A model construction method and system based on virtual scene

Abstract

The present invention relates to the technical field of virtual scene construction, and discloses a model construction method and system based on virtual scenes, including: collecting construction requirements of virtual scenes; inferring a complete demand model based on the construction requirements; and constructing virtual scenes based on the inferred demand model. It can accurately capture user needs and improve the accuracy of demand capture. The initial population is composed of randomly selected element information, and a preliminary scene that meets the needs is quickly generated through an intelligent algorithm, and is continuously optimized during the iterative process, thereby realizing efficient virtual scene construction. When guiding the user's demand description, through diversified input methods and flexible interaction rules, the user can adjust the demand at any time. The scene can be optimized according to different demand dimensions, so that the virtual scene finally constructed is not only fully functional, but also reasonably laid out.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual scene construction, and in particular to a model construction method and system based on virtual scenes. Background Art

[0002] Virtual scene construction technology is widely used in game development, virtual reality, architectural design, and smart city planning. With the continuous development of computer graphics and artificial intelligence technology, the accuracy and complexity of virtual scene construction have been significantly improved. However, current virtual scene construction methods still face many challenges, including inaccurate capture of user needs, unreasonable construction of scene elements, lack of flexible interaction methods, and low construction efficiency.

[0003] Traditional virtual scene construction methods usually rely on manual design and layout by professionals, which is not only time-consuming and labor-intensive, but also difficult to meet the personalized needs of different users. In order to improve efficiency and accuracy, researchers have begun to introduce automation technology and intelligent algorithms to generate and optimize virtual scenes through computer-aided design tools. For example, basic scenes are generated using rule engines and template libraries, and scenes are refined through parametric modeling tools. However, these methods still seem to be unable to cope with complex scenes and diverse needs. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing virtual scene construction method has the problems of high cost and low efficiency.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a model construction method based on a virtual scene, comprising:

[0007] Collect requirements for building virtual scenes;

[0008] Based on the construction requirements, a complete demand model is inferred;

[0009] Build virtual scenarios based on the inferred demand model.

[0010] As a preferred solution of the virtual scene-based model construction method described in the present invention, the construction requirements include guiding the user to clearly describe the requirements through a guided question-answering system, and the specific guidance content is: scene type, functional requirements in the scene, and aesthetic requirements;

[0011] Among them, the required input methods include text input and voice input.

[0012] As a preferred solution of the virtual scene-based model construction method described in the present invention, the construction requirements also include using automatic speech recognition technology to convert voice input content into text form, and extracting and vectorizing the text input content and the voice input content in text form;

[0013] The specific steps of content extraction and vectorization are:

[0014] Collect the input text of different guide contents in all successful simulation records and get it after vectorization ; Get the standard reference content in the question-answering system and vectorize it to get ;use and Together they constitute a reference vector set;

[0015] ,

[0016] in, Indicates the number of The vector representation of the guided content, BERT means vectorization using the BERT model;

[0017] ,

[0018] in, Indicates the standard reference in the question-answering system A vector representation of the content;

[0019] Construct a reference vector set:

[0020] ,

[0021] in, represents a reference vector set, which contains vectors of all historical records and standard references in the question-answering system; n represents the number of all successful simulation records; m represents the number of standard references in the question-answering system;

[0022] Calculate the similarity score:

[0023] ,

[0024] Where D represents the dimension of the vector; Representation vector The value in the dth dimension; Representation vector The value in the dth dimension; Represents a reference vector set The rth vector in ; Represents the vectorization of input text t; =1, and Represents the weight coefficient in different guide contents and The values ​​are different;

[0025] Iterate over a collection of reference vectors , calculate the similarity score between each reference vector and the text vector, and get the value with the highest similarity score;

[0026] ,

[0027] in, Indicates the value with the highest similarity score;

[0028] If the guiding content is scene type or aesthetic requirements, output The reference vector in is used as the demand simulation target; if the guidance content is the functional requirement in the scene, the preset threshold and and ,when , then the content of the text input satisfies the judgment, and the output The reference vector in is used as the demand simulation target; when Then the content of the text input does not meet the judgment, and the output All reference vectors of are used as demand simulation targets;

[0029] The successful simulation record includes that the user evaluates the model construction result of the virtual scene as excellent, and there is no behavior of changing the input construction requirements during the simulation process;

[0030] The standard reference includes storing standard input text in the guidance content of the guided question-answering system, and by identifying the standard input text, it is possible to directly match the corresponding virtual scene content.

[0031] As a preferred solution of the model building method based on virtual scene of the present invention, wherein: the inference of the complete demand model includes, using the demand simulation target, matching the element information in the virtual scene database;

[0032] If the required simulation target is the standard reference, the corresponding virtual scene content is directly matched as the element information in the virtual scene; if the required simulation target is the input text in the successful simulation record, all virtual scene contents matched in the successful simulation record are used as the element information in the virtual scene;

[0033] Classify and integrate the element information matched by each guide content to obtain the complete requirements for the model:

[0034] ,

[0035] Among them, H represents the complete requirements for the model, C represents the element information of the scene type in the virtual scene, G represents the element information of the functional requirements in the virtual scene, and M represents the element information of the aesthetic requirements in the virtual scene.

[0036] As a preferred solution of the model building method based on virtual scenes described in the present invention, wherein: the building of virtual scenes for the inferred demand model includes building the virtual scenes using an improved annealing algorithm;

[0037] Suppose the element set of scene types in the virtual scene database is , the element set of functional requirements is , the set of elements of aesthetic requirements is ;

[0038] Individuals in the initial population , which is composed of randomly selected scene type elements, all functional requirement elements, and randomly selected aesthetic requirement elements;

[0039] Initial population:

[0040] ,

[0041] in, Indicates any number of elements randomly selected from the element information C of the scene type; Indicates any number of elements randomly selected from the element information C of aesthetic requirements;

[0042] Element position initialization:

[0043] The position of each element in virtual space is represented by , the initial position is in the range Random assignment within ;

[0044] Where W represents the range length of the virtual space, Represents the u-th element.

[0045] As a preferred solution of the model building method based on virtual scene described in the present invention, wherein: the improved annealing algorithm also includes defining the initial temperature and cooling rate: assuming the initial temperature T0 and the cooling rate α;

[0046] Calculate the fitness of the current virtual scene:

[0047] Combining the vector representations of scene type, functional requirements, and aesthetic requirements and their spatial locations into an input vector;

[0048] ,

[0049] Among them, the element position representing the scene type, the element position representing the functional requirement, the element position representing the aesthetic requirement element;

[0050] The neural network outputs a fitness score through multiple non - linear transformations ;

[0051] ,

[0052] Generate a new solution by changing the position and random perturbation:

[0053] ,

[0054] Among them, represents the perturbation intensity;

[0055] Decide whether to accept the new solution according to the Metropolis criterion:

[0056] ,

[0057] Among them, is the fitness difference between the new solution and the current solution, represents the current temperature; if L < P(ΔE), then accept the new solution, otherwise keep the current solution;

[0058] Continue to cool down until the temperature reaches the lowest or the maximum number of iterations is reached. The cooling process updates the temperature through exponential decay;

[0059] ,

[0060] Among them, represents the cooling rate.

[0061] As a preferred solution of the model construction method based on virtual scenes described in the present invention, among them: after completing the solution process of one round of the annealing algorithm, randomly re - select the scene type elements and aesthetic requirement elements to generate a completely new initial population, and use the annealing algorithm to solve;

[0062] After completing the generation combinations of all initial populations, for each combination, sort the fitness values obtained by the annealing algorithm from largest to smallest, and construct the model of the virtual scene;

[0063] According to the sorting, preferentially map the virtual scenes with earlier sequences. If the user selects to update the scene, then recursively map the virtual scenes of the next sequence backward according to the sorting; if the user selects to change the input construction requirements, then re - execute the model construction of the virtual scene according to the input requirements;

[0064] After completing the simulation task of the virtual scene, the simulation results are evaluated as excellent, good or poor through human-computer interaction.

[0065] A model building system based on a virtual scene using the method of the present invention, wherein:

[0066] A collection unit collects the construction requirements of the virtual scene;

[0067] An analysis unit, based on the construction requirements, infers a complete demand model;

[0068] The construction unit constructs a virtual scene based on the inferred demand model.

[0069] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein: the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0070] A computer-readable storage medium stores a computer program, wherein: when the computer program is executed by a processor, the steps of any one of the methods of the present invention are implemented.

[0071] Beneficial effects of the present invention: The model construction method based on virtual scenes provided by the present invention can accurately capture user needs, including scene types, functional requirements and aesthetic requirements. Through content extraction and vectorization processing of text and voice input, the comprehensiveness and accuracy of demand information are ensured, and the accuracy of demand capture is improved. The initial population is composed of randomly selected element information, and a preliminary scene that meets the needs is quickly generated through an intelligent algorithm, and is continuously optimized during the iterative process, thereby realizing efficient virtual scene construction. When guiding the user's demand description, through diversified input methods and flexible interaction rules, users can adjust their needs at any time. The system dynamically adjusts and optimizes the scene based on user feedback and evaluation, ensuring the flexibility of the scene construction process and user satisfaction. The scene can be optimized according to different demand dimensions, so that the final constructed virtual scene is not only fully functional, but also reasonably laid out. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0073] Figure 1 An overall flow chart of a model building method based on a virtual scene provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0075] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a model building method based on a virtual scene, comprising:

[0076] S1: Collect the requirements for building virtual scenes.

[0077] Furthermore, through a guided question-and-answer system, users are guided to clearly describe their needs. The specific guidance content is: scene type, functional requirements in the scene, and aesthetic requirements; among them, the input methods of requirements include text input and voice input.

[0078] Specifically, automatic speech recognition technology is used to convert voice input content into text form, and content extraction and vectorization are performed on the text input content and the voice input content in text form.

[0079] The specific steps of content extraction and vectorization are:

[0080] Collect the input text of different guide contents in all successful simulation records and get it after vectorization ; Get the standard reference content in the question-answering system and vectorize it to get ;use and Together they constitute a reference vector set.

[0081] ,

[0082] in, Indicates the number of The vector representation of the guided content, BERT means vectorization using the BERT model.

[0083] ,

[0084] in, Indicates the standard reference in the question-answering system A vector representation of the content.

[0085] Construct a reference vector set:

[0086] ,

[0087] in, represents a reference vector set, which contains vectors of all historical records and standard references in the question-answering system; n represents the number of all successful simulation records; m represents the number of standard references in the question-answering system.

[0088] Calculate the similarity score:

[0089] ,

[0090] Where D represents the dimension of the vector; Representation vector The value in the dth dimension; Representation vector The value in the dth dimension; Represents a reference vector set The rth vector in ; Represents the vectorization of input text t; =1, and Represents the weight coefficient in different guide contents and The values ​​are different.

[0091] Iterate over a collection of reference vectors , calculate the similarity score between each reference vector and the text vector, and get the value with the highest similarity score.

[0092] ,

[0093] in, Indicates the value with the highest similarity score.

[0094] If the guiding content is scene type or aesthetic requirements, output The reference vector in is used as the demand simulation target; if the guidance content is the functional requirement in the scene, the preset threshold and and ,when , then the content of the text input satisfies the judgment, and the output The reference vector in is used as the demand simulation target; when Then the content of the text input does not meet the judgment, and the output All reference vectors are used as demand simulation targets.

[0095] It should be noted that the successful simulation record includes that the user evaluates the model construction result of the virtual scene as excellent, and there is no behavior of changing the input construction requirements during the simulation process; the standard reference includes that the standard input text is stored in the guidance content of the guided question-and-answer system, and by identifying the standard input text, it can be directly matched to the corresponding virtual scene content.

[0096] It is also important to know that by converting voice input into text through automatic speech recognition technology, combining text input content, and using the BERT model for content extraction and vectorization, user demand information can be accurately extracted to avoid information omission or misunderstanding. The input text of different guide contents in the successful simulation record and the standard reference content in the question-answering system are collected and vectorized by the BERT model to form a reference vector set. This enables the system to use historical data and standard references for high-quality demand matching and improve the accuracy of demand understanding. By calculating the similarity score between the input text vector and each vector in the reference vector set, the system can quickly identify the scene elements that best match the user's needs. This step ensures that the elements selected during the scene construction process can meet user needs to the greatest extent. The system can select and match the most appropriate scene elements according to the specific needs of different guide contents (such as scene type, functional requirements, and aesthetic requirements), output personalized demand simulation targets, and ensure that the generated virtual scene meets the personalized needs of users.

[0097] By setting different weight coefficients, the system can flexibly adjust the weight of similarity calculation according to the importance of different guidance contents, and realize multi-dimensional demand assessment and optimization.

[0098] S2: Based on the construction requirements, infer a complete demand model.

[0099] The element information in the virtual scene database is matched by using the required simulation target. If the required simulation target is the standard reference, the corresponding virtual scene content is directly matched as the element information in the virtual scene; if the required simulation target is the input text in the successful simulation record, all virtual scene contents matched in the successful simulation record are used as the element information in the virtual scene.

[0100] Classify and integrate the element information matched by each guide content to obtain the complete requirements for the model:

[0101] ,

[0102] Among them, H represents the complete requirements for the model, C represents the element information of the scene type in the virtual scene, G represents the element information of the functional requirements in the virtual scene, and M represents the element information of the aesthetic requirements in the virtual scene.

[0103] It is to be said that through matching and classification, an accurate and detailed demand model HHH is obtained, which provides an accurate demand basis for the construction of virtual scenes. This step ensures that the subsequent virtual scene construction can be based on comprehensive and reliable demand data. By utilizing the element information in the standard reference and successful simulation records, the high quality of the matched scene elements in terms of consistency and rationality is guaranteed. This lays the foundation for the construction of virtual scenes and ensures that the final generated virtual scenes can meet user needs. The classified and integrated demand model HHH contains all the necessary scene element information and provides basic data support for intelligent scene construction. Subsequent steps can be optimized based on this information through intelligent algorithms (such as annealing algorithms and neural networks) to generate high-quality virtual scenes that meet user needs. Through detailed demand analysis and information classification, the uncertainty and repeated operations in the subsequent scene construction process are reduced, and the construction efficiency is improved. The classified element information C, G, and M can be directly used for the layout and optimization of scene elements to improve the construction effect.

[0104] S3: Construct a virtual scenario based on the inferred demand model.

[0105] The virtual scene is constructed using the improved annealing algorithm.

[0106] Suppose the element set of scene types in the virtual scene database is , the element set of functional requirements is , the set of elements of aesthetic requirements is .

[0107] Individuals in the initial population , which is composed of randomly selected scene type elements, all functional requirement elements and randomly selected aesthetic requirement elements.

[0108] Initial population:

[0109] ,

[0110] in, Indicates any number of elements randomly selected from the element information C of the scene type; In the element information C of aesthetic requirements, any number of elements randomly selected are shown.

[0111] Element position initialization:

[0112] The position of each element in the virtual space is represented by , the initial position is in the range Random assignment within .

[0113] Where W represents the range length of the virtual space, Represents the u-th element.

[0114] It should be said that the improved annealing algorithm is a global optimization algorithm that can effectively avoid the local optimal problem. By simulating the physical annealing process and introducing random perturbations and mutation mechanisms, the virtual scene can be gradually optimized to find the global optimal solution. The initial population is constructed by randomly selecting elements of scene types and aesthetic requirements and combining all functional requirements elements. The diversity of the initial population helps to enhance the algorithm's search ability and improve the quality and diversity of the final generated scene.

[0115] The position of each element in the virtual space is initialized as a random assignment, so that each individual in the initial population has a certain randomness in spatial layout, avoiding excessive concentration and limitations of the initial state, and providing a wider search space for the subsequent optimization process.

[0116] The diversity and randomness of the initial population provide a rich optimization basis for the annealing algorithm, which helps the algorithm to effectively search globally and find the best virtual scene layout solution. Through the intelligent optimization algorithm, individuals in the initial population can be quickly screened and optimized, shortening the time of virtual scene construction and improving construction efficiency. The optimization algorithm repeatedly adjusts and evaluates the position and combination of elements to ensure that the final generated virtual scene is more reasonable in layout, more functional and more beautiful.

[0117] Define initial temperature and cooling rate: Let initial temperature T0 and cooling rate α.

[0118] Calculate the fitness of the current virtual scene:

[0119] The vector representations of scene type, functional requirements, and aesthetic requirements and their spatial locations are combined into an input vector.

[0120] ,

[0121] in, The element position indicating the scene type, The location of the element that represents the functional requirement, Indicates the element position of the aesthetically required element.

[0122] The neural network outputs a fitness score through multiple layers of nonlinear transformations .

[0123] ,

[0124] Generate new solutions by replacing positions and random perturbations:

[0125] ,

[0126] Among them, represents the perturbation intensity.

[0127] Decide whether to accept the new solution according to the Metropolis criterion:

[0128] ,

[0129] Among them, is the fitness difference between the new solution and the current solution, represents the current temperature; if L < P(ΔE), then accept the new solution, otherwise keep the current solution.

[0130] Continue to cool down until the temperature reaches the lowest or the maximum number of iterations. The cooling process updates the temperature through exponential decay.

[0131] ,

[0132] Among them, represents the cooling rate.

[0133] It should be noted that using the improved annealing algorithm to construct the virtual scene can efficiently find the global optimal solution and generate a high-quality virtual scene that meets the user's needs. This step ensures the flexibility and stability of the algorithm through the intelligent optimization algorithm and the random perturbation mechanism, and improves the efficiency and effect of virtual scene construction.

[0134] Furthermore, after completing a round of solution of the annealing algorithm, randomly select the scene type elements and aesthetic requirement elements again to generate a new initial population, and use the annealing algorithm to solve.

[0135] After completing the generation combination of all initial populations, for each combination, sort the fitness of the solution obtained by the annealing algorithm from largest to smallest, and perform the model construction of the virtual scene.

[0136] According to the sorting, map the virtual scenes with higher sequences first. If the user chooses to update the scene, then recursively map the virtual scenes of the next sequence backward; if the user chooses to change the input construction requirements, then re-perform the model construction of the virtual scene according to the input requirements.

[0137] After completing the simulation task of the virtual scene, evaluate the simulation results as excellent, good, or poor through human-computer interaction.

[0138] It should be said that by re-randomly selecting scene types and aesthetic elements, a new initial population is generated, and multiple rounds of annealing algorithm solutions are performed. The scenes generated in each round may have different element combinations and layout schemes, thereby improving the diversity of scene construction. The fitness of the annealing algorithm solution under each combination is sorted, and the virtual scenes at the front of the sequence are mapped first. This step selects the optimal scene through intelligent sorting to improve the efficiency and effect of scene construction. By prioritizing the mapping of scenes with high fitness, unnecessary repeated calculations and adjustments are reduced, the construction efficiency is improved, and the quality of the scene is guaranteed. Through human-computer interaction, users evaluate the simulation results and further optimize the scene construction algorithm and parameter settings based on the evaluation results.

[0139] On the other hand, this embodiment also provides a model building system based on a virtual scene, which includes:

[0140] The collection unit collects the construction requirements of the virtual scene.

[0141] The analysis unit infers a complete demand model based on the construction requirements.

[0142] The construction unit constructs a virtual scene based on the inferred demand model.

[0143] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0144] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0145] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0146] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0147] Example 2, below, is an embodiment of the present invention, which provides a model construction method based on a virtual scene. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0148] Through simulation experiment 1, the performance of this method is compared with that of traditional methods. Among them, the traditional methods include: manual design: professional designers manually build virtual scenes according to user needs. Automated tools: using existing automated tools to perform preliminary layout and combination of scene elements, without intelligent optimization and user feedback mechanism.

[0149] The scene construction time and user satisfaction score of each method are recorded (both are average values), and the fitness scores of the three methods are calculated by the fitness calculation method of the present invention. The experimental data are recorded in Table 1.

[0150] Table 1 Experimental data table

[0151]

[0152] From Table 1, it can be seen that the method of the present invention is superior to the traditional method in terms of fitness score, scene construction time and user satisfaction score. The average fitness score in the traditional method-manual design is 0.65. Due to the complete reliance on manual design, the rationality of the scene layout and element combination is greatly affected by the experience of the designer, and there are subjectivity and limitations. The average fitness score in the traditional method-automated tool is 0.72. Although the automated tool is used, the lack of intelligent optimization and multi-round solution mechanism leads to limited improvement in the fitness score. The average fitness score in this method is 0.85. Through the improved annealing algorithm, multiple rounds of solution and optimization are performed, which effectively avoids the local optimum and significantly improves the rationality of the scene layout and element combination. The scene construction time in the traditional method-manual design is 180 minutes. Manual design is time-consuming, inefficient, and difficult to quickly respond to changes in user needs. The scene construction time in the traditional method-automated tool is 120 minutes. The automated tool has improved efficiency to a certain extent, but lacks intelligent optimization and the construction time is still long. The scene construction time in this method is 60 minutes. Through intelligent optimization algorithms and automated processes, the efficiency of scene construction is greatly improved and the construction time is shortened. The user satisfaction score in the traditional method-manual design is 3.5. Since the scene design relies on manual experience and is highly subjective, user satisfaction is relatively low. The user satisfaction score in the traditional method-automated tool is 4.0. Automated tools have improved the rationality of scene design to a certain extent, but there is a lack of user feedback mechanism, and the improvement of user satisfaction is limited. The user satisfaction score in this method is 4.8. Through the guided question-and-answer system, intelligent optimization algorithm and user feedback-driven optimization mechanism, the generated virtual scene is more in line with user needs, and user satisfaction is significantly improved.

[0153] Simulation experiment 2 will compare the performance of the method of the present invention and the automation tools in the traditional method under different simulation requirements. 30 users with different backgrounds were recruited to test the traditional method and the present method respectively. The data are recorded in Table 2.

[0154] Each user constructs a virtual scene using the traditional method and the method of the present invention respectively.

[0155] Users evaluate the simulation results and record user satisfaction, number of refreshes, and number of demand changes.

[0156] Each user simulated each method three times to ensure the reliability and stability of the data.

[0157] Table 2 Data record table

[0158]

[0159] It can be seen from Table 2 that under different simulation requirements, the user satisfaction score of the method of the present invention is significantly higher than that of the traditional method, and the number of refresh times and demand replacement times are significantly reduced.

[0160] The present invention significantly improves user satisfaction and reduces the number of refreshes and demand changes under different simulation requirements through intelligent optimization algorithms, user feedback-driven optimization mechanisms, and automated processes. Compared with traditional methods, the present invention has obvious advantages in construction efficiency, user satisfaction, and flexibility.

[0161] Specifically, the present invention shows excellent performance in different scene types such as entertainment areas, commercial areas, residential areas, educational areas and industrial areas. The user satisfaction score is significantly higher than that of traditional methods, showing the powerful ability of intelligent optimization algorithms in meeting user needs. At the same time, the significant reduction in the number of refreshes and demand changes proves the excellent performance of the present method in terms of construction efficiency and flexibility.

[0162] Through the above comparative analysis, the present invention demonstrates its innovation and practicality in the field of virtual scene construction, and has broad application prospects and market potential. The experimental data fully proves the advantages and effectiveness of the method of the present invention in practical applications.

[0163] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A model construction method based on a virtual scene, characterized in that: include: Collect requirements for building virtual scenes; Based on the construction requirements, a complete demand model is inferred; Construct virtual scenarios based on the inferred demand model; The construction of requirements includes guiding the user to clearly describe the requirements through a guided question-and-answer system, and the specific guidance content is: scene type, functional requirements in the scene, and aesthetic requirements; Among them, the required input methods include text input and voice input; The construction requirements also include using automatic speech recognition technology to convert voice input content into text form, and extracting and vectorizing the text input content and the voice input content in text form; The specific steps of content extraction and vectorization are: Collect the input text of different guide contents in all successful simulation records and get it after vectorization Get the content of the standard reference in the question-answering system and vectorize it to get use and Together they constitute a reference vector set; in, Represents the vector representation of the i-th guide content in the history record, and BERT represents vectorization using the BERT model; in, The vector representation of the jth content in the standard reference in the question-answering system; Construct a reference vector set: in, represents a reference vector set, which contains vectors of all historical records and standard references in the question-answering system; n represents the number of all successful simulation records; m represents the number of standard references in the question-answering system; Calculate the similarity score: Where D represents the dimension of the vector; Represents vector V t The value in the dth dimension; Represents vector V r The value in the dth dimension; V r Represents a reference vector set The rth vector in V t =BERT(t) represents the vectorization of the input text t; α+β=1, α and β represent weight coefficients, and the values ​​of α and β in different guide contents are different; Iterate over a collection of reference vectors Calculate the similarity score between each reference vector and the text vector, and get the value with the highest similarity score; Among them, S * Indicates the value with the highest similarity score; If the guiding content is scene type or aesthetic requirement, output S * The reference vector in is used as the demand simulation target; if the guidance content is the functional requirement in the scene, the thresholds δ and θ are preset and δ>θ, when S * >δ, then the content of the text input satisfies the judgment, and output S * The reference vector in is used as the demand simulation target; when S * ≤δ, then the text input content does not meet the judgment, and output S(V t ,V r )>θ as the demand simulation target; The successful simulation record includes that the user evaluates the model construction result of the virtual scene as excellent, and there is no behavior of changing the input construction requirements during the simulation process; The standard reference includes storing standard input text in the guidance content of the guided question-answering system, and by identifying the standard input text, it is possible to directly match the corresponding virtual scene content.

2. The model building method based on virtual scene according to claim 1, characterized in that: The inferring a complete demand model includes using the demand simulation target to match element information in a virtual scene database; If the required simulation target is the standard reference, the corresponding virtual scene content is directly matched as the element information in the virtual scene; If the required simulation target is the input text in the successful simulation record, all virtual scene contents matched in the successful simulation record are used as element information in the virtual scene; Classify and integrate the element information matched by each guide content to obtain the complete requirements for the model: H = {C, G, M} Among them, H represents the complete requirements for the model, C represents the element information of the scene type in the virtual scene, G represents the element information of the functional requirements in the virtual scene, and M represents the element information of the aesthetic requirements in the virtual scene.

3. The model building method based on virtual scene according to claim 2, characterized in that: The constructing of a virtual scene for the inferred demand model includes constructing the virtual scene using an improved annealing algorithm; Assume that the element set of scene types in the virtual scene database is E C , the element set of functional requirements is E G , the set of elements of aesthetic requirements is E M ; Individual P in the initial population k , which is composed of randomly selected scene type elements, all functional requirement elements, and randomly selected aesthetic requirement elements; Initial population: in, Indicates any number of elements randomly selected from the element information C of the scene type; Indicates any number of elements randomly selected from the element information C of aesthetic requirements; Element position initialization: The position of each element in the virtual space is represented by pos(e u )=(x u ,y u ), the initial position is randomly assigned in the range [0,W]×[0,W], x u ,y u ∈[0,W]; Where W represents the range length of the virtual space, e u Represents the u-th element.

4. The model building method based on virtual scene according to claim 3, characterized in that: The improved annealing algorithm further includes defining an initial temperature and a cooling rate: assuming an initial temperature T0 and a cooling rate α; Calculate the fitness of the current virtual scene: Combining the vector representations of scene type, functional requirements, and aesthetic requirements and their spatial locations into an input vector; x=[P k ,pos(C k ),pos(G),pos(M k )] Among them, pos(C k ) indicates the element position of the scene type, pos(G) indicates the element position of the functional requirement, pos(M k ) represents the element position of the aesthetic requirement element; The neural network outputs the fitness score F through multiple layers of nonlinear transformation total ; F total =NN(x) Generate new solutions by replacing positions and randomly perturbing: pose u ) new =pos(and u )+σ,σ∈[-∈,∈] Among them, ∈ represents the disturbance intensity; Decide whether to accept the new solution according to the Metropolis criteria: where ΔE = F new - F current is the fitness difference between the new solution and the current solution, and T represents the current temperature; if L < P(ΔE), the new solution is accepted, otherwise the current solution is maintained; Continue cooling until the temperature reaches the minimum or the maximum number of iterations is reached. The cooling process updates the temperature through exponential decrease; T new =α·T current Here, α∈(0,1) represents the cooling rate.

5. The model building method based on virtual scene according to claim 4, characterized in that: After completing a round of annealing algorithm solution process, randomly select scene type elements and aesthetic requirement elements again to generate a new initial population and use annealing algorithm to solve; After all the combinations of initial populations are generated, the fitness of the annealing algorithm under each combination is sorted from large to small, and the model of the virtual scene is constructed; According to the order, the virtual scenes at the front of the sequence are mapped first. If the user chooses to update the scene, the order is recursively followed to map the virtual scenes in the next sequence. If the user chooses to change the input construction requirements, the model construction of the virtual scene is re-executed according to the input requirements. After completing the simulation task of the virtual scene, the simulation results are evaluated as excellent, good or poor through human-computer interaction.

6. A model building system based on a virtual scene using the method according to any one of claims 1 to 5, characterized in that: A collection unit collects the construction requirements of the virtual scene; An analysis unit, based on the construction requirements, infers a complete demand model; The construction unit constructs a virtual scene based on the inferred demand model.

7. A computer device comprising: Memory and processor; The memory stores a computer program, wherein the processor implements the steps of any one of the methods of claims 1-5 when executing the computer program.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images