A cloud edge collaboration-based disaster recovery data transmission method and system

By aligning disaster recovery data in edge node slices and using a consistent hash function to determine anchor nodes, non-anchor nodes only upload differing data, thus solving the problem of redundant data transmission in distributed systems and improving transmission efficiency and data recovery capabilities under link failures.

CN122578611BActive Publication Date: 2026-11-10JIAJIE TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611067412.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-17
Publication Date
2026-11-10
Estimated Expiration
2046-07-17

AI Technical Summary

Technical Problem

In application scenarios such as distributed smart grids or wide-area vehicle-road cooperation, regional physical events cause multiple edge nodes to concurrently transmit a large amount of disaster recovery data with global redundancy, which increases the amount of data transmitted over the wide area network and prolongs the synchronization cycle to the central cloud node.

Method used

By slicing and aligning data when disaster recovery data traffic surges at edge nodes, spatiotemporal features of events are extracted, and anchor nodes are determined using a consistent hash function. Non-anchor nodes only upload differential data, while anchor nodes uniformly upload redundant data. By combining hash tree comparison and differential upload strategies, duplicate data transmission is reduced.

Benefits of technology

It reduces the amount of globally redundant data transmitted concurrently across edge nodes, improves the efficiency of disaster recovery data transmission, reduces network bandwidth consumption, and enables rapid data recovery and high-quality transmission in the event of link failure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122578611B_ABST
    Figure CN122578611B_ABST
Patent Text Reader

Abstract

The application discloses a cloud edge coordination-based disaster data transmission method and system, and relates to the field of digital data processing.In the method, when the edge node monitors that the data flow exceeds a threshold value, the local disaster data is sliced and aligned, the event space-time features are extracted and standardized, and then the anchor node is independently determined through consistent hash mapping.The non-anchor node calculates the data slice fingerprint and sends it to the anchor node; the anchor node identifies the global redundant data across nodes and the difference data specific to each node through centralized comparison of the fingerprints, and issues a differentiated upload strategy accordingly.Finally, the non-anchor node only uploads the difference data to the central cloud node, and the anchor node uniformly uploads the global redundant data and the local difference data.The application is used for reducing the amount of global redundant data transmitted across edge nodes in parallel, reducing the overall transmission bandwidth consumption of the wide area network, thereby shortening the transmission period of regional disaster data synchronization to the central cloud node.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of electronic digital data processing, and in particular relates to a disaster recovery data transmission method and system based on cloud-edge collaboration. Background Technology

[0002] With the rapid development of information technology, the security and business continuity of massive amounts of data have become increasingly important, and disaster recovery systems have become a core infrastructure for ensuring data security. Traditional disaster recovery data transmission typically involves directly uploading data generated by local terminals to a central cloud node for centralized storage. However, when dealing with high-frequency data generated by massive numbers of devices, this direct end-to-cloud transmission mode consumes significant wide area network bandwidth resources, and long-distance data routing and forwarding can easily lead to high network latency, resulting in reduced transmission efficiency of disaster recovery data.

[0003] Related technologies have introduced a cloud-edge collaborative architecture for processing and transmitting disaster recovery data. Edge computing nodes are deployed close to the data source at the edge, where they prioritize receiving and caching disaster recovery data generated by local terminal devices. The edge nodes then utilize their computing power to perform preprocessing operations such as deduplication and compression on this data. The preprocessed disaster recovery data is then transmitted in batches from the edge nodes to the central cloud nodes for persistent storage as needed. This cloud-edge collaborative architecture reduces the amount of data transmitted across the wide area network by offloading some data processing tasks to the edge, thereby reducing network bandwidth pressure and improving overall data transmission efficiency.

[0004] However, in applications such as distributed smart grids or wide-area vehicle-road cooperative systems, when a regional physical event occurs, multiple adjacent edge nodes within the same geographical area simultaneously receive a large amount of highly spatially correlated device status data. When related technologies are implemented in these scenarios, each edge node independently performs deduplication and compression preprocessing on the received data based on its local cache, and then establishes uplinks to transmit it to the central cloud node. Under this mutually isolated vertical processing and transmission mechanism, similar cross-node data triggered by the same regional event is treated as independent valid data and processed and uploaded by each edge node. This results in a large amount of globally redundant disaster recovery data being concurrently transmitted across multiple edge-cloud uplinks during a regional event, thereby increasing the overall data transmission volume of the wide area network and prolonging the transmission cycle of regional disaster recovery data synchronization to the central cloud node. Summary of the Invention

[0005] This application provides a disaster recovery data transmission method and system based on cloud-edge collaboration, which reduces the amount of globally redundant data transmitted concurrently across edge nodes, reduces the overall transmission bandwidth consumption of the wide area network, and thus shortens the transmission cycle of regional disaster recovery data synchronization to the central cloud node.

[0006] In the first aspect, this application provides a disaster recovery data transmission method based on cloud-edge collaboration. When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it slices and aligns the locally cached disaster recovery data based on a preset time window and extracts the event spatiotemporal features of the disaster recovery data respectively. Each edge node performs spatiotemporal gridding dimensionality reduction on the event spatiotemporal features to obtain standardized event spatiotemporal features; Each edge node inputs the standardized event spatiotemporal features into a consistent hash function, which maps them in a virtual ring composed of edge nodes, and independently determines the anchor node corresponding to the event spatiotemporal features. If the current edge node is not an anchor node, calculate the data fingerprint of each disaster recovery data slice after alignment and send the data fingerprint to the anchor node; Anchor nodes compare the received data fingerprints with local data fingerprints to determine the global redundant data shared across nodes and the differential data unique to each edge node, and then issue differentiated upload strategies to the corresponding non-anchor edge nodes. Non-anchor edge nodes upload the differential data from the locally cached disaster recovery data to the central cloud node according to the differential upload strategy; Anchor nodes upload global redundant data and local differential data to the central cloud node.

[0007] When disaster recovery data traffic surges, each edge node slices and aligns the data based on a preset time window and extracts the event's spatiotemporal features, enabling standardized processing of dispersed data within the same spatiotemporal dimension. By inputting the standardized event spatiotemporal features into a consistent hash function to determine anchor nodes, deduplication tasks for the same sudden event can be dynamically and evenly distributed across specific edge nodes, avoiding single-point computational bottlenecks caused by centralized processing. Non-anchor nodes only send data fingerprints, not the original data, to anchor nodes. Anchor nodes can identify globally redundant data shared across nodes and node-specific differences by comparing fingerprints at the edge. Based on this, a differentiated upload strategy is issued, ensuring that non-anchor nodes only upload differentiated data, while anchor nodes uniformly upload redundant data. This cloud-edge collaborative edge deduplication mechanism reduces the amount of duplicate data transmitted to the central cloud node, lowers network bandwidth consumption between the cloud and the edge, and improves the overall transmission efficiency of disaster recovery data in scenarios with sudden traffic surges.

[0008] In conjunction with some implementation methods of the first aspect, in some implementation methods, the locally cached disaster recovery data is sliced ​​and aligned based on a preset time window, and the event spatiotemporal features of the disaster recovery data are extracted respectively, specifically including: When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it calculates the peak traffic time of the locally cached disaster recovery data during the traffic burst and uses the peak traffic time as the event alignment anchor point. Based on the event alignment anchor point, a preset time window is constructed by extending forward and backward according to the preset time span; The local cached disaster recovery data is truncated and divided into blocks according to a preset time window to obtain disaster recovery data slices after slice alignment. Extract the device source geographic coordinate set and event alignment anchor point of the disaster recovery data slice, and combine them to generate a multi-dimensional spatiotemporal vector; Dimensionality reduction encoding is performed on multidimensional spatiotemporal vectors to obtain the event spatiotemporal characteristics of disaster recovery data.

[0009] By adopting the above technical solution, each edge node can accurately locate the core moment of data surge caused by the same disaster event at different nodes by calculating the peak traffic time during the traffic burst as the event alignment anchor. A preset time window is constructed forward and backward based on this anchor, and then truncated and segmented, effectively filtering background data unrelated to the core event and ensuring the temporal correlation of sliced ​​data between different nodes. Furthermore, a multi-dimensional spatiotemporal vector is generated by combining the set of device source geographic coordinates with the event alignment anchor, closely linking the physical spatial source of the data with the time of occurrence, truly reflecting the spatiotemporal distribution attributes of disaster recovery data. This multi-dimensional spatiotemporal vector is then dimensionality-reduced, compressing the feature volume while retaining core spatiotemporal correlation information. This reduces the computational complexity and communication overhead of subsequent hash mapping and inter-node feature interaction, improving the accuracy and processing efficiency of distributed disaster recovery data slice alignment and feature representation.

[0010] In conjunction with some implementations of the first aspect, in some implementations, the anchor node compares the received data fingerprints with its local data fingerprints to determine the globally redundant data shared across nodes and the differential data unique to each edge node, and then issues a differentiated upload strategy to the corresponding non-anchor edge node, specifically including: Anchor nodes construct local hash trees using local data fingerprints as leaf nodes and extract hash values ​​from nodes at each level of the local hash tree; Each received data fingerprint is compared with the hash value of each node at each level of the local hash tree to determine the target data fingerprint that matches the comparison. The disaster recovery data corresponding to the target data fingerprint is identified as globally redundant data shared across nodes, and the disaster recovery data corresponding to the unmatched remaining data fingerprint is identified as differential data unique to each edge node. Based on the sequence position of the remaining data fingerprint in the corresponding non-anchor edge node, a differentiated upload strategy consisting of binary bits is generated; The differentiated upload strategy is distributed to the corresponding non-anchor edge nodes.

[0011] By adopting the above technical solution, a structured fingerprint indexing system is formed by constructing a local hash tree with local data fingerprints as leaf nodes and extracting hash values ​​from nodes at each level. The received data fingerprints are compared layer by layer with the local hash tree. This allows for the rapid matching of high-level hash values, directly eliminating a large number of irrelevant branches and avoiding the inefficient global traversal and comparison of massive amounts of data fingerprints. This enables the rapid and accurate separation of globally redundant data shared across nodes and unique, differing data. Based on the sequence position of the remaining data fingerprints in the corresponding nodes, a differentiated upload strategy composed of binary bits is generated. Utilizing the high-density information carrying capacity of binary bits, complex upload instructions are compressed into extremely small data packets. This approach not only improves the execution speed of multi-node massive data fingerprint comparison and reduces the memory and computing power consumption of anchor nodes, but also reduces the downlink bandwidth usage when issuing control strategies.

[0012] In conjunction with some implementations of the first aspect, in some implementations, after issuing the differentiated upload strategy to the corresponding non-anchor edge nodes, the method further includes: Anchor node calculations determine the proportion of differential data to total data volume under the current standardized event spatiotemporal characteristics. If the ratio exceeds the preset conflict threshold, the anchor node sends an upload suspension command to the corresponding non-anchor edge node to pause the transmission of differential data to the central cloud node. Anchor nodes generate feature extraction configuration parameters and distribute these parameters to each non-anchor edge node that has cached differential data. Each non-anchor edge node performs envelope extraction and low-frequency component sampling on the locally cached differential data according to the feature extraction configuration parameters, generates a structural feature vector, and feeds the structural feature vector back to the anchor node; The anchor node calculates the dynamic time warping distance between the received structural feature vectors, and associates and maps the difference data with a dynamic time warping distance less than a preset distance threshold into a semantic redundancy dataset in memory; The anchor node obtains the signal-to-noise ratio (SNR) parameters of each non-anchor edge node corresponding to the semantically redundant dataset, and determines the non-anchor edge node with the largest SNR parameter as the target node. Anchor nodes generate and distribute secondary differentiated upload strategies to control target nodes to upload corresponding data blocks in the semantically redundant dataset, and to control other non-anchor edge nodes to upload residual data blocks in the local differentiated data after removing the semantically redundant dataset.

[0013] By adopting the above technical solution, anchor nodes suspend transmission when the proportion of discrepancies exceeds a preset conflict threshold, avoiding network congestion caused by the simultaneous uploading of a large amount of non-identical but highly similar data. By controlling non-anchor edge nodes to perform envelope extraction and low-frequency component sampling to generate structural feature vectors from the discrepancies, high-frequency random noise and minor fluctuations during data acquisition are filtered out, preserving the underlying physical change trends of the data. Calculating the dynamic time warping distance between structural feature vectors effectively overcomes time axis distortion caused by sampling clock deviations or signal transmission delays between different nodes, accurately identifying semantically redundant datasets with different fingerprints but highly consistent semantic content. The node with the highest signal-to-noise ratio is selected as the target node to upload this semantically redundant data, while other nodes are controlled to only upload residual data blocks, further mining deeper levels of similar data, reducing the amount of invalid data transmission caused by noise interference, and improving the overall quality of disaster recovery data received by the central cloud node.

[0014] In conjunction with some implementations of the first aspect, in some implementations, the anchor node calculates the dynamic time-warped distance between the received structural feature vectors, specifically including: Anchor node acquisition provides the geospatial distance between two non-anchor edge nodes that provide structural feature vectors; Anchor nodes obtain the corresponding physical signal propagation speed based on the event type in the current standardized event spatiotemporal characteristics, and calculate the maximum propagation delay of the physical signal between two non-anchor edge nodes by combining the geographic spatial distance. The anchor node maps the maximum propagation delay to the index offset of the feature sequence based on the sampling frequency of the structural feature vector, and generates dynamic warping window parameters. Anchor nodes, based on dynamic warping window parameters, constrain the boundaries of the optimization path in the alignment matrix of the structural feature vectors, and calculate the dynamic time warping distance only within the boundaries.

[0015] By employing the aforementioned technical solution, the maximum propagation delay of a physical signal between two nodes can be calculated by obtaining the geospatial distance between two non-anchor edge nodes and combining it with the propagation speed of physical signals for a specific event type. This maximum propagation delay is then mapped to an index offset based on the sampling frequency of the structural feature vector to generate a dynamic warping window parameter. This transforms the propagation constraints of physical space into mathematical constraints for sequence alignment. Based on this dynamic warping window parameter, the boundary of the optimization path is constrained in the alignment matrix of the structural feature vector, ensuring that the calculation of the dynamic time warping distance is performed only within a reasonable delay range that conforms to physical laws, avoiding invalid calculations for extreme offset paths that are unlikely to have physical association. This calculation method, which combines physical and spatiotemporal constraints, reduces the search space of the dynamic time warping algorithm, lowers the computational resources and time consumption of anchor nodes when calculating feature vector similarity, and eliminates mismatch interference from irrelevant signal peaks, thereby improving the accuracy and processing efficiency of semantically redundant dataset identification.

[0016] In conjunction with some implementations of the first aspect, in some implementations, after the anchor node generates and distributes the secondary differentiated upload strategy, the method further includes: If the anchor node detects that the uplink of the target node has been interrupted, the anchor node extracts the truncation timestamp of the last successfully uploaded data block before the interruption. Anchor nodes are selected from the remaining non-anchor edge nodes, with the second largest signal-to-noise ratio, as relay nodes. Anchor node retrieval is the feature alignment path matrix between the target node and the relay node generated during the calculation of dynamic time warping distance; The anchor node maps the truncated timestamp into the feature alignment path matrix to determine the corresponding synchronization continuation index of the truncated timestamp in the local cache data of the relay node; The anchor node sends a flow stitching control command to the relay node to control the relay node to stop uploading residual data blocks and to control the relay node to start uploading complete local cached data to the central cloud node from the synchronization continuation index, and to encapsulate the phase compensation identifier generated based on the feature alignment path matrix in the uploaded data header.

[0017] By adopting the above technical solution, when the anchor node detects an uplink interruption at the target node, it can quickly respond to link failures while ensuring data quality by extracting the truncation timestamp of the last successfully uploaded data block and selecting the node with the second highest signal-to-noise ratio as the relay node. Mapping the truncation timestamp using the feature alignment path matrix between the target node and the relay node accurately converts the physical interruption time of the target node into a synchronization continuation index in the local cached data sequence of the relay node, achieving precise alignment of heterogeneous timelines across nodes. Sending a flow stitching control command containing this index and a phase compensation identifier to the relay node allows it to directly resume uploading complete cached data from the corresponding data breakpoint. This cross-node breakpoint continuation mechanism avoids the complete discarding and retransmission of uploaded semantically redundant data due to a single node link failure, reduces ineffective network bandwidth consumption in unstable link scenarios, and improves the transmission continuity and overall recovery efficiency of disaster recovery data in complex network environments.

[0018] In conjunction with some implementations of the first aspect, in some implementations, the anchor node maps the truncated timestamp input feature alignment path matrix to determine the synchronization continuation index corresponding to the truncated timestamp in the relay node's local cache data, specifically including: The anchor node converts the truncated timestamp into the truncated index value of the local feature sequence of the target node based on the data sampling frequency of the target node; In the feature alignment path matrix, anchor nodes extract all alignment coordinate pairs with the truncated index value as the base dimension to generate a candidate mapping set. If there are multiple alignment coordinate pairs with different target dimension coordinates in the candidate mapping set, the anchor node will determine the target dimension coordinates corresponding to the alignment coordinate pair with the smallest target dimension coordinates as the synchronization continuation index.

[0019] By adopting the above technical solution, the truncated timestamp is converted into a truncated index value based on the data sampling frequency of the target node. Then, all alignment coordinate pairs with this index value as the base dimension are extracted from the feature alignment path matrix to generate a candidate mapping set. This allows for a comprehensive acquisition of all possible associated positions on the relay node's timeline at the moment of the target node's interruption. When the dynamic time warping algorithm generates a one-to-many mapping relationship, resulting in multiple candidate mapping sets with different target dimension coordinates, the anchor node determines the coordinates corresponding to the alignment coordinate pair with the smallest target dimension coordinate as the synchronization continuation index. This ensures that the relay node resumes uploading from the earliest associated data frame. This prevents delays in the continuation point due to signal timeline stretching and deformation, avoids missing some associated data slices during cross-node flow stitching, reduces the risk of data loss during link switching, and improves the integrity of the disaster recovery data ultimately received and stitched by the central cloud node.

[0020] Secondly, embodiments of this application provide a cloud-edge collaborative disaster recovery data transmission system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and one or more processors call the computer instructions to cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a system, cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer program product that, when run on a system, causes the system to execute the method described in any possible implementation of the first aspect.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application provides a cloud-edge collaborative disaster recovery data transmission method. When disaster recovery data traffic surges, each edge node slices and aligns the data based on a preset time window and extracts the event's spatiotemporal features, enabling the unification of scattered data into a unified spatiotemporal dimension for standardized processing. By inputting the standardized event spatiotemporal features into a consistent hash function to determine anchor nodes, deduplication tasks for the same sudden event can be dynamically and evenly distributed across specific edge nodes, avoiding single-point computational bottlenecks caused by centralized processing. Non-anchor nodes only send data fingerprints, not the original data, to anchor nodes. Anchor nodes can identify globally redundant data shared across nodes and node-specific differences by comparing fingerprints at the edge. Based on this, a differentiated upload strategy is issued, ensuring that non-anchor nodes only upload differentiated data, while anchor nodes uniformly upload redundant data. This cloud-edge collaborative edge deduplication mechanism reduces the amount of duplicate data transmitted to the central cloud node, lowers network bandwidth consumption between the cloud and edge, and improves the overall transmission efficiency of disaster recovery data in traffic surge scenarios.

[0024] 2. This application provides a disaster recovery data transmission method based on cloud-edge collaboration. Anchor nodes pause transmission when the proportion of discrepancies exceeds a preset conflict threshold, avoiding network congestion caused by the simultaneous uploading of a large amount of highly similar but not completely identical data. By controlling non-anchor edge nodes to perform envelope extraction and low-frequency component sampling on the discrepancies to generate structural feature vectors, high-frequency random noise and minor fluctuations during data acquisition are filtered out, preserving the underlying physical change trends of the data. Calculating the dynamic time warping distance between structural feature vectors effectively overcomes time axis distortion caused by sampling clock deviations or signal transmission delays between different nodes, accurately identifying semantically redundant datasets with different fingerprints but highly consistent semantic content. The node with the highest signal-to-noise ratio is selected as the target node to upload this semantically redundant data, while other nodes are controlled to only upload residual data blocks, further mining deeper levels of similar data, reducing the amount of invalid data transmission caused by noise interference, and improving the overall quality of disaster recovery data received by the central cloud node.

[0025] 3. This application provides a cloud-edge collaborative disaster recovery data transmission method. When the anchor node detects an uplink interruption at the target node, it extracts the truncation timestamp of the last successfully uploaded data block and selects the node with the second-highest signal-to-noise ratio as the relay node. This enables rapid response to link failures while ensuring data quality. By retrieving the feature alignment path matrix between the target node and the relay node to map the truncation timestamp, the physical interruption time of the target node can be accurately converted into a synchronization continuation index in the local cached data sequence of the relay node, achieving precise alignment of heterogeneous timelines across nodes. A flow stitching control command containing this index and a phase compensation identifier is sent to the relay node, enabling the relay node to directly resume uploading complete cached data from the corresponding data breakpoint. This cross-node breakpoint continuation mechanism avoids the complete discarding and retransmission of uploaded semantically redundant data due to a single node link failure, reduces ineffective network bandwidth consumption in unstable link scenarios, and improves the transmission continuity and overall recovery efficiency of disaster recovery data in complex network environments. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a disaster recovery data transmission method based on cloud-edge collaboration in an embodiment of this application.

[0027] Figure 2 This is another flowchart illustrating a disaster recovery data transmission method based on cloud-edge collaboration in an embodiment of this application.

[0028] Figure 3 This is another flowchart illustrating a disaster recovery data transmission method based on cloud-edge collaboration in this application.

[0029] Figure 4This is a schematic diagram of the physical device structure of a disaster recovery data transmission system based on cloud-edge collaboration provided in an embodiment of this application. Detailed Implementation

[0030] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0031] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0032] The following example is used in conjunction with Figure 1 The present application describes a disaster recovery data transmission method based on cloud-edge collaboration in its embodiments: Please see Figure 1 This is a flowchart illustrating a disaster recovery data transmission method based on cloud-edge collaboration in an embodiment of this application.

[0033] S101. When each edge node detects that the received disaster recovery data traffic exceeds the preset threshold, it performs slice alignment on the locally cached disaster recovery data based on the preset time window and extracts the event spatiotemporal features of the disaster recovery data respectively. When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it slices and aligns the locally cached disaster recovery data based on a preset time window, and extracts the event spatiotemporal features of the disaster recovery data. This can be achieved in the following ways: When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it calculates the peak traffic time point of the locally cached disaster recovery data during the traffic burst and uses the peak traffic time point as the event alignment anchor point. Based on the event alignment anchor point, a preset time window is constructed by extending forward and backward according to a preset time span. The locally cached disaster recovery data is truncated and segmented according to the preset time window to obtain disaster recovery data slices after slice alignment. The set of device source geographic coordinates and event alignment anchor points of the disaster recovery data slices are extracted and combined to generate a multi-dimensional spatiotemporal vector. The multi-dimensional spatiotemporal vector is dimensionality reduced and encoded to obtain the event spatiotemporal features of the disaster recovery data.

[0034] Edge nodes can include, but are not limited to, devices such as edge gateways, edge servers, or access routers deployed at the network edge. Disaster recovery data traffic refers to the data stream transmitted for disaster recovery and data backup, such as system log backups and database snapshots. A preset threshold is the traffic criterion for triggering emergency handling; its source is typically based on a combination of historical normal business traffic statistics baselines and the physical bandwidth limit of the edge node's network interface. A preset time window refers to the time interval used for data extraction and alignment. Event spatiotemporal features are vectorized expressions that simultaneously reflect the geographical location and time attributes of data generation. When disaster recovery data traffic bursts, different edge nodes may receive the same event data at slightly different times. Slicing and aligning the data based on the preset time window can unify data scattered across different nodes into the same spatiotemporal dimension for standardized processing. Extracting event spatiotemporal features can accurately characterize the spatiotemporal source attributes of data, providing a high-dimensional feature basis for subsequent judgments on whether different nodes have received redundant data of the same event, improving the accuracy of cross-node data feature matching, and reducing the computational overhead of subsequent processing.

[0035] This can be achieved using sliding window statistics and principal component analysis algorithms. When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it uses a sliding window to statistically analyze the traffic value per unit time in real time, calculates the traffic derivative, and records the moment when the traffic derivative turns from positive to negative as the peak traffic time point during the traffic burst, using this peak traffic time point as the event alignment anchor point. Using this event alignment anchor point as the reference center, a closed preset time window is constructed by extending forward and backward according to a preset fixed time span. Based on the start and end timestamps of this preset time window, the locally cached disaster recovery data stream is truncated and processed into blocks of fixed byte lengths to obtain sliced ​​disaster recovery data slices. The network protocol header of the disaster recovery data slice is parsed to extract the included set of device source geographic coordinates, which is then concatenated with the timestamp value of the event alignment anchor point to generate a multi-dimensional spatiotemporal vector. Principal component analysis (PCA) is used to calculate the covariance matrix of the multidimensional spatiotemporal vector. Several principal components with the largest eigenvalues ​​are extracted and linearly transformed to complete the dimensionality reduction encoding of the multidimensional spatiotemporal vector, and finally the event spatiotemporal features of disaster recovery data are obtained.

[0036] Alternatively, adaptive threshold detection and an autoencoder model can be used. Each edge node calculates the moving average and standard deviation of disaster recovery data traffic in real time. When the instantaneous traffic exceeds the average plus a certain multiple of the standard deviation, burst detection is triggered. The maximum traffic moment is extracted from the high-traffic interval as the traffic peak time point and used as the event alignment anchor point. A preset time span is dynamically calculated based on the average duration of similar historical bursts, and a preset time window is constructed by extending forward and backward from the event alignment anchor point. The locally cached disaster recovery data is truncated according to the preset time window and divided into variable-length blocks based on the logical boundaries of the data, resulting in sliced ​​disaster recovery data slices after slice alignment. The device source geographic coordinate set and event alignment anchor points of the disaster recovery data slices are extracted and arranged in a preset dimensional order to generate multi-dimensional spatiotemporal vectors. These multi-dimensional spatiotemporal vectors are input into a pre-trained autoencoder model. Because the autoencoder model learns the potential low-dimensional data distribution patterns of massive spatiotemporal data through its internal encoder neural network hierarchical structure, it can effectively compress high-dimensional spatiotemporal information. The output vector of the hidden layer of the autoencoder model is extracted as the dimensionality reduction encoding result, thus obtaining the event spatiotemporal features of the disaster recovery data.

[0037] S102. Each edge node performs spatiotemporal gridding dimensionality reduction processing on the spatiotemporal features of the event to obtain standardized spatiotemporal features of the event. Event spatiotemporal features are the dimensionality-reduced encoding results containing geographic coordinates and time information extracted in the preceding steps. Spatiotemporal gridding dimensionality reduction is a spatial discretization technique that divides a continuous spatiotemporal coordinate space into discrete grid cells and maps features to the corresponding grid cells. Standardized event spatiotemporal features refer to uniform format features that have eliminated minor coordinate fluctuations and time offsets after gridding processing. Because event spatiotemporal features typically have continuous and irregular spatiotemporal distribution characteristics, directly performing similarity calculations or hash mappings is difficult and prone to bias. Through spatiotemporal gridding dimensionality reduction, similar spatiotemporal features can be normalized to the same grid cell, filtering out minor spatiotemporal noise. This processing method not only further reduces the data dimensionality of features but also makes similar event features from different edge nodes highly consistent, improving the stability of subsequent hash mappings and the success rate of cross-node event matching.

[0038] This can be achieved using a fixed-scale 3D mesh generation method. Each edge node acquires the geographic latitude and longitude coordinates and timestamp information from the event's spatiotemporal features. A 3D spatiotemporal mesh coordinate system with fixed spatial and temporal resolution is predefined in system memory. The latitude and longitude coordinates from the event's spatiotemporal features are mapped to the corresponding 2D spatial mesh cells through a floor operation. Simultaneously, the timestamp information is divided by the temporal resolution and rounded down, mapped to the corresponding 1D temporal mesh cells, thus obtaining a 3D mesh index coordinate system. The frequency of the event's spatiotemporal features falling within each 3D mesh cell is counted, and the mesh index coordinates with frequencies greater than zero are extracted. These index coordinates of non-empty meshes are serialized and concatenated according to spatial and temporal chronology to form a dimensionality-reduced discrete feature sequence. This discrete feature sequence is used as the standardized event spatiotemporal features.

[0039] Alternatively, a density-based adaptive spatiotemporal clustering gridding method can be used. Each edge node utilizes a density-based spatial clustering algorithm to perform clustering analysis on the spatiotemporal features of the event, calculating the neighborhood density of each feature point in the spatiotemporal dimension, and identifying multiple core regions with dense spatiotemporal distribution. The spatiotemporal grid is dynamically divided according to the density distribution characteristics of each core region; a finer-grained grid is used for regions with high feature point density, and a coarser-grained grid is used for regions with low feature point density. The spatiotemporal features of the event are assigned to the corresponding adaptive grids, and the geometric center coordinates and average timestamp of all feature points within each grid are calculated, serving as the representative features of that grid. All representative features of the grids containing data are arranged according to a preset hash sorting rule to form a dimensionality-reduced feature vector, which is then output as the standardized spatiotemporal features of the event.

[0040] S103. Each edge node inputs the standardized event spatiotemporal features into a consistent hash function, maps them in a virtual ring composed of each edge node, and independently determines the anchor node corresponding to the event spatiotemporal features. Consistent hashing is a special hashing algorithm that minimizes the number of remapped features when the number of nodes changes. A virtual ring is a logically circular address space used to evenly distribute nodes and data. Anchor nodes are centralized processing nodes selected on the virtual ring, responsible for coordinating the processing of disaster recovery data corresponding to specific event spatiotemporal features. Standardized event spatiotemporal features are input into a consistent hashing function to determine anchor nodes, enabling deduplication tasks for the same sudden event to be dynamically and evenly distributed across specific edge nodes. This distributed routing mechanism reduces the risk of single-point computational bottlenecks caused by centralized processing, while allowing different edge nodes to independently and consistently direct data for the same event to the same anchor node without mutual communication and negotiation, improving the data routing efficiency of the cloud-edge collaborative network and the overall fault tolerance of the system.

[0041] In distributed scenarios for disaster recovery data acquisition and transmission, the network topology of edge nodes may change due to equipment failures or dynamic expansion. Traditional hash modulo algorithms can lead to large-scale data remapping when the number of nodes changes, causing network congestion. By introducing a consistent hash function, the feature identifiers of data blocks are mapped to the same logical ring with the network identifiers of edge nodes, and matching is performed according to specific directional rules. This ensures that when a node joins or leaves, only the data mapping relationship of adjacent nodes is affected. This computational logic reduces the amount of data migration when the system topology changes, and improves the stability and scalability of distributed storage and global deduplication comparison for disaster recovery data.

[0042] The specific calculation formula of a consistent hash function mainly includes three parts: node hash calculation, data hash calculation, and addressing mapping calculation. Let the number of bits in the hash space be m, usually 32, to construct a closed hash ring space. For any given edge node identifier and disaster recovery data block identifiers The formulas for calculating their hash values ​​are as follows: in, and These represent the string lengths of the node identifier and the data block identifier, respectively. This represents the underlying encoding value of the extracted character, where p is a preset prime constant.

[0043] After obtaining the hash values ​​of the nodes and data, the data block Mapped to target node The addressing calculation formula is defined as: If set If empty, it maps to the node with the smallest hash value: This can be achieved using a standard secure hash algorithm combined with virtual node mapping. Each edge node pre-collects network identifier information from all participating physical edge nodes, generating multiple virtual node identifiers with incrementally increasing numbers for each physical edge node. All virtual node identifiers are input into the standard secure hash algorithm to calculate the corresponding node hash value, and then mapped to a closed virtual ring with a fixed numerical range in ascending order of hash value. When it is necessary to determine the anchor node, each edge node converts the standardized event spatiotemporal characteristics into a byte sequence, and inputs it into the same standard secure hash algorithm to calculate a feature hash value. Starting from the position of this feature hash value on the virtual ring, the search proceeds clockwise, extracting the first virtual node encountered, querying the physical edge node mapped to that virtual node, and independently identifying it as the anchor node corresponding to that event spatiotemporal characteristic.

[0044] Alternatively, a weighted Cyclic Redundancy Check (CRBC) hash algorithm can be used. Each edge node calculates a comprehensive weight value based on its CPU computing power, remaining storage capacity, and available network bandwidth, and allocates a corresponding number of hash slots in logical memory according to this weight value. The CRBC hash algorithm is used to calculate the media access control address of each edge node and distribute it into the corresponding hash slots, forming a virtual ring. Each edge node uses standardized event spatiotemporal features as input data and calculates a feature check code using the same CRBC hash algorithm. This feature check code is moduloed by the total number of slots in the virtual ring to obtain a target slot index. The edge node currently belonging to the target slot index is found in the virtual ring and independently identified as the anchor node corresponding to the event spatiotemporal feature.

[0045] S104. If the current edge node is not an anchor node, calculate the data fingerprint of each disaster recovery data slice after alignment and send the data fingerprint to the anchor node. The current edge node refers to the physical node currently executing processing logic. Non-anchor nodes are edge nodes that, after consistent hashing, were not selected as the node for coordinating the processing of event data. A data fingerprint is a fixed-length, unique identifier string calculated using a specific algorithm to represent the original content of the data. Non-anchor nodes only send data fingerprints to anchor nodes, rather than the original disaster recovery data. This approach significantly reduces the volume of data transmitted between edge nodes. By transmitting lightweight data fingerprints instead of massive amounts of original data, the network bandwidth consumption on the edge side is reduced, and data transmission latency is shortened. Simultaneously, the data fingerprint retains the unique characteristics of the original data, enabling anchor nodes to quickly compare data based on the fingerprint, improving the overall efficiency of redundant data identification on the edge side.

[0046] This can be implemented using a message digest algorithm. If the current edge node discovers through consistent hashing that it is not the anchor node corresponding to the spatiotemporal characteristics of the event, it iterates through the disaster recovery data slices in its local cache after slice alignment. For each disaster recovery data slice, it extracts all its binary data content and inputs it into a preset message digest algorithm model. The message digest algorithm performs operations such as end padding, block processing, and multiple rounds of nonlinear logical transformations on the slice data to generate a fixed-length hash value as the data fingerprint of the disaster recovery data slice. The current edge node packages all the generated data fingerprints according to the original order of the slices and sends the packaged data fingerprint set to the corresponding anchor node through the secure transport layer protocol communication link between edge nodes.

[0047] Alternatively, a Bloom filter combined with a block hash algorithm can be used. If the current edge node is not the anchor node, further micro-blocking is performed on each aligned disaster recovery data slice. Multiple independent hash values ​​are calculated for each micro-block using various lightweight hash algorithms, and these hash values ​​are mapped to the bit array of the Bloom filter. The state vector of the Bloom filter is extracted as a coarse data fingerprint for that disaster recovery data slice. Simultaneously, a precise hash value is calculated for the entire disaster recovery data slice using a circular hash algorithm as the precise data fingerprint. The current edge node combines the coarse and precise data fingerprints into a composite data fingerprint, attaches the sequence number of the corresponding disaster recovery data slice, encapsulates it into a remote procedure call protocol message, and sends it to the anchor node through the edge network.

[0048] S105. The anchor node compares the received data fingerprints with the local data fingerprints to determine the global redundant data shared across nodes and the differential data unique to each edge node, and sends the differential upload strategy to the corresponding non-anchor edge node. Anchor nodes compare the received data fingerprints with their local data fingerprints to determine the globally redundant data shared across nodes and the unique difference data specific to each edge node. They then distribute differentiated upload strategies to the corresponding non-anchor edge nodes. This can be achieved through the following methods: Anchor nodes construct local hash trees using local data fingerprints as leaf nodes and extract hash values ​​from each level of the local hash tree. Each received data fingerprint is compared layer by layer with the hash values ​​of each level of the local hash tree to identify the target data fingerprint that matches. The disaster recovery data corresponding to the target data fingerprint is identified as globally redundant data shared across nodes, while the disaster recovery data corresponding to the unmatched remaining data fingerprints is identified as differential data unique to each edge node. Based on the sequence position of the remaining data fingerprints in the corresponding non-anchor edge nodes, a differentiated upload strategy consisting of binary bits is generated. The differentiated upload strategy is then distributed to the corresponding non-anchor edge nodes.

[0049] Global redundant data refers to identical disaster recovery data that exists in backups across multiple edge nodes. Differentiated data refers to unique disaster recovery data that exists only in a specific edge node. A differentiated upload strategy is a set of control instructions that guides non-anchor nodes to upload only specific data. A local hash tree is a tree-like data structure built on hash values. Leaf nodes store basic data fingerprints, and the hash values ​​of level nodes are calculated from the hash values ​​of their child nodes. The target data fingerprint is the fingerprint that matches the comparison, while the remaining data fingerprints are the unmatched fingerprints. Binary bits are the smallest data unit constituting the strategy. By constructing a local hash tree and performing layer-by-layer comparisons, anchor nodes can quickly narrow down the search range, reducing the computational complexity of comparing massive amounts of fingerprints. Transforming the comparison results into a differentiated upload strategy composed of binary bits provides a command issuance method with extremely high compression ratio. This mechanism allows non-anchor nodes to clearly know which data needs to be uploaded, reducing the amount of duplicate data transmitted to the central cloud node and improving network bandwidth utilization.

[0050] This can be achieved using Merkle tree construction and a layer-by-layer recursive comparison method. The anchor node extracts the data fingerprint of the local disaster recovery data slice as a leaf node. It combines and concatenates the data fingerprints of two adjacent leaf nodes, calculates the hash value to generate a parent node, and recursively constructs the Merkle tree layer by layer upwards until a unique root hash value is generated, thus extracting the hash values ​​of each level of the local hash tree. The anchor node also constructs a receiving Merkle tree for each received data fingerprint. It first compares the root hash values ​​of the two trees; if the root hash values ​​are different, it recursively compares the hash values ​​of the child nodes downwards until it reaches the leaf node level to determine the matching target data fingerprint. The disaster recovery data corresponding to the target data fingerprint is identified as globally redundant data shared across nodes, and the disaster recovery data corresponding to the unmatched remaining data fingerprints is identified as differential data unique to each edge node. Based on the position index of the remaining data fingerprint in the original slice sequence of the corresponding non-anchor edge node, a binary bitmap of the same length as the total number of slices is constructed. The binary bits corresponding to the differential data are set to 1, and the binary bits corresponding to the redundant data are set to 0, generating a differentiated upload strategy composed of binary bits. Finally, a transmission control protocol connection is established, and the differentiated upload strategy is distributed to the corresponding non-anchor edge nodes.

[0051] Alternatively, a balanced binary search hash tree and parallel comparison method can be used. Anchor nodes construct a balanced binary search hash tree based on local data fingerprints, sorting them by fingerprint value. Each non-leaf node stores the aggregate hash value of all fingerprints under its subtree, extracting hash values ​​from each level of nodes. Anchor nodes enable a multi-threaded processing mechanism, distributing received data fingerprints to multiple parallel processing threads. Each thread performs a layer-by-layer parallel search and comparison of the received data fingerprint with the hash values ​​of each level of the local hash tree, leveraging the characteristics of binary search to quickly locate the matching target data fingerprint. The disaster recovery data corresponding to the target data fingerprint is marked as globally redundant data shared across nodes, while the disaster recovery data corresponding to the unmatched remaining data fingerprints is marked as differential data unique to each edge node. The logical sequence positions of the remaining data fingerprints in the corresponding non-anchor edge nodes are statistically analyzed, and these sequence positions are compressed and encoded using run-length encoding to generate a differentiated upload strategy composed of binary bits. This differentiated upload strategy is encapsulated in a lightweight control message and sent to the corresponding non-anchor edge nodes via a message queue mechanism.

[0052] S106. Non-anchor edge nodes upload the differential data in the locally cached disaster recovery data to the central cloud node according to the differential upload strategy. A central cloud node refers to a cluster of cloud computing servers or a cloud storage platform deployed in the core data center, responsible for the centralized storage, management, and global recovery of massive disaster recovery data. Non-anchor edge nodes only upload differentiated data according to policies, filtering out globally redundant data already identified by anchor nodes. This cloud-edge collaborative end-side deduplication upload mechanism significantly reduces the volume of data streams transmitted over long distances from the edge to the central cloud. By reducing the amount of duplicate data transmitted to the central cloud node, not only is bandwidth consumption on the wide area network reduced, alleviating congestion pressure on the core network, but the storage space occupied by the central cloud node is also reduced, improving the overall transmission efficiency of disaster recovery data and the utilization rate of cloud resources in scenarios of sudden traffic surges.

[0053] A selective upload method based on bitmap parsing can be used. After receiving the differentiated upload strategy from the anchor node, the non-anchor edge node parses the binary bitmap sequence. The non-anchor edge node traverses the locally cached disaster recovery data slice sequence in chronological order, searching for the corresponding identifier bit in the binary bitmap based on the current slice's sequence number. If the identifier bit represents a specific value indicating differentiated data, the disaster recovery data slice is extracted and placed into the upload queue in memory; if the identifier bit represents redundant data, the slice is skipped. Finally, the non-anchor edge node establishes a secure shell protocol tunnel connection with the central cloud node, performs lossless compression and packaging of the differentiated data in the upload queue, attaches a global identifier for the incident, and the slice sequence number, and uploads it uniformly to the central cloud node.

[0054] Alternatively, an asynchronous transmission method based on instruction stream parsing can be used. After receiving the differentiated upload strategy, the non-anchor edge node decodes its internal compression encoding to reconstruct the specific physical storage addresses or logical offset list of the differentiated data in its local cache disk. The non-anchor edge node then initiates an asynchronous input / output read task, directly and accurately reading the corresponding differentiated data blocks from the local storage medium based on the parsed address list, reducing the computational overhead of traversing the entire dataset. The read differentiated data blocks are encapsulated into multiple independent data packets. Utilizing the User Datagram Protocol (UDP) combined with application-layer reliability assurance mechanisms, a multi-threaded concurrent channel is opened to asynchronously upload these differentiated data packets to the central cloud node, and integrity verification is performed after transmission.

[0055] S107, the anchor node uploads global redundant data and local differential data to the central cloud node.

[0056] Global redundant data represents the duplicate data received by multiple edge nodes during the same emergency, while local differential data is unique to the anchor node. As the coordinator for handling the spatiotemporal characteristics of a specific event, the anchor node is responsible for uploading the global redundant data on behalf of all relevant edge nodes. By having the anchor node uniformly upload redundant data, the central cloud node only needs to receive one complete copy of the shared data, greatly optimizing the cloud's receiving logic. Simultaneously, the anchor node also uploads its own local differential data, improving the integrity of the disaster recovery dataset. This mechanism clarifies the data transmission division of labor among nodes in the cloud-edge collaborative architecture, reduces network bandwidth consumption between the cloud and edge, and improves the data synchronization efficiency of the entire disaster recovery system.

[0057] This can be achieved using data concatenation and streaming. After completing the comparison of all data fingerprints, the anchor node extracts globally redundant data slices marked as shared across nodes and locally differentiable data slices that exist only in the local cache from its local cache. Based on the timestamp order or logical sequence number of these data slices in the original disaster recovery data stream, the globally redundant data and locally differentiable data are sequentially concatenated in memory to reconstruct a complete local disaster recovery data stream. The anchor node establishes a secure socket layer connection with the central cloud node and uploads the reconstructed disaster recovery data stream to the central cloud node in real time and continuously using streaming. Data type tags are attached to the transmitted data stream for the central cloud node to identify during persistent storage.

[0058] Alternatively, a classification and packaging method combined with multi-channel concurrent transmission can be used. Anchor nodes store the identified global redundant data and local differential data in separate memory buffers. A high-intensity dictionary compression algorithm is applied to the data in the global redundant data buffer to generate a redundant data compressed package; a conventional fast compression algorithm is applied to the data in the local differential data buffer to generate a differential data compressed package. The anchor node requests the establishment of multiple parallel multipath transmission control protocol channels from the central cloud node, allocating the redundant data compressed package and the differential data compressed package to different transmission channels for concurrent uploading. Custom metadata information is added to the header of the uploaded message to indicate whether the data packet belongs to global redundant data or local differential data, so that the central cloud node can classify, store, and index it upon receipt.

[0059] In the above embodiments, when disaster recovery data traffic surges, each edge node slices and aligns the data based on a preset time window and extracts the event's spatiotemporal features, enabling the dispersed data to be standardized under the same spatiotemporal dimension. By inputting the standardized event spatiotemporal features into a consistent hash function to map and determine the anchor node, the deduplication task for the same sudden event can be dynamically and evenly distributed to specific edge nodes, avoiding the single-point computation bottleneck caused by centralized processing. Non-anchor nodes only send data fingerprints to anchor nodes instead of the original data. Anchor nodes can identify globally redundant data shared across nodes and the unique difference data of each node by comparing fingerprints at the edge side. Based on this, a differentiated upload strategy is issued, so that non-anchor nodes only upload the difference data, while anchor nodes uniformly upload redundant data. This cloud-edge collaborative edge deduplication mechanism reduces the amount of duplicate data transmitted to the central cloud node, reduces network bandwidth consumption between the cloud and the edge, and improves the overall transmission efficiency of disaster recovery data in traffic surge scenarios.

[0060] In the above embodiments, effective deduplication of completely consistent globally redundant data is achieved through data fingerprint comparison. However, in the actual disaster recovery data acquisition and transmission process, due to noise interference in the physical environment of each edge node or slight deviations in sensor sampling clocks, disaster recovery data with highly consistent semantic content often differs at the underlying binary level, thus failing to be effectively identified through accurate data fingerprint comparison. To further mine deeper similar data and avoid network congestion caused by the simultaneous uploading of a large amount of non-completely consistent but semantically redundant data, this application proposes another disaster recovery data transmission method based on cloud-edge collaboration, building upon the above embodiments. The following is a combination of... Figure 2 Another disaster recovery data transmission method based on cloud-edge collaboration in the embodiments of this application is described below: Please see Figure 2 This is another flowchart illustrating a disaster recovery data transmission method based on cloud-edge collaboration in this application.

[0061] S201. Anchor node calculation: Under the current standardized event spatiotemporal characteristics, the proportion of differential data to the total data volume. The total data volume refers to the sum of disaster recovery data received by all relevant edge nodes under the current standardized spatiotemporal characteristics of the event, including globally redundant data identified in the preliminary steps and node-specific differential data. The volume of differential data refers to the size of the set of data that could not be preliminarily determined to be unique to each node through precise data fingerprint matching. Anchor nodes calculate this ratio to macroscopically assess the degree of heterogeneity of data under the current emergency. In highly correlated spatiotemporal events, if the proportion of differential data is too high, it often indicates the existence of pseudo-differential data with the same physical semantics but altered underlying binary due to environmental noise or sampling bias. By calculating this ratio, an objective quantitative evaluation indicator is provided for subsequent transmission control, enabling the system to intelligently perceive abnormal states in data distribution, reducing the risk of blindly uploading massive amounts of pseudo-differential data to the cloud, and improving the edge computing cluster's ability to understand complex data.

[0062] This can be achieved using a precise statistical method based on data block size. The anchor node iterates through all disaster recovery data slices marked as differential data under the current event's spatiotemporal characteristics, reads the byte length field from the header information of each disaster recovery data slice, and sums the byte lengths of all differential data slices to obtain the total number of bytes of differential data. Simultaneously, the anchor node uses the same method to count the total number of bytes of all disaster recovery data slices marked as globally redundant data. The total number of bytes of differential data is added to the total number of bytes of globally redundant data to obtain the total data volume under the current event's spatiotemporal characteristics. Finally, the total number of bytes of differential data is divided by the total data volume to calculate the precise proportion of differential data to the total data volume.

[0063] Alternatively, a fast estimation method based on a fixed number of slices can be used. When aligning disaster recovery data slices at each edge node, if a fixed-length slice partitioning strategy is used, the anchor node can directly count the total number of slices classified as differential data under the current event's spatiotemporal characteristics. Then, the anchor node counts the total number of all slices included under the current event's spatiotemporal characteristics, i.e., the sum of the number of differential data slices and the number of globally redundant data slices. Dividing the total number of differential data slices by the total number of all slices allows for a quick estimation of the proportion of differential data to the total data volume, thereby improving calculation speed while maintaining a certain level of accuracy.

[0064] S202. If the ratio is greater than the preset conflict threshold, the anchor node sends an upload suspension command to the corresponding non-anchor edge node to suspend the transmission of differential data to the central cloud node. The preset conflict threshold is a numerical limit used to determine whether the proportion of discrepancies in data is abnormal. It is typically derived from a statistical baseline of real discrepancies in similar historical events and dynamically set in conjunction with the available uplink bandwidth capacity of the current edge network. The upload suspension command is a specific network control signaling used to temporarily interrupt the data transmission process. When the proportion of discrepancies in data exceeds the preset conflict threshold, it indicates that a large amount of potentially semantically redundant data is about to be uploaded, which can easily cause congestion on the WAN link. The anchor node actively sends the upload suspension command to pause transmission, providing the system with a time window for in-depth semantic analysis. This proactive flow control mechanism improves the utilization of network bandwidth resources, reduces the impact risk of sudden traffic surges on the central cloud node, and ensures the transmission stability of critical disaster recovery data.

[0065] This can be achieved using an out-of-band signaling mechanism based on the Transmission Control Protocol (TCP). The anchor node compares the calculated ratio in memory with a pre-configured conflict threshold in the system parameter table. If the ratio exceeds the threshold, the anchor node constructs an out-of-band signaling data packet with the highest network priority. A specific suspend opcode and an identifier representing the spatiotemporal characteristics of the target event are written into the control payload of this packet. The anchor node then forwards this out-of-band signaling data packet to all non-anchor edge nodes preparing to upload the event's difference data via the TCP management channel established between edge nodes. Upon receiving this signaling, the non-anchor edge nodes immediately interrupt the transmission process of their corresponding data queues.

[0066] Alternatively, a publish-subscribe control method based on message middleware can be used. Anchor nodes continuously monitor the proportion of discrepancies. When the proportion exceeds a preset conflict threshold, the anchor node, acting as a message publisher, publishes a control message containing an upload suspension command to the message queue middleware deployed in the edge cluster. The publication topic of this control message is deeply bound to the current standardized event spatiotemporal characteristics. Each non-anchor edge node, acting as a subscriber, listens to the message stream under this topic in real time. Once it receives the upload suspension command, the non-anchor edge node calls the application programming interface of its local operating system's network interface controller to modify the socket state for discrepancy data transmission to the central cloud node to blocking mode, thereby pausing data transmission.

[0067] S203. The anchor node generates feature extraction configuration parameters and sends the feature extraction configuration parameters to each non-anchor edge node that has cached differential data. Feature extraction configuration parameters are a set of technical specifications that guide edge nodes on how to perform signal processing and feature transformation on raw data, including but not limited to the window size for envelope extraction, the cutoff frequency of the low-pass filter, and the resampling rate. Because different non-anchor edge nodes may have hardware heterogeneity and differences in default configurations, allowing each node to extract features independently would lead to misalignment of the generated structural feature vectors in terms of dimension and scale. Anchor nodes uniformly generate and distribute feature extraction configuration parameters, forcing all relevant nodes to adopt a consistent signal processing standard. This centralized parameter distribution mechanism ensures that the distributed extracted feature vectors are within the same metric space, improving the scientific rigor and accuracy of cross-node feature comparisons and reducing feature alignment errors caused by inconsistent preprocessing standards.

[0068] This can be achieved using a static parameter template matching method based on event type. Anchor nodes parse the standardized spatiotemporal features of the current event and extract the corresponding emergency event type identifier from the disaster recovery data. Anchor nodes then retrieve the corresponding preset feature extraction template from their locally stored configuration database based on this event type identifier. This template defines in detail parameters such as the optimal envelope extraction window length, the cutoff frequency of the low-pass filter, and the resampling rate for this type of data. The anchor nodes package these extracted parameters into a feature extraction configuration parameter file in Extensible Markup Language (XML) format and distribute it to all non-anchor edge nodes that cache the differential data via a secure transmission link in the edge network.

[0069] Alternatively, a dynamic parameter adaptive generation method based on data distribution characteristics can be used. Anchor nodes send requests to each non-anchor edge node to obtain sample fragments of a small amount of differential data. Anchor nodes perform Fast Fourier Transform on these sample fragments to analyze their spectral energy distribution characteristics and identify the low-frequency band range where the main physical signal energy is concentrated. Based on this band range, the optimal low-frequency component sampling rate and the smoothing coefficient required for the envelope extraction algorithm are dynamically calculated. Anchor nodes combine these dynamically calculated values ​​to generate customized feature extraction configuration parameters and serialize them into a lightweight data exchange format message, which is then sent to the corresponding non-anchor edge nodes.

[0070] S204. Each non-anchor edge node performs envelope extraction and low-frequency component sampling on the locally cached differential data according to the feature extraction configuration parameters, generates a structural feature vector, and feeds the structural feature vector back to the anchor node. Envelope extraction is a signal processing technique that filters out high-frequency oscillations and outlines the macroscopic fluctuations of a signal. Low-frequency component sampling refers to discarding high-frequency abrupt changes in the data, retaining only the basic data points that reflect the core changing trends of physical events. The structural feature vector is a one-dimensional or multi-dimensional mathematical array composed of these sampling points, representing the essential form of the data. The original difference data often contains minute fluctuations caused by sensor thermal noise or environmental interference, which are the root cause of the failure of accurate hash fingerprints. By performing envelope extraction and low-frequency component sampling, these surface noise interferences can be effectively removed, restoring the underlying physical change patterns of disaster recovery data. Transforming the processed trend into a structural feature vector significantly compresses the data dimensionality, improves the system's sensitivity to identifying false difference data, and reduces the resource consumption for subsequent similarity calculations of anchor nodes.

[0071] This can be achieved using a combination of Hilbert transform and digital filtering downsampling. After receiving the feature extraction configuration parameters, each non-anchor edge node extracts the locally cached difference data sequence, applies the Hilbert transform to calculate the analytic signal, and extracts the instantaneous amplitude of the analytic signal to obtain the envelope sequence of the original difference data. Next, based on the cutoff frequency specified in the configuration parameters, a finite impulse response low-pass filter is constructed to filter the envelope sequence and remove residual high-frequency noise. Finally, the filtered sequence is sampled at equal intervals for low-frequency components according to the resampling rate specified in the configuration parameters. The sampled data points are arranged in chronological order to generate a one-dimensional structural feature vector, which is then fed back to the anchor node via a remote procedure call protocol.

[0072] Alternatively, a sliding window extreme value extraction and piecewise averaging method can be used. Each non-anchor edge node resolves its feature extraction configuration parameters to obtain the specified sliding window length and step size. This sliding window is moved across the local difference data sequence, and local maxima are found and extracted within each window interval. These local maxima are then connected using a cubic spline interpolation algorithm to fit and generate the upper envelope of the data. Subsequently, based on the low-frequency component sampling requirements in the configuration parameters, the upper envelope is divided into multiple fixed-length time segments, and the arithmetic mean of the data points within each time segment is calculated. These piecewise averages are combined into a discrete numerical sequence, which serves as the structural feature vector of the difference data, and then packaged and sent to the anchor nodes.

[0073] S205. The anchor node calculates the dynamic time warping distance between the received structural feature vectors and associates and maps the difference data with a dynamic time warping distance less than a preset distance threshold into a semantic redundancy dataset in memory. Anchor nodes calculate the dynamic time warping distance between the received structural feature vectors and associate and map the difference data with dynamic time warping distances less than a preset distance threshold into a semantic redundancy dataset in memory. Specifically, the calculation of the dynamic time warping distance between the received structural feature vectors by the anchor nodes includes: Anchor nodes obtain the geospatial distance between two non-anchor edge nodes that provide structural feature vectors; based on the event type in the current standardized event spatiotemporal features, anchor nodes obtain the corresponding physical signal propagation speed and, combined with the geospatial distance, calculate the maximum propagation delay of the physical signal between the two non-anchor edge nodes; anchor nodes map the maximum propagation delay to the index offset of the feature sequence based on the sampling frequency of the structural feature vectors, generating dynamic warping window parameters; based on the dynamic warping window parameters, anchor nodes constrain the boundary of the optimization path in the alignment matrix of the structural feature vectors and calculate the dynamic time warping distance only within the boundary.

[0074] Dynamic time warping distance is an algorithmic metric that measures the similarity between two sequences of different lengths or with time-axis distortion. A preset distance threshold is a quantified boundary for determining whether two feature vectors belong to the same semantic event, typically derived from training a machine learning clustering model on historical redundant datasets. Semantic redundant datasets refer to data sets with different underlying binary fingerprints but highly consistent macroscopic physical semantics. Geospatial distance refers to the spatial span between the physical locations of edge nodes. Physical signal propagation speed refers to the rate at which the physical phenomenon causing the data propagates in space. Maximum propagation delay is the theoretical maximum time difference for a signal to reach different nodes. Index offset is the difference in the subscripts of the feature sequence array converted from the time difference. The dynamic warping window parameter is a constraint used to limit the search range of the alignment matrix. The alignment matrix is ​​a two-dimensional table recording the point-to-point distances of sequences. The optimal path is the route with the minimum cumulative distance found in the matrix. Standard dynamic time warping algorithms traverse the entire alignment matrix, which is not only computationally intensive but also prone to forcibly aligning signal segments with no physical connection. By incorporating geospatial distance and physical signal propagation speed to calculate the maximum propagation delay, and converting it into an index offset to constrain the boundary of the optimization path, strict physical spatiotemporal limitations are imposed on the algorithm. This constraint mechanism based on physical laws significantly reduces ineffective search space, improves the efficiency of dynamic time warping distance calculation, reduces the false positive rate, and ensures the rigor of semantically redundant dataset partitioning.

[0075] This can be achieved using a constrained dynamic time warping algorithm based on great circle distance and a fixed propagation model. Anchor node parsing provides the network identifiers of two non-anchor edge nodes of the structural feature vector. The latitude and longitude coordinates of both nodes are retrieved from the global node registry, and the actual geographic spatial distance between the two nodes is calculated using the great circle distance formula. The anchor node identifies the event type based on the spatiotemporal characteristics of the current event and retrieves the fixed physical signal propagation speed corresponding to that type of event from the built-in physical attribute table. The geographic spatial distance is divided by the physical signal propagation speed to calculate the maximum propagation delay of the physical signal between the two nodes. The anchor node multiplies this maximum propagation delay by the sampling frequency of the structural feature vector and rounds down to obtain the index offset of the feature sequence, which is used as the dynamic warping window parameter. When constructing the alignment matrix of the two structural feature vectors, the anchor node extends the width of this index offset to both sides of the matrix diagonal to form a strip boundary. When performing dynamic programming to find the shortest path, only the matrix elements within this strip boundary are calculated and updated, ultimately yielding the dynamic time warped distance under constraints.

[0076] Alternatively, a constrained dynamic time warping algorithm based on 3D spatial distance and a dynamic medium model can be used. The anchor node acquires the 3D spatial coordinates (including altitude) of two non-anchor edge nodes and calculates the straight-line geographic distance using the Euclidean distance formula. The anchor node invokes the dynamic medium propagation model based on the event type, inputting the current ambient temperature and humidity parameters to dynamically estimate the corrected physical signal propagation speed. The maximum propagation delay is obtained by dividing the 3D spatial distance by the corrected propagation speed. The anchor node calculates the index offset using the maximum propagation delay and sampling frequency, generating dynamic warping window parameters. The anchor node uses these parameters to construct a parallelogram-shaped constraint boundary in the alignment matrix, limiting the search range of the optimal path. By performing iterative calculations of the cumulative distance within the parallelogram boundary, the optimal alignment path is found, and the dynamic time warping distance is calculated.

[0077] S206. The anchor node obtains the signal-to-noise ratio (SNR) parameters of each non-anchor edge node corresponding to the semantic redundancy dataset, and determines the non-anchor edge node with the largest SNR parameter as the target node. Signal-to-noise ratio (SNR) is a quantitative indicator used to measure the signal quality of disaster recovery data. It represents the ratio between the effective physical signal power and the power of useless background noise in the original acquired data. It can be obtained through various methods, such as hardware-level reading or software algorithm evaluation. A target node is a specific edge node selected by the system from among many nodes with the same semantic data to represent the entire cluster in performing a complete data upload task to the central cloud node. Anchor nodes acquire and compare these SNR parameters to filter out the highest quality data source from semantically redundant datasets. Due to the differences in the physical environments of different edge nodes, the noise interference levels of their acquired data vary. Determining the non-anchor edge node with the highest SNR parameter as the target node ensures that the data held by the selected node has the highest fidelity and can most accurately recreate the true physical state of the sudden event. This data quality-based optimization mechanism improves the overall clarity and usability of the disaster recovery data ultimately received by the central cloud node, reduces the risk of subsequent cloud analysis distortion due to uploaded, severely noise-contaminated, low-quality data, and enhances the disaster recovery capabilities of the disaster recovery system under the cloud-edge collaborative architecture.

[0078] This can be achieved using an active query method based on the hardware-level driver interface. The anchor node extracts a list of network addresses of all associated non-anchor edge nodes based on the semantic redundancy dataset mapped in memory. The anchor node sends a signal-to-noise ratio (SNR) parameter retrieval request to each non-anchor edge node in this list via a control channel. Upon receiving the request, each non-anchor edge node calls the hardware abstraction layer (HAL) application programming interface of its local operating system to directly read the real-time SNR value recorded in the registers of the sensor hardware or baseband processing chip responsible for data acquisition. Each non-anchor edge node encapsulates the read hardware-level SNR value in a response message and sends it back to the anchor node. After receiving the response, the anchor node sorts all the returned SNR values ​​in descending order in memory, extracts the largest SNR value at the top of the list, and updates the network identifier of the non-anchor edge node providing the largest value to the target node's state attribute.

[0079] Alternatively, a software evaluation method based on data fragment variance analysis can be used. The anchor node issues instructions to each associated non-anchor edge node to calculate the signal-to-noise ratio (SNR) of the current data using a software algorithm. Each non-anchor edge node extracts the original difference data fragments from its local cache and calculates the overall variance of these fragments as the total signal power. Subsequently, each non-anchor edge node uses a high-pass filter to extract the high-frequency fluctuation components from the original data and calculates the variance of these high-frequency fluctuation components as the background noise power. Each non-anchor edge node subtracts the background noise power from the total signal power to obtain the effective signal power, then divides the effective signal power by the background noise power to calculate the SNR parameter at the software evaluation level, and sends it to the anchor node. The anchor node sets up a numerical comparator to sequentially read the received SNR parameters, compares them pairwise, continuously updates the current maximum value record, and after traversal, sets the non-anchor edge node corresponding to the final maximum SNR parameter as the target node.

[0080] S207. The anchor node generates and distributes a secondary differentiated upload strategy to control the target node to upload the corresponding data block in the semantically redundant dataset, and to control the other non-anchor edge nodes to upload the residual data block in the local differentiated data after removing the semantically redundant dataset.

[0081] The secondary differentiated upload strategy refers to a refined transmission control scheme tailored to the data redundancy status of different nodes after completing in-depth semantic feature comparison. It typically manifests as a configuration file or instruction set containing specific execution rules. A data block is the basic logical unit for dividing disaster recovery data during network transmission and local storage. A residual data block refers to the set of truly unique data remaining after the semantically redundant data portion has been removed from the locally cached differentiated data of each non-anchor edge node. The anchor node generates and distributes this strategy to transform the preliminary semantic analysis results into substantial network traffic scheduling actions. By directing target nodes to upload complete, high-quality, representative data while restricting other nodes to upload only their unique residual portions, this differentiated scheduling mechanism maximizes the reduction of invalid data load in the network link. This step improves the overall uplink bandwidth utilization of the edge cluster, reduces the storage overhead and data processing pressure of the central cloud node, and enhances the overall efficiency of disaster recovery data synchronization.

[0082] This can be achieved using a policy distribution method based on access control lists and bitmap masks. Anchor nodes generate an access control rule allowing full upload for the target node and filtering rules containing bitmap masks for other non-anchor edge nodes. In this bitmap mask, the positions of data blocks corresponding to semantically redundant datasets are marked as logical zero, and the positions of residual data blocks are marked as logical one. Anchor nodes combine these rules into a secondary differentiated upload policy file and distribute it to each node via a secure socket layer channel. After parsing the policy, the target node directly pushes the corresponding semantically redundant data block into the sending queue and starts transmission. Other non-anchor edge nodes, after parsing the policy, perform a bitwise AND operation between the locally cached differentiated data block index and the received bitmap mask. The system automatically discards data blocks with a logical zero result, extracting only the residual data blocks with a logical one result and packaging them for upload to the central cloud node.

[0083] Alternatively, an instruction scheduling method based on task scheduling queues and dynamic data pointers can be used. Anchor nodes construct a distributed task scheduling instruction set as a secondary differentiated upload strategy. For target nodes, anchor nodes issue upload instructions containing a range of continuous data pointers, instructing them to read and upload the entire semantically redundant dataset. For other non-anchor edge nodes, anchor nodes calculate the physical offset address and length of the residual data blocks in their local storage medium based on the semantic comparison results, generate fragmented upload instructions containing multiple discrete data pointers, and issue them. Upon receiving the instructions, target nodes directly perform direct memory access to read and send large blocks of data based on the continuous data pointers. Other non-anchor edge nodes, upon receiving the instructions, call the file system's random read interface, strictly following the discrete data pointers specified in the instructions, skipping the semantically redundant data portion, accurately locating and reading the residual data blocks, and then re-encapsulating these fragmented residual data blocks into new transmission protocol data units for upload.

[0084] In the above embodiments, the anchor node pauses transmission when the proportion of discrepancies exceeds a preset conflict threshold, avoiding network congestion caused by the simultaneous uploading of a large amount of non-identical but highly similar data. By controlling non-anchor edge nodes to perform envelope extraction and low-frequency component sampling on the discrepancies to generate structural feature vectors, high-frequency random noise and minor fluctuations during data acquisition are filtered out, while preserving the underlying physical change trends of the data. Calculating the dynamic time warping distance between structural feature vectors effectively overcomes the time axis distortion caused by sampling clock deviations or signal transmission delays between different nodes, accurately identifying semantically redundant datasets with different fingerprints but highly consistent semantic content. The node with the highest signal-to-noise ratio is selected as the target node to upload this semantically redundant data, while the remaining nodes are controlled to only upload residual data blocks, further mining deeper levels of similar data, reducing the amount of invalid data transmission caused by noise interference, and improving the overall quality of disaster recovery data received by the central cloud node.

[0085] In the above embodiments, by selecting the target node with the highest signal-to-noise ratio to uniformly upload semantically redundant data, the overall quality of data received by the central cloud node is effectively improved and the amount of invalid transmission is reduced. However, in actual disaster recovery scenarios, the network environment on the edge side is usually harsh and unstable, and the target node is prone to sudden uplink interruption failures during the upload task. If the uploaded semantically redundant data is discarded directly after the link is interrupted and the full transmission is retried, it will cause serious bandwidth waste and recovery delay. In order to ensure the continuity of transmission in complex network environments, this application further proposes another disaster recovery data transmission method based on cloud-edge collaboration, based on the above embodiments. The following is combined with Figure 3 This application describes yet another disaster recovery data transmission method based on cloud-edge collaboration: Please see Figure 3 This is another flowchart illustrating a disaster recovery data transmission method based on cloud-edge collaboration in this application.

[0086] S301. If the anchor node detects that the uplink of the target node has been interrupted, the anchor node extracts the truncation timestamp of the last successfully uploaded data block before the interruption. An uplink refers to the network transmission channel through which edge nodes send data to the central cloud node, such as cellular network links, satellite communication links, or microwave transmission links. An interruption refers to an unexpected disconnection during data transmission due to network congestion, signal attenuation, or hardware failure. A truncated timestamp is the timestamp corresponding to the last data block successfully received and acknowledged by the central cloud node before the link interruption occurs at the target node. Anchor nodes extract this timestamp after detecting a link interruption at the target node, aiming to accurately pinpoint the breakpoint of the data transmission task on the timeline. Extracting the truncated timestamp provides an absolute time reference for subsequent cross-node task relay. This breakpoint recording mechanism reduces the risk of transmitted data becoming invalid due to network failures, minimizes bandwidth waste caused by re-uploading the entire dataset, and improves the recovery agility of disaster recovery systems in harsh network environments.

[0087] This can be achieved using a detection and extraction method based on a heartbeat keep-alive mechanism and Transmission Control Protocol (TCP) acknowledgment messages. The anchor node continuously listens for heartbeat keep-alive data packets sent by the target node in the control channel established with both the target node and the central cloud node. When the anchor node does not receive a heartbeat packet from the target node within a set timeout period, it determines that its uplink has been interrupted. Subsequently, the anchor node sends a status query request to the central cloud node to obtain the last TCP acknowledgment message for the target node's data stream recorded in the central cloud node's network protocol stack. The anchor node parses the sequence number in this acknowledgment message and, based on the mapping table between data block sequence numbers and timestamps maintained internally by the system, reverse-engineers the truncation timestamp of the last successfully uploaded data block before the target node's interruption.

[0088] Alternatively, a detection and extraction method based on edge gateway traffic monitoring and application layer log parsing can be used. The anchor node calls the edge network gateway's traffic monitoring application programming interface to statistically analyze the data throughput of the target node's uplink port in real time. When the data throughput drops to zero or below the minimum threshold for maintaining the basic connection within multiple consecutive sampling periods, the anchor node triggers a link interruption alarm. Then, the anchor node remotely logs into the target node's operating system via the Secure Shell protocol and reads the application layer runtime logs of the disaster recovery data upload agent. The anchor node uses regular expressions to match the last transmission record with a successful transmission marker in the log file, extracts the truncation timestamp information inherent in that data block, and stores it in the anchor node's local memory for later retrieval.

[0089] S302. The anchor node is selected from the remaining non-anchor edge nodes, and the node with the second largest signal-to-noise ratio is selected as the relay node. The node with the second-highest signal-to-noise ratio (SNR) refers to an edge node whose data signal quality assessment value is second only to the original target node in a semantically redundant dataset. A relay node is a replacement node dynamically selected by the system to take over the remaining data upload task when the original upload node fails; examples include backup edge servers or auxiliary acquisition terminals. Anchor nodes are selected from the remaining non-anchor edge nodes as relay nodes, aiming to quickly find the best-quality replacement among currently available nodes in the event of a sudden failure of the primary node. By selecting the node with the second-highest SNR, the continued upload of disaster recovery data maintains high clarity and fidelity. This dynamic node switching mechanism based on data quality priority reduces the negative impact of link failures on the accuracy of cloud data restoration and improves the robustness and fault tolerance of the cloud-edge collaborative architecture in complex environments.

[0090] This can be achieved using a selection method based on memory state table lookup and dynamic sorting. When selecting the initial target node, the anchor node has already obtained the signal-to-noise ratio (SNR) parameters of all non-anchor edge nodes. The anchor node reads the SNR state table residing in its local memory, first marking the status of target nodes whose links have been interrupted as unavailable and removing them from the candidate list. Then, the anchor node calls a quicksort algorithm to sort the SNR parameters of all remaining non-anchor edge nodes in descending order. The anchor node directly extracts the node at the top of the sorted results—that is, the node with the highest SNR among currently available nodes, and the node with the second highest global SNR—updates its network identifier and media access control address to the relay node, and allocates system resources for the relay transmission task to it.

[0091] Alternatively, a selection method based on real-time probing and distributed consensus can be used. Considering that dynamic changes in the edge physical environment may cause historical signal-to-noise ratio (SNR) data to become invalid, the anchor node, upon detecting an interruption, broadcasts a real-time SNR probing request to all active non-anchor edge nodes. Upon receiving the request, each node immediately calls the underlying hardware interface to re-evaluate the SNR of the currently collected data and feeds the result back to the anchor node. After collecting all feedback, the anchor node broadcasts this data as a proposal to all nodes in the edge cluster, initiating the distributed consensus algorithm. Each node cross-validates and compares the SNR values ​​in the proposal, ultimately electing the non-anchor edge node with the highest current real-time SNR (i.e., the second-highest node after the original target node fails) as the successor node. The anchor node records this consensus result and triggers the subsequent succession process.

[0092] S303. Anchor node retrieval is the feature alignment path matrix between the target node and the relay node generated when calculating the dynamic time warp distance. Dynamic time warping distance (DTT) is a quantitative metric used to measure the similarity between two time series that may differ in length. The feature alignment path matrix (MAS) is a two-dimensional data structure, such as a warp matrix or cost backtracking matrix, that records the optimal matching relationship between data points of the two time series during the calculation of this distance. Anchor nodes retrieve the MAS between the target node and the relay node to obtain the timeline mapping relationship formed between these two nodes during historical data acquisition. Since data sampling from different nodes may have phase differences or time offsets, directly using absolute timestamps for continuation can lead to data misalignment. Retrieving this matrix provides a fundamental mathematical model for subsequent accurate breakpoint mapping. This mechanism of reusing existing calculation results reduces the computational overhead of re-performing complex sequence alignments during the fault recovery phase, thus improving the response speed of the relay node.

[0093] This can be achieved using a direct read method based on a local distributed caching system. During the initial semantic redundancy dataset assessment and feature comparison, the anchor node serializes the various feature alignment path matrices it generates and stores them in the distributed memory cache system of the edge cluster. The anchor node generates a composite query key value according to a preset string concatenation rule, based on the unique device identifiers of the target node and the relay node. The anchor node then uses this composite query key value to initiate a read request to the distributed cache system, directly retrieving the corresponding serialized matrix data into its local working memory. Subsequently, the anchor node calls a deserialization function to restore the read byte stream into a two-dimensional array structure of feature alignment path matrices that can be directly addressed and traversed by the program.

[0094] Alternatively, a graph database-based relational query method can be used. Anchor nodes maintain a lightweight graph database locally, where vertices represent edge nodes, edges represent redundant data relationships between nodes, and the feature alignment path matrix is ​​compressed and stored as attribute data on the corresponding edges. Anchor nodes construct graph database query statements, using the target node and relay node as the starting and ending vertices, to retrieve specific relational edges connecting these two vertices. The graph database engine executes the query, locates the relational edge, and extracts its associated matrix attribute fields. Anchor nodes perform a decompression algorithm on the extracted compressed matrix data to recover the complete feature alignment path matrix, which is then loaded into a cache for subsequent truncation timestamp mapping operations to quickly read matrix elements.

[0095] S304. The anchor node maps the truncated timestamp into the feature alignment path matrix to determine the corresponding synchronization continuation index of the truncated timestamp in the local cache data of the relay node. The anchor node maps the truncated timestamp into the feature alignment path matrix to determine the corresponding synchronization continuation index of the truncated timestamp in the local cache data of the relay node. This can be achieved in the following way: The anchor node converts the truncated timestamp into a truncated index value of the target node's local feature sequence based on the target node's data sampling frequency. In the feature alignment path matrix, the anchor node extracts all alignment coordinate pairs with the truncated index value as the base dimension to generate a candidate mapping set. If there are multiple alignment coordinate pairs with different target dimension coordinates in the candidate mapping set, the anchor node determines the target dimension coordinates corresponding to the alignment coordinate pair with the smallest target dimension coordinates as the synchronization continuation index.

[0096] Mapping refers to the process of transforming elements in one data space to their corresponding positions in another data space according to specific rules. Synchronization continuation index refers to the starting data read position in the local cached data sequence of the relay node that is physically aligned with the interruption time of the target node, such as an array index or memory offset. Data sampling frequency refers to the number of times the sensor collects data per unit time. Truncated index value is the relative position number of the truncated timestamp in the target node's own data sequence. Baseline dimension is the row or column direction representing the target node's time axis in the matrix. Alignment coordinate pair is the combination of row and column indices in the matrix that records the matching relationship between two sequence points. Candidate mapping set is a temporary cluster of all possible matching coordinate pairs. Target dimension coordinates are the index values ​​representing the relay node's time axis in the matrix. The anchor node performs this mapping operation to solve the alignment problem of heterogeneous time axes across nodes. Since the dynamic time warping algorithm allows many-to-one matching, by selecting the alignment coordinate pair with the smallest target dimension coordinates, it ensures that the relay node starts uploading from the earliest valid matching point. This precise indexing and transformation mechanism reduces the probability of data overlap or omission during data stream stitching, and improves the temporal continuity of data received in the cloud.

[0097] The above-mentioned refined technical solution can be implemented using a mapping method based on array index operations and matrix traversal. The anchor node obtains the initial sampling time point of the target node, subtracts this initial time point from the truncated timestamp to obtain the time difference, multiplies the time difference by the data sampling frequency of the target node, and rounds down to calculate the truncated index value of the target node's local feature sequence. Next, the anchor node loads the feature alignment path matrix into a two-dimensional array, using the truncated index value as the row index (i.e., the baseline dimension), and traverses all columns in that row. The anchor node extracts all column indices marked as valid alignment paths in that row, combines these row and column indices into alignment coordinate pairs, and stores them in the candidate mapping set. Subsequently, the anchor node traverses the candidate mapping set, compares the column indices (i.e., target dimension coordinates) of all alignment coordinate pairs, selects the column index with the smallest value, and directly assigns it to the synchronous continuation index variable.

[0098] The above-mentioned refined technical solution can also be implemented using a mapping method based on hash tables and priority queues. When parsing the truncated timestamp, the anchor node calls a preset clock conversion function and, combined with the data sampling frequency, directly outputs the corresponding truncated index value. To accelerate the query, the anchor node has previously converted the feature alignment path matrix into a hash table, where the key is the baseline dimension index of the target node, and the value is a minimum priority queue that stores the target dimension coordinates of all relay nodes aligned with it. The anchor node uses the calculated truncated index value as the key to query the hash table and directly obtains the corresponding minimum priority queue as a candidate mapping set. If the queue is not empty and contains multiple elements, the anchor node directly performs a dequeue operation, obtaining the top element of the priority queue. Due to the characteristics of the minimum priority queue, this element is the value with the smallest target dimension coordinate, and the anchor node determines it as the synchronization continuation index.

[0099] S305. The anchor node sends a flow stitching control command to the relay node to control the relay node to stop uploading residual data blocks and to control the relay node to upload complete local cached data to the central cloud node starting from the synchronization continuation index, and to encapsulate the phase compensation identifier generated based on the feature alignment path matrix in the uploaded data header.

[0100] Stream stitching control commands are control signaling generated and issued by the anchor node to instruct relay nodes to switch data streams and resume interrupted data flow. Examples include remote procedure call commands or message queue control messages. Residual data blocks are unique data remaining in the relay node's local cache after removing semantically redundant parts. Phase compensation identifiers are metadata recording the minute phase difference or time offset between the target node and the relay node in the data acquisition timing, such as time offset fields or delay calibration parameters. The anchor node sends this command to the relay node to achieve a smooth handover of the data upload task. Controlling the relay node to stop uploading residual data blocks and switch to uploading complete data ensures that the cloud can receive the missing redundant data. Encapsulating the phase compensation identifier provides calibration parameters for subsequent data fusion at the central cloud node. This command scheduling mechanism enhances the self-healing capability of the edge cluster in response to single-point failures, reduces the complexity of cloud data stitching, and improves the overall disaster recovery efficiency of the disaster recovery system.

[0101] This can be achieved using a command-based approach using a publish-subscribe model and a custom application-layer protocol. The anchor node, based on the calculated synchronization continuation index and the phase difference value extracted from the feature alignment path matrix, serializes the data into a generic data format message containing control opcodes, an index field, and a phase compensation identifier, according to the custom application-layer protocol format. The anchor node, acting as the publisher, publishes this message to the relay node's dedicated command subscription topic in the edge cluster message middleware. Upon receiving a new message, the relay node's daemon immediately parses the message, triggering an interrupt signal to pause the thread currently responsible for uploading residual data blocks. Subsequently, based on the synchronization continuation index in the message, the relay node calls the file system's pointer offset function to precisely locate the read pointer to the corresponding position of the locally cached complete data. It then writes the phase compensation identifier into the header of the generated data packet and starts a new transmission thread to upload data to the central cloud node.

[0102] Alternatively, a control method based on edge container orchestration and dynamic updates of environment variables can be used. The anchor node locates the container instance responsible for data upload on the relay node by calling the container orchestration application programming interface of the edge computing platform. The anchor node formats the synchronization continuation index and phase compensation identifier into key-value pairs and dynamically updates the configuration mapping table or environment variables of the container instance through the interface. When the monitoring script inside the relay node container detects a change in environment variables, it performs a hot reload operation, gracefully shutting down the current residual data upload process and releasing the network port. Then, the monitoring script restarts the full data upload process based on the new environment variable parameters. During initialization, this process reads the synchronization continuation index to set the starting read address of the memory-mapped file and injects the phase compensation identifier into the extended header of the transport protocol, thus beginning the continuous push of complete local cached data to the central cloud node.

[0103] In the above embodiments, when the anchor node detects an uplink interruption at the target node, it extracts the truncation timestamp of the last successfully uploaded data block and selects the node with the second-highest signal-to-noise ratio as the relay node. This allows for a rapid response to link failures while ensuring data quality. Mapping the truncation timestamp using the feature alignment path matrix between the target node and the relay node accurately converts the physical interruption time of the target node into a synchronization continuation index in the relay node's local cached data sequence, achieving precise alignment across heterogeneous timelines. Sending a flow stitching control command containing this index and a phase compensation identifier to the relay node enables it to directly resume uploading complete cached data from the corresponding data breakpoint. This cross-node breakpoint continuation mechanism avoids the complete discarding and retransmission of uploaded semantically redundant data due to a single node link failure, reduces ineffective network bandwidth consumption in unstable link scenarios, and improves the transmission continuity and overall recovery efficiency of disaster recovery data in complex network environments.

[0104] The system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference needed]. Figure 4 This is a schematic diagram of the physical device structure of a disaster recovery data transmission system based on cloud-edge collaboration, provided in an embodiment of this application.

[0105] It should be noted that, Figure 4 The structure of the system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0106] like Figure 4 As shown, the system includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 402 or a program loaded from storage portion 408 into Random Access Memory (RAM) 403, such as executing the methods described in the above embodiments. The RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via a bus 404. An Input / Output (I / O) interface 405 is also connected to the bus 404.

[0107] The following components are connected to I / O interface 405: input section 406 including a camera, infrared sensor, etc.; output section 407 including a liquid crystal display (LCD) and speakers, etc.; storage section 408 including a hard disk, etc.; and communication section 409 including a network interface card such as a LAN (Local Area Network) card and a modem, etc. Communication section 409 performs communication processing via a network such as the Internet. Drive 410 is also connected to I / O interface 405 as needed. Removable media 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 410 as needed so that computer programs read from it can be installed into storage section 408 as needed.

[0108] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit (CPU) 401, it performs the various functions defined in the present invention.

[0109] It should be noted that the computer-readable medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. The transmitted data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof.

[0110] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0111] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the system described in the above embodiments; or it may exist independently and not assembled into the system. The storage medium carries one or more computer programs that, when executed by a processor of a system, cause the system to implement the methods provided in the above embodiments.

[0112] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0113] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0114] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0115] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A disaster recovery data transmission method based on cloud-edge collaboration, applied to distributed smart grids or wide-area vehicle-road cooperative application scenarios, implemented through a system including a central cloud node and multiple edge nodes, characterized in that... The method includes: When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it slices and aligns the locally cached disaster recovery data based on a preset time window, and extracts the event spatiotemporal features of the disaster recovery data, specifically including: When each edge node detects that the received disaster recovery data traffic exceeds a preset threshold, it calculates the peak traffic time point of the locally cached disaster recovery data during the traffic burst, and uses the peak traffic time point as the event alignment anchor point. Based on the event alignment anchor point, a preset time window is constructed by extending forward and backward according to a preset time span; The local cached disaster recovery data is truncated and divided into blocks according to the preset time window to obtain disaster recovery data slices after slice alignment. Extract the set of device source geographic coordinates of the disaster recovery data slice and the event alignment anchor point, and combine them to generate a multi-dimensional spatiotemporal vector; The multidimensional spatiotemporal vector is dimensionality reduced and encoded to obtain the event spatiotemporal features of the disaster recovery data; Each edge node performs spatiotemporal gridding dimensionality reduction processing on the event spatiotemporal features to obtain standardized event spatiotemporal features; Each edge node inputs the standardized event spatiotemporal features into a consistent hash function, maps them in a virtual ring composed of the edge nodes, and independently determines the anchor node corresponding to the event spatiotemporal features. If the current edge node is not the anchor node, calculate the aligned disaster recovery data slice data fingerprint and send the data fingerprint to the anchor node; The anchor node compares the received data fingerprints with the local data fingerprints to determine the global redundant data shared across nodes and the differential data unique to each edge node, and then sends a differentiated upload strategy to the corresponding non-anchor edge node. The non-anchor edge node uploads the differential data from the locally cached disaster recovery data to the central cloud node according to the differential upload strategy; The anchor node uploads the global redundant data and local differential data to the central cloud node.

2. The method according to claim 1, characterized in that, The anchor node compares the received data fingerprints with its local data fingerprints to determine the globally redundant data shared across nodes and the unique difference data of each edge node, and then issues a differentiated upload strategy to the corresponding non-anchor edge node, specifically including: The anchor node constructs a local hash tree using the local data fingerprint as leaf nodes, and extracts the hash values ​​of each level node of the local hash tree; Each received data fingerprint is compared layer by layer with the hash value of each node at each level of the local hash tree to determine the target data fingerprint that matches the comparison. The disaster recovery data corresponding to the target data fingerprint is determined as globally redundant data shared across nodes, and the disaster recovery data corresponding to the unmatched remaining data fingerprint is determined as differential data unique to each edge node. Based on the sequence position of the remaining data fingerprint in the corresponding non-anchor edge node, a differentiated upload strategy consisting of binary bits is generated; The differentiated upload strategy is then distributed to the corresponding non-anchor edge nodes.

3. The method according to claim 1, characterized in that, After issuing the differentiated upload strategy to the corresponding non-anchor edge node, the method further includes: The anchor node calculates the proportion of the differential data to the total data volume under the current standardized event spatiotemporal characteristics. If the ratio is greater than a preset conflict threshold, the anchor node sends an upload suspension command to the corresponding non-anchor edge node to suspend the transmission of the difference data to the central cloud node. The anchor node generates feature extraction configuration parameters and sends the feature extraction configuration parameters to each non-anchor edge node that caches the difference data; Each of the non-anchor edge nodes performs envelope extraction and low-frequency component sampling on the locally cached difference data according to the feature extraction configuration parameters, generates a structural feature vector, and feeds the structural feature vector back to the anchor node; The anchor node calculates the dynamic time warping distance between each received structural feature vector, and associates and maps the difference data where the dynamic time warping distance is less than a preset distance threshold into a semantic redundancy dataset in memory; The anchor node obtains the signal-to-noise ratio (SNR) parameters of each non-anchor edge node corresponding to the semantic redundancy dataset, and determines the non-anchor edge node with the largest SNR parameter as the target node. The SNR parameter is a quantitative indicator used to measure the signal quality of disaster recovery data, and represents the ratio between the effective physical signal power and the useless background noise power in the original collected data. The anchor node generates and distributes a secondary differentiated upload strategy to control the target node to upload the corresponding data block in the semantic redundancy dataset, and to control the other non-anchor edge nodes to upload the residual data block in the local differentiated data after removing the semantic redundancy dataset.

4. The method according to claim 3, characterized in that, The anchor node calculates the dynamic time-warped distance between the received structural feature vectors, specifically including: The anchor node obtains the geographic spatial distance between two non-anchor edge nodes that provide the structural feature vector; The anchor node obtains the corresponding physical signal propagation speed based on the event type in the current standardized event spatiotemporal characteristics, and calculates the maximum propagation delay of the physical signal between the two non-anchor edge nodes in combination with the geographic spatial distance. The anchor node maps the maximum propagation delay to the index offset of the feature sequence based on the sampling frequency of the structural feature vector, and generates dynamic warping window parameters. The anchor node constrains the boundary of the optimization path in the alignment matrix of the structural feature vector based on the dynamic warping window parameters, and calculates the dynamic time warping distance only within the boundary.

5. The method according to claim 3, characterized in that, After the anchor node generates and distributes the secondary differentiated upload strategy, the method further includes: If the anchor node detects that the uplink of the target node has been interrupted, the anchor node extracts the truncation timestamp of the last successfully uploaded data block before the interruption. The anchor node selects the node with the second largest signal-to-noise ratio from the remaining non-anchor edge nodes as the relay node; The anchor node retrieves the feature alignment path matrix between the target node and the relay node, which was generated when calculating the dynamic time warp distance. The anchor node inputs the truncated timestamp into the feature alignment path matrix for mapping, and determines the synchronization continuation index corresponding to the truncated timestamp in the local cache data of the relay node; The anchor node sends a flow stitching control command to the relay node to control the relay node to stop uploading residual data blocks and to control the relay node to upload complete local cached data to the central cloud node starting from the synchronization continuation index, and to encapsulate the phase compensation identifier generated based on the feature alignment path matrix in the uploaded data header.

6. The method according to claim 5, characterized in that, The anchor node maps the truncated timestamp into the feature alignment path matrix to determine the synchronization continuation index corresponding to the truncated timestamp in the local cache data of the relay node, specifically including: The anchor node converts the truncated timestamp into a truncated index value of the local feature sequence of the target node according to the data sampling frequency of the target node; The anchor node extracts all alignment coordinate pairs with the truncation index value as the base dimension from the feature alignment path matrix to generate a candidate mapping set. If there are multiple alignment coordinate pairs with different target dimension coordinates in the candidate mapping set, the anchor node determines the target dimension coordinates corresponding to the alignment coordinate pair with the smallest target dimension coordinates as the synchronization continuation index.

7. A disaster recovery data transmission system based on cloud-edge collaboration, characterized in that, The system includes: One or more processors and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described in any one of claims 1-6.

8. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the system, the system performs the method as described in any one of claims 1-6.

9. A computer program product, characterized in that, When the computer program product is run on the system, the system performs the method as described in any one of claims 1-6.

Citation Information

Patent Citations

  • Underground cable safety assessment method and system based on cloud edge cooperation and multi-agent deep reinforcement learning

    CN119337227A

  • Versatile data reduction for internet of things

    US20220129426A1