Configuration method and system for large-scale 3D model dynamic loading strategy in digital twinborn visualization platform

Abstract

The invention relates to the technical field of terminal computing power visualization, in particular to a configuration method and system for a dynamic loading strategy of a large-scale 3D model in a digital twin visualization platform, and the method comprises the following steps: collecting front-end burying point data, uploading user data, preprocessing the user data, constructing a user behavior database, and selecting a machine learning model. Model construction and training, machine learning prediction, dynamic loading strategy configuration and loading strategy optimization are carried out; the method has the beneficial effects that the user behavior data is collected through the front-end burying point, and the user concern area is predicted by applying machine learning, so that the dynamic optimization of the loading strategy is realized. According to the method, the loading efficiency of the large-scale 3D model is effectively improved, the waiting time of the user is shortened, meanwhile, according to different attention degrees of the user, the mode of combining preloading and loading as needed is adopted, the smooth experience of the attention area of the user is guaranteed, and system resources are saved.

Description

Technical Field

[0001] The present invention relates to the technical field of terminal computing power visualization, and specifically to a configuration method and system for a dynamic loading strategy of large-scale 3D models in a digital twin visualization platform. Background Technique

[0002] In the application of digital twin technology, the loading efficiency of large-scale 3D models and the user experience are key issues.

[0003] The existing loading strategies mainly adopt static or fixed loading modes, without considering the actual needs and behavior habits of users, resulting in resource waste and a decline in user experience. With the rapid development of digital twin technology, its applications in various fields are becoming increasingly widespread, but the loading and rendering of large-scale 3D models have always been the key factors restricting the platform performance and user experience. Traditional loading strategies mainly adopt static or fixed methods and cannot be flexibly adjusted according to the actual needs and behavior habits of users, resulting in low loading efficiency, long user waiting time, and seriously affecting the interactive experience and practicality of the platform.

[0004] Therefore, there is an urgent need for a method that can dynamically adjust the loading strategy according to user behavior. Summary of the Invention

[0005] The purpose of the present invention is to provide a configuration method and system for a dynamic loading strategy of large-scale 3D models in a digital twin visualization platform to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A configuration method for a dynamic loading strategy of large-scale 3D models in a digital twin visualization platform, the method comprising the following steps:

[0007] S1. Front-end buried point data collection: Set buried points in the front-end application of the digital twin visualization platform for collecting user interaction behavior data;

[0008] S2. User data upload: Real-time upload the user data collected in step S1 to the back-end server for subsequent machine learning model training and analysis;

[0009] S3. User data preprocessing: Perform data deduplication, data verification, missing value and outlier processing on the user data received in S2 to ensure data validity;

[0010] S4. Build a user behavior database, design the database table structure according to the structure and characteristics of the user behavior data, including a user information table and a user behavior and area ID record table, and import the preprocessed data into the database;

[0011] S5. Select a machine learning model. The supervised learning method will be adopted, and the logistic regression algorithm will be selected for prediction;

[0012] S6. Model construction and training: Use the scikit-learn library function of the logistic regression algorithm and introduce the LogisticRegression model class to create a model instance; Use the training dataset in S3 to call the fit method of the model to train the model;

[0013] S7. Machine learning prediction: Use the test dataset in S3 to call the predict method of the model, return the prediction results, and the output is the 3D model area ID, interaction method, and probability value that the user may be interested in in the future;

[0014] S8. Configure the dynamic loading strategy;

[0015] S9. Optimize the loading strategy: Continuously iterate and improve the dynamic loading strategy according to the continuous accumulation of user behavior data and the in-depth analysis results.

[0016] Preferably, in step S1, the interaction behavior data includes, but is not limited to, the movement trajectory of the mouse on the 3D model, click events, drag events, stay time, and the data information of the corresponding 3D model.

[0017] Preferably, step S3 further includes:

[0018] Label the user data, such as the user interaction behavior type and the area ID where the user is located;

[0019] Extract features from the user data, such as the user's basic information, interaction behavior data, and historical behavior records as the feature matrix;

[0020] Encode the user's labels, use the 3D model area ID that the user may be interested in in the future as the label vector, format the data into a data form that the model can recognize, and divide the data into a training dataset and a test dataset.

[0021] Preferably, the specific implementation of the predict method in step S7 is: Accept the feature matrix of the test dataset, calculate the linear combination of the feature data, and convert the result of the linear combination through a logistic function to obtain the predicted probability value of each area ID.

[0022] Preferably, step S8 further includes:

[0023] Set a model loading threshold D. According to the real-time prediction result data of the machine learning model in step S6, when the probability of the user's attention area is greater than the threshold D, preload the 3D model area to reduce the user's waiting time. At the same time, when the overview of the user's attention area is less than the threshold D, load the 3D model area on demand to save system resources.

[0024] A configuration system for a dynamic loading strategy of large-scale 3D models in a digital twin visualization platform, which is applied to a configuration method for a dynamic loading strategy of large-scale 3D models in a digital twin visualization platform. The system includes:

[0025] A front-end data collection module, which is used to set data collection points in the front-end application of the digital twin visualization platform and collect the user's interaction behavior data;

[0026] A user data upload module, which is used to upload the user data collected by the front-end data collection module to the back-end server in real time;

[0027] A user data preprocessing module, which is used to perform data deduplication, data verification, missing value and outlier processing on the uploaded user data to ensure data validity;

[0028] A user behavior database construction module, which is used to design the database table structure according to the structure and characteristics of the user behavior data and import the preprocessed data into the database;

[0029] A machine learning model configuration module, which is used to select the logistic regression algorithm in the supervised learning method for prediction, construct and train a machine learning model, and use the trained model for prediction to output the 3D model area ID, interaction method and probability value that the user may focus on in the future;

[0030] A dynamic loading strategy configuration module, which is used to configure the dynamic loading strategy according to the prediction results of the machine learning model.

[0031] Preferably, the interaction behavior data collected by the front-end data collection module includes, but is not limited to, the mouse movement trajectory, click event, drag event, stay time on the 3D model, and the data information of the corresponding 3D model.

[0032] Preferably, the user data preprocessing module further includes:

[0033] A user data annotation unit, which is used to annotate the user data, including the user interaction behavior type and the area ID where it is located;

[0034] A feature extraction unit, which is used to extract features from the user data to form a feature matrix. The features include the user's basic information, interaction behavior data and historical behavior records;

[0035] A label encoding unit, which is used to encode labels for users, use the 3D model area IDs that users may be interested in in the future as label vectors, and format the data to meet the input requirements of machine learning models. At the same time, the data is divided into a training data set and a test data set.

[0036] Preferably, the prediction function in the machine learning model configuration module is specifically implemented as follows:

[0037] Create an instance of the LogisticRegression model using the scikit-learn library function of the logistic regression algorithm;

[0038] Train the model using the training data set;

[0039] Use the test data set to call the predict method of the model. This method accepts the feature matrix of the test data set, calculates the linear combination of the feature data, and obtains the predicted probability values of each area ID through a logistic function transformation.

[0040] Preferably, the dynamic loading strategy configuration module further includes:

[0041] Set the model loading threshold D;

[0042] According to the real-time prediction results of the machine learning model, when the probability of the user's interested area is greater than the threshold D, preload the 3D model area to reduce the user's waiting time;

[0043] When the probability of the user's interested area is less than or equal to the threshold D, load the 3D model area on demand to save system resources.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] The configuration method and system of the large-scale 3D model dynamic loading strategy in the digital twin visualization platform proposed by the present invention collect user behavior data through front-end data collection, and use machine learning to predict the user's interested areas, realizing the dynamic optimization of the loading strategy. This method effectively improves the loading efficiency of large-scale 3D models, reduces the user's waiting time, and at the same time, according to the different degrees of user attention, adopts a combination of preloading and on-demand loading, which not only ensures a smooth experience in the user's interested areas, but also saves system resources. Generally speaking, the present invention effectively optimizes the user interaction experience of the digital twin visualization platform, improves the overall performance and user satisfaction. Description of the Drawings

[0046] Figure 1 It is a flow chart of the method of the present invention. Detailed Embodiments

[0047] In order to clearly and completely describe the objectives, technical solutions of the present invention and make the advantages more clearly understood, the following further details the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are some, but not all, embodiments of the present invention, and are only used to explain the embodiments of the present invention, not to limit the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0048] Example 1, please refer to Figure 1 , the present invention provides a technical solution: a configuration method for a dynamic loading strategy of large-scale 3D models in a digital twin visualization platform, the method comprising the following steps:

[0049] S1. Front-end buried point data collection: Set buried points in the front-end application of the digital twin visualization platform to collect user interaction behavior data, where the interaction behavior data includes, but is not limited to, the movement trajectory of the mouse on the 3D model, click events, drag events, residence time, and data information of the corresponding 3D model;

[0050] S2. User data upload: Real-time upload the user data collected in step S1 to the back-end server for subsequent machine learning model training and analysis;

[0051] S3. User data preprocessing: Perform data deduplication, data verification, missing value and outlier processing on the user data received in S2 to ensure data validity; then, label the user data, such as the type of user interaction behavior and the area ID; then, extract features from the user data, such as user basic information, interaction behavior data, and historical behavior records, etc. as a feature matrix. Finally, perform label encoding on the user, use the area ID of the 3D model that the user may be interested in in the future as a label vector, format the data into a data form that the model can recognize, and divide the data into a training data set and a test data set;

[0052] S4. Build a user behavior database, design the database table structure according to the structure and characteristics of the user behavior data, including a user information table, a user behavior and area ID record table, etc., and import the preprocessed data into the database;

[0053] S5. Select a machine learning model. Since the supervised learning method will be used, algorithms such as linear regression, logistic regression, decision tree, or random forest can be selected for prediction. In this paper, the logistic regression algorithm is selected.

[0054] S6. Model construction and training: Use the scikit-learn library function of the logistic regression algorithm and introduce the LogisticRegression model class to create a model instance; use the training dataset (feature matrix and label vector) in S3 to call the fit method of the model to train the model;

[0055] S7. Machine learning prediction: Use the test dataset (feature matrix) in S3 to call the predict method of the model, return the prediction results, and the output is the 3D model area ID, interaction method, and probability value that the user may be interested in in the future (such as the probability value that the user clicks on a certain ID area). Specifically, the predict method is implemented as follows: Accept the feature matrix of the test dataset, calculate the linear combination of the feature data (such as Formula 1), and convert the result of the linear combination through a logistic function (sigmoid function) to obtain the predicted probability value of each area ID (such as Formula 2). Formula 1 is as follows:

[0056] Z = X_ new *W + b

[0057] Where W is the weight vector and b is the bias term.

[0058] Formula 2 is as follows:

[0059]

[0060] Where x is the input of the function and σ(x) is the output of the function.

[0061] S8. Dynamic loading strategy configuration: Set the model loading threshold D; according to the real-time prediction result data of the machine learning model in step S6, when the probability of the user's interested area is greater than the threshold D, pre-load the 3D model area to reduce the user's waiting time, and at the same time, when the overview of the user's interested area probability is less than the threshold D, load the 3D model area on demand to save system resources;

[0062] S9. Loading strategy optimization: Continuously iterate and improve the dynamic loading strategy according to the continuous accumulation of user behavior data and the in-depth analysis results.

[0063] Embodiment 2, based on Embodiment 1, proposes: A configuration system for a large-scale 3D model dynamic loading strategy in a digital twin visualization platform, which is applied to a configuration method for a large-scale 3D model dynamic loading strategy in a digital twin visualization platform. The system includes:

[0064] The front-end data collection module is used to set up data points in the front-end application of the digital twin visualization platform to collect users' interaction behavior data. The interaction behavior data collected by the front-end data collection module includes, but is not limited to, the movement trajectory of the mouse on the 3D model, click events, drag events, dwell time, and the data information of the corresponding 3D model.

[0065] The user data upload module is used to upload the user data collected by the front-end data collection module to the back-end server in real time. It also includes:

[0066] The user data annotation unit is used to annotate user data, including the type of user interaction behavior and the area ID where it occurs.

[0067] The feature extraction unit is used to extract features from user data to form a feature matrix. The features include user basic information, interaction behavior data, and historical behavior records.

[0068] The label encoding unit is used to encode user labels, use the area ID of the 3D model that the user may be interested in in the future as the label vector, and format the data to make the data meet the input requirements of the machine learning model. At the same time, the data is divided into a training data set and a test data set.

[0069] The user data preprocessing module is used to perform data deduplication, data verification, missing value and outlier processing on the uploaded user data to ensure data validity.

[0070] The user behavior database construction module is used to design the database table structure according to the structure and characteristics of user behavior data, and import the preprocessed data into the database.

[0071] The machine learning model configuration module is used to select the logistic regression algorithm in the supervised learning method for prediction, construct and train a machine learning model, and use the trained model for prediction to output the area ID, interaction method, and probability value of the 3D model that the user may be interested in in the future. The specific implementation of the prediction function is as follows:

[0072] Create an instance of the LogisticRegression model using the scikit-learn library function of the logistic regression algorithm.

[0073] Train the model using the training data set.

[0074] Call the predict method of the model using the test data set. This method accepts the feature matrix of the test data set, calculates the linear combination of the feature data, and obtains the predicted probability value of each area ID through the logistic function transformation.

[0075] The dynamic loading policy configuration module is used to configure the dynamic loading policy according to the prediction results of the machine learning model. It also includes:

[0076] Set the model loading threshold D;

[0077] According to the real-time prediction result of the machine learning model, when the probability of the user's attention area is greater than the threshold D, preload the 3D model area to reduce the user's waiting time;

[0078] When the probability of the user's attention area is less than or equal to the threshold D, load the 3D model area on demand to save system resources.

[0079] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A configuration method for a large-scale 3D model dynamic loading strategy in a digital twin visualization platform, characterized in that: The method comprises the following steps: S1. Front-end tracking data collection: Set tracking points in the front-end application of the digital twin visualization platform to collect user interaction behavior data; S2, user data upload: upload the user data collected in step S1 to the backend server in real time for subsequent machine learning model training and analysis; S3, user data preprocessing: perform data deduplication, data verification, missing value and outlier processing on the user data received from S2 to ensure data validity; S4. Build a user behavior database. According to the structure and characteristics of user behavior data, design the database table structure, including user information table, user behavior and area ID record table, and import the pre-processed data into the database; S5. Select the machine learning model, which will use the supervised learning method and select the logistic regression algorithm for prediction; S6, model construction and training: Use the scikit-learn library function of the logistic regression algorithm and introduce the LogisticRegression model class to create a model instance; use the training data set in S3 to call the model's fit method to train the model; S7, machine learning prediction: Use the test data set in S3 to call the model's predict method and return the prediction result. The output is the 3D model area ID, interaction method, and probability value that the user may pay attention to in the future. S8, dynamic loading strategy configuration; S9. Loading strategy optimization: Based on the continuous accumulation of user behavior data and in-depth analysis results, the dynamic loading strategy is continuously iterated and improved.

2. According to claim 1, a configuration method for a large-scale 3D model dynamic loading strategy in a digital twin visualization platform is characterized in that: In step S1, the interactive behavior data includes but is not limited to the movement track of the mouse on the 3D model, click events, drag events, dwell time and data information of the corresponding 3D model.

3. The method for configuring a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 1, characterized in that: Step S3 also includes: Label user data, such as user interaction behavior type and region ID; Extract features from user data, such as user basic information, interaction behavior data, and historical behavior records as feature matrices; The user labels are encoded, the 3D model area IDs that the user may pay attention to in the future are used as label vectors, the data is formatted into a data form that the model can recognize, and the data is divided into training data sets and test data sets.

4. The method for configuring a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 1, characterized in that: The predict method in step S7 is specifically implemented as follows: accepting the feature matrix of the test data set, calculating the linear combination of the feature data, converting the result of the linear combination through a logical function, and obtaining the predicted probability value of each region ID.

5. The method for configuring a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 1, characterized in that: Step S8 also includes: A model loading threshold D is set. According to the real-time prediction result data of the machine learning model in step S6, when the probability of the user's attention area is greater than the threshold D, the 3D model area is preloaded to reduce the user's waiting time. At the same time, when the user's attention area overview is less than the threshold D, the 3D model area is loaded on demand to save system resources.

6. A configuration system for a large-scale 3D model dynamic loading strategy in a digital twin visualization platform, applied to a configuration method for a large-scale 3D model dynamic loading strategy in a digital twin visualization platform as described in any one of claims 1 to 5, characterized in that: The system comprises: The front-end data collection module is used to set up tracking points in the front-end application of the digital twin visualization platform to collect user interaction behavior data; The user data upload module is used to upload the user data collected by the front-end data collection module to the back-end server in real time; User data preprocessing module, used to perform data deduplication, data verification, missing value and outlier processing on uploaded user data to ensure data validity; User behavior database construction module, used to design the database table structure according to the structure and characteristics of user behavior data, and import the pre-processed data into the database; The machine learning model configuration module is used to select the logistic regression algorithm in the supervised learning method for prediction, build and train the machine learning model, and use the trained model for prediction, outputting the 3D model area ID, interaction method and probability value that the user may pay attention to in the future; The dynamic loading strategy configuration module is used to configure the dynamic loading strategy according to the prediction results of the machine learning model.

7. The configuration system of a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 6 is characterized by: The interactive behavior data collected by the front-end data collection module includes but is not limited to the movement trajectory of the mouse on the 3D model, click events, drag events, dwell time, and data information of the corresponding 3D model.

8. The configuration system of a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 6, characterized in that: The user data preprocessing module also includes: A user data labeling unit, used to label user data, including the user interaction behavior type and the area ID; A feature extraction unit is used to extract features from user data to form a feature matrix, where the features include basic user information, interactive behavior data, and historical behavior records; The label encoding unit is used to encode user labels, use the 3D model area ID that the user may pay attention to in the future as the label vector, and format the data to make the data meet the input requirements of the machine learning model, and divide the data into training data sets and test data sets.

9. The configuration system of a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 6, characterized in that: The prediction function in the machine learning model configuration module is specifically implemented as follows: Use the scikit-learn library function of the logistic regression algorithm to create a LogisticRegression model instance; Train the model using the training dataset; The predict method of the model is called using the test data set. This method accepts the feature matrix of the test data set, calculates the linear combination of the feature data, and obtains the predicted probability value of each region ID through the logical function conversion.

10. The configuration system of a large-scale 3D model dynamic loading strategy in a digital twin visualization platform according to claim 6, characterized in that: The dynamic loading strategy configuration module also includes: Set the model loading threshold D; According to the real-time prediction results of the machine learning model, when the probability of the user's attention area is greater than the threshold D, the 3D model area is preloaded to reduce the user's waiting time; When the probability of the user paying attention to the area is less than or equal to the threshold D, the 3D model area is loaded on demand to save system resources.

Images