Multi-scene simulation method and system based on digital twin technology

Abstract

The invention discloses a multi-scene simulation method and system based on a digital twinning technology, and belongs to the field of digital twinning technology.The multi-scene simulation system based on the digital twinning technology comprises a physical entity sensing module, a data processing and managing module, a digital twinning model building module and a multi-scene simulation module, the physical entity sensing module is responsible for collecting and transmitting various data from real physical entities through various sensors, and the data processing and managing module is responsible for processing and cleaning the data transmitted by the physical entity sensing module. According to the method, a new simulation scene is quickly established through the cooperation among the sub-modules in the digital twinborn model construction module, the accuracy and applicability of the model are continuously improved through real-time data updating and model optimization, and the modeling period is greatly shortened.

Description

Technical Field

[0001] The present invention belongs to the field of digital twin technology, and specifically relates to a multi-scenario simulation method and system based on digital twin technology. Background Art

[0002] The multi-scenario simulation system is a platform that combines software and hardware. It digitally models and simulates a variety of scenarios in the real world, such as industrial production processes, traffic conditions, military combat environments, and medical surgical processes. Through preset rules, parameters, and algorithms, the system can generate highly realistic scenario environments, and dynamically adjust and display the evolution of the scenarios based on user input or preset change conditions. Multi-scenario simulation systems are often used for production line design and optimization. By simulating different production process scenarios, companies can discover potential bottleneck problems in advance. For example, automobile manufacturers simulate the scenarios of automobile assembly production lines, set different production rhythms, equipment layouts, and staffing parameters, and observe the operating efficiency and product quality of the production line, so as to determine the optimal production plan, improve production efficiency, and reduce costs.

[0003] Existing multi-scenario simulation systems are usually built based on theoretical models and limited data. The objects and environmental characteristics in the simulated scenes may deviate from the actual situation, resulting in reduced accuracy and reliability of the simulation results. At the same time, the existing systems are not detailed enough and cannot accurately simulate some key physical phenomena and complex system behaviors, affecting the true reproduction and in-depth understanding of the actual scenes. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a multi-scenario simulation method and system based on digital twin technology.

[0005] The technical solution adopted to solve the above technical problems is: a multi-scenario simulation system based on digital twin technology, including four subsystems: physical entity perception module, data processing and management module, digital twin model construction module and multi-scenario simulation module. The physical entity perception module is responsible for collecting and transmitting various types of data from real physical entities through multiple sensors. The data processing and management module is responsible for processing and cleaning the data transmitted by the physical entity perception module. The digital twin model construction module is responsible for constructing the corresponding three-dimensional geometric model according to the geometric characteristics of the physical entity. The multi-scenario simulation module is responsible for providing a visual interface for users to configure different simulation scenarios.

[0006] The physical entity perception module includes a temperature sensor, a pressure sensor, a displacement sensor, a camera, and a lidar. The physical entity perception module transmits the collected data to the data processing center through wired and wireless communication methods. For example, in a factory environment, the temperature sensor will collect the temperature of the equipment during operation in real time, and the camera will capture the image information of each link in the production process, providing a rich and accurate source of raw data for subsequent construction of the digital twin model.

[0007] To improve the success rate of information recognition, the data processing and management module preprocesses the data information sent by the physical entity perception module to remove noise data and irrelevant information. The specific formula is:

[0008] Let the original data information sent by the physical entity perception module be O = {o1, o2,..., o n}, where o j represents a character in the data information. Let the set of noise characters be N = {n1, n2,..., n m}, and the cleaned data information I clean is obtained through the following method:

[0009]

[0010] Principal Component Analysis (PCA) feature extraction:

[0011] Let the feature matrix of the module data information be X, with dimensions n×p, where n is the number of samples (i.e., the number of device identification information), and p is the number of original features:

[0012] First, calculate the covariance matrix:

[0013]

[0014] Then, solve for the eigenvalues λ1, λ2,..., λ p and the corresponding eigenvectors υ1, υ2,..., υ p , satisfying ∑υ i = λ i υ i , i = 1, 2,..., p

[0015] Sort the eigenvalues in descending order and select the first k eigenvectors (k < p) to form the projection matrix W = {υ1, υ2,..., υ k}

[0016] The extracted principal component feature matrix Z is: Z = XW, with dimensions n×k. These principal components can more effectively represent the features of the module for subsequent recognition;

[0017] Then, a feature extraction algorithm is used to mine more representative device features. The accuracy of device classification is improved by using a machine learning algorithm. The support vector machine (SVM) algorithm is used to distinguish different types of devices by constructing a hyperplane. In the training phase, the data information of known module types is used as training data to let the SVM learn the boundaries between different types of devices. For newly connected devices, the SVM determines the type of device it belongs to based on the position of its identification information in the feature space. The specific formula is:

[0018] Suppose the training data set is {(x1, y1), (x2, y2), …, (x n ,y n )}, where x1 is the feature vector of the device identification information after PCA extraction, y i ∈{-1,1} indicates the category to which the device belongs;

[0019] The goal of SVM is to find a hyperplane w T x+b=0, so that the interval between the two types of data points is maximized. The optimization problem is expressed as:

[0020]

[0021] By solving the Lagrange duality, we can get the Lagrange multiplier α, and then get the parameters of the hyperplane and b = y j -w T x j (For the support vector x j );

[0022] For new module data information x new , its category prediction y new for:

[0023] y new =sign(W T x new +b)

[0024] Improve the recognition success rate through adaptive parameter adjustment. The specific formula is as follows:

[0025] Assume that the classification threshold of module data recognition is T, the initial recognition success rate is R0, and when the recognition success rate drops below the lower limit R min When , adjust the threshold;

[0026] Assume that there is a functional relationship R=f(T) between the recognition success rate R and the threshold T (this functional relationship is obtained through experiments or historical data fitting);

[0027] When R <R min When , the threshold T is updated according to the derivative f'(T) of f(T)min =T+ΔT, where the size of ΔT is determined by f'(T) and the expected improvement in success rate:

[0028]

[0029] In this way, the threshold is adaptively adjusted to improve the success rate of data recognition processing.

[0030] The data processing and management module includes a data storage and management submodule, which stores the processed data in a database according to certain rules and structures. The database type is selected as a MySQL database to store structured data according to requirements.

[0031] The digital twin model construction module includes a geometric modeling submodule, a physical modeling submodule and a behavioral modeling submodule. The geometric modeling submodule usually adopts an algorithm based on a convolutional neural network.

[0032] Spatial positioning and basic shape setting: construct a 3D geometric body for each node, generate spatial positioning through node data, and set the basic shape and size of the geometric body;

[0033] Geometry connection: connect the geometry between nodes to generate geometric structures. According to the formula,

[0034]

[0035] Calculate the connection information of 3D geometry, where y ij is an element of the output feature map, x i+u,j+v is the element of the input feature map, w uv is the weight of the convolution kernel, b k is the bias term, d i+u,j+v is the distance compensation from the node to the center, α is the weighted coefficient of distance compensation, and λ is the regularization coefficient;

[0036] Parameter correction and model integration: adjust the physical parameters and functional parameters in the model one by one, make adjustments through parameter simulation and verification, verify the consistency of multiple parameters with actual applications, conduct model verification tests, generate parameter optimization structures, integrate multiple levels and internal structures of the model, perform inter-layer connections and structural reinforcement, build an integrated model through level integration, conduct model verification, and generate a model structure.

[0037] The physical modeling submodule adopts finite element analysis method.

[0038] Discretization process: Divide the physical entity into a finite number of units and nodes, assuming the total number of units is n e , the total number of nodes is n n ;

[0039] Unit characteristic analysis: For each unit, according to its geometry, material properties and the type of physical problem, the unit stiffness matrix [K e ], mass matrix [M e ]’s feature matrix;

[0040] Assemble the overall matrix: Assemble the characteristic matrices of each unit into [K] overall stiffness matrix and overall mass matrix [M] according to the connection relationship of the nodes. The assembly process is expressed as,

[0041] Apply boundary conditions and loads: According to the actual situation of the physical problem, boundary conditions are applied to the boundary nodes of the model, and external loads are applied to the corresponding nodes. Let the external load vector be {F}

[0042] Solve the system of equations: For dynamic problems, we get the equations of motion

[0043]

[0044] in, and {u} are the node acceleration vector, velocity vector and displacement vector respectively, [C] is the damping matrix,

[0045] For statics problems, the equation simplifies to

[0046] [K]{u}={F}

[0047] The equations are solved by direct and iterative methods to obtain the displacement field of the node, and then the stress field and strain field are further calculated based on the displacement field.

[0048] The behavior modeling submodule is based on the behavior modeling of the hidden Markov model. The hidden Markov model introduces hidden states on the basis of the Markov chain. The observed behavior is generated by the hidden states, and the hidden states satisfy the Markov property.

[0049] Algorithm construction process: Hidden state space S = {s1, s2, ..., s n}, observation state space O = {o1, o2, ..., o m};

[0050] Determine the initial probability distribution: π = (π1, π2, ..., π n ), represents the probability of being in each hidden state at the initial moment;

[0051] Determine the transition probability matrix: A = (a ij ), where a ij Represents the hidden state s i Transfer to hidden state sj The probability of

[0052] Determine the transmission probability matrix: B = (b jk ), where b jk Indicates that in the hidden state s j Observe the observation state o k Probability

[0053] Model training: Use the Baum-Welch algorithm to train the model and estimate the parameters of the model π, A and B through a large amount of observation data;

[0054] Behavior prediction and analysis: Given an observation sequence, the Viterbi algorithm is used to decode the most likely hidden state sequence to analyze and predict the behavior.

[0055] The multi-scenario simulation module includes a scenario configuration submodule: providing a visual interface for users to configure different simulation scenarios, allowing users to select the physical entity range to be simulated, set the simulation time span, and select the external environmental factors involved in the simulation. It supports the rapid call of existing mature scenario configurations from the preset scenario template library, and also allows users to customize and create new scenarios according to their own needs, and save and share customized scenario configurations;

[0056] Simulation execution submodule: According to the conditions set by the scenario configuration submodule, the digital twin model is started for simulation operation. During the operation, the relevant data in the data processing and management module is called in real time as the input of the model to drive the digital twin model to dynamically evolve according to the set physical laws and behavior patterns, and can simulate the actual operation process and state changes of the physical entity in the corresponding scenario;

[0057] Result analysis and visualization submodule: Analyze the simulation results, compare the simulation results in different scenarios, analyze the changes in key indicators, dig out valuable information, and present the analysis results to users in an intuitive and visual way;

[0058] The specific steps include:

[0059] Step 1: Deploy various sensors for the simulation objects to collect real-time operating data of temperature, pressure and speed. Use laser scanning and photogrammetry to obtain its geometric shape and spatial position information, collect environmental data of the simulation scene, mine and integrate relevant historical data, clean the collected data, remove noise and outliers, fill in missing values ​​through data interpolation and smoothing methods, and then standardize and normalize data of different formats and sources, and store them in the database;

[0060] Step 2: Based on the collected geometry and position information, accurately construct a 3D model of the physical entity. Based on physical principles and mathematical formulas, establish a physical model that describes the behavior of the physical entity. Use machine learning and deep learning algorithms to analyze historical data and real-time data, explore the behavior patterns and laws of the physical entity, and build a behavior prediction model. Organically integrate the geometric model, physical model, and behavior model to form a complete digital twin model.

[0061] Step 3: According to the simulation purpose and needs, different scene categories are divided, specific parameters are set for each scene, the initial state of each scene is clarified, the configured scene parameters and initial conditions are input into the digital twin model, and simulation calculations are performed using a high-performance computing platform. During the simulation process, the model state is updated in real time to simulate the dynamic interaction between the physical entity and the environment.

[0062] The beneficial effects of the present invention are as follows: the present invention quickly builds new simulation scenarios through the cooperation between the various sub-modules in the digital twin model construction module, and continuously improves the accuracy and applicability of the model through real-time data updates and model optimization, greatly shortening the modeling cycle, and obtaining a large amount of data on physical entities in real time through the Internet of Things and sensors, and can construct a virtual model that is highly consistent with the real world, accurately restoring the appearance, structure, performance and behavior of the physical entities, making the simulation scene more realistic. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic diagram of the module of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0065] like Figure 1 As shown, a multi-scenario simulation system based on digital twin technology in this embodiment includes four subsystems: a physical entity perception module, a data processing and management module, a digital twin model construction module and a multi-scenario simulation module. The physical entity perception module is responsible for collecting and transmitting various types of data from real physical entities through multiple sensors. The data processing and management module is responsible for processing and cleaning the data transmitted by the physical entity perception module. The digital twin model construction module is responsible for constructing the corresponding three-dimensional geometric model according to the geometric features of the physical entity. The multi-scenario simulation module is responsible for providing a visual interface for users to configure different simulation scenarios.

[0066] The physical entity perception module includes temperature sensors, pressure sensors, displacement sensors, cameras, and lidar. The physical entity perception module transmits the collected data to the data processing center through wired and wireless communication methods. For example, in a factory environment, the temperature sensor will collect the temperature of the equipment during operation in real time, and the camera will capture the image information of each link in the production process, providing a rich and accurate source of raw data for the subsequent construction of the digital twin model. For example, when a fire occurs, the temperature sensors, smoke sensors, pressure sensors, and cameras distributed inside and outside the building collect fire-related data in real time, including environmental parameters, the status of fire-fighting facilities, the location of personnel, etc. Through the Internet of Things communication technology, the data collected by the sensors is transmitted to the server or edge computing node of the system.

[0067] To improve the success rate of information recognition, the data processing and management module preprocesses the data information sent by the physical entity perception module to remove noise data and irrelevant information. The specific formula is:

[0068] Let the original data information sent by the physical entity perception module be O = {o1, o2,..., o n}, where o j represents a character in the data information. Let the noise character set be N = {n1, n2,..., n m}, and the cleaned data information I clean is obtained through the following method:

[0069]

[0070] Principal Component Analysis (PCA) feature extraction:

[0071] Let the feature matrix of the module data information be X, and its dimension is n×p, where n is the number of samples (i.e., the number of device identification information), and p is the number of original features:

[0072] First, calculate the covariance matrix:

[0073]

[0074] Then, solve the eigenvalues λ1, λ2,..., λ p and the corresponding eigenvectors υ1, υ2,..., υ p , satisfying ∑υ i = λ i υ i , i = 1, 2,..., p

[0075] Sort according to the eigenvalue size, and select the first k eigenvectors (k < p) to form the projection matrix W = {υ1, υ2,..., υ k}

[0076] The extracted principal component feature matrix Z is: Z = XW, with a dimension of n × k. These principal components can more effectively represent the characteristics of the module for subsequent identification;

[0077] Then, a feature extraction algorithm is used to mine more representative device features. The accuracy of device classification is improved by using a machine learning algorithm. The support vector machine (SVM) algorithm is used to distinguish different types of devices by constructing a hyperplane. In the training phase, the data information of known module types is used as training data to let the SVM learn the boundaries between different types of devices. For newly connected devices, the SVM determines the type of device it belongs to based on the position of its identification information in the feature space. The specific formula is:

[0078] Suppose the training data set is {(x1, y1), (x2, y2), …, (x n ,y n )}, where x1 is the feature vector of the device identification information after PCA extraction, y i ∈{-1,1} indicates the category to which the device belongs;

[0079] The goal of SVM is to find a hyperplane w T x+b=0, so that the interval between the two types of data points is maximized. The optimization problem is expressed as:

[0080]

[0081] By solving the Lagrange duality, we can get the Lagrange multiplier α, and then get the parameters of the hyperplane and b = y j -w T x j (For the support vector x j );

[0082] For new module data information x new , its category prediction y new for:

[0083] y new =sign(w T x new +b)

[0084] Improve the recognition success rate through adaptive parameter adjustment. The specific formula is as follows:

[0085] Assume that the classification threshold of module data recognition is T, the initial recognition success rate is R0, and when the recognition success rate drops below the lower limit R min When , adjust the threshold;

[0086] Assume that there is a functional relationship R=f(T) between the recognition success rate R and the threshold T (this functional relationship is obtained through experiments or historical data fitting);

[0087] When R <R min When , the threshold T is updated according to the derivative f'(T) of f(T) min =T+ΔT, where the size of ΔT is determined by f'(T) and the expected improvement in success rate:

[0088]

[0089] In this way, the threshold is adaptively adjusted to improve the success rate of data recognition processing.

[0090] The data processing and management module includes a data storage and management submodule. The data storage and management submodule stores the processed data in the database according to certain rules and structures. The database type selects MySQL database to store structured data according to needs.

[0091] The digital twin model building module includes a geometric modeling submodule, a physical modeling submodule, and a behavioral modeling submodule. The geometric modeling submodule usually adopts an algorithm based on a convolutional neural network.

[0092] Spatial positioning and basic shape setting: construct a 3D geometric body for each node, generate spatial positioning through node data, and set the basic shape and size of the geometric body;

[0093] Geometry connection: connect the geometry between nodes to generate geometric structures. According to the formula,

[0094]

[0095] Calculate the connection information of 3D geometry, where y ij is an element of the output feature map, x i+u,j+v is the element of the input feature map, w uv is the weight of the convolution kernel, b k is the bias term, d i+u,j+v is the distance compensation from the node to the center, α is the weighted coefficient of distance compensation, and λ is the regularization coefficient;

[0096] Parameter correction and model integration: adjust the physical parameters and functional parameters in the model one by one, make adjustments through parameter simulation and verification, verify the consistency of multiple parameters with actual applications, conduct model verification tests, generate parameter optimization structures, integrate multiple levels and internal structures of the model, perform inter-layer connections and structural reinforcement, build an integrated model through level integration, conduct model verification, and generate a model structure.

[0097] The physical modeling submodule uses finite element analysis method.

[0098] Discretization process: Divide the physical entity into a finite number of units and nodes, assuming the total number of units is n e , the total number of nodes is n n ;

[0099] Unit characteristic analysis: For each unit, according to its geometry, material properties and the type of physical problem, the unit stiffness matrix [K e ], mass matrix [M e ]’s feature matrix;

[0100] Assemble the overall matrix: Assemble the characteristic matrices of each unit into [K] overall stiffness matrix and overall mass matrix [M] according to the connection relationship of the nodes. The assembly process is expressed as,

[0101] Apply boundary conditions and loads: According to the actual situation of the physical problem, boundary conditions are applied to the boundary nodes of the model, and external loads are applied to the corresponding nodes. Let the external load vector be {F}

[0102] Solve the system of equations: For dynamic problems, we get the equations of motion

[0103]

[0104] in, and {u} are the node acceleration vector, velocity vector and displacement vector respectively, [C] is the damping matrix,

[0105] For statics problems, the equation simplifies to

[0106] [K]{u}={F}

[0107] The equations are solved by direct and iterative methods to obtain the displacement field of the node, and then the stress field and strain field are further calculated based on the displacement field.

[0108] The behavior modeling submodule is based on the behavior modeling of the hidden Markov model. The hidden Markov model introduces hidden states on the basis of the Markov chain. The observed behavior is generated by the hidden states, and the hidden states satisfy the Markov property.

[0109] Algorithm construction process: Hidden state space S = {s1, s2, ..., s n}, observation state space O = {o1, o2, ..., o m};

[0110] Determine the initial probability distribution: π = (π1, π2, ..., π n), represents the probability of being in each hidden state at the initial moment;

[0111] Determine the transition probability matrix: A = (a ij ), where a ij Represents the hidden state s i Transfer to hidden state s j The probability of

[0112] Determine the transmission probability matrix: B = (b jk ), where b jk Indicates that in the hidden state s j Observe the observation state o k Probability

[0113] Model training: Use the Baum-Welch algorithm to train the model and estimate the parameters of the model π, A and B through a large amount of observation data;

[0114] Behavior prediction and analysis: Given an observation sequence, use the Viterbi algorithm to decode the most likely hidden state sequence to analyze and predict the behavior;

[0115] In case of fire, algorithms can be used to establish mathematical models of physical processes such as fire spread, smoke diffusion, and personnel evacuation. When fire-fighting facilities are maintained or modified, the relevant information in the model can be updated in a timely manner to ensure the consistency of the model with the actual situation.

[0116] The multi-scenario simulation module includes a scenario configuration submodule: it provides a visual interface for users to configure different simulation scenarios. Users select the physical entity range to be simulated, set the simulation time span, and select the external environmental factors involved in the simulation. It supports the rapid call of existing mature scenario configurations from the preset scenario template library, and also allows users to create new scenarios according to their own needs, and save and share customized scenario configurations.

[0117] Simulation execution submodule: According to the conditions set by the scenario configuration submodule, the digital twin model is started for simulation operation. During the operation, the relevant data in the data processing and management module is called in real time as the input of the model to drive the digital twin model to dynamically evolve according to the set physical laws and behavior patterns, and can simulate the actual operation process and state changes of the physical entity in the corresponding scenario;

[0118] Result analysis and visualization submodule: Analyze the results of the simulation, compare the simulation results in different scenarios, analyze the changes in key indicators, dig out valuable information, and present the analysis results to users in an intuitive and visual way;

[0119] According to the actual fire situation or preset scenario, the starting position of the fire, type of fire source, size of the fire and other parameters are set. The multi-scenario simulation module starts the fire simulation program. Based on the physical model, it simulates the spread of the fire, the diffusion path and concentration distribution of the smoke, and calculates parameters such as temperature changes and oxygen content in different areas. Based on the fire simulation results, it simulates the startup and operation process of fire-fighting facilities, such as the spraying of water by the automatic sprinkler fire-fighting system, the forced landing of the fire elevator, and the closing of the fire shutter door.

[0120] The specific steps include:

[0121] Step 1: Deploy various sensors for the simulation objects to collect real-time operating data of temperature, pressure and speed. Use laser scanning and photogrammetry to obtain its geometric shape and spatial position information, collect environmental data of the simulation scene, mine and integrate relevant historical data, clean the collected data, remove noise and outliers, fill in missing values ​​through data interpolation and smoothing methods, and then standardize and normalize data of different formats and sources, and store them in the database;

[0122] Step 2: Based on the collected geometry and position information, accurately construct a 3D model of the physical entity. Based on physical principles and mathematical formulas, establish a physical model that describes the behavior of the physical entity. Use machine learning and deep learning algorithms to analyze historical data and real-time data, explore the behavior patterns and laws of the physical entity, and build a behavior prediction model. Organically integrate the geometric model, physical model, and behavior model to form a complete digital twin model.

[0123] Step 3: According to the simulation purpose and needs, different scene categories are divided, specific parameters are set for each scene, the initial state of each scene is clarified, the configured scene parameters and initial conditions are input into the digital twin model, and simulation calculations are performed using a high-performance computing platform. During the simulation process, the model state is updated in real time to simulate the dynamic interaction between the physical entity and the environment.

[0124] The above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention.

Claims

1. A multi-scenario simulation system based on digital twin technology, characterized by: It includes four major subsystems: physical entity perception module, data processing and management module, digital twin model construction module and multi-scenario simulation module. The physical entity perception module is responsible for collecting and transmitting various types of data from real physical entities through various sensors. The data processing and management module is responsible for processing and cleaning the data transmitted by the physical entity perception module. The digital twin model construction module is responsible for constructing the corresponding three-dimensional geometric model based on the geometric characteristics of the physical entity. The multi-scenario simulation module is responsible for providing a visual interface for users to configure different simulation scenarios.

2. According to claim 1, a multi-scenario simulation system based on digital twin technology is characterized in that: The physical entity perception module includes a temperature sensor, a pressure sensor, a displacement sensor, a camera and a laser radar. The physical entity perception module transmits the collected data to a data processing center through wired and wireless communication methods.

3. According to claim 1, a multi-scenario simulation system based on digital twin technology is characterized in that: In order to improve the success rate of information recognition, the data processing and management module pre-processes the data information sent by the physical entity perception module to remove noise data and irrelevant information. The specific formula is: Assume that the original data information sent by the physical entity perception module is O = {o1, o2, ..., o n }, where o j Represents a character in the data information, and the noise character set is N = {n1, n2, ..., n m }, cleaned data information I clean Obtained by: Principal Component Analysis (PCA) Feature Extraction: Suppose the feature matrix of the module data information is X, whose dimension is n×p, where n is the number of samples (i.e. the number of device identification information) and p is the number of original features: First calculate the covariance matrix: Then solve the eigenvalues ​​λ1, λ2, …, λ of the covariance matrix ∑ p and the corresponding eigenvectors υ1,υ2,…,υ p , satisfy,∑υ i =λ i υ i , i=1,2,…,p; Sorted by the eigenvalue magnitude, select the first k eigenvectors (k < p) to form the projection matrix W = {υ1, υ2, …, υ k}; The extracted principal component feature matrix Z is: Z = XW, with a dimension of n × k. These principal components can more effectively represent the characteristics of the module for subsequent identification; Then, a feature extraction algorithm is used to mine more representative device features. The accuracy of device classification is improved by using a machine learning algorithm. The support vector machine (SVM) algorithm is used to distinguish different types of devices by constructing a hyperplane. In the training phase, the data information of known module types is used as training data to let the SVM learn the boundaries between different types of devices. For newly connected devices, the SVM determines the type of device it belongs to based on the position of its identification information in the feature space. The specific formula is: Suppose the training data set is {(x1, y1), (x2, y2), …, (x n ,y n )}, where x1 is the feature vector of the device identification information after PCA extraction, y i ∈{-1,1} indicates the category to which the device belongs; The goal of SVM is to find a hyperplane w T x+b=0, so that the interval between the two types of data points is maximized. The optimization problem is expressed as: By solving the Lagrange duality, we can get the Lagrange multiplier α, and then get the parameters of the hyperplane (For the support vector x j ); For new module data information x new , its category prediction y new for: y new =sign(w T x new +b) Improve the recognition success rate through adaptive parameter adjustment. The specific formula is as follows: Assume that the classification threshold of module data recognition is T, the initial recognition success rate is R0, and when the recognition success rate drops below the lower limit R min When , adjust the threshold; Assume that there is a functional relationship R=f(T) between the recognition success rate R and the threshold T (this functional relationship is obtained through experiments or historical data fitting); When R <R min When , the threshold T is updated according to the derivative f'(T) of f(T) min =T+ΔT, where the size of ΔT is determined by f'(T) and the expected improvement in success rate: In this way, the threshold is adaptively adjusted to improve the success rate of data recognition processing.

4. According to claim 3, a multi-scenario simulation system based on digital twin technology is characterized in that: The data processing and management module includes a data storage and management submodule, which stores the processed data in a database according to certain rules and structures. The database type is selected as a MySQL database to store structured data according to requirements.

5. According to claim 1, a multi-scenario simulation system based on digital twin technology is characterized in that: The digital twin model construction module includes a geometric modeling submodule, a physical modeling submodule and a behavioral modeling submodule. The geometric modeling submodule usually adopts an algorithm based on a convolutional neural network. Spatial positioning and basic shape setting: construct a 3D geometric body for each node, generate spatial positioning through node data, and set the basic shape and size of the geometric body; Geometry connection: connect the geometry between nodes to generate geometric structures. According to the formula, Calculate the connection information of 3D geometry, where y ij is an element of the output feature map, x i+u,j+v is the element of the input feature map, w uv is the weight of the convolution kernel, b k is the bias term, d i+u,j+v is the distance compensation from the node to the center, α is the weighted coefficient of distance compensation, and λ is the regularization coefficient; Parameter correction and model integration: adjust the physical parameters and functional parameters in the model one by one, make adjustments through parameter simulation and verification, verify the consistency of multiple parameters with actual applications, conduct model verification tests, generate parameter optimization structures, integrate multiple levels and internal structures of the model, perform inter-layer connections and structural reinforcement, build an integrated model through level integration, conduct model verification, and generate a model structure.

6. According to claim 5, a multi-scenario simulation system based on digital twin technology is characterized in that: The physical modeling submodule adopts finite element analysis method. Discretization process: Divide the physical entity into a finite number of units and nodes, assuming the total number of units is n e , the total number of nodes is n n ; Unit characteristic analysis: For each unit, according to its geometry, material properties and the type of physical problem, the unit stiffness matrix [K e ], mass matrix [M e ]’s feature matrix; Assemble the overall matrix: Assemble the characteristic matrices of each unit into [K] overall stiffness matrix and overall mass matrix [M] according to the connection relationship of the nodes. The assembly process is expressed as, Apply boundary conditions and loads: According to the actual situation of the physical problem, boundary conditions are applied to the boundary nodes of the model, and external loads are applied to the corresponding nodes. Let the external load vector be {F}; Solve the system of equations: For dynamic problems, we get the equations of motion, Among them, {ü}, and {u} are the node acceleration vector, velocity vector and displacement vector respectively, [C] is the damping matrix, For statics problems, the equation simplifies to, [K]{u}={F} The equations are solved by direct and iterative methods to obtain the displacement field of the node, and then the stress field and strain field are further calculated based on the displacement field.

7. The multi-scenario simulation system based on digital twin technology according to claim 5 is characterized in that: The behavior modeling submodule is based on the behavior modeling of the hidden Markov model. The hidden Markov model introduces hidden states on the basis of the Markov chain. The observed behavior is generated by the hidden states, and the hidden states satisfy the Markov property. Algorithm construction process: Hidden state space S = {s1, s2, ..., s n }, observation state space O = {o1, o2, ..., o m }; Determine the initial probability distribution: π = (π1, π2, ..., π n ), represents the probability of being in each hidden state at the initial moment; Determine the transition probability matrix: A = (a ij ), where a ij Represents the hidden state s i Transfer to hidden state s j probability; Determine the transmission probability matrix: B = (b jk ), where b jk Indicates that in the hidden state s j Observe the observation state o k Probability Model training: Use the Baum-Welch algorithm to train the model and estimate the parameters of the model π, A and B through a large amount of observation data; Behavior prediction and analysis: Given an observation sequence, the Viterbi algorithm is used to decode the most likely hidden state sequence to analyze and predict the behavior.

8. The smart home device control system based on the Internet of Things operating system according to claim 1 is characterized in that: The multi-scenario simulation module includes a scenario configuration submodule: providing a visual interface for users to configure different simulation scenarios, allowing users to select the physical entity range to be simulated, set the simulation time span, and select the external environmental factors involved in the simulation. It supports the rapid call of existing mature scenario configurations from the preset scenario template library, and also allows users to customize and create new scenarios according to their own needs, and save and share customized scenario configurations; Simulation execution submodule: According to the conditions set by the scenario configuration submodule, the digital twin model is started for simulation operation. During the operation, the relevant data in the data processing and management module is called in real time as the input of the model to drive the digital twin model to dynamically evolve according to the set physical laws and behavior patterns, and can simulate the actual operation process and state changes of the physical entity in the corresponding scenario; Result analysis and visualization submodule: Analyze the results of the simulation, dig out valuable information by comparing the simulation results in different scenarios and analyzing the changes in key indicators, and present the analysis results to the user in an intuitive and visual way.

9. A multi-scenario simulation method based on digital twin technology according to any one of claims 1 to 8, characterized in that: The specific steps include: Step 1: Deploy various sensors for the simulation objects to collect real-time operating data of temperature, pressure and speed. Use laser scanning and photogrammetry to obtain its geometric shape and spatial position information, collect environmental data of the simulation scene, mine and integrate relevant historical data, clean the collected data, remove noise and outliers, fill in missing values ​​through data interpolation and smoothing methods, and then standardize and normalize data of different formats and sources, and store them in the database; Step 2: Based on the collected geometry and position information, accurately construct a 3D model of the physical entity. Based on physical principles and mathematical formulas, establish a physical model that describes the behavior of the physical entity. Use machine learning and deep learning algorithms to analyze historical data and real-time data, explore the behavior patterns and laws of the physical entity, and build a behavior prediction model. Organically integrate the geometric model, physical model, and behavior model to form a complete digital twin model. Step 3: According to the simulation purpose and needs, different scene categories are divided, specific parameters are set for each scene, the initial state of each scene is clarified, the configured scene parameters and initial conditions are input into the digital twin model, and simulation calculations are performed using a high-performance computing platform. During the simulation process, the model state is updated in real time to simulate the dynamic interaction between the physical entity and the environment.

Images