Low-latency transmission method and system for realizing the virtual-real integration of intelligent Internet of Things and the metaverse

Abstract

The present invention provides a low-latency transmission method and system for realizing the virtual-real integration of intelligent Internet of Things and the metaverse, which relates to the technical field of the metaverse level, including acquiring physical world data, classifying and sorting it, establishing a mapping relationship between the data stream and the virtual space, constructing a two-layer data processing architecture, setting different transmission priorities, and real-time monitoring the network quality and starting a data compensation mechanism. This method can realize the real-time interaction between the physical world and the virtual space, improve the data transmission efficiency and stability, reduce the system latency, and enhance the user's immersive experience.

Description

Technical Field

[0001] The present invention relates to metaverse technology, and in particular to a low-latency transmission method and system for realizing the virtual-real fusion of intelligent IoT and the metaverse. Background Art

[0002] With the rapid development of Internet of Things (IoT) and metaverse technologies, low-latency transmission for realizing the virtual-real fusion of intelligent IoT and the metaverse has become a current research hotspot. Intelligent IoT terminals can collect a large amount of physical-world data, while the metaverse provides a virtualized display and interaction platform for this data. By organically combining the two, seamless connection between the physical world and the virtual world can be achieved, bringing immersive experiences to users.

[0003] However, there are still some technical challenges in the process of realizing the integration of intelligent IoT and the metaverse. First, the transmission of a large amount of IoT data brings huge pressure to the network, easily causing transmission delays and affecting user experiences. Second, there is a lack of an effective mapping mechanism between physical-world data and the metaverse virtual space, making it difficult to achieve precise correspondence and real-time interaction between the two worlds. Finally, existing data processing architectures are often single, unable to perform differential processing on different types of data, resulting in low resource utilization efficiency.

[0004] To solve these problems, there is an urgent need for a new low-latency transmission method that can effectively process and transmit a large amount of IoT data, achieve precise mapping between the physical world and the metaverse virtual space, and improve transmission efficiency through a multi-level data processing architecture, thereby providing users with a smooth and real-time virtual-real fusion experience. Summary of the Invention

[0005] Embodiments of the present invention provide a low-latency transmission method and system for realizing the virtual-real fusion of intelligent IoT and the metaverse, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention,

[0007] A low-latency transmission method for realizing the virtual-real fusion of intelligent IoT and the metaverse is provided, including:

[0008] Obtaining physical-world data sent by multiple intelligent IoT terminals, where the physical-world data includes terminal identification information;

[0009] Classifying and sorting the physical-world data according to the terminal identification information to generate corresponding data streams, and establishing a mapping relationship between the data streams and the metaverse virtual space;

[0010] Construct a two - layer data processing architecture based on the mapping relationship. Set the target state data obtained through the state prediction algorithm in the first layer; set the scenario collaborative processing unit in the second layer to construct a scenario fusion model in combination with the target state data;

[0011] Set different transmission priorities for the target state data and the scenario fusion model. Divide the target state data into first - priority data streams and the scenario fusion model into second - priority data streams; monitor the network quality of each priority data stream in real time. When network fluctuations are detected, start the data compensation mechanism and perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm.

[0012] Classify and sort the physical world data according to the terminal identification information, generate corresponding data streams, and establish the mapping relationship between the data streams and the meta - universe virtual space, including:

[0013] Classify the physical world data according to the terminal type and data type according to the terminal identification information to obtain an initial data set;

[0014] Extract the data information in the initial data set to generate data feature labels, convert the data feature labels into corresponding semantic description information based on preset semantic mapping rules, and cluster the data in the initial data set according to the semantic description information to generate data streams with semantic associations;

[0015] Map the data stream to the meta - universe virtual space to establish the mapping relationship between the physical world and the virtual space.

[0016] Construct a two - layer data processing architecture based on the mapping relationship. Set the target state data obtained through the state prediction algorithm in the first layer; set the scenario collaborative processing unit in the second layer to construct a scenario fusion model in combination with the target state data, including:

[0017] Obtain the historical state sequence of the IoT terminal, where the historical state sequence includes a position vector, a velocity vector, and an acceleration vector;

[0018] Divide the historical state sequence into multiple state sequence segments according to a preset time window size; input the state sequence segments into a long short - term memory network. The long short - term memory network generates the hidden state at the current moment based on the state sequence segment at the current moment and the hidden state at the previous moment, and obtains the initial predicted state through the linear transformation of the weight matrix and the bias term according to the hidden state;

[0019] Obtain the real-time observation data of the IoT terminal, construct a prediction error covariance matrix and an observation noise covariance matrix based on the real-time observation data, and perform matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix to obtain the Kalman gain;

[0020] Correct the initial prediction state according to the Kalman gain, which specifically includes: calculating the deviation between the real-time observation data and the initial prediction state, and adding the product of the deviation and the Kalman gain to the initial prediction state to obtain the corrected target state data;

[0021] Send the corrected target state data to the IoT terminal to update the state at the next moment, and obtain a scene fusion model.

[0022] Constructing a prediction error covariance matrix and an observation noise covariance matrix based on the real-time observation data, and performing matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix to obtain the Kalman gain includes:

[0023] Obtain the true state and the predicted state of the IoT terminal, subtract the predicted state from the true state to obtain the prediction error, multiply the prediction error by the transpose of the prediction error, and find the expected value of the result of the transpose multiplication to obtain the prediction error covariance matrix;

[0024] Calculate the deviation between the real-time observation data and the predicted state to obtain an innovation sequence, and perform a cumulative summation operation on the innovation sequence within a preset sliding window to obtain the observation noise covariance matrix;

[0025] Add the prediction error covariance matrix and the observation noise covariance matrix to obtain the innovation covariance; divide the prediction error covariance matrix by the observation noise covariance matrix to obtain an adaptive weight factor;

[0026] Multiply the prediction error covariance matrix by the observation noise covariance matrix to obtain a seventh product, multiply the seventh product by the inverse matrix of the innovation covariance to obtain the initial Kalman gain, and multiply the initial Kalman gain by the adaptive weight factor to obtain the corrected Kalman gain.

[0027] Real-time monitor the network quality of each of the priority data streams. When network fluctuations are detected, start a data compensation mechanism, and perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm, including:

[0028] Construct a network state vector including the end-to-end delay and the packet loss rate, and add the product of the network state vector and the smoothing factor to the product of the smoothing state value at the previous moment and the smoothing factor to obtain the smoothing state value at the current moment;

[0029] Subtract the smoothed state value at the previous moment from the smoothed state value at the current moment to obtain a difference vector, calculate the two-norm of the difference vector to obtain a fluctuation detection value, and add the smoothed state value at the current moment to the product of the fluctuation detection value and a preset adjustment coefficient to construct a multi-level fluctuation threshold;

[0030] Input the fluctuation detection value and the multi-level fluctuation threshold into a compensation mapping function to calculate the compensation intensity;

[0031] Based on dividing the packet loss rate by the compensation intensity to obtain a third product, multiplying the fluctuation detection value by a preset fluctuation compensation coefficient to obtain a fourth product, adding the third product and the fourth product and rounding up to obtain the number of first-priority compensation data packets;

[0032] Divide the end-to-end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain a calculation result; compare the calculation result with a minimum window value, and take the product greater than a preset window value threshold as the compensation time window;

[0033] Perform compensation operations on data streams with different priorities according to the number of first-priority compensation data packets, the number of redundant packets, and the compensation time window respectively.

[0034] Dividing the end-to-end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain a calculation result; comparing the calculation result with a minimum window value, and taking the product greater than a preset window value threshold as the compensation time window includes:

[0035] Use the end-to-end delay, the product of the fluctuation detection value and the preset adjustment coefficient as the divisor for division operation to obtain an initial compensation time;

[0036] Compare the initial compensation time with a preset minimum window value. When the initial compensation time is greater than the preset minimum window value, use the initial compensation time as the reference compensation window value;

[0037] Multiply the reference compensation window value by a preset smoothing coefficient to obtain a first product, multiply the compensation window value at the previous moment by the complement of the preset smoothing coefficient to obtain a second product, and add the first product and the second product to obtain a smoothed compensation window value;

[0038] Multiply the smoothed compensation window value by the reference compensation window value to obtain a corrected compensation window value; calculate the current resource utilization rate and the weighted quality of service value based on the corrected compensation window value;

[0039] Construct an objective function by subtracting the product of the current resource utilization rate and a preset balance factor from the weighted quality of service value, and solve for the maximum value of the objective function under the constraint that the correction compensation window value is between a preset minimum window value and a preset maximum window value to obtain an optimal compensation window value;

[0040] Calculate a performance evaluation value based on the optimal compensation window value, and update the preset adjustment coefficient according to the performance evaluation value; use the optimal compensation window value as the final compensation time window value.

[0041] In the second aspect of the embodiments of the present invention,

[0042] Provide a low-latency transmission system for realizing the virtual-real fusion of intelligent IoT and the metaverse, including:

[0043] A first unit for obtaining physical world data sent by multiple intelligent IoT terminals, where the physical world data includes terminal identification information;

[0044] A second unit for classifying and sorting the physical world data according to the terminal identification information, generating corresponding data streams, and establishing a mapping relationship between the data streams and the virtual space of the metaverse;

[0045] A third unit for constructing a two-layer data processing architecture based on the mapping relationship, setting target state data obtained by a state prediction algorithm in the first layer; setting a scene collaboration processing unit in the second layer to construct a scene fusion model in combination with the target state data;

[0046] A fourth unit for setting different transmission priorities for the target state data and the scene fusion model, classifying the target state data into a first-priority data stream, and classifying the scene fusion model into a second-priority data stream; real-time monitoring the network quality of each priority data stream, and when network fluctuations are detected, starting a data compensation mechanism to perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm.

[0047] In the third aspect of the embodiments of the present invention,

[0048] Provide an electronic device, including:

[0049] A processor;

[0050] A memory for storing instructions executable by the processor;

[0051] Wherein, the processor is configured to call the instructions stored in the memory to execute the foregoing method.

[0052] In the fourth aspect of the embodiments of the present invention,

[0053] Provided is a computer-readable storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the foregoing method is implemented.

[0054] The beneficial effects of this application are as follows:

[0055] By establishing a mapping relationship between the data stream in the physical world and the virtual space of the metaverse, and constructing a two-layer data processing architecture, the efficient integration of the physical world and the virtual world is achieved. This architecture can better process and integrate data from different intelligent IoT terminals, improving the efficiency and accuracy of data processing.

[0056] By adopting a state prediction algorithm and a scene fusion model, the data transmission delay can be effectively reduced. By predicting the target state data and constructing a scene fusion model, the system can respond to changes more quickly, reducing the time for data processing and transmission, thereby enhancing the user experience.

[0057] By setting different transmission priorities and real-time monitoring of network quality, combined with a data compensation mechanism, the stability and reliability of the system are significantly improved. When network fluctuations occur, the system can intelligently adjust the data transmission strategy to ensure the timely transmission of critical data, effectively reducing the impact of network fluctuations on system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic flowchart of a low-latency transmission method for realizing the virtual-real fusion of intelligent IoT and the metaverse in an embodiment of the present invention;

[0059] Figure 2 It is a flowchart of the mapping relationship between physical world data and the virtual space of the metaverse in an embodiment of the present invention;

[0060] Figure 3 It is a flowchart of constructing a two-layer data processing architecture and a scene fusion model in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on 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.

[0062] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0063] Figure 1 This is a schematic flowchart of a low-latency transmission method for realizing the virtual-real integration of intelligent IoT and the metaverse in an embodiment of the present invention. As Figure 1 shown, the method includes:

[0064] Obtain physical world data sent by multiple intelligent IoT terminals, where the physical world data includes terminal identification information;

[0065] Classify and organize the physical world data according to the terminal identification information, generate corresponding data streams, and establish a mapping relationship between the data streams and the virtual space of the metaverse;

[0066] Construct a two-layer data processing architecture based on the mapping relationship. Set the target state data obtained by the state prediction algorithm in the first layer; set a scenario collaborative processing unit in the second layer to construct a scenario fusion model in combination with the target state data;

[0067] Set different transmission priorities for the target state data and the scenario fusion model, divide the target state data into a first-priority data stream, and divide the scenario fusion model into a second-priority data stream; monitor the network quality of each priority data stream in real time. When network fluctuations are detected, start a data compensation mechanism, and perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm.

[0068] In an alternative implementation, classifying and organizing the physical world data according to the terminal identification information, generating corresponding data streams, and establishing a mapping relationship between the data streams and the virtual space of the metaverse includes:

[0069] Classify the physical world data according to the terminal type and data type according to the terminal identification information to obtain an initial data set;

[0070] Extract the data information in the initial data set to generate data feature labels, convert the data feature labels into corresponding semantic description information based on a preset semantic mapping rule, and cluster the data in the initial data set according to the semantic description information to generate data streams with semantic associations;

[0071] Map the data stream to the virtual space of the metaverse and establish a mapping relationship between the physical world and the virtual space.

[0072] When obtaining terminal identification information, the unique identifiers of the terminal device (such as MAC address, device serial number, IMEI code, etc.), device type information (such as smart phone, VR headset, smart watch, etc.), and data collection capabilities (such as camera parameters, sensor types, data sampling rates, etc.) can be collected through the device registration module. These information are stored in the terminal information database to form a portrait of the terminal device. For example, for a VR headset device equipped with a depth camera, its terminal identification information may include: "Device type: VR headset; Manufacturer: Brand A; Device ID: VR20250410001; Sensors: 6DoF position tracking, gesture recognition; Resolution: 4K×2; Refresh rate: 120Hz".

[0073] When classifying physical world data according to the terminal identification information, the received data is preliminarily classified through the data classification engine. In specific implementation, a terminal type mapping table and a data type classification rule library can be established. For example, data from a smart phone can be classified into geographical location data, acceleration data, image data, etc.; data from smart home devices can be classified into temperature data, humidity data, electrical appliance status data, etc. Through the matching of identification information and rules, an initial data set is generated, which contains data source information, data type labels, and the content of the original data. Taking the walking trajectory collected by a smart phone as an example, the initial data set may contain information such as: "Data source: Smart phone A; Data type: Location trajectory; Original data: {timestamp, longitude, latitude, altitude, accuracy}".

[0074] When extracting features from the initial data set and generating data feature labels, a feature extractor is used to process the initial data. For location data, features such as moving speed, moving direction, and staying time can be extracted; for image data, features such as object category, scene type, and human posture can be extracted; for audio data, features such as speech content and ambient sound classification can be extracted. Taking the location trajectory data as an example, the extracted feature labels may be: "Moving speed: 5km / h; Movement mode: Walking; Trajectory shape: Circular; Staying points: 3; Duration: 30 minutes".

[0075] When converting data feature labels into semantic description information based on preset semantic mapping rules, the semantic mapping module converts the feature labels understood by the machine into description information with semantic understanding according to the predefined mapping rule library. For example, the "Moving speed: 5km / h; Movement mode: Walking; Trajectory shape: Circular; Staying points: 3" of the location trajectory can be converted into "The user carried out a 30-minute walking exercise in the park, walked around the park once, and stayed at the fitness area, rest pavilion, and lake on the way". The semantic mapping rule library realizes the mapping from features to semantics by associating data features with real-world semantic concepts.

[0076] When clustering the initial data set according to semantic description information, the data clustering engine groups the data based on semantic similarity calculation. In specific implementation, semantic distance measurement methods can be used to calculate the semantic similarity degree between different data, and algorithms such as hierarchical clustering or density clustering are used to form data clusters with semantic associations. For example, the park walking data of multiple users within the same time period may be clustered into a "park morning exercise activity" data stream; the location data, computer usage records, and meeting room reservation information in the office area may be clustered into a "work activity" data stream.

[0077] When establishing a mapping relationship by mapping the data stream to the metaverse virtual space, the virtual space mapping engine is responsible for converting the data stream into an expression form in the virtual space. Specific implementations include: spatial mapping, temporal mapping, entity mapping, and interaction mapping. Spatial mapping determines how the geographical location in the physical world corresponds to the coordinates in the virtual world; temporal mapping deals with the relationship between the time in the physical world and the time in the virtual world; entity mapping defines the representation method of objects in the physical world in the virtual world; interaction mapping stipulates how the behaviors in the physical world are transformed into interactions in the virtual world. Taking the park walking activity of a user as an example, after mapping, it may be shown in the virtual world that the virtual image of the user walks along the same path in the virtual park, the stopping points are highlighted, the walking speed is consistent with the actual speed, and even the real-world weather and environmental sounds can be perceived.

[0078] During the mapping process, semantic-level mapping rules can be used to enhance the conversion from the physical world to the virtual world. For example, when the action of "a user raises an arm to wave" is detected in the physical world, not only can the same action be reproduced in the virtual world, but it can also be transformed into a social behavior of "a virtual character greets" according to the context semantics; the semantic scene of multiple people gathering in a meeting room in the physical world can automatically create a meeting environment and deploy relevant tools in the virtual space.

[0079] To ensure the accuracy and real-time performance of the mapping, the system adopts an incremental update mechanism to process the continuously generated physical world data. After new data arrives, the corresponding elements in the virtual space are quickly updated according to the established mapping rules, and only the changed parts are processed, reducing unnecessary computing and transmission overhead.

[0080] Figure 2 The following is the flowchart of the mapping relationship between the physical world data and the metaverse virtual space in the embodiment of the present invention:

[0081] This figure shows a complete process of data processing and mapping. Starting from the terminal identification information and physical world data as input sources, the system classifies and organizes the data according to the terminal type and data type based on the terminal identification information, forming an initial data set. This initial data set is then divided into two processing branches: one branch extracts data information and generates data feature tags, and then converts these features into semantic description information based on semantic mapping rules; the other branch directly performs clustering processing on the initial data set to generate a semantic association data stream. The processing results of these two branches are finally converged together and jointly mapped into the metaverse virtual space. The entire process reflects the conversion process from physical world data to virtual space, realizing the structured processing and virtual-real mapping of data through semantic mapping and feature extraction, and constituting a complete data processing and space mapping framework. This processing method not only ensures the integrity and relevance of data, but also realizes the effective integration of the physical world and the virtual space, which is a typical data-driven virtual-real integration processing process.

[0082] The mapping between the physical world and the virtual world is usually achieved based on direct geometric relationships and predefined rules, lacking processing at the semantic understanding level. For example, traditional VR / AR systems mainly focus on the geometric correspondence from the physical space to the virtual space. After the user's physical actions are collected by sensors, they are directly mapped to virtual characters, lacking the understanding and conversion of action semantics. Most existing metaverse platforms use preset rules to convert user behaviors into virtual behaviors, which are difficult to process multi-modal data in complex scenarios and cannot achieve semantic association of cross-domain data.

[0083] The improvement starting point of this application lies in introducing a semantic understanding layer, elevating the processing of physical world data from pure data classification to the semantic understanding level. By establishing mapping rules between data features and semantic descriptions, a complete conversion link from physical world information to semantic understanding and then to virtual space expression is realized. This data processing method based on semantic understanding can better understand the actual meaning of user behaviors and environmental changes, thus providing a more realistic semantic expression in the virtual space.

[0084] The improvement effects of this application are mainly reflected in three aspects: First, the accuracy of physical-virtual mapping is improved through semantic understanding, making the expressions in the virtual world more in line with user intentions; Second, the semantic-based data clustering method realizes the association of cross-device and cross-modal data, can integrate information from different terminals, and form a more complete context understanding; Finally, the semantic-level mapping mechanism reduces the number of predefined rules, improves the adaptability and scalability of the system, and enables it to process more diverse physical world inputs.

[0085] In an alternative embodiment, a two - layer data processing architecture is constructed based on the mapping relationship. At the first layer, target state data is obtained through a state prediction algorithm; at the second layer, a scenario collaborative processing unit constructs a scenario fusion model in combination with the target state data, including:

[0086] Obtain the historical state sequence of the IoT terminal, where the historical state sequence includes a position vector, a velocity vector, and an acceleration vector;

[0087] Divide the historical state sequence into multiple state sequence segments according to a preset time window size; input the state sequence segments into a long short - term memory network. The long short - term memory network generates the hidden state at the current moment based on the state sequence segment at the current moment and the hidden state at the previous moment, and obtains an initial predicted state through a linear transformation of the weight matrix and the bias term according to the hidden state;

[0088] Obtain the real - time observation data of the IoT terminal, construct a prediction error covariance matrix and an observation noise covariance matrix based on the real - time observation data, and perform matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix to obtain a Kalman gain;

[0089] Correct the initial predicted state according to the Kalman gain, specifically including: calculating the deviation between the real - time observation data and the initial predicted state, and superimposing the product of the deviation and the Kalman gain on the initial predicted state to obtain the corrected target state data;

[0090] Send the corrected target state data to the IoT terminal, update the state at the next moment, and obtain a scenario fusion model.

[0091] Construct a two - layer data processing architecture according to a pre - established mapping relationship. At the first layer, a state prediction algorithm is set to obtain target state data. At the second layer, a scenario collaborative processing unit is set to construct a scenario fusion model in combination with the target state data.

[0092] Obtain the historical state sequence of the IoT terminal, including a position vector, a velocity vector, and an acceleration vector. For example, for a mobile robot, its historical state sequence can be:

[0093] Position vector: (x, y, z)=(1, 2, 3), (2, 3, 4), (3, 4, 5);

[0094] Velocity vector: (vx, vy, vz)=(0.1, 0.2, 0.3), (0.2, 0.3, 0.4), (0.3, 0.4, 0.5);

[0095] Acceleration vector: (ax,ay,az)=(0.01,0.02,0.03),(0.02,0.03,0.04),(0.03,0.04,0.05).

[0096] The historical state sequence is divided into multiple state sequence segments according to the preset time window size. For example, if the time window size is set to 3, the state sequence segments can be obtained: [(1,2,3), (2,3,4), (3,4,5)].

[0097] The state sequence fragments are input into the long short-term memory network (LSTM). The input layer of the LSTM network contains 3 neurons, corresponding to the three features of position, speed, and acceleration. The hidden layer is set with 64 neurons, and the output layer contains 3 neurons, corresponding to the predicted position, speed, and acceleration at the next moment. LSTM generates the hidden state at the current moment based on the state sequence fragment at the current moment and the hidden state at the previous moment. Specifically, LSTM controls the flow of information through the input gate, forget gate, and output gate to update the cell state and hidden state. Then, LSTM obtains the initial predicted state based on the hidden state through the linear transformation of the weight matrix and the bias term.

[0098] Obtain the real-time observation data of the IoT terminal, and construct the prediction error covariance matrix and the observation noise covariance matrix based on the real-time observation data. For example, suppose the real-time observed position is (3.1, 4.1, 5.1), which deviates from the initial prediction state (3, 4, 5). Based on the historical prediction error, the prediction error covariance matrix can be constructed as a diagonal matrix with diagonal elements of 0.01, 0.01, and 0.01. The observation noise covariance matrix can also be set as a diagonal matrix with diagonal elements of 0.005, 0.005, and 0.005.

[0099] The Kalman gain is obtained by performing matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix. The observation noise covariance matrix is a 3x3 unit matrix in this example. Through matrix operations, the Kalman gain is obtained as a diagonal matrix with diagonal elements of approximately 0.67, 0.67, 0.67.

[0100] The initial predicted state is corrected according to the Kalman gain. The specific steps include: calculating the deviation between the real-time observation data and the initial predicted state, which is (0.1, 0.1, 0.1) in this case. The product of the deviation and the Kalman gain is superimposed on the initial predicted state to obtain the corrected target state data. The corrected target state data is approximately (3.067, 4.067, 5.067).

[0101] The corrected target state data is sent to the IoT terminal to update the state at the next moment, thereby obtaining a scene fusion model.

[0102] The scenario collaborative processing unit receives the corrected target state data from the first layer. For example, the received data is position (3.067, 4.067, 5.067), velocity (0.3, 0.4, 0.5), and acceleration (0.03, 0.04, 0.05).

[0103] The scenario collaborative processing unit obtains environmental perception data. These data may come from various sensors, such as cameras, lidars, etc. For example, it obtains the point cloud data of the surrounding environment and identifies an obstacle nearby at position (5, 6, 7).

[0104] The scenario collaborative processing unit fuses the target state data with the environmental perception data. Specifically, an occupancy grid map can be used to represent the environment. The current position and predicted trajectory of the IoT terminal are marked on the grid map, and at the same time, the identified obstacles are also marked on the map.

[0105] Based on the fused data, the scenario collaborative processing unit performs path planning and decision-making. For example, considering that the predicted trajectory may collide with an obstacle, the system can plan a new path to avoid the obstacle. Suppose the newly planned path is a series of position points: (3.067, 4.067, 5.067), (4, 5, 6), (6, 7, 8).

[0106] The scenario collaborative processing unit generates control instructions and sends them to the IoT terminal. The control instructions may include speed adjustment, direction change, etc. For example, the generated control instructions are: adjust the speed to (0.2, 0.3, 0.4) and change the direction angle to 30 degrees.

[0107] After receiving the control instructions, the IoT terminal executes the corresponding actions and sends its state data to the first layer again at the next time step for prediction and correction, and so on, continuously optimizing the scenario fusion model.

[0108] Through the above double-layer data processing architecture, the present invention realizes the accurate prediction of the IoT terminal state and the dynamic fusion of the scenario. The state prediction algorithm of the first layer combines LSTM and Kalman filtering, which can effectively process time-series data and correct the prediction results. The scenario collaborative processing unit of the second layer combines the predicted state with the environmental perception data, realizing a higher-level scenario understanding and decision-making control. This architecture design enables the system to adapt to complex and changing environments, improving the intelligence level and operation efficiency of the IoT terminal.

[0109] Figure 3 The flowchart of the double-layer data processing architecture and scenario fusion model construction for the embodiment of the present invention is as follows:

[0110] The figure shows a two-layer state prediction and scenario collaborative processing architecture. In the first-layer state prediction algorithm, starting from the mapping relationship, the target state data is obtained through processing and used to construct a scenario fusion model. The second-layer scenario collaborative processing unit includes two parallel processing paths: the left path first obtains the historical state sequences of the IoT terminals (including location, speed, and acceleration information), divides these sequences into state sequence segments according to a preset time window, and then inputs them into a long short-term memory network (LSTM) for processing. Finally, the initial predicted state is obtained through linear transformation; the right path obtains the real-time observation data of the IoT terminals, constructs a prediction error covariance matrix and an observation noise covariance matrix, and calculates the Kalman gain. The processing results of the two paths will be used to calculate the deviation, and the initial predicted state will be corrected based on the Kalman gain. Finally, the state of the IoT terminals is updated and a scenario fusion model is generated. This two-layer architecture design not only ensures the accuracy of state prediction but also realizes the dynamic collaborative processing of scenarios, reflecting the organic combination of the prediction algorithm and real-time observation data, and constituting a complete prediction-correction closed-loop system.

[0111] In an alternative embodiment, constructing a prediction error covariance matrix and an observation noise covariance matrix based on the real-time observation data, and performing matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix to obtain the Kalman gain includes:

[0112] Obtain the real state and the predicted state of the IoT terminal, subtract the predicted state from the real state to obtain the prediction error, multiply the prediction error by the transpose of the prediction error, and find the expected value of the result of the transpose multiplication to obtain the prediction error covariance matrix;

[0113] Calculate the deviation between the real-time observation data and the predicted state to obtain the innovation sequence, and perform cumulative summation operations on the innovation sequence within a preset sliding window to obtain the observation noise covariance matrix;

[0114] Add the prediction error covariance matrix and the observation noise covariance matrix to obtain the innovation covariance; divide the prediction error covariance matrix by the observation noise covariance matrix to obtain the adaptive weight factor;

[0115] Multiply the prediction error covariance matrix by the observation noise covariance matrix to obtain a seventh product, multiply the seventh product by the inverse matrix of the innovation covariance to obtain the initial Kalman gain, and multiply the initial Kalman gain by the adaptive weight factor to obtain the corrected Kalman gain.

[0116] Obtain the true state and predicted state of the IoT terminal. Taking a smartphone as an example, the true state can be the actual position coordinates (x, y, z) obtained through the GPS module, and the predicted state is the position coordinates (x', y', z') predicted based on the previous state and the motion model.

[0117] Calculate the prediction error, that is, subtract the predicted state from the true state. For example, the prediction error is (Δx, Δy, Δz) = (x - x', y - y', z - z'). Then perform a self-multiplication operation on the prediction error to obtain the squared terms of the prediction error. For three-dimensional coordinates, 9 squared terms can be obtained.

[0118] Find the expected value of these squared terms to obtain the prediction error covariance matrix. In actual operation, the expected value can be approximated by calculating the average value of these squared terms through multiple samplings. For example, perform 100 samplings, sum the squared terms for each time and divide by 100 to obtain a 3x3 matrix, which is the prediction error covariance matrix P.

[0119] Calculate the deviation between the real-time observation data and the predicted state to obtain the innovation sequence. Taking smartphone positioning as an example, the real-time observation data can be the position estimate (x_obs, y_obs, z_obs) obtained by converting the base station signal strength, and the innovation sequence is (x_obs - x', y_obs - y', z_obs - z').

[0120] Perform a cumulative summation operation on the innovation sequence within a preset sliding window to obtain the observation noise covariance matrix R. The size of the sliding window can be set according to actual needs, for example, taking 10 sampling points. Calculate the sum of squares of the innovation sequences for these 10 points to obtain a 3x3 matrix R.

[0121] Add the prediction error covariance matrix P and the observation noise covariance matrix R to obtain the innovation covariance S. This step is actually to add the corresponding elements of two 3x3 matrices to obtain a new 3x3 matrix S.

[0122] Calculate the adaptive weight factor. Divide each element of the prediction error covariance matrix P by the corresponding element of the observation noise covariance matrix R to obtain a new 3x3 matrix, which is the adaptive weight factor W.

[0123] Multiply the prediction error covariance matrix P by the observation noise covariance matrix H to obtain an intermediate result matrix. The observation noise covariance matrix H reflects the relationship between the state variable and the observed quantity, and can be simplified to a 3x3 identity matrix in this example. Therefore, the result of this step is still the matrix P.

[0124] Multiply the intermediate result matrix obtained in the previous step by the inverse matrix of the innovation covariance S to obtain the initial Kalman gain K. Matrix inversion can be achieved through numerical calculation methods such as the Gauss-Jordan elimination method. The multiplication operation is the standard matrix multiplication, resulting in a new 3x3 matrix K.

[0125] Multiply the corresponding elements of the initial Kalman gain K by the adaptive weight factor W to obtain the corrected Kalman gain K'. This step adaptively adjusts the Kalman gain to adapt to different observation noise environments.

[0126] The following is a specific data case to illustrate the above process:

[0127] Suppose at a certain moment, the true position of the smartphone is (10, 20, 30), the predicted position is (9, 21, 31), and the observed position is (11, 19, 29).

[0128] Calculate the prediction error: (1, -1, -1);

[0129] The square term of the prediction error: ((1, -1, -1), (-1, 1, 1), (-1, 1, 1));

[0130] Suppose the prediction error covariance matrix P obtained after multiple samplings is:

[0131] ((1.2, -0.1, -0.1), (-0.1, 1.1, 0.1), (-0.1, 0.1, 1.3));

[0132] The innovation sequence: (2, -2, -2);

[0133] Suppose the observation noise covariance matrix R accumulated within a 10-point sliding window is:

[0134] ((0.8, 0, 0), (0, 0.9, 0), (0, 0, 1.0));

[0135] The innovation covariance S:

[0136] ((2.0, -0.1, -0.1), (-0.1, 2.0, 0.1), (-0.1, 0.1, 2.3));

[0137] The adaptive weight factor W:

[0138] ((1.5, -0.11, -0.1), (-0.11, 1.22, 0.1), (-0.1, 0.11, 1.3));

[0139] Initial Kalman gain K (assumed to be obtained through numerical calculation):

[0140] ((0.6, -0.03, -0.026), (-0.03, 0.55, 0.022), (-0.026, 0.022, 0.565));

[0141] Corrected Kalman gain K':

[0142] ((0.9, -0.0033, -0.0026), (-0.0033, 0.671, 0.0022), (-0.0026,0.00242, 0.7345));

[0143] In an alternative embodiment, the network quality of each of the priority data streams is monitored in real time. When network fluctuations are detected, a data compensation mechanism is started, and the compensation processing for different transmission priorities based on the prediction results of the state prediction algorithm includes:

[0144] Construct a network state vector including end-to-end delay and packet loss rate, and add the product of the network state vector and the smoothing factor to the product of the smoothing state value at the previous moment and the smoothing factor to obtain the smoothing state value at the current moment;

[0145] Subtract the smoothing state value at the previous moment from the smoothing state value at the current moment to obtain a difference vector, calculate the two-norm of the difference vector to obtain a fluctuation detection value, and add the product of the smoothing state value at the current moment and the fluctuation detection value and a preset adjustment coefficient to construct a multi-level fluctuation threshold;

[0146] Input the fluctuation detection value and the multi-level fluctuation threshold into a compensation mapping function to calculate the compensation intensity;

[0147] Based on the packet loss rate divided by the compensation intensity to obtain a third product, multiply the fluctuation detection value by a preset fluctuation compensation coefficient to obtain a fourth product, add the third product and the fourth product and round up to obtain the number of first-priority compensation data packets;

[0148] Divide the end-to-end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain a calculation result; compare the calculation result with a minimum window value, and take the product greater than the preset window value threshold as the compensation time window;

[0149] Perform compensation operations on data streams with different priorities according to the number of first-priority compensation data packets, the number of redundant packets, and the compensation time window.

[0150] Construct a network state vector. This vector contains two metrics: end-to-end delay and packet loss rate. For example, if the end-to-end delay at the current moment is 50 ms and the packet loss rate is 2%, then the network state vector can be represented as [50, 2].

[0151] Next, calculate the smoothed state value. Select a smoothing factor α, for example, α = 0.8. Assume that the smoothed state value at the previous moment is [48, 1.8], then the smoothed state value at the current moment is calculated as follows:

[0152] Current smoothed state value = α * [50, 2] + (1 - α) * [48, 1.8] = [49.6, 1.96];

[0153] Calculate the fluctuation detection value. Subtract the smoothed state value at the previous moment from the current smoothed state value to obtain a difference vector [1.6, 0.16]. Calculate the two-norm of this vector, that is, the fluctuation detection value is 1.61.

[0154] Construct a multi-level fluctuation threshold. Assume that the preset adjustment coefficient β = 0.5, then the multi-level fluctuation threshold is calculated as follows:

[0155] Multi-level fluctuation threshold = [49.6, 1.96] + 1.61 * 0.5 = [50.405, 2.765];

[0156] Input the fluctuation detection value and the multi-level fluctuation threshold into the compensation mapping function to calculate the compensation intensity. The compensation mapping function can be a piecewise function that returns different compensation intensities according to the size of the fluctuation detection value. For example, when the fluctuation detection value is less than 1, the compensation intensity is 1; when the fluctuation detection value is between 1 and 2, the compensation intensity is 1.5; when the fluctuation detection value is greater than 2, the compensation intensity is 2. In this example, the fluctuation detection value is 1.61, and the corresponding compensation intensity is 1.5.

[0157] Calculate the number of first-priority compensation data packets. Assume that the preset fluctuation compensation coefficient γ = 0.8, then:

[0158] Number of first-priority compensation data packets = ceil((2% / 1.5) + (1.61 * 0.8)) = ceil(1.33 + 1.288) = 3;

[0159] Calculate the compensation time window. Assume that the minimum window value is 100 ms and the preset window value threshold is 1.2, then:

[0160] Calculation result = 50 ms / (1.61 * 0.5) = 62.11 ms;

[0161] Compensation time window = max(62.11ms, 100ms) * 1.2 = 120ms;

[0162] Based on the calculated number of compensation data packets and the compensation time window, perform compensation operations on data streams with different priorities.

[0163] For the data stream with the first priority, send 3 additional redundant data packets within a 120ms time window.

[0164] For the data stream with the second priority, the compensation intensity can be reduced according to the actual situation. For example, send 2 additional redundant data packets within a 120ms time window.

[0165] For the data stream with the third priority, the compensation intensity can be further reduced. For example, send 1 additional redundant data packet within a 120ms time window.

[0166] Differentiated compensation processing can be performed on data streams with different priorities according to the real-time fluctuations of network quality, thereby improving the overall transmission quality and reliability.

[0167] Various parameters can be adjusted according to specific scenarios and requirements. For example, the value of the smoothing factor α can be adjusted according to the stability of the network environment; the value of the preset adjustment coefficient β can be adjusted according to the sensitivity to network fluctuations; the value of the preset fluctuation compensation coefficient γ can be adjusted according to the requirements of the service for transmission reliability.

[0168] The compensation mapping function can also be customized according to actual needs. For example, a more complex non-linear function can be adopted to more accurately reflect the relationship between network fluctuations and compensation intensity.

[0169] In the implementation process, the following points need to be noted:

[0170] 1. The construction of the network state vector should be based on a reliable network quality monitoring mechanism to ensure the measurement accuracy of end-to-end delay and packet loss rate.

[0171] 2. The calculation of the smoothed state value helps to filter out short-term fluctuations, but may also lead to a slower response to sudden changes. Therefore, the selection of the smoothing factor α needs to be balanced between the smoothing effect and the response speed.

[0172] 3. The calculation of the fluctuation detection value uses the second norm, which can comprehensively consider the changes of multiple indicators. However, in some scenarios, different weights may need to be assigned to different indicators.

[0173] 4. The construction method of the multi-level fluctuation threshold can be adjusted according to actual needs. For example, multiple threshold levels can be set, corresponding to different compensation strategies.

[0174] 5. The design of the compensation mapping function directly affects the compensation effect. A large amount of experimental data can be used to optimize this function to better adapt to a specific network environment.

[0175] 6. The calculation of the compensation time window needs to consider the real-time state of the network and the tolerance of the service. If the window is too small, the compensation effect may not be obvious; if the window is too large, the latency may increase.

[0176] 7. The compensation processing for data flows with different priorities should match the business importance. Different compensation strategies can be set for different priorities according to the actual application scenario.

[0177] 8. When implementing the compensation operation, the limitation of network bandwidth needs to be considered to avoid network congestion caused by over-compensation.

[0178] 9. The entire compensation mechanism should be adaptive and able to automatically adjust various parameters according to the long-term network conditions to adapt to different network environments.

[0179] 10. During the implementation process, attention should be paid to the efficiency of the algorithm to ensure that it can be quickly executed in a real-time system. Technologies such as parallel computing can be considered to improve the processing speed.

[0180] In an alternative implementation, dividing the end-to-end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain a calculation result; comparing the calculation result with the minimum window value, and taking the product greater than the preset window value threshold as the compensation time window includes:

[0181] Performing a division operation with the product of the end-to-end delay, the fluctuation detection value, and the preset adjustment coefficient as the divisor to obtain an initial compensation time;

[0182] Comparing the initial compensation time with the preset minimum window value, and when the initial compensation time is greater than the preset minimum window value, taking the initial compensation time as the reference compensation window value;

[0183] Multiplying the reference compensation window value by the preset smoothing coefficient to obtain a first product, multiplying the previous moment's compensation window value by the complement of the preset smoothing coefficient to obtain a second product, and adding the first product and the second product to obtain a smoothed compensation window value;

[0184] Multiplying the smoothed compensation window value by the reference compensation window value to obtain a corrected compensation window value; calculating the current resource utilization rate and the weighted quality of service value based on the corrected compensation window value;

[0185] Construct an objective function by subtracting the product of the current resource utilization rate and a preset balance factor from the weighted quality of service value, and solve for the maximum value of the objective function under the constraint that the correction compensation window value is between a preset minimum window value and a preset maximum window value to obtain the optimal compensation window value;

[0186] Calculate a performance evaluation value based on the optimal compensation window value, and update the preset adjustment coefficient according to the performance evaluation value; Use the optimal compensation window value as the final compensation time window value.

[0187] Obtain the end-to-end delay, the fluctuation detection value, and the preset adjustment coefficient. The end-to-end delay can be measured by a network test tool and may be, for example, 100 ms. The fluctuation detection value can be obtained by statistically analyzing network jitter and may be, for example, 0.2. The preset adjustment coefficient can be set according to experience and may be, for example, 1.5.

[0188] Divide the end-to-end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain the initial compensation time. Specifically, divide 100 ms by the product of 0.2 and 1.5 to obtain the initial compensation time of 333.33 ms.

[0189] Compare the initial compensation time with the preset minimum window value. The preset minimum window value can be set according to system requirements and may be, for example, 200 ms. Since 333.33 ms is greater than 200 ms, 333.33 ms is used as the reference compensation window value.

[0190] Introduce smoothing processing. Set the preset smoothing coefficient to 0.8, then its complement is 0.2. Multiply the reference compensation window value of 333.33 ms by 0.8 to get the first product of 266.66 ms, multiply the previous moment's compensation window value (assumed to be 300 ms) by 0.2 to get the second product of 60 ms, and add the two to get the smoothed compensation window value of 326.66 ms.

[0191] For further correction, multiply the smoothed compensation window value by the reference compensation window value to obtain the corrected compensation window value of 108888.89 ms².

[0192] Based on the corrected compensation window value, calculate the current resource utilization rate and the weighted quality of service value. Assume the current resource utilization rate is 0.7 and the weighted quality of service value is 0.9.

[0193] Construct the objective function: Subtract the product of the current resource utilization rate of 0.7 and the preset balance factor (assumed to be 0.5), which is 0.35, from the weighted quality of service value of 0.9 to obtain an objective function value of 0.55. Under the constraint that the correction compensation window value is between the preset minimum window value of 200ms and the preset maximum window value (assumed to be 500ms), solve for the maximum value of the objective function through an iterative optimization algorithm to obtain the optimal compensation window value, which may be 400ms for example.

[0194] Based on the optimal compensation window value of 400ms, calculate the performance evaluation value. The weighted average method can be used to comprehensively evaluate the resource utilization rate and the quality of service index to obtain the performance evaluation value, which may be 0.85 for example.

[0195] According to the performance evaluation value of 0.85, update the preset adjustment coefficient. An adaptive adjustment strategy can be adopted. When the performance evaluation value is higher than expected, appropriately reduce the adjustment coefficient, and vice versa. For example, the adjustment coefficient can be adjusted from 1.5 to 1.4.

[0196] Take the optimal compensation window value of 400ms as the final compensation time window value.

[0197] To improve the calculation efficiency, parallel processing technology can be adopted. For example, while calculating the initial compensation time, the update of the fluctuation detection value can be carried out in parallel. In addition, caching technology can be used to store intermediate calculation results to avoid repeated calculations.

[0198] Under extreme network conditions, this method can still maintain stability. For example, when the end-to-end delay surges to 1000ms due to severe network congestion, through the adjustment of the fluctuation detection value and the adjustment coefficient, a reasonable compensation time window can still be obtained to avoid the system from overreacting.

[0199] This method can also be combined with machine learning technology. By training a model with historical data, predict the optimal compensation window value. For example, a long short-term memory network (LSTM) model can be used. Input historical network state data and output the predicted compensation window value to further improve the calculation accuracy and efficiency.

[0200] During the implementation process, attention needs to be paid to the numerical precision problem. It is recommended to use double-precision floating-point numbers for calculations to ensure the accuracy of the calculation results. At the same time, appropriate exception handling should be carried out for possible division-by-zero errors.

[0201] To facilitate system management and tuning, a visual interface can be designed to display the change trends of various parameters in real time. For example, line charts can be used to show the historical changes of indicators such as end-to-end delay, fluctuation detection value, and compensation time window, to help administrators intuitively understand the system status.

[0202] In a multi - user scenario, different quality - of - service weights can be set for different users to achieve differential compensation time window calculations. For example, for paying users, higher quality - of - service weights can be set to obtain better compensation time windows.

[0203] This method can also be used in combination with other network optimization technologies, such as congestion control, load balancing, etc., to form a comprehensive network optimization solution. For example, the congestion window size can be dynamically adjusted according to the calculated compensation time window to achieve more refined traffic control.

[0204] A distributed architecture can be adopted to disperse the computing tasks to multiple nodes for execution, improving the scalability and fault tolerance of the system. For example, distributed computing frameworks such as Hadoop can be used to process large - scale compensation time window calculation tasks in parallel.

[0205] By comprehensively considering multiple factors such as end - to - end delay, network fluctuations, resource utilization, and quality of service, this method dynamically calculates the optimal compensation time window, which can effectively improve network performance and user experience. By introducing technologies such as smoothing processing, adaptive adjustment, and machine learning, the robustness and adaptability of the method are further enhanced, making it applicable to various complex network environments.

[0206] In the second aspect of the embodiments of the present invention,

[0207] A low - latency transmission system for realizing the virtual - real fusion of intelligent IoT and the metaverse is provided, including:

[0208] A first unit for obtaining physical world data sent by multiple intelligent IoT terminals, where the physical world data includes terminal identification information;

[0209] A second unit for classifying and sorting the physical world data according to the terminal identification information, generating corresponding data streams, and establishing a mapping relationship between the data streams and the virtual space of the metaverse;

[0210] A third unit for constructing a two - layer data processing architecture based on the mapping relationship, setting target state data obtained by a state prediction algorithm in the first layer; and setting a scene collaboration processing unit in the second layer to construct a scene fusion model in combination with the target state data;

[0211] A fourth unit for setting different transmission priorities for the target state data and the scene fusion model, dividing the target state data into first - priority data streams, and dividing the scene fusion model into second - priority data streams; real - time monitoring the network quality of each priority data stream, and when network fluctuations are detected, starting a data compensation mechanism to perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm.

[0212] In a third aspect of the embodiments of the present invention,

[0213] there is provided an electronic device, comprising:

[0214] a processor;

[0215] a memory for storing instructions executable by the processor;

[0216] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0217] In a fourth aspect of the embodiments of the present invention,

[0218] there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0219] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are loaded.

[0220] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A low-latency transmission method for realizing the virtual-real fusion of intelligent Internet of Things and the metaverse, characterized in that Including: Obtain physical world data sent by multiple intelligent Internet of Things terminals, where the physical world data includes terminal identification information; Classify and organize the physical world data according to the terminal identification information, generate corresponding data streams, and establish a mapping relationship between the data streams and the metaverse virtual space; Based on the mapping relationship, construct a two-layer data processing architecture. Set the target state data obtained by the state prediction algorithm in the first layer; set the scenario collaborative processing unit in the second layer to construct a scenario fusion model in combination with the target state data; Set different transmission priorities for the target state data and the scenario fusion model, divide the target state data into the first-priority data stream, and divide the scenario fusion model into the second-priority data stream; Real-time monitor the network quality of each priority data stream. When network fluctuations are detected, start the data compensation mechanism, and perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm; Construct a network state vector including end-to-end delay and packet loss rate, add the product of the network state vector and the smoothing factor to the product of the previous moment's smoothed state value and the smoothing factor to obtain the smoothed state value at the current moment; Subtract the smoothed state value at the previous moment from the smoothed state value at the current moment to obtain a difference vector, calculate the two-norm of the difference vector to obtain a fluctuation detection value, and add the product of the smoothed state value at the current moment and the fluctuation detection value and the preset adjustment coefficient to construct a multi-level fluctuation threshold; Input the fluctuation detection value and the multi-level fluctuation threshold into the compensation mapping function to calculate the compensation intensity; Based on the packet loss rate divided by the compensation intensity to obtain a third product, multiply the fluctuation detection value by the preset fluctuation compensation coefficient to obtain a fourth product, add the third product and the fourth product and round up to obtain the number of first-priority compensation data packets; Divide the end-to-end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain a calculation result; compare the calculation result with the minimum window value, and take the product greater than the preset window value threshold as the compensation time window; Perform compensation operations on data streams of different priorities according to the number of first-priority compensation data packets, the number of redundant packets, and the compensation time window respectively.

2. The method according to claim 1, wherein Classify and organize the physical world data according to the terminal identification information, generate corresponding data streams, and establish a mapping relationship between the data streams and the metaverse virtual space, including: Classify the physical world data according to the terminal type and data type according to the terminal identification information to obtain an initial data set; Extract the data information in the initial data set to generate data feature labels, convert the data feature labels into corresponding semantic description information based on preset semantic mapping rules, and cluster the data in the initial data set according to the semantic description information to generate data streams with semantic associations; Map the data stream to the metaverse virtual space and establish a mapping relationship between the physical world and the virtual space.

3. The method according to claim 1, wherein Construct a two - layer data processing architecture based on the mapping relationship. Set the target state data obtained by the state prediction algorithm in the first layer; set a scene collaborative processing unit in the second layer to construct a scene fusion model in combination with the target state data, including: Obtain the historical state sequence of the IoT terminal, where the historical state sequence includes a position vector, a velocity vector, and an acceleration vector; Divide the historical state sequence into multiple state sequence segments according to a preset time window size; input the state sequence segments into a long short - term memory network. The long short - term memory network generates the hidden state at the current moment based on the state sequence segment at the current moment and the hidden state at the previous moment, and obtains the initial predicted state through a linear transformation of the weight matrix and the bias term according to the hidden state; Obtain the real - time observation data of the IoT terminal, construct a prediction error covariance matrix and an observation noise covariance matrix based on the real - time observation data, and perform matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix to obtain the Kalman gain; Correct the initial predicted state according to the Kalman gain, specifically including: calculating the deviation between the real - time observation data and the initial predicted state, and adding the product of the deviation and the Kalman gain to the initial predicted state to obtain the corrected target state data; Send the corrected target state data to the IoT terminal, update the state at the next moment, and obtain the scene fusion model.

4. The method according to claim 3, wherein Construct a prediction error covariance matrix and an observation noise covariance matrix based on the real - time observation data, and perform matrix operations on the product of the prediction error covariance matrix and the observation noise covariance matrix to obtain the Kalman gain, including: Obtain the real state and the predicted state of the IoT terminal, subtract the predicted state from the real state to obtain the prediction error, multiply the prediction error by the transpose of the prediction error, and find the expected value of the result of the transpose multiplication to obtain the prediction error covariance matrix; Calculate the deviation between the real - time observation data and the predicted state to obtain the innovation sequence, and perform a cumulative summation operation on the innovation sequence within a preset sliding window to obtain the observation noise covariance matrix; Add the prediction error covariance matrix and the observation noise covariance matrix to obtain the innovation covariance; divide the prediction error covariance matrix by the observation noise covariance matrix to obtain the adaptive weight factor; Multiply the prediction error covariance matrix by the observation noise covariance matrix to obtain the seventh product, multiply the seventh product by the inverse matrix of the innovation covariance to obtain the initial Kalman gain, and multiply the initial Kalman gain by the adaptive weight factor to obtain the corrected Kalman gain.

5. The method according to claim 1, characterized in that, Divide the end - to - end delay by the product of the fluctuation detection value and the preset adjustment coefficient to obtain the calculation result; Compare the calculation result with the minimum window value, and take the product greater than the preset window value threshold as the compensation time window, including: Use the end - to - end delay, the product of the fluctuation detection value and the preset adjustment coefficient as the divisor for division operation to obtain the initial compensation time; Compare the initial compensation time with a preset minimum window value. When the initial compensation time is greater than the preset minimum window value, use the initial compensation time as the reference compensation window value; Multiply the reference compensation window value by a preset smoothing coefficient to obtain a first product, multiply the previous moment's compensation window value by the complement of the preset smoothing coefficient to obtain a second product, and add the first product and the second product to obtain a smoothed compensation window value; Multiply the smoothed compensation window value by the reference compensation window value to obtain a corrected compensation window value; calculate the current resource utilization rate and the weighted quality of service value based on the corrected compensation window value; Construct an objective function by subtracting the product of the current resource utilization rate and a preset balance factor from the weighted quality of service value. Under the constraint condition that the corrected compensation window value is between the preset minimum window value and the preset maximum window value, solve the maximum value of the objective function to obtain the optimal compensation window value; Calculate a performance evaluation value based on the optimal compensation window value, and update the preset adjustment coefficient according to the performance evaluation value; use the optimal compensation window value as the final compensation time window value.

6. A low-latency transmission system for realizing the virtual-real integration of intelligent Internet of Things and the metaverse, which is used to implement the method described in any one of the foregoing claims 1-5, and is characterized in that, Comprising: A first unit for obtaining physical world data sent by multiple intelligent Internet of Things terminals, where the physical world data includes terminal identification information; A second unit for classifying and sorting the physical world data according to the terminal identification information, generating corresponding data streams, and establishing a mapping relationship between the data streams and the metaverse virtual space; A third unit for constructing a two-layer data processing architecture based on the mapping relationship, setting target state data obtained by a state prediction algorithm in the first layer; setting a scene collaborative processing unit in the second layer to construct a scene fusion model in combination with the target state data; A fourth unit for setting different transmission priorities for the target state data and the scene fusion model, dividing the target state data into first-priority data streams, and dividing the scene fusion model into second-priority data streams; Monitor the network quality of each of the priority data streams in real time. When network fluctuations are detected, start a data compensation mechanism, and perform compensation processing on different transmission priorities based on the prediction results of the state prediction algorithm.

7. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 5.

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

Images