Incremental update method, apparatus, equipment and medium for digital twins

Abstract

This application relates to a method, apparatus, device, and medium for incremental updates of digital twins, belonging to the field of digital twin technology. It includes: using an adaptive sliding window algorithm and a Hausdorff distance algorithm to filter unchanged TB-level static data or unmodified models, extracting only a very small amount of changed data, reducing the data processing volume, alleviating the storage and computing pressure on edge nodes, and avoiding resource exhaustion and system paralysis; using multimodal data fusion technology to perform differential calculations on time-series sensor data and 3D model point cloud data respectively, which is more efficient than the general full reconstruction and significantly shortens the data capture time; introducing a semantic analysis network based on a scaling dot product attention mechanism to identify the importance of changes, and using a multi-dimensional weighted formula for priority scheduling, prioritizing high-score tasks and delaying low-score tasks, ensuring that critical business operations can obtain computing resources preferentially under limited resources, meeting the low-latency requirements of critical scenarios.

Description

Technical Field

[0001] This application relates to the field of digital twin technology, specifically to a method, apparatus, device, and medium for incremental updating of a digital twin. Background Technology

[0002] With the deepening development of Industry 4.0 and the concept of the metaverse, digital twin technology has become a core bridge connecting the physical world and the digital space. In a typical edge-cloud collaborative architecture, in order to maintain a high degree of consistency between the digital twin and the physical entity, the digital twin needs to undergo high-frequency data updates and state synchronization.

[0003] However, with the increasing complexity of application scenarios, the data volume of digital twins for individual industrial equipment has surged from the GB level to the TB level. Existing digital twin update technologies typically reconstruct the entire digital twin at fixed time intervals. Faced with massive data, limited resources, and high real-time requirements, edge nodes are constrained by hardware costs, physical space, and power consumption, resulting in relatively limited computing and storage resources. This makes it difficult to support the full reconstruction and frequent updates of TB-level large-scale digital twins. Forcing the deployment of a full update mechanism could easily lead to the exhaustion of edge device resources, system paralysis, and an inability to meet the sub-100ms real-time requirements of scenarios such as autonomous driving and telemedicine. Summary of the Invention

[0004] In view of this, this application provides a method, apparatus, device, and medium for incremental updates of digital twins. The main objective is to address the problem that current digital twin update technologies typically reconstruct the entire digital twin at fixed time intervals. When faced with massive amounts of data, limited resources, and high real-time requirements, edge nodes are constrained by hardware costs, physical space, and power consumption, resulting in relatively limited computing and storage resources. This makes it difficult to support the full reconstruction and frequent updates of TB-level large-scale digital twins. Forcibly deploying a full update mechanism can easily lead to the exhaustion of edge device resources, system paralysis, and an inability to meet the sub-100ms real-time requirements of scenarios such as autonomous driving and telemedicine.

[0005] Firstly, this application provides a method for incremental updating a digital twin, including: Real-time acquisition of multimodal data of physical entities, wherein the multimodal data includes at least time-series sensor data and 3D model point cloud data; The time-series sensor data is dynamically differentially calculated using an adaptive sliding window algorithm to obtain a first state change result. The 3D model point cloud data is geometrically altered using a Hausdorff distance algorithm to obtain a second state change result. The first and second state change results are then subjected to multimodal fusion decision processing based on a weight allocation strategy to generate an initial change set. The initial change set is transformed into a current change feature vector, which is then input into a semantic analysis network based on a scaled dot product attention mechanism. By calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, a change importance score is determined. Based on the change importance score, each change item in the initial change set is prioritized to generate a semantically labeled change set. Each change item in the semantically labeled change set is taken as a change task to be scheduled. Based on the change importance score, business urgency coefficient and estimated resource consumption, the priority score of each change task is calculated. Combined with the resource constraints of the current edge node, the optimal incremental update execution sequence is generated. The priority score is positively correlated with the change importance score and the business urgency coefficient, and negatively correlated with the estimated resource consumption. The incremental update data in the incremental update execution sequence is subject to triple verification. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verifications include data integrity verification, logical rationality verification, and performance impact assessment.

[0006] Secondly, this application provides a digital twin incremental update device, comprising: The acquisition module is used to acquire multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and three-dimensional model point cloud data; The first calculation module is used to perform dynamic differential calculation on the time-series sensor data using an adaptive sliding window algorithm to obtain a first state change result, to perform geometric change identification on the three-dimensional model point cloud data using a Hausdorff distance algorithm to obtain a second state change result, and to perform multimodal fusion decision processing on the first state change result and the second state change result based on a weight allocation strategy to generate an initial change set. The tagging module is used to transform the initial change set into a current change feature vector, input it into a semantic analysis network based on a scaled dot product attention mechanism, determine the change importance score by calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, and prioritize each change item in the initial change set according to the change importance score to generate a semantically tagged change set. The second calculation module is used to take each change item in the semantically labeled change set as a change task to be scheduled, calculate the priority score of each change task based on the change importance score, the business urgency coefficient and the estimated resource consumption, and generate the optimal incremental update execution sequence in combination with the resource constraints of the current edge node. The priority score is positively correlated with the change importance score and the business urgency coefficient, and negatively correlated with the estimated resource consumption. The verification module is used to perform triple verification on the incremental update data in the incremental update execution sequence. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verification includes data integrity verification, logical rationality verification, and performance impact assessment.

[0007] Thirdly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the digital twin incremental update method described in the first aspect.

[0008] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the digital twin incremental update method described in the first aspect.

[0009] By employing the above technical solutions, this application provides a digital twin incremental update method, apparatus, device, and medium. Compared with existing technologies, this application can acquire multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and 3D model point cloud data; use an adaptive sliding window algorithm to perform dynamic differential calculation on the time-series sensor data to obtain a first state change result; use the Hausdorff distance algorithm to perform geometric change recognition on the 3D model point cloud data to obtain a second state change result; and perform multimodal fusion decision processing on the first and second state change results based on a weight allocation strategy to generate an initial change set; transform the initial change set into a current change feature vector, input it into a semantic analysis network based on a scaled dot product attention mechanism, and determine the change importance score by calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library. The system prioritizes each change item in the initial change set based on its importance score, generating a semantically labeled change set. Each change item in this semantically labeled change set is then treated as a scheduled change task. Based on the change importance score, business urgency coefficient, and estimated resource consumption, a priority score is calculated for each change task. Combined with the resource constraints of the current edge node, an optimal incremental update execution sequence is generated. The priority score is positively correlated with the change importance score and business urgency coefficient, and negatively correlated with the estimated resource consumption. Triple verification is performed on the incremental update data in the execution sequence. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any verification fails, a compensation mechanism is triggered to ensure system consistency. The triple verification includes data integrity verification, logical rationality verification, and performance impact assessment.

[0010] By employing the above technical solutions, this application utilizes an adaptive sliding window algorithm and a Hausdorff distance algorithm to accurately filter out TB-level static data or unchanged models, extracting only a very small amount of changed data (initial change set). This reduces the amount of data to be processed from TB-level to MB or even KB-level, fundamentally alleviating the storage and computing pressure on edge nodes and avoiding resource exhaustion and system paralysis. This application also employs multimodal data fusion technology to perform differential calculations on time-series sensor data and 3D model point cloud data separately. This targeted algorithm is more efficient than general full-scale reconstruction, significantly shortening data acquisition time.

[0011] Even under resource constraints, this application does not employ a simple "first-come, first-served" or "full-processing" approach. Instead, it introduces a semantic analysis network based on a scaled dot product attention mechanism. The system can identify the importance of changes (e.g., obstacle recognition data in autonomous driving has higher priority than background environment data) and prioritizes tasks using a multi-dimensional weighted formula. By prioritizing high-scoring tasks and deferred low-scoring tasks, it ensures that critical business operations receive priority computing resources under limited resources, thereby meeting the low-latency requirements of critical scenarios.

[0012] When generating the update sequence, this application dynamically determines the resource constraints of the current edge nodes. Updates are only executed when the remaining resources meet the task consumption estimate; otherwise, they are marked as pending. This resource-constrained strategy prevents system blocking or crashes caused by resource contention.

[0013] To address network uncertainties, this application designs a consistency verification and compensation mechanism. If verification fails, the system will choose between "full update remediation" or "degradation strategy (maintaining core functions)" based on network and time conditions. This ensures that the system can maintain the real-time performance of core functions even during network fluctuations or temporary resource shortages, preventing complete system failure.

[0014] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a digital twin incremental update method provided in this application embodiment; Figure 2 A logical architecture diagram of a digital twin incremental update method provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a digital twin incremental update device provided in an embodiment of this application. Detailed Implementation

[0018] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] The following description, with reference to the accompanying drawings, describes a digital twin incremental update method, apparatus, device, and medium according to embodiments of this application.

[0020] This application provides a method, apparatus, device, and medium for incremental updates of digital twins. Its main purpose is to address the problem that current digital twin update technologies typically reconstruct the entire digital twin at fixed time intervals. When faced with massive amounts of data, limited resources, and high real-time requirements, edge nodes are constrained by hardware costs, physical space, and power consumption, resulting in relatively limited computing and storage resources. This makes it difficult to support the full reconstruction and frequent updates of TB-level large-scale digital twins. Forcing a full update mechanism can easily lead to resource exhaustion of edge devices and system paralysis, failing to meet the sub-100ms real-time requirements of scenarios such as autonomous driving and telemedicine.

[0021] like Figure 1 As shown, embodiments of this application provide a digital twin incremental update method, including: Step 101: Collect multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and 3D model point cloud data.

[0022] In this embodiment, as Figure 2 As shown, the system can acquire multimodal data in real time through data layer acquisition modules deployed on physical entities (such as industrial robots, autonomous vehicles, or medical devices). The multimodal data includes at least time-series sensor data and 3D model point cloud data.

[0023] Among them, time-series sensor data may include scalar or vector data such as temperature, pressure, vibration acceleration, current and voltage that change continuously over time, and usually exist in the form of high-frequency sampling streams.

[0024] Three-dimensional model point cloud data may include a set of geometric coordinates of the physical entity surface obtained by LiDAR, depth camera or structured light scanner, which can be used to characterize the spatial morphology and structural features of the physical entity.

[0025] The data acquisition module can synchronously transmit the above data to the edge nodes for timestamp alignment and format standardization, in preparation for subsequent differential calculations.

[0026] Step 102: Use the adaptive sliding window algorithm to perform dynamic differential calculation on the time-series sensor data to obtain the first state change result. Use the Hausdorff distance algorithm to perform geometric change recognition on the 3D model point cloud data to obtain the second state change result. Then, based on the weight allocation strategy, perform multimodal fusion decision processing on the first and second state change results to generate the initial change set.

[0027] like Figure 2 As shown, this step can be executed by the state difference capture module, which is the core component of the digital twin incremental algorithm and is responsible for real-time identification of fine-grained state changes of physical entities. This module can employ multimodal data fusion technology, simultaneously processing temporal sensor data and 3D model point cloud data. Through an improved sliding window algorithm and Hausdorff distance calculation, it achieves efficient and accurate change detection.

[0028] For time-series sensor data, traditional fixed-window methods struggle to balance real-time performance and stability. This embodiment employs an improved sliding-window algorithm to perform dynamic differential calculations on the time-series sensor data, obtaining the first state change result. The specific process is as follows: First, the average value of time-series sensor data within the current time window can be calculated in real time. and standard deviation ; The sliding window size is dynamically adjusted based on the mean and standard deviation using a formula for dynamic window size adjustment. The formula for dynamically adjusting the size of the sliding window is shown below:

[0029] In the formula, To adjust the sliding window size, This represents the standard deviation (reflecting the degree of data fluctuation). The mean (representing the average level of the data). This is the sensitivity adjustment coefficient (used to control the response strength to changes in data, for example, set to 2.0). Set the base window size (to ensure the minimum sampling size, for example, set it to 10). To round up; Based on the adjusted sliding window size, statistical features (such as peak value, root mean square value, trend slope, etc.) within the window are extracted and compared with the statistical features of the previous window. If the feature difference value exceeds the preset difference threshold, it is determined that a state change has occurred at the sensor level, and a first state change result is generated. The preset difference threshold can be dynamically configured according to the data type, and the first state change result can include the change type, change magnitude, and occurrence time.

[0030] For 3D model point cloud data, a local comparison algorithm based on feature points can be used. The specific process is as follows: First, the feature point set A of the 3D model at the previous time step and the feature point set B of the 3D model at the current time step can be obtained. Feature points can be extracted using algorithms such as FPFH (Fast Point Feature Histogram). Calculate the Euclidean distance from each point in the 3D model feature point set A at the previous time step to the nearest point in the 3D model feature point set B at the current time step, and take the supremum (i.e., the maximum value) of the Euclidean distance to obtain the positive distance, denoted as: ;in, Let be the positive Hausdorff distance between the feature point set A of the 3D model at the previous time step and the feature point set B of the 3D model at the current time step. This is to take the supremum (i.e. the maximum value) of all points a in the feature point set A of the 3D model at the previous time step. To find the infimum (i.e. the minimum value) of all points b in the feature point set B of the 3D model at the current moment. Let be the Euclidean distance between points a and b.

[0031] Calculate the Euclidean distance from each point in the current 3D model feature point set B to the nearest point in the previous 3D model feature point set A, and take the supremum (i.e., the maximum value) of the Euclidean distance to obtain the back-to-back distance, denoted as: ;in, Let be the inverse Hausdorff distance from the current 3D model feature point set B to the previous 3D model feature point set A. To find the supremum (i.e. the maximum value) of all points b in the feature point set B of the 3D model at the current moment. To find the infimum (i.e. minimum value) of all points a in the feature point set A of the 3D model at the previous time step. Let be the Euclidean distance between points a and b.

[0032] Extract the maximum value between the forward distance and the reverse distance as the final Hausdorff distance. The specific formula is as follows:

[0033] In the formula, Let be the Hausdorff distance between the feature point set A of the 3D model at the previous time step and the feature point set B of the 3D model at the current time step. Let be the positive Hausdorff distance between the feature point set A of the 3D model at the previous time step and the feature point set B of the 3D model at the current time step. Let be the inverse Hausdorff distance between the current set of 3D model feature points B and the previous set of 3D model feature points A.

[0034] If the Hausdorff distance is greater than the preset geometric change threshold (e.g., 5.0 mm), it is determined that the 3D model has undergone significant geometric changes (such as component displacement or deformation), and a second state change result is generated. The second state change result includes the Hausdorff distance value, the coordinates of the changed position, and the degree of change.

[0035] To avoid false positives or false negatives caused by a single data source, this embodiment can fuse the two results mentioned above based on a weight allocation strategy: It can assess the reliability confidence of time-series sensor data based on sensor health status and signal-to-noise ratio, and assess the accuracy confidence of 3D model point cloud data based on scanning resolution and point cloud density; If the first state change result is consistent with the second state change result (i.e., both are detected or neither is detected), the state change conclusion is directly confirmed. When the results of the first state change and the second state change are inconsistent (e.g., the sensor shows an abnormality but the model remains unchanged, or vice versa), the reliability confidence and the accuracy confidence are compared. If the reliability confidence is greater than the accuracy confidence, the detection result corresponding to the reliability confidence is taken as the state change conclusion. If the accuracy confidence is greater than the reliability confidence, the detection result corresponding to the accuracy confidence is taken as the state change conclusion. Based on the final confirmed state change conclusions, a comprehensive state difference report is generated, which includes the change type (such as mechanical displacement, parameter mutation), change degree (quantitative value), change location (spatial coordinates or sensor ID), and timestamp. This report is then output as the initial change set to the next module.

[0036] This application balances the requirements of detection sensitivity and stability through a dynamic adjustment mechanism of sliding window, and uses Hausdorff distance to robustly detect geometric changes in the 3D model, effectively solving the difficulty of different sensor data representation differences in the process of multimodal data fusion.

[0037] Step 103: Transform the initial change set into the current change feature vector, input it into the semantic analysis network based on the scaling dot product attention mechanism, calculate the correlation between the current change feature vector and the key-value pairs in the historical change pattern library, determine the change importance score, and prioritize each change item in the initial change set according to the change importance score to generate a semantically labeled change set.

[0038] like Figure 2 As shown, this step is performed by the change semantic analysis module, which serves as the intelligent decision-making component of the digital twin incremental algorithm, responsible for assessing the importance of changes. This network, based on an attention mechanism and a multi-head attention layer, is capable of learning and identifying key change patterns from multi-dimensional features.

[0039] Each change item in the initial change set can be transformed into a current change feature vector. Feature encoding can include: change type encoding, numerical feature encoding, and spatiotemporal feature encoding.

[0040] Among them, change type encoding can use One-hot encoding; numerical feature encoding can normalize change magnitude, Hausdorff distance value, etc. and map them into vectors; spatiotemporal feature encoding can use sine and cosine functions to handle time periodicity and directly use three-dimensional coordinates as spatial features.

[0041] The current changed feature vector can be input into a semantic analysis network based on the Scaled Dot-ProductAttention mechanism. The semantic analysis network internally maintains a historical change pattern library, which can contain historical feature vectors (as keys) and corresponding historical importance labels (as values).

[0042] The current change feature vector can be used as the query vector. The historical feature vectors in the historical change pattern library are encoded into key vector matrices. The importance labels of historical changes (such as manually marked scores of 1-10 or fault levels) are encoded into value vector matrices. ; Calculate query vector With key vector matrix dot product results Since the dot product result increases with increasing dimension, it may cause the Softmax function to enter the gradient flattening region. Therefore, a scaling factor can be used. Scaling the dot product results prevents gradient vanishing and makes the training process more stable; The scaled dot product result is normalized using the Softmax function to obtain attention weight values, which represent the degree of correlation between the current change and each historical pattern. The Softmax function can convert any real vector into a probability distribution, ensuring that the sum of all scores is 1 and that each value is in the interval (0,1), so that each score can be interpreted as the "weight" of the correlation between the current change and each historical pattern.

[0043] The importance score is obtained by performing a weighted summation operation on the attention weight values ​​and the value vector matrix. The specific formula is as follows:

[0044] In the formula, To change the importance score, The current change feature vector, i.e., the query vector Query, is an abstract representation of the detected state change (such as feature point displacement or sudden sensor reading changes), with dimensions of [dimensional value missing]. , Encode the historical feature vectors in the historical change pattern library, such as the "identifier" or "index" vectors of all patterns in the historical change pattern library, i.e., the key vector matrix Key, with dimensions of . , Scaling factor The importance labels for historical changes are encoded as value vector matrices. For example, the "specific content" or "value" vector of historical change patterns associated with each Key typically contains deep semantic information such as the importance tags and scope of impact of historical changes, with dimensions of [missing information]. .

[0045] The core logic of the algorithm in this application is to infer the importance of the current change by comparing its similarity to historical patterns. The more similar the query is to a certain key, the higher the corresponding attention weight, thereby extracting more semantic information related to the pattern (such as high importance) from the value, and finally synthesizing a score vector representing the degree of importance of the current change.

[0046] Based on the calculated change importance score, and comparing it with preset thresholds (e.g., a high priority threshold of 0.8 and a medium priority threshold of 0.5), each change item in the initial change set is prioritized, generating a semantically tagged change set. Simultaneously, the system can output the historical pattern ID with the highest matching degree, providing interpretable evidence for decision-making.

[0047] Step 104: Take each change item in the semantically labeled change set as a change task to be scheduled. Calculate the priority score of each change task based on the change importance score, business urgency coefficient, and estimated resource consumption. Combine the resource constraints of the current edge node to generate the optimal incremental update execution sequence. The priority score is positively correlated with the change importance score and business urgency coefficient, and negatively correlated with the estimated resource consumption.

[0048] like Figure 2 As shown, this step can be executed by the priority scheduling module, which is the core scheduling component of the digital twin incremental algorithm. It can intelligently schedule and allocate computing tasks based on the importance of changes, business urgency, and resource constraints.

[0049] Specifically, each change item in the semantically tagged change set can be considered as a task to be scheduled. For the i-th change task, its change importance score, business urgency coefficient (defined by business rules, such as 1.0 for real-time interactive tasks and 0.5 for batch tasks), and estimated resource consumption (CPU, memory, or bandwidth consumption predicted based on historical data) are comprehensively considered. A multi-dimensional weighted formula is used to calculate the priority score, whereby the multi-dimensional weighted formula is as follows:

[0050] In the formula, Let i be the priority score for the i-th change task. As an importance weighting factor, Let i be the change importance score for the i-th change task. As an urgency weighting factor, Let i be the business urgency coefficient of the i-th change task. As a resource consumption penalty factor, This represents the estimated resource consumption for the i-th change task.

[0051] Each change task can be assigned a corresponding virtual weight according to its priority score, and the change tasks can be sorted in descending order of virtual weight to ensure that high-priority tasks get more scheduling opportunities, while low-priority tasks are not completely starved.

[0052] One possible approach is to map priority scores to scheduling weights (either directly or by scaling), generate priority scheduling queues based on the scheduling weights, and ensure that high-weight tasks are scheduled more often within a given period.

[0053] The change task can be selected sequentially from the head of the priority scheduling queue as the current candidate change task; check whether the remaining available resources of the current edge node (such as the remaining number of CPU cores and memory capacity) meet the estimated resource consumption of the current candidate change task. If the conditions are met, the current candidate change task is added to the incremental update execution sequence, and the actual consumed resources are deducted from the total resources (updating the remaining available resources); if the conditions are not met, the current candidate change task is marked as waiting or downgraded (e.g., delayed execution), and the next change task is selected from the priority scheduling queue as the new current candidate change task. The above process is repeated until the priority scheduling queue is traversed or the current scheduling cycle ends.

[0054] Step 105: Perform triple verification on the incremental update data in the incremental update execution sequence. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verifications include data integrity verification, logical rationality verification, and performance impact assessment.

[0055] like Figure 2 As shown, this step can be performed by the consistency verification module, which provides comprehensive data verification functions for the incremental algorithm of digital twins based on edge cloud scenarios, ensuring that the updated digital twin is consistent with the physical entity.

[0056] Each incremental update data item in the incremental update execution sequence can be verified. To ensure that data has not been tampered with or corrupted during transmission or processing, integrity verification can be performed on the incremental update data.

[0057] Incremental update data can be used as input to generate a fixed-length (256-bit) unique digital fingerprint, i.e., a SHA-256 hash value, using the SHA-256 algorithm. During verification, the HA-256 hash value can be used. Compared with the raw hash value provided by the data source or pre-stored Comparison, if and only if Data integrity is only confirmed when the hash value is changed; otherwise, it is determined that the data may have been tampered with or corrupted. This is because even the slightest alteration will result in a significantly different hash value, thus effectively identifying data anomalies.

[0058] The system can perform logical rationality checks on incremental update data in the incremental update execution sequence based on constraints set by business rules. This check focuses on whether the data is correct in its business context. This is achieved through a set of predefined business rules. In practice, these rules are a set of logical judgments. Their judgment logic can be formalized as follows:

[0059] In the formula, It is a universal quantifier. For the i-th business rule, For a set of business rules, Update data incrementally.

[0060] That is, for the set of business rules Each business rule in Updated data All conditions must be met to make the logical judgment true. For example, the rule might be that the battery temperature is less than a safety threshold or that the device status is true. {Normal operation, standby, shutdown}, etc. The core of this process is to ensure that all business constraints are met and that the logical state of the data is valid.

[0061] Performance impact assessment and verification can predict potential system performance changes caused by updates through simulation. Its core is a lightweight simulation model F, with the specific formula shown below:

[0062] In the formula, To quantify the performance impact indicators, It is a lightweight simulation model. To update data incrementally, This refers to the current system environment and status.

[0063] The quantified performance impact metrics can include expected response time latency, resource utilization improvement, etc. The purpose is to provide a quantitative assessment, offering a basis for decision-making regarding how system performance will change if this update is executed. If the quantified performance impact metrics are within acceptable limits (e.g., latency increase <10ms), the update passes verification.

[0064] If all three verifications pass, the update is confirmed to be effective, the data is written to the digital twin database, and feedback is sent to the physical entity as needed (such as sending control commands).

[0065] If any of the triple checks fails, a compensation mechanism is immediately triggered to ensure system consistency. The specific logic is as follows: The system monitors the current network communication status (such as bandwidth and packet loss rate) and the remaining time before the next scheduled regular update in real time. It compares the remaining time with the preset time threshold required for a full update. If the network communication status is good (i.e., the full transmission bandwidth requirement is met) and the remaining time is greater than the time threshold required for a full update, it determines that the current incremental data is unreliable and there is a time window for repair. At this time, the incremental update data that failed the verification can be discarded, and a full data update operation can be directly requested from the central cloud or local backup to replace the current digital twin state with the complete dataset, so as to completely eliminate the risk of inconsistency. If the network communication is poor (i.e., the full transmission requirement is not met), or if the remaining time is less than or equal to the time threshold required for a full update, i.e., there is not enough time to complete the full update, then a degradation strategy will be executed. The degradation strategy may include: masking abnormal data, i.e., masking incremental update data that failed in this verification; maintaining the old state, i.e., keeping the digital twin in the valid state of the previous moment to ensure that the system does not present an erroneous state; logging and retrying, i.e., logging the abnormal log of verification failure, which may include the reason for failure, data fingerprint, etc., and marking this update task as pending synchronization. A timed retry mechanism is initiated, and when network communication is detected to have recovered or the next update cycle has arrived, the data synchronization request is re-initiated (this may involve attempting to fetch incremental packets again or switching to low-precision mode synchronization).

[0066] Finally, a consistency verification report containing detailed results of each verification can be generated for auditing and problem tracing. Through this rigorous process, the consistency verification module provides key assurance for the reliable evolution of digital twins.

[0067] In summary, according to the incremental update method for digital twins provided in this application, compared with the existing technology, this application can collect multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and 3D model point cloud data; use an adaptive sliding window algorithm to perform dynamic difference calculation on the time-series sensor data to obtain a first state change result; use the Hausdorff distance algorithm to perform geometric change recognition on the 3D model point cloud data to obtain a second state change result; and perform multimodal fusion decision processing on the first and second state change results based on a weight allocation strategy to generate an initial change set; transform the initial change set into a current change feature vector, input it into a semantic analysis network based on a scaled dot product attention mechanism, and determine the change importance score by calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, and then apply the change... Importance scores are used to prioritize each change item in the initial change set, generating a semantically labeled change set. Each change item in the semantically labeled change set is treated as a change task to be scheduled. Based on the change importance score, business urgency coefficient, and estimated resource consumption, the priority score of each change task is calculated. Combined with the resource constraints of the current edge node, an optimal incremental update execution sequence is generated. The priority score is positively correlated with the change importance score and business urgency coefficient, and negatively correlated with the estimated resource consumption. The incremental update data in the incremental update execution sequence undergoes triple verification. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verifications include data integrity verification, logical rationality verification, and performance impact assessment.

[0068] By employing the above technical solutions, this application utilizes an adaptive sliding window algorithm and a Hausdorff distance algorithm to accurately filter out TB-level static data or unchanged models, extracting only a very small amount of changed data (initial change set). This reduces the amount of data to be processed from TB-level to MB or even KB-level, fundamentally alleviating the storage and computing pressure on edge nodes and avoiding resource exhaustion and system paralysis. This application also employs multimodal data fusion technology to perform differential calculations on time-series sensor data and 3D model point cloud data separately. This targeted algorithm is more efficient than general full-scale reconstruction, significantly shortening data acquisition time.

[0069] Even under resource constraints, this application does not employ a simple "first-come, first-served" or "full-processing" approach. Instead, it introduces a semantic analysis network based on a scaled dot product attention mechanism. The system can identify the importance of changes (e.g., obstacle recognition data in autonomous driving has higher priority than background environment data) and prioritizes tasks using a multi-dimensional weighted formula. By prioritizing high-scoring tasks and deferred low-scoring tasks, it ensures that critical business operations receive priority computing resources under limited resources, thereby meeting the low-latency requirements of critical scenarios.

[0070] When generating the update sequence, this application dynamically determines the resource constraints of the current edge nodes. Updates are only executed when the remaining resources meet the task consumption estimate; otherwise, they are marked as pending. This resource-constrained strategy prevents system blocking or crashes caused by resource contention.

[0071] To address network uncertainties, this application designs a consistency verification and compensation mechanism. If verification fails, the system will choose between "full update remediation" or "degradation strategy (maintaining core functions)" based on network and time conditions. This ensures that the system can maintain the real-time performance of core functions even during network fluctuations or temporary resource shortages, preventing complete system failure.

[0072] Based on the above Figure 1 The specific implementation of the method shown in this embodiment provides a digital twin incremental update device, such as... Figure 3 As shown, the device includes: a data acquisition module 31, a first calculation module 32, a marking module 33, a second calculation module 34, and a verification module 35; The acquisition module 31 is used to acquire multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and three-dimensional model point cloud data; The first calculation module 32 is used to perform dynamic differential calculation on the time-series sensor data using an adaptive sliding window algorithm to obtain a first state change result, to perform geometric change identification on the three-dimensional model point cloud data using a Hausdorff distance algorithm to obtain a second state change result, and to perform multimodal fusion decision processing on the first state change result and the second state change result based on a weight allocation strategy to generate an initial change set. The tagging module 33 is used to transform the initial change set into a current change feature vector, input it into a semantic analysis network based on a scaling dot product attention mechanism, determine the change importance score by calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, and prioritize each change item in the initial change set according to the change importance score to generate a change set with semantic tags. The second calculation module 34 is used to take each change item in the semantically labeled change set as a change task to be scheduled, calculate the priority score of each change task based on the change importance score, the business urgency coefficient and the estimated resource consumption, and generate the optimal incremental update execution sequence in combination with the resource constraints of the current edge node. The priority score is positively correlated with the change importance score and the business urgency coefficient, and the priority score is negatively correlated with the estimated resource consumption. The verification module 35 is used to perform triple verification on the incremental update data in the incremental update execution sequence. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verification includes data integrity verification, logical rationality verification, and performance impact assessment.

[0073] In specific application scenarios, the first calculation module 32 can be used to calculate the mean and standard deviation of the time-series sensor data within the current time window in real time. The sliding window size is dynamically adjusted based on the mean and the standard deviation using a dynamic window size adjustment formula. Based on the adjusted sliding window size, the statistical features within the window are extracted and compared with the statistical features of the previous window. If the feature difference value exceeds a preset difference threshold, the first state change result is generated. The formula for dynamically adjusting the size of the sliding window is as follows:

[0074] In the formula, To adjust the sliding window size, The standard deviation is... The mean, This is the sensitivity adjustment coefficient. Base window size, This is for rounding up.

[0075] In specific application scenarios, the first calculation module 32 can be used to obtain the feature point set of the three-dimensional model at the previous moment and the feature point set of the three-dimensional model at the current moment. The upper bound of the distance from each point in the 3D model feature point set of the previous time step to the nearest point in the 3D model feature point set of the current time step is calculated to obtain the positive distance; Calculate the supremum of the distance from each point in the current 3D model feature point set to the nearest point in the previous 3D model feature point set, and obtain the inverse distance; The maximum value between the forward distance and the reverse distance is extracted as the Hausdorff distance; If the Hausdorff distance is greater than a preset geometric change threshold, the 3D model is determined to have undergone a significant change, and the second state change result is generated.

[0076] In specific application scenarios, the first calculation module 32 can be used to obtain the reliability confidence level of the time-series sensor data and the accuracy confidence level of the three-dimensional model point cloud data; When the first state change result is consistent with the second state change result, the state change conclusion is directly confirmed; When the first state change result is inconsistent with the second state change result, the reliability confidence level and the accuracy confidence level are compared. If the reliability confidence level is greater than the accuracy confidence level, the detection result corresponding to the reliability confidence level is taken as the state change conclusion. If the accuracy confidence level is greater than the reliability confidence level, the detection result corresponding to the accuracy confidence level is taken as the state change conclusion. Based on the status change conclusions, a comprehensive status differential report containing change type, change degree, change location, and timestamp is generated as the initial change set.

[0077] In specific application scenarios, the tagging module 33 can be used to use the current change feature vector as a query vector, encode the historical feature vector in the historical change pattern library into a key vector matrix, and encode the importance tags of historical changes in the historical change pattern library into a value vector matrix. Calculate the dot product of the query vector and the key vector matrix, and scale the dot product using a scaling factor; The scaled dot product result is normalized using the Softmax function to obtain the attention weight values; The importance score of the change is obtained by performing a weighted summation operation on the attention weight value and the value vector matrix.

[0078] In specific application scenarios, the second calculation module 34 can be used to calculate the priority score of each change task based on the change importance score, business urgency coefficient and estimated resource consumption using a multi-dimensional weighted formula. The changed tasks are sorted in descending order of priority score to construct a priority scheduling queue; The change task is selected sequentially from the head of the priority scheduling queue as the current candidate change task; Detect whether the remaining available resources of the current edge node meet the estimated resource consumption of the current candidate change task; If the conditions are met, the current candidate change task is added to the incremental update execution sequence, and the remaining available resources are updated. If the conditions are not met, the current candidate change task is marked as waiting or downgraded, and the next change task is selected from the priority scheduling queue as the new current candidate change task until the priority scheduling queue is traversed or the scheduling cycle ends. The multi-dimensional weighting formula is as follows:

[0079] In the formula, Let i be the priority score for the i-th change task. As an importance weighting factor, Let i be the change importance score for the i-th change task. As an urgency weighting factor, Let i be the business urgency coefficient of the i-th change task. As a resource consumption penalty factor, This represents the estimated resource consumption for the i-th change task.

[0080] In specific application scenarios, the verification module 35 can be used to detect the current network communication status and the remaining time until the next scheduled regular update in real time; Compare the remaining time with the preset time threshold required for a full update; If the network communication status meets the full transmission requirement and the remaining time is greater than the time threshold required for the full update, then the incremental update data that failed the current verification is discarded, and a full data update operation is performed to replace the current digital twin state. If the network communication status does not meet the requirements for full transmission, or if the remaining time is less than or equal to the time threshold required for the full update, a degradation strategy is executed. The degradation strategy includes: blocking the incremental update data that failed the verification, controlling the digital twin to maintain the valid state of the previous moment, recording the abnormal log of the verification failure, marking the current update task as pending synchronization, and starting a timed retry mechanism. When the network communication status is detected to be restored or the next update cycle is reached, the data synchronization request is re-initiated.

[0081] It should be noted that other corresponding descriptions of the functional units involved in the digital twin incremental update device provided in this embodiment can be found in [reference]. Figure 1 The corresponding description in [the document] will not be repeated here.

[0082] Based on the above, Figure 1 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 The method shown.

[0083] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0084] Based on the above, Figure 1 The method shown, and Figure 3 To achieve the above objectives, the present application also provides an electronic device, comprising a storage medium and a processor; the storage medium for storing a computer program; and the processor for executing the computer program to implement the above-described virtual device embodiments. Figure 1 The method shown.

[0085] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0086] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0087] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the execution of the digital twin incremental update program and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the digital twin incremental update physical device.

[0088] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. By applying the solution of this embodiment, compared with the existing technology, this application can collect multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and three-dimensional model point cloud data; use an adaptive sliding window algorithm to perform dynamic differential calculation on the time-series sensor data to obtain a first state change result; use the Hausdorff distance algorithm to perform geometric change recognition on the three-dimensional model point cloud data to obtain a second state change result; and perform multimodal fusion decision processing on the first state change result and the second state change result based on a weight allocation strategy to generate an initial change set; convert the initial change set into a current change feature vector, input it into a semantic analysis network based on a scaled dot product attention mechanism, and determine the change importance score by calculating the correlation between the current change feature vector and the key-value pairs in the historical change pattern library, and then adjust the initial change set according to the change importance score. Each change item in the initial change set is prioritized to generate a semantically tagged change set. Each change item in the semantically tagged change set is treated as a change task to be scheduled. Based on the change importance score, business urgency coefficient, and estimated resource consumption, the priority score of each change task is calculated. Combined with the resource constraints of the current edge node, an optimal incremental update execution sequence is generated. The priority score is positively correlated with the change importance score and business urgency coefficient, and negatively correlated with the estimated resource consumption. The incremental update data in the incremental update execution sequence undergoes triple verification. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verifications include data integrity verification, logical rationality verification, and performance impact assessment.

[0089] By employing the above technical solutions, this application utilizes an adaptive sliding window algorithm and a Hausdorff distance algorithm to accurately filter out TB-level static data or unchanged models, extracting only a very small amount of changed data (initial change set). This reduces the amount of data to be processed from TB-level to MB or even KB-level, fundamentally alleviating the storage and computing pressure on edge nodes and avoiding resource exhaustion and system paralysis. This application also employs multimodal data fusion technology to perform differential calculations on time-series sensor data and 3D model point cloud data separately. This targeted algorithm is more efficient than general full-scale reconstruction, significantly shortening data acquisition time.

[0090] Even under resource constraints, this application does not employ a simple "first-come, first-served" or "full-processing" approach. Instead, it introduces a semantic analysis network based on a scaled dot product attention mechanism. The system can identify the importance of changes (e.g., obstacle recognition data in autonomous driving has higher priority than background environment data) and prioritizes tasks using a multi-dimensional weighted formula. By prioritizing high-scoring tasks and deferred low-scoring tasks, it ensures that critical business operations receive priority computing resources under limited resources, thereby meeting the low-latency requirements of critical scenarios.

[0091] When generating the update sequence, this application dynamically determines the resource constraints of the current edge nodes. Updates are only executed when the remaining resources meet the task consumption estimate; otherwise, they are marked as pending. This resource-constrained strategy prevents system blocking or crashes caused by resource contention.

[0092] To address network uncertainties, this application designs a consistency verification and compensation mechanism. If verification fails, the system will choose between "full update remediation" or "degradation strategy (maintaining core functions)" based on network and time conditions. This ensures that the system can maintain the real-time performance of core functions even during network fluctuations or temporary resource shortages, preventing complete system failure.

[0093] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0094] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for incremental updating a digital twin, characterized in that, The method includes: Real-time acquisition of multimodal data of physical entities, wherein the multimodal data includes at least time-series sensor data and 3D model point cloud data; The time-series sensor data is dynamically differentially calculated using an adaptive sliding window algorithm to obtain a first state change result. The 3D model point cloud data is geometrically altered using a Hausdorff distance algorithm to obtain a second state change result. The first and second state change results are then subjected to multimodal fusion decision processing based on a weight allocation strategy to generate an initial change set. The initial change set is transformed into a current change feature vector, which is then input into a semantic analysis network based on a scaled dot product attention mechanism. By calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, a change importance score is determined. Based on the change importance score, each change item in the initial change set is prioritized to generate a semantically labeled change set. Each change item in the semantically labeled change set is taken as a change task to be scheduled. Based on the change importance score, business urgency coefficient and estimated resource consumption, the priority score of each change task is calculated. Combined with the resource constraints of the current edge node, the optimal incremental update execution sequence is generated. The priority score is positively correlated with the change importance score and the business urgency coefficient, and negatively correlated with the estimated resource consumption. The incremental update data in the incremental update execution sequence is subject to triple verification. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verifications include data integrity verification, logical rationality verification, and performance impact assessment.

2. The method according to claim 1, characterized in that, The step of using an adaptive sliding window algorithm to perform dynamic differential calculation on the time-series sensor data to obtain the first state change result includes: Calculate the mean and standard deviation of the time-series sensor data within the current time window in real time; The sliding window size is dynamically adjusted based on the mean and the standard deviation using a dynamic window size adjustment formula. Based on the adjusted sliding window size, the statistical features within the window are extracted and compared with the statistical features of the previous window. If the feature difference value exceeds a preset difference threshold, the first state change result is generated. The formula for dynamically adjusting the size of the sliding window is as follows: In the formula, To adjust the sliding window size, The standard deviation is... The mean, This is the sensitivity adjustment coefficient. Base window size, This is for rounding up.

3. The method according to claim 1, characterized in that, The step of using the Hausdorff distance algorithm to identify geometric changes in the 3D model point cloud data to obtain the second state change result includes: Obtain the feature point set of the 3D model at the previous time step and the feature point set of the 3D model at the current time step; The upper bound of the distance from each point in the 3D model feature point set of the previous time step to the nearest point in the 3D model feature point set of the current time step is calculated to obtain the positive distance; Calculate the supremum of the distance from each point in the current 3D model feature point set to the nearest point in the previous 3D model feature point set, and obtain the inverse distance; The maximum value between the forward distance and the reverse distance is extracted as the Hausdorff distance; If the Hausdorff distance is greater than a preset geometric change threshold, the 3D model is determined to have undergone a significant change, and the second state change result is generated.

4. The method according to claim 1, characterized in that, The multimodal fusion decision processing based on the weight allocation strategy for the first state change result and the second state change result generates an initial change set, including: Obtain the reliability confidence level of the time-series sensor data and the accuracy confidence level of the 3D model point cloud data; When the first state change result is consistent with the second state change result, the state change conclusion is directly confirmed; When the first state change result is inconsistent with the second state change result, the reliability confidence level and the accuracy confidence level are compared. If the reliability confidence level is greater than the accuracy confidence level, the detection result corresponding to the reliability confidence level is taken as the state change conclusion. If the accuracy confidence level is greater than the reliability confidence level, the detection result corresponding to the accuracy confidence level is taken as the state change conclusion. Based on the status change conclusions, a comprehensive status differential report containing change type, change degree, change location, and timestamp is generated as the initial change set.

5. The method according to claim 1, characterized in that, The process of transforming the initial change set into a current change feature vector and inputting it into a semantic analysis network based on a scaled dot product attention mechanism, and determining the change importance score by calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, includes: The current change feature vector is used as the query vector, the historical feature vectors in the historical change pattern library are encoded as key vector matrices, and the importance tags of historical changes in the historical change pattern library are encoded as value vector matrices. Calculate the dot product of the query vector and the key vector matrix, and scale the dot product using a scaling factor; The scaled dot product result is normalized using the Softmax function to obtain the attention weight values; The importance score of the change is obtained by performing a weighted summation operation on the attention weight value and the value vector matrix.

6. The method according to claim 1, characterized in that, The step involves calculating the priority score for each change task based on the change importance score, business urgency coefficient, and estimated resource consumption. Combined with the resource constraints of the current edge node, an optimal incremental update execution sequence is generated, including: Based on the change importance score, business urgency coefficient, and estimated resource consumption, a priority score for each change task is calculated using a multi-dimensional weighted formula. The changed tasks are sorted in descending order of priority score to construct a priority scheduling queue; The change task is selected sequentially from the head of the priority scheduling queue as the current candidate change task; Detect whether the remaining available resources of the current edge node meet the estimated resource consumption of the current candidate change task; If the conditions are met, the current candidate change task is added to the incremental update execution sequence, and the remaining available resources are updated. If the conditions are not met, the current candidate change task is marked as waiting or downgraded, and the next change task is selected from the priority scheduling queue as the new current candidate change task until the priority scheduling queue is traversed or the scheduling cycle ends. The multi-dimensional weighting formula is as follows: In the formula, Let i be the priority score for the i-th change task. As an importance weighting factor, Let i be the change importance score for the i-th change task. As an urgency weighting factor, Let be the business urgency coefficient for the i-th change task. As a resource consumption penalty factor, This represents the estimated resource consumption for the i-th change task.

7. The method according to claim 1, characterized in that, If any of the three checks fails, a compensation mechanism is triggered to ensure system consistency, including: Real-time monitoring of current network communication status and remaining time until the next scheduled routine update; Compare the remaining time with the preset time threshold required for a full update; If the network communication status meets the full transmission requirement and the remaining time is greater than the time threshold required for the full update, then the incremental update data that failed the current verification is discarded, and a full data update operation is performed to replace the current digital twin state. If the network communication status does not meet the requirements for full transmission, or if the remaining time is less than or equal to the time threshold required for the full update, a degradation strategy is executed. The degradation strategy includes: blocking the incremental update data that failed the verification, controlling the digital twin to maintain the valid state of the previous moment, recording the abnormal log of the verification failure, marking the current update task as pending synchronization, and starting a timed retry mechanism. When the network communication status is detected to be restored or the next update cycle is reached, the data synchronization request is re-initiated.

8. A digital twin incremental update device, characterized in that, include: The acquisition module is used to acquire multimodal data of physical entities in real time, wherein the multimodal data includes at least time-series sensor data and three-dimensional model point cloud data; The first calculation module is used to perform dynamic differential calculation on the time-series sensor data using an adaptive sliding window algorithm to obtain a first state change result, to perform geometric change identification on the three-dimensional model point cloud data using a Hausdorff distance algorithm to obtain a second state change result, and to perform multimodal fusion decision processing on the first state change result and the second state change result based on a weight allocation strategy to generate an initial change set. The tagging module is used to transform the initial change set into a current change feature vector, input it into a semantic analysis network based on a scaled dot product attention mechanism, determine the change importance score by calculating the correlation between the current change feature vector and key-value pairs in the historical change pattern library, and prioritize each change item in the initial change set according to the change importance score to generate a semantically tagged change set. The second calculation module is used to take each change item in the semantically labeled change set as a change task to be scheduled, calculate the priority score of each change task based on the change importance score, the business urgency coefficient and the estimated resource consumption, and generate the optimal incremental update execution sequence in combination with the resource constraints of the current edge node. The priority score is positively correlated with the change importance score and the business urgency coefficient, and negatively correlated with the estimated resource consumption. The verification module is used to perform triple verification on the incremental update data in the incremental update execution sequence. If all three verifications pass, the update is confirmed to be effective and feedback is sent to the physical entity or digital twin. If any of the three verifications fails, a compensation mechanism is triggered to ensure system consistency. The triple verification includes data integrity verification, logical rationality verification, and performance impact assessment.

9. An electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the digital twin incremental update method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the digital twin incremental update method according to any one of claims 1 to 7.

Images