference feature vectors, with each dimension corresponding to a feature item, the value of which is the user's preference strength for that feature, ranging from 0 to 1.

[0090] The preference correction coefficient is calculated using Bayesian inference: Based on historical selection records and the attribute features of candidate 3D models, the probability of a user selecting a candidate 3D model given the preference features is calculated. Obtain the probability that a user selects a particular candidate 3D model. The probability of a certain preference feature occurring And the probability of this preference feature occurring under the conditions of this model. Calculate the preference correction coefficient : The preference correction coefficient is typically between 0.8 and 1.2. If the preference correction coefficient is greater than 1, it can be considered that the user has a positive preference for the attribute features of the candidate 3D model; if the preference correction coefficient is less than 1, it can be considered that the user has a negative preference for the attribute features of the candidate 3D model.

[0091] Based on initial matching score And preference correction coefficient To obtain an optimized matching score: .

[0092] Specifically, the output layer includes a model ranking module 310 and a recommendation list generation module 320, which are used to rank candidate 3D models according to the optimization matching score and generate a 3D model recommendation list.

[0093] The model ranking module 310 ranks the candidate 3D models according to the optimized matching score. In this embodiment, the models are arranged in descending order of optimized matching score. For candidate 3D models with the same optimized matching score, a secondary ranking is performed based on the component complexity, arranged in ascending order of component complexity. The quantitative index of component complexity is a comprehensive score of the number of sub-components and parameters contained in the 3D model.

[0094] The recommendation list generation module 320 selects the top N 3D models from the sorted candidate 3D models to generate a 3D model recommendation list, where N is typically 10. The 3D model recommendation list also includes a model thumbnail corresponding to each 3D model, the model name, the matching percentage, and the model's main parameter information. The model thumbnail displays the model's appearance as a pre-rendered thumbnail; the model name follows standard naming conventions and includes a combined description of part type, theme specifications, and configuration features; the main parameter information displays the model's key shape and structural parameters in tabular form.

[0095] The 3D model recommendation list is rendered and displayed through front-end interface components in the output layer, employing a responsive layout design that is compatible with both desktop and mobile browsers. The list supports user operations such as filtering by matching degree, filtering by part type, and sorting by update time.

[0096] In this embodiment, the data processing layer further includes a preference update module 290, used to record user behavior data of selecting 3D models in the 3D model recommendation list, and write this data into the user preference data for updating the preference model. The behavior data includes the following fields: a selection timestamp recording the precise time of the selection operation, in the format "year-month-day hour:minute:second"; a selected model identifier recording the unique identifier of the selected model, encoded using an alphanumeric combination; the original sorting position in the recommendation list recording the ranking number of the selected model in the recommendation list; and a browsing history recording the list of model identifiers browsed by the user before making the final selection and their browsing duration. The recording of behavior data is triggered by the event listening module of the front-end interface, sending the data to the back-end data processing module via asynchronous communication. After parsing and formatting the behavior data, the back-end data processing module updates the historical selection record table in the user preference data. The preference model update adopts an incremental update strategy; each time a new selection record is appended to the historical record queue, the retraining process of the preference model is triggered. During retraining, the preference learning optimization module re-counts the updated historical selection records, generating updated preference feature vectors and preference correction coefficients. The updated preference model takes effect immediately and participates in subsequent recommendation score calculations.

[0097] Specifically, it also includes a verification module 200, which automatically generates elevator design constraint rules based on key features and detects in real time whether the current operation violates these rules during the 3D model recommendation and selection process. If a violation of the elevator design constraint rules is detected, an alarm is triggered. The rule generation engine, based on parameters such as rated load capacity, elevator type, and lifting height in the key features, queries a preset elevator design specification knowledge base to automatically generate a set of constraint rules relevant to the current design task. The elevator design specification knowledge base contains structured clauses of national elevator design standards, organized according to safety requirements, performance requirements, and dimensional requirements.

[0098] The alarm triggering mechanism adopts a three-level progressive mode: Level 1 alarm is the prompt level. When it detects that the current operation may violate design constraints, a warning icon and prompt text are displayed on the interface, describing the specific violation, such as "door width is lower than standard requirements." Level 1 alarms do not interrupt the operation process, and users can choose to ignore them and continue. Level 2 alarm is the warning level. Based on Level 1 alarms, a confirmation dialog box pops up, requiring the user to select "Confirm Modification" or "Cancel Modification." When "Confirm Modification" is selected, the system records the operation as an exception and executes the modification. When "Cancel Modification" is selected, the system undoes the current modification and restores the state before the modification. Level 3 alarm is the prohibition level. When it detects that the current operation seriously violates safety regulations, the system directly blocks the operation, does not pop up a confirmation dialog box, and displays a prohibition icon and prohibition instructions on the interface. Level 3 alarms are typically used for mandatory clauses involving personal safety, such as the minimum car area requirement corresponding to the rated load capacity.

[0099] The database comprises a historical design database 410, a 3D model database 420, and a user preference database 430, used to store historical design data, 3D model data, and user preference data, and to enable data reading and writing. The historical design database 410 stores elevator design project data accumulated by the company. Each project record includes information such as project identifier, design parameters, selected model list, design change history, and completion time. The 3D model database 420 includes an elevator 3D model library, storing standardized 3D model files and metadata. The model files are stored in a unified data format, and the metadata includes information such as model identifier, component type, version number, creation time, associated parameters, and file size. The user preference database 430 stores user information and preference data for each user. User information includes employee ID, name, department, position, and start date. Preference data includes historical selection records, preference feature vectors, recent access time, and frequently used parameter settings.

[0100] A technician from an elevator manufacturing company needs to design a machine-room-less freight elevator. Key input features include: elevator type is machine-room-less freight elevator, rated load capacity is 2000kg, lifting height is 25 meters, number of floors and doors is 4, hoistway width is 2400mm, and hoistway depth is 2300mm. Based on these key features, the system initially determines the elevator's basic configuration, automatically setting the car dimensions to approximately 1800mm x 2000mm and the rated speed to 1.0m / s.

[0101] The user inputs a natural language description in the supplementary feature input module: "The car door needs to be widened to 2100mm, adopt a center-split opening, and have a safety light curtain and voice prompt function. The door panel material is stainless steel." The semantic recognition engine processes this description, and the natural language parsing module segments the natural language text into the following words: "Car door / needs / widening / to / 2100mm / , / adopts / center-split / opening / , / with / safety / light curtain / and / voice / prompt / function / , / door panel / material / is / stainless steel." The intent classification and recognition module identifies the intent tags as "door width adjustment," "door opening method selection," "safety configuration," and "material requirements." The entity extraction module extracts the entities: component name "car door," size parameter "2100mm," structure type "center-split," accessories "safety light curtain," "voice prompt," and material "stainless steel." The relation mapping module maps entities to parameters: shape parameters (door opening width 2100mm, door panel thickness 50mm), structural parameters (door opening method enumeration value "center split"), safety device enumeration value "safety light curtain", voice prompt enumeration value "yes"), and material enumeration value "stainless steel". Based on the semantic analysis results, the required 3D model is a 3D model of the car door category.

[0102] The weight normalization module receives user-input parameter weight values: the original weight value for door opening width is 0.7, the original weight value for door opening method is 0.5, the original weight value for material is 0.3, and the original weight value for safety device is 0.2. After normalization, the normalized weight values ​​are 0.401, 0.287, 0.172, and 0.140, respectively. The sum of the normalized weights is exactly equal to 1.0, ensuring the rationality of the weight allocation.

[0103] The matching factor extraction module filters candidate 3D models for the car door category from the 3D model library. Assuming there are 50 3D car door models in the library, the matching factors for each candidate model are extracted. Taking one candidate model as an example: the core component type of the candidate model is a car door, with a semantic similarity of 0.98 to the target component type; the door opening width of the candidate model is 2150mm, with a relative deviation of 2.4% from the target's 2100mm, resulting in a matching degree of 0.90; the door opening method of the candidate model is a split type, perfectly matching the target, with a matching degree of 1.0; the material of the candidate model is stainless steel, perfectly matching the target material, with a matching degree of 1.0; the safety device configuration associated with the candidate model includes a safety light curtain but does not include voice prompts, resulting in a partial matching degree of 0.5 with the target configuration. The initial matching score is calculated as the sum of the products of each matching factor and the normalized weight, resulting in a score of 0.94.

[0104] The preference learning and optimization module queries the user's preference data. The profile records show that in the last 30 designs, the user selected stainless steel material 25 times, a center-part door 20 times, and a safety light curtain configuration 18 times. The preference learning and optimization module calculates the user's preference feature vector: stainless steel material preference strength is 0.83, center-part door preference strength is 0.67, and safety light curtain preference strength is 0.60. Based on Bayesian inference, the preference correction coefficient is calculated to be 1.12, indicating that the user tends to choose models with higher matching scores. The optimized matching score is calculated to be 0.94 × 1.12 = 1.05.

[0105] The system calculates optimization matching scores for each of the 50 candidate 3D car door models, sorts them in descending order of score, and generates a recommended list of the top 10 models. The top-ranked candidate model in the recommended list has an optimization score of 1.08, is named "Center-part Stainless Steel Car Door (2100mm)," and displays a matching degree of "98.2% match." Its key parameters include a door opening width of 2100mm, a door panel thickness of 50mm, stainless steel material, and features such as a safety light curtain and voice prompts. The recommended list is displayed in a card-style list format on the front-end interface, with each card containing a model thumbnail, model name, matching degree, and a summary of key parameters.

[0106] After browsing the recommended list, the user selected the fourth-ranked model, "Center-Splitting Stainless Steel Car Door (2100mm)". Upon receiving the selection, the system displays a detailed information page for the selected model, including a 3D preview view, a complete parameter list, technical specifications, and a list of related components. Simultaneously, the system records the selection behavior data: for example, the selection timestamp is 10:15:32 AM on March 8, 2024; the selected model identifier is CDOOR-2100-SUS-0007; the original ranking position in the recommended list is 4; and the user browsed all models ranked 1 to 6 before making the selection. This behavioral data is then updated in the user's preference data, triggering an incremental update to the preference model. The updated preference model will reflect the user's new preferences for stainless steel material and voice prompt functionality.

[0107] During model selection and design modification, the system automatically generates design constraint rules based on key features. For a freight elevator with a rated load capacity of 2000kg, the generated constraint rules include: the car door width must not be less than 1500mm, the door operator's rated power must not be less than 1.5kW, and the safety light curtain response time must not exceed 30ms. If the user modifies the door width parameter to 1400mm, the system detects that this operation violates the rule of "the car door width must not be less than 1500mm" and immediately triggers a Level 1 alarm, displaying a red warning icon and the text message "Door width is lower than standard requirements" on the interface. If the user confirms and continues to modify, the system pops up a Level 2 alarm confirmation dialog box, requiring confirmation of the modification intention. If the user insists on modifying, the system records this modification as an exception operation and automatically adds an exception description to the design document. If the system detects that the modified door width is 1200mm and is lower than the minimum allowable value specified by the safety standard, the system directly triggers a Level 3 alarm to prevent the modification operation.

[0108] In another embodiment, when a user needs to design multiple elevators with similar configurations simultaneously, they enter batch design mode: In batch design mode, the system acquires the key and supplementary feature descriptions of the first elevator and generates a recommended list of 3D models for it. After selecting the 3D model for the first elevator, users can use the "Batch Application" function to copy the model configuration of the first elevator to a target number of other elevator projects. During batch copying, the system performs a difference analysis on the key features of each target elevator compared to the key features of the first elevator and automatically adjusts relevant parameters.

[0109] Specifically, it also includes a difference analysis module, which compares the key feature vectors of other elevators in the target number with the key feature vector of the main elevator item by item to identify the feature items that differ. For the feature items that differ, the natural language parsing module, intent classification and recognition module, entity extraction module, and relation mapping module are called in sequence to generate corresponding shape parameters and structural parameters for the difference items. If there are no feature items that differ, the elevator directly copies the parameter configuration of the first elevator. For example, if the target elevator and the first elevator have the same rated load capacity but different lifting heights, only the parameters related to the lifting height are adjusted, and the parameters of components related to the lifting height, such as guide rail length parameters, wire rope parameters, and speed governor parameters, are automatically updated.

[0110] After adjustments are made, a recommended list of 3D models of the target elevator is generated. Specifically, the preference learning optimization module analyzes the consistency of component selection among elevators in batch projects, encouraging the use of models with the same or similar configurations within the same batch of projects to improve manufacturing efficiency and maintenance convenience. The consistency weight is dynamically adjusted based on the number of batch projects; the more projects, the higher the consistency weight. Based on the preference correction coefficient, consistency weight, and initial matching score, the optimized matching score is calculated: Optimized matching score = Preference correction coefficient × Consistency weight × Initial matching score, where the consistency weight ranges from 0.95 to 1.05.

[0111] In this embodiment, the verification module is also used to perform batch rule verification in batch design mode, allowing users to check the rule compliance of all target elevators at once, and to summarize and display the rule violations of each target elevator in a list form, which is convenient for centralized processing and correction.

[0112] In another embodiment, when multiple users collaborate on a large elevator design project, they enter a collaborative design mode: The collaborative design mode also includes a project management module and an access control module. The project management module maintains the structural information of the design project, including the project's organizational structure, a list of participating collaborators, the allocation of design responsibilities for each person, and the progress of design tasks. The access control module assigns different operational permissions based on user roles, including viewing permissions, editing permissions, approval permissions, and version control permissions.

[0113] In collaborative design mode, the lead designer first creates a design project, defining the key features and design objectives of the elevator project. The lead designer can assign multiple collaborators to participate in the design, each responsible for different subsystems of the elevator; for example, one person might be responsible for the elevator car subsystem, another for the elevator shaft subsystem, and yet another for the elevator electrical subsystem. Subsystem designers input supplementary features of their respective subsystems through the corresponding supplementary feature input module. The system then executes steps S1 to S5 of the aforementioned method for each subsystem, generating a recommended list of 3D models for each subsystem.

[0114] After each subsystem is designed, an integration verification is performed to check whether the coordination between subsystems meets the elevator design constraints, such as the coordination between car dimensions and hoistway dimensions, and the coordination between door operator travel and door width. If parameter conflicts are detected between subsystems, a conflict report is automatically generated, indicating the conflict location and type, and a conflict notification is sent to the relevant subsystem designers. The conflict report includes the name of the conflicting component, parameter name, parameter values ​​for each subsystem, and conflict level. After the designers adjust the parameters according to the conflict report, the system recalculates the optimization matching score and re-ranks the candidate 3D models.

[0115] In the collaborative design model, the preference learning optimization module is also used to implement team preference learning, analyze the overall preference data of the project team, and extract the team's preference features. For example, if the project team has repeatedly selected a certain brand of door operator products in past projects, the team preference features will record that brand preference, and subsequent recommendations will prioritize products from that brand. Team preference features and individual preference features can be used simultaneously to optimize the matching score calculation. The final optimized matching score combines the individual preference correction coefficient and the team preference correction coefficient through weighted fusion.

[0116] The collaborative design pattern's rule verification alerts utilize a project-level rule set. The system generates project-level design constraint rules based on the key characteristics of the overall project, and simultaneously generates subsystem-level rules based on the parameters of each subsystem. Project-level rules have higher priority; violations of project-level rules by subsystem designers will directly trigger a level-three alert, preventing unauthorized operations. Subsystem-level rules can be exception-applied for by subsystem designers with authorization from the lead designer. Exception applications require a written explanation of the reason and an alternative plan, and take effect upon approval.

[0117] This invention discloses a method and system for recommending 3D elevator models. Through semantic enhancement, it automatically converts the user's natural language description into precise component shape and structural parameters, solving the problem of inaccurate recommendations caused by imprecise feature information. Through the synergistic effect of a natural language parsing module, an intent classification and recognition module, an entity extraction module, and a relationship mapping module, it accurately understands the user's true design intent, significantly improving the matching degree between the recommendation results and actual needs. Through preference learning optimization, it dynamically adjusts the matching score based on the user's previous selection behavior, making the recommendation results more in line with user habits and reducing the time users spend filtering 3D models. By introducing a preference correction coefficient, the recommendation results can adapt to the operating habits and preference characteristics of different users, providing personalized recommendation services.

[0118] Normalization ensures that the weights of each parameter meet the preset normalization conditions, avoiding recommendation bias caused by improper weight allocation. Normalization allows user-defined weights and system-preset weights to be calculated and compared on the same scale, guaranteeing the fairness and rationality of weight allocation. Intelligent recommendation reduces the time users spend manually searching and selecting models, significantly improving the efficiency of non-standard elevator design. Through the synergistic effect of semantic enhancement and preference learning, users no longer need to perform tedious keyword searches and manual filtering; the system automatically recommends the most suitable candidate models, greatly shortening design preparation time.

[0119] Based on the same inventive concept, the present invention also provides an electronic device, comprising: a chip, a processor, and a memory, wherein the memory is used to store computer program code, the computer program code including computer instructions, and the electronic device executes an elevator 3D model recommendation method when the chip executes the computer instructions.

[0120] refer to Figure 5 The electronic device includes a processor 510, a memory 520, an input device 530, and an output device 540. The processor 510, memory 520, input device 530, and output device 540 are coupled together via connectors, which may include various interfaces, transmission lines, or buses, etc., and are not limited in this embodiment of the invention. It should be understood that in the various embodiments of the invention, coupling refers to mutual connection in a specific way, including direct connection or indirect connection through other devices, such as through various interfaces, transmission lines, buses, etc.

[0121] The processor 510 can be one or more graphics processing units (GPUs). If the processor 510 is a GPU, the GPU can be a single-core GPU or a multi-core GPU. Optionally, the processor 510 can be a processor group composed of multiple GPUs, with the multiple processors coupled to each other via one or more buses. Optionally, the processor can also be other types of processors, etc., and this embodiment of the invention is not limited thereto.

[0122] The memory 520 can be used to store computer program instructions, as well as various types of computer program code, including program code for executing the present invention. Optionally, the memory includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or compact disc read-only memory (CD-ROM), which is used for related instructions and data.

[0123] Input device 530 is used to input data and / or signals, and output device 540 is used to output data and / or signals. Output device 540 and input device 530 can be independent devices or an integrated device. The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments and claims of this application are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that, unless otherwise stated, “a plurality” and “several” refer to two or more; “and / or” refers to and includes any or all possible combinations of one or more associated listed items; “first,” “second,” “third,” etc., are used only to distinguish and not to describe a particular order or sequence, nor should they be construed as indicating or implying relative importance. When the above description relates to drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. In the description of this application, those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0124] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0125] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for recommending 3D elevator models, characterized in that, Includes the following steps: S1. Obtain the supplementary features of the elevator to be designed. Based on semantic recognition, perform semantic parsing on the supplementary features of the elevator to be designed to obtain the component names, shape parameters and structural parameters of the components, as well as the weight values ​​of the shape parameters and structural parameters input by the user. The supplementary features are the natural language text input by the user. S2. Normalize the shape parameter weight values, structure parameter weight values ​​and default weight values ​​to generate a normalized weight vector. S3. Based on the component name, select several candidate 3D models from the elevator 3D model library, extract the matching factor of each candidate 3D model, and calculate the initial matching score of the candidate 3D model by combining the component's shape and structural parameters, matching factor, preset weighting factor and normalized weight vector. S4. Optimize and correct the initial matching score based on user preference data to generate an optimized matching score; S5. Sort the candidate 3D models according to the optimization matching score and generate a 3D model recommendation list.

2. The elevator 3D model recommendation method according to claim 1, characterized in that, Also includes: Obtain the key features of the elevator to be designed, and determine the basic configuration and shape structure of the elevator to be designed based on the key features; Key features include the elevator type, rated load capacity, and number of doors per floor of the elevator to be designed; Based on key features, generate elevator design constraint rules; If a user's design operation or selection operation is detected to violate the elevator design constraints, a corresponding alarm operation will be triggered.

3. The elevator 3D model recommendation method according to claim 2, characterized in that, Step S1 includes: Perform word segmentation, part-of-speech tagging, and syntactic analysis on natural language text, and output the language processing results; Input the language processing results into the design intent classification model, and output the intent label and confidence score; Key entity information is identified and extracted from part-of-speech-tagged text using named entity recognition, and the entity extraction results are output. Based on intent tags, the extracted entities are semantically mapped to the elevator component parameter library to generate shape parameters and structural parameters. The system identifies natural language text and obtains the weight values ​​of shape parameters and structure parameters input by the user.

4. The elevator 3D model recommendation method according to claim 1, characterized in that, Step S4 includes: Acquire user preference data, which includes historical 3D model recommendation lists, historical model selections, and historical list browsing behavior. Based on user preference data and the attribute characteristics of each candidate 3D model, calculate the probability that a user will select a particular candidate 3D model. The probability of a certain preference feature occurring And the probability of this preference feature occurring under the conditions of this model. Obtain the preference correction coefficient : ; Based on initial matching score and preference adjustment coefficient This will result in an optimized matching score.

5. The elevator 3D model recommendation method according to claim 2, characterized in that, Also includes: If there are multiple elevators to be designed, obtain the key features and supplementary features of the first elevator, execute steps S1-S5, and generate a recommended list of three-dimensional models of the first elevator. Based on the user's selection behavior of the recommended list of 3D models of the first elevator, the 3D model configuration of the first elevator is updated to the target number of other elevators, and the key features of the other elevators are analyzed to identify the differences between the key features of the first elevator and the key features of the other elevators. Based on semantic recognition, semantic parsing is performed on the differences to generate the component names, shape parameters, and structural parameters of other elevators; Perform steps S3-S5 to generate a recommended list of 3D models for other elevators within the target number.

6. The elevator 3D model recommendation method according to claim 5, characterized in that, Also includes: Based on the number of elevators to be designed, a consistency weight is generated, where the consistency weight is positively correlated with the number of elevators; The initial matching score is optimized and corrected based on user preference data and consistency weights to generate an optimized matching score.

7. The elevator 3D model recommendation method according to claim 2, characterized in that, Also includes: If multiple users participate in the elevator design, obtain the key characteristics of the elevator, and divide the elevator system into several subsystems, the subsystems include the elevator car subsystem, the elevator shaft subsystem, and the elevator electrical subsystem; Obtain the supplementary features of each subsystem, execute steps S2-S5, and generate a 3D model recommendation list for each subsystem. Based on the selection behavior of each user on the recommended list of 3D models of each subsystem, the coordination relationship between each subsystem is verified to meet the elevator design constraints. If there is a conflict between parameters of the subsystems, a conflict report is generated and fed back to the relevant users. The conflict report includes the name of the conflicting component, the name of the parameter, the parameter value of each subsystem and the conflict level, and marks the conflict location and conflict type.

8. The elevator 3D model recommendation method according to claim 7, characterized in that, The process of obtaining supplementary features for each subsystem and executing steps S2-S5 to generate a 3D model recommendation list for each subsystem includes: Obtain the current user's preference data and generate the current user's preference correction coefficient; Obtain the team's preference data, which includes multiple users, and generate a team preference correction coefficient. The current user preference correction coefficient and the team preference correction coefficient are processed by combining weights. Based on the initial matching score and the weighted preference correction coefficient, the optimized matching score is obtained. Based on the optimized matching score, the candidate 3D models are sorted to generate a recommended list of 3D models.

9. An elevator 3D model recommendation system, used to implement the elevator 3D model recommendation method according to any one of claims 1-8, characterized in that, It includes an input layer, a data processing layer, an output layer, and a database. The input layer includes a key feature input module and a supplementary feature input module, which are used to obtain the key features and supplementary features of the elevator to be designed. The data processing layer is used to determine the basic configuration and shape structure of the elevator to be designed based on key features; Based on semantic recognition, the supplementary features of the elevator to be designed are semantically parsed to obtain the component names, shape parameters and structural parameters of the elevator to be designed, as well as the weight values ​​of the shape parameters and structural parameters input by the user. The shape parameter weight values, structure parameter weight values, and default weight values ​​are normalized to generate a normalized weight vector. Based on the component name, several candidate 3D models are selected from the elevator 3D model library. The matching factor of each candidate 3D model is extracted. Combining the component's shape and structural parameters, the matching factor, the preset weighting factor, and the normalized weight vector, the initial matching score of the candidate 3D model is calculated. The initial matching score is optimized and corrected based on user preference data to generate an optimized matching score. Key features include the elevator type and rated load capacity of the elevator to be designed, and supplementary features are natural language text. The output layer is used to sort the candidate 3D models based on the optimization matching score and generate a recommended list of 3D models; The database includes a historical design data database, a 3D model database, and a user preference database, which are used to store historical design data, elevator 3D model data, and user preference data, and enable data reading and writing.

10. An electronic device, characterized in that, include: The electronic device comprises a chip, a processor, and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein, when the chip executes the computer instructions, the electronic device performs an elevator 3D model recommendation method as described in any one of claims 1 to 8.

Images