Data migration method and device, computer equipment and storage medium

Abstract

The invention discloses a data migration method and device, computer equipment and a storage medium, and realizes a data migration processing mode oriented to an unstable network environment by combining neural network prediction, decision tree anomaly recognition and a distributed cache mechanism. Therefore, the data transmission stability and the migration task completion rate in the cross-network environment are obviously improved. Firstly, a pre-trained neural network is utilized to carry out deep prediction on a real-time network state, and a potential performance reduction trend is identified in advance, so that a migration process can actively adjust a path or a strategy before a risk occurs. And secondly, by performing decision tree identification on a data packet sending sequence, the system can position a potential interruption point based on fine-grained time sequence characteristics, and quantify the risk in an interpretable manner, so that migration control can be accurate to a specific data packet level. Finally, under the condition that the risk is unacceptable, the system automatically activates a distributed cache mechanism, and the durability and the restorability of the data are guaranteed in a redundant copy storage mode.

Description

Technical Field

[0001] This application belongs to the field of data migration technology, specifically relating to a data migration method, apparatus, computer equipment, and storage medium. Background Technology

[0002] In the context of information technology development and enterprise digital transformation, data migration plays a crucial role in ensuring business continuity and system upgrades. Whether it's updating enterprise systems or integrating data across platforms, data migration directly impacts the integrity of business data and the stability of the system. However, current data migration methods often reveal deep-seated shortcomings when dealing with complex environments and dynamic business needs, failing to meet the modern enterprise's requirements for efficiency, security, and seamless transition.

[0003] Existing methods often fail to address the challenges of data migration across multiple environments and scenarios due to a lack of proactive assessment of network performance and system compatibility, leading to unforeseen delays or failures during the migration process. Furthermore, many solutions lack effective contingency mechanisms for handling exceptional circumstances, increasing the risk of business interruptions or data inconsistencies. This problem stems not merely from limitations in technical tools, but from insufficient attention to environmental complexity and dynamic changes during migration planning and execution, ultimately impacting the overall reliability of the migration.

[0004] Solving this problem first requires accurate prediction of network environment and transmission speed. Failure to assess network stability in advance can lead to prolonged delays or interruptions during data migration. This uncertainty further exacerbates another critical issue: how to quickly restore and maintain the normal operation of the original system in the event of migration failure or data anomalies. For example, in some business scenarios, if data synchronization with the new system fails and a timely rollback to the original system is not possible, the enterprise's core business processes may come to a standstill, or even result in the risk of data loss.

[0005] Therefore, how to achieve efficient and stable data migration in complex network environments, while ensuring business continuity under abnormal circumstances, has become a key issue that urgently needs to be addressed. Summary of the Invention

[0006] The purpose of this application is to provide a data migration method, apparatus, computer equipment, and storage medium to address the issues of high efficiency and stability in data migration under complex network environments, while ensuring business continuity under abnormal circumstances.

[0007] To address the aforementioned technical problems, this application provides a data migration method, employing the following technical solution: A data migration method, comprising: Real-time network transmission data is collected, and a pre-trained neural network model is used to extract features from the network transmission data and predict network performance to obtain estimated values ​​of network performance indicators. If the estimated value of the network performance indicator is lower than the preset indicator threshold, the backup transmission path is activated and the path configuration of the backup transmission path is obtained. Obtain the data packet sending sequence of the data to be migrated from the path configuration, and use the decision tree algorithm to identify potential interruption points in the data packet sending sequence; Determine the interruption risk level of each potential interruption point and compare the interruption risk level with the preset risk level standard; If the interruption risk level is lower than the preset risk level standard, the data to be migrated will be encapsulated into data packets and sent to the target migration database based on the data packet sending sequence. To address the aforementioned technical problems, this application also provides a data migration apparatus, which employs the following technical solution: A data migration apparatus, comprising: The network prediction module is used to collect real-time network transmission data, and uses a pre-trained neural network model to extract features from the network transmission data and predict network performance to obtain estimated values ​​of network performance indicators. The backup transmission module is used to activate the backup transmission path and obtain the path configuration of the backup transmission path when the estimated value of the network performance index is lower than the preset index threshold. The interruption identification module is used to obtain the data packet sending sequence of the data to be migrated from the path configuration, and to identify potential interruption points in the data packet sending sequence using a decision tree algorithm; The risk assessment module is used to determine the interruption risk level of each potential interruption point and compare the interruption risk level with the preset risk level standard. The data migration module is used to encapsulate the data to be migrated into data packets when the interruption risk level is less than the preset risk level standard, and send the data packets to be migrated to the target migration database based on the data packet sending sequence. The temporary cache module is used to activate the preset data caching mechanism when the interruption risk level is greater than or equal to the preset risk level standard, and temporarily store the data to be migrated in the distributed buffer to obtain a cached data copy.

[0008] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: A computer device includes a memory and a processor, the memory storing computer-readable instructions, the processor executing the computer-readable instructions to implement the steps of the data migration method as described in any of the preceding claims.

[0009] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below: A computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of the data migration method as described in any one of the preceding descriptions.

[0010] Compared with the prior art, the embodiments of this application have the following main advantages: This application discloses a data migration method, apparatus, computer equipment, and storage medium. By combining neural network prediction, decision tree anomaly identification, and distributed caching mechanisms, it achieves an intelligent, adaptive, and highly reliable data migration approach for unstable network environments, significantly improving the stability of data transmission across network environments and the completion rate of migration tasks. First, a pre-trained neural network is used to perform deep prediction of real-time network conditions, identifying potential performance degradation trends in advance. This allows the migration process to proactively adjust paths or strategies before risks occur, avoiding the limitations of traditional methods that can only passively handle failures after they happen. Second, by using decision tree identification on the data packet transmission sequence, the system can locate potential interruption points based on fine-grained temporal characteristics and quantify risks in an interpretable manner, enabling migration control to be precise down to the specific data packet level, thereby reducing unnecessary global interruptions. Finally, in cases where the risk is unacceptable, the system automatically activates the distributed caching mechanism, using redundant copy storage to ensure data persistence and recoverability, achieving continuity and fault tolerance in the migration process. This application enables the system to possess predictive, adaptive, and highly available characteristics through the synergy of neural network prediction, decision tree anomaly identification, and distributed caching mechanisms. It can maintain a high success rate of stable migration in complex and ever-changing network environments, reduce the impact of network fluctuations on core business, and has significant engineering value and practical application effects. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied; Figure 2 A flowchart of one embodiment of the data migration method according to this application is shown; Figure 3 It shows Figure 2A flowchart of one embodiment of step S201; Figure 4 A schematic diagram of one embodiment of the data migration apparatus according to this application is shown; Figure 5 It shows Figure 4 A schematic diagram of a embodiment of the network prediction module 401; Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0017] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0020] It should be noted that the data migration method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the data migration device is generally set in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0022] Continue to refer to Figure 2 A flowchart of an embodiment of a data migration method according to this application is shown. The data migration method includes the following steps: S201: Collect real-time network transmission data, use a pre-trained neural network model to extract features from the network transmission data and predict network performance, and obtain an estimated value of network performance indicators. Specifically, the system needs to collect multi-dimensional network transmission data from network devices (such as switches, routers, and host network interface cards) and application layer proxies, including instantaneous bandwidth utilization, end-to-end latency (RTT), round-trip latency distribution, packet loss rate, jitter statistics, TCP / UDP retransmission events, traffic burst indicators, connection establishment / disconnection frequency, and link layer error counts. In addition, it can collect environmental context information, such as the current routing path topology identifier, link bandwidth limit, QoS label, time window (period), and historical statistical aggregation values.

[0023] To ensure the consistency of model input, network-transmitted data should undergo time alignment (timestamp normalization), missing value imputation (forward filling or model-based interpolation), outlier detection and removal (based on Z-score or IQR), and unit normalization / standardization before flowing into the model.

[0024] Secondly, pre-trained neural network models can employ hybrid architectures to simultaneously capture short-term transient features and long-term temporal dependencies: for example, first extracting morphological features within a local time window using a set of one-dimensional convolutional (1D-CNN) or residual convolutional layers at the input layer, then modeling long-term dependencies through gated recurrent units (GRU) or long short-term memory (LSTM) modules, or using Transformer self-attention modules to capture the interactive effects between different time locations. The model output can be either a regressive prediction of continuous network performance metrics (such as predicting packet loss rate / latency in the next n seconds) or a categorical quality rating (such as "high / medium / low"). The large-scale offline dataset used in the pre-training phase should cover various network conditions and fault types, and loss functions (such as weighted MSE, Huber loss, or classification cross-entropy with chi-square correction) should be used to balance the contribution of rare but critical fault samples.

[0025] Finally, online inference requires low latency and stability, which is typically achieved through model distillation, quantization (int8 or lower), pruning, and near real-time batch processing (micro-batch) to reduce computational load and response latency. The model's output estimates should also include confidence or uncertainty estimates (e.g., using Monte Carlo Dropout, Bayesian approximation, or output distribution parameterization).

[0026] S202, If the estimated value of the network performance indicator is lower than the preset indicator threshold, then the backup transmission path is started and the path configuration of the backup transmission path is obtained. Specifically, the preset metric thresholds can be either fixed values ​​(e.g., latency > 200ms or packet loss rate > 1%) or relative / adaptive thresholds (e.g., a percentage decrease of more than 20% compared to the historical baseline). The preset metric threshold settings must be based on the tolerance of business SLAs and migration operations, and can be tiered according to the priority of the migration data (critical business data, non-critical batch processing, etc.). The system typically maintains a threshold policy library, and when a trigger condition occurs, it determines whether to immediately switch, downgrade and retry, or observe and wait based on policy priority to avoid oscillations caused by frequent switching due to misjudgments. To reduce false triggers, decisions are usually made by combining model confidence and short-term sliding window averaging: the backup path is only triggered when the estimated value continuously exceeds the threshold and the model uncertainty is within an acceptable range.

[0027] Backup paths can be pre-calculated and maintained by the network control plane (such as an SDN controller), or they can be generated instantly through real-time routing protocols or intelligent path calculation (based on constraint shortest path and multi-path fault tolerance). The startup process includes submitting a switchover request to the path calculation module, querying the set of available paths, evaluating the capacity of candidate paths, estimating end-to-end latency, and performing security and policy compliance checks on the paths (such as whether they cross restricted network domains). Path configuration acquisition should not only involve the next-hop list but also include per-hop MTU, bandwidth reservation, priority / QoS markings, congestion control recommendations, and possible encryption / tunneling information. If the system is an SDN architecture, flow table rules, flow rate limiting policies, and monitoring probe configurations also need to be issued to continue collecting performance data for closed-loop verification after the backup path goes online.

[0028] Path switching should be atomic as much as possible to avoid data inconsistencies caused by fragmentation or duplicate transmissions. If a failure occurs during the switching process (insufficient backup path capacity or configuration conflicts), the system should support a quick rollback to the primary path or attempt the next candidate path, while recording the switching event for offline analysis. Furthermore, to reduce switching latency and service interruption risks, a "gradual switching" mode can be pre-established: first migrate non-critical segment traffic to the backup path and observe its stability, then gradually increase the proportion of migrated traffic. All these actions need to be closely coordinated with the management plane, caching mechanisms, and risk identification modules to ensure that the decision to trigger the backup path is both timely and robust.

[0029] S203: Obtain the data packet sending sequence of the data to be migrated from the path configuration, and use the decision tree algorithm to identify potential interruption points in the data packet sending sequence; Specifically, path configuration typically includes metadata such as bandwidth, latency, queue information, and forwarding rules for each hop link. From this, the sending window and acknowledgment (ACK) sequence of data packets in the time dimension can be derived. The data to be migrated forms a sending sequence at the packet level, recording the timestamp, sequence number, size, retransmission flag, acknowledgment time of the sender and receiver, and any error codes from the link layer or transport layer for each packet. This time-series data needs to be used to construct a feature vector table in a unified format (e.g., the sending interval of each packet, the sending rate gradient, the packet loss / retransmission count of neighboring packets, and RTT samples), and further sliding window statistical features (such as the average packet loss rate, maximum latency, and jitter variance within the window) can be constructed.

[0030] Decision tree algorithms (or their ensembles such as random forests and gradient boosting trees) excel at handling mixed-type data and highly interpretable splitting rules, making them well-suited for identifying potential outage points. Key characteristics include: a sudden increase in packet loss rate within a short period, an abnormally large increase in transmission intervals (potentially indicating congestion or blocking), abrupt changes in ACK latency, a continuous retransmission count threshold, and instantaneous bandwidth requirements that do not match the capacity of a hop in the path configuration. During the training phase, historical migration logs and simulated fault injection data, labeled "outage / non-outage" or "high-risk / low-risk," can be used for supervised learning. Model optimization requires attention to sample imbalance (scarcity of outage events), typically employing undersampling or cost-sensitive learning to improve recall for the minority class.

[0031] Decision tree inference should support real-time evaluation at the packet-by-packet or sliding window level: calculate the split path for each packet or packet group to be sent and determine whether it hits the interruption judgment rule; once triggered, record the location as a "potential interruption point". Simultaneously, retain the decision path and feature contribution information to facilitate risk level assessment and root cause localization by manual or automated modules (e.g., indicating whether it is caused by link congestion, increased link error rate, or routing jitter). Furthermore, to reduce the false alarm rate, the system can combine threshold buffering (e.g., requiring all k consecutive windows to be judged as abnormal) or fuse neural network prediction results (the confidence level of S201) to adopt a hybrid judgment strategy to improve stability and accuracy.

[0032] S204, determine the interruption risk level of each potential interruption point and compare the interruption risk level with the preset risk level standard; Specifically, the risk level should be a multi-dimensional, adjustable quantitative representation, typically including a probabilistic risk score (e.g., a probability prediction between 0 and 1), a graded label (e.g., low / medium / high / severe), and a confidence interval. Risk assessment can comprehensively utilize multi-source information: the feature contribution of decision tree labels, the uncertainty of neural network predictions of short-term network performance (e.g., packet loss rate within the next 30 seconds), the health indicators of each hop link along the path, and historical failure rates weighted by business importance. The assessment model can be a lightweight weighted scorer (linear or non-linear), or further employ a Bayesian fusion method to combine the outputs of multiple models (decision tree, neural network, statistical threshold) into a final risk probability and provide a confidence interval.

[0033] Preset risk level standards should simultaneously consider business tolerance (e.g., short-term packet loss is acceptable for certain migration tasks, but data loss is not), cost / latency trade-offs, and overall system stability strategies. Standards can be static thresholds (e.g., a risk probability ≥ 0.7 is considered "high risk") or dynamic thresholds (adaptively adjusted based on current system load, the abundance of available backup paths, or time sensitivity). The comparison process, in addition to simple numerical comparisons, should include entropy / confidence assessment: when the risk probability approaches the threshold and the model confidence is low, an "observation + reassessment" strategy should be triggered, or a secondary assessment should be initiated (e.g., waiting for the next time window or conducting more detailed probing) to avoid frequent erroneous actions due to transient noise.

[0034] The risk level assessment directly determines whether to continue sending, trigger caching, or change the transmission strategy. Therefore, auditability and traceability must be guaranteed: each risk assessment should record input characteristics, intermediate outputs of each model, final fusion weights, and decision rationale (e.g., "due to three consecutive retransmissions and a predicted future packet loss probability > 0.8"). Furthermore, an alarm and notification mechanism should be implemented to report high-risk events to the operations and maintenance console, and manual policy overriding should be supported. To improve system robustness, it is recommended to also retain a "security confidence fallback" mechanism: if the assessment component or its data source is abnormal (e.g., monitoring is missing or the model fails), a more conservative threshold should be used or a caching mechanism should be directly implemented to prevent data loss.

[0035] S205 If the interruption risk level is less than the preset risk level standard, the data to be migrated will be encapsulated into data packets and sent to the target migration database in data packet form based on the data packet sending sequence.

[0036] Specifically, before entering the network, data needs to be encapsulated according to the target database's receiving protocol and network transmission best practices. This includes fragmentation size (considering path MTU and link characteristics), sequence number embedding, checksum / hash (such as CRC or SHA) to ensure end-to-end integrity, and optional forward error correction (FEC) encoding to reduce the impact of minor packet loss on the overall migration. Regarding fragmentation strategies, adaptive fragmentation is recommended: dynamically adjusting packet size to match the current estimated bandwidth and latency, reducing retransmission overhead caused by fragmentation, while ensuring high throughput of the transmission channel. If business requirements permit, aggregation can be enabled to reduce header overhead, but the retransmission cost of aggregation must be balanced.

[0037] Data transmission should strictly adhere to the previously generated packet transmission sequence, including transmission timestamps, priority ordering, and retransmission strategies. Congestion control algorithms (such as improved TCP-based algorithms or application-layer rate control) should be employed to avoid path congestion. In multi-path or SR-TE scenarios, traffic distribution strategies can be used to allocate different fragments to different sub-paths according to priority or size. Acknowledgment and retransmission mechanisms also need to be designed: the receiver sends back an acknowledgment (ACK) after successfully writing to the target migration database, and the sender maintains a sliding window and triggers retransmission upon timeout or receipt of a duplicate ACK.

[0038] Even under low-risk conditions, the system must maintain end-to-end consistency guarantees: segmented integrity checks are implemented during transmission, and atomic write strategies are employed at the receiving end (e.g., writing to a temporary table before switching metadata pointers), with data verification (e.g., hash comparison) performed after writing. Key metrics (bandwidth usage, actual latency, packet loss, ACK latency) should be monitored in real-time during transmission, and this monitoring data should be fed back to the risk assessment module for closed-loop processing. If monitoring indicates a sudden performance degradation, the system must support immediate cancellation of packet transmission and switching to a buffer or backup path. This improves transmission efficiency and enables rapid response to changes in risk, ensuring data security and consistency.

[0039] S206 If the interruption risk level is greater than or equal to the preset risk level standard, the preset data caching mechanism is activated, and the data to be migrated is temporarily stored in the distributed buffer to obtain a cached data copy.

[0040] Specifically, when the risk level reaches or exceeds a preset standard, the system prioritizes data integrity and recoverability rather than forcibly continuing the migration. The goal of the caching mechanism is to retain copies of the migrated data at the network edge or controlled nodes in a low-latency, highly reliable manner, and then safely complete the migration once network conditions are restored. Triggering conditions should consider not only the risk level but also cache resource availability, expected cache duration, and data priority. If cache node resources are insufficient, priority scheduling or tiered caching (hot caching / cold caching) strategies can be used to ensure that critical data is protected first.

[0041] Distributed buffers typically consist of multiple cache nodes, which can be deployed in different network locations to increase fault tolerance. Before data is written, it needs to be chunked, with each data block accompanied by metadata (source task ID, sequence number, checksum, timestamp, priority, etc.). Replication strategies can employ master-slave replication, multi-master replication, or erasure coding to balance space and fault tolerance: master-slave replication is simple but has high storage overhead, while erasure coding has high storage efficiency but higher read / write latency. The choice of strategy should be determined based on the expected cache duration, network bandwidth, and recovery time objective (RTO). The write process should use a reliable transport protocol (such as TLS+ acknowledgment mechanism) and perform consistency checks (such as hash comparison) after successful write to ensure replica integrity.

[0042] After each successful write, the system generates a cache index record (including the node where each replica resides, verification information, storage duration, expiration policy, etc.) and writes the index to a centralized or distributed cache management table for quick location. Cache management also needs to include lifecycle management (expiration / reclaiming by policy), space alerts, replication health checks, and read / write performance monitoring. When the network recovers or the risk decreases, the recovery module selects the optimal recovery source (lowest latency and best consistency) based on the index and retransmits the data to the target migration database according to the original sending sequence or a new sending sequence; after successful recovery, the corresponding cache replicas should be cleaned up to release resources. The entire caching mechanism must also meet security and compliance requirements: cached data should be encrypted both statically and during transmission, access should be controlled, and complete audit logs should be maintained to meet regulatory or retrospective needs.

[0043] Further, please refer to Figure 3 The steps involved in collecting real-time network transmission data, using a pre-trained neural network model to extract features from the network transmission data and predict network performance to obtain estimated network performance metrics include: S301 collects multi-dimensional parameters from real-time network transmission data and performs time-series processing on these parameters, which include bandwidth utilization, network latency, packet loss rate, and jitter. S302, preprocess the multi-dimensional parameters after time serialization, and input the preprocessed multi-dimensional parameters into the input layer of the neural network model to generate the input representation vector; S303 uses the feature extraction layer of the neural network model to perform nonlinear mapping on the input representation vector, extracting deep features across the time dimension to obtain the network state feature vector; S304, input the network state feature vector into the prediction layer of the neural network model, and achieve the prediction of network performance indicators through the combination of feature weighting and activation function; S305 quantifies the predicted output of the neural network model and generates estimated values ​​of the network performance metrics.

[0044] In this embodiment, the neural network model relies on high-quality data representation before predicting network performance. Therefore, in S301, not only are core network performance indicators such as bandwidth utilization, network latency, packet loss rate, and jitter collected, but the original collected data is also serialized using timestamps to ensure that the temporal correlation of network behavior is fully preserved. In this way, the model can perceive the dynamic changes in network state within a continuous time window, rather than making judgments based solely on isolated instantaneous data. In the preprocessing stage of S302, normalization, noise filtering, and missing value imputation are performed on the sequence data to ensure the stability and trainability of the model input and generate an input representation vector that accurately reflects the network's changing trends. Subsequently, in S303, the model's feature extraction layer (e.g., one-dimensional convolution, recurrent neural network units, or self-attention structures) transforms the input representation vector into deep network state features that can characterize network congestion patterns, latency fluctuations, and short-term trends through multi-layer nonlinear mapping, enabling the system to understand the network's operating state from a higher dimension. In S304, these deep features are fed into the prediction layer, where feature weighting, linear transformation, and activation function combinations are used to output the future network performance trend. Finally, in S305, the prediction results are smoothed, discretized, or mapped to intervals using quantization strategies to form a network performance metric estimate that can be used for decision-making, facilitating path switching, risk identification, and data transmission strategy selection.

[0045] By following the steps above, potential risks can be identified in advance before network performance deteriorates, enabling accurate prediction of network status and significantly improving the intelligence and reliability of path selection, risk avoidance, and transmission scheduling during data migration.

[0046] Furthermore, the steps of inputting the network state feature vector into the prediction layer of the neural network model, and predicting the network performance index through a combination of feature weighting and activation functions, specifically include: The network state feature vector is input into the fully connected weight matrix of the prediction layer. The network state feature vector is then linearly weighted by the fully connected weight matrix to generate the first intermediate prediction value. Based on the first predicted intermediate value, a bias parameter is introduced for bias compensation to obtain the second predicted intermediate value for nonlinear mapping; The activation function of the neural network model is applied to the second predicted intermediate value to generate a nonlinear response value that characterizes the trend of network performance. The nonlinear response values ​​are input into the prediction output layer. By weighting and fusing the nonlinear response values ​​and performing a linear transformation, preliminary prediction results of network performance indicators are generated. Based on the preliminary prediction results, the output calibration parameters saved during the model training phase are called to perform scale calibration and error compensation on the predicted values, and the final network performance index prediction results are obtained.

[0047] In this embodiment, the network state feature vector serves as the input to the neural network prediction layer. First, a linear weighted operation is performed through the fully connected weight matrix to form a first intermediate prediction value. This step combines different feature dimensions according to their weights, highlighting the features that have the greatest impact on network performance changes. The second intermediate prediction value, obtained after introducing bias parameters, adjusts the first intermediate value to compensate for potential system bias or historical mean deviations in feature combinations, making the model output more stable and accurate. In the activation function processing stage, the model transforms the linear weighted result into a nonlinear response value that reflects complex network behavior patterns through nonlinear mapping. This step captures nonlinear features such as sudden changes in network latency, bandwidth fluctuations, and instantaneous changes in packet loss rate. When the nonlinear response value is input to the prediction output layer, the model generates preliminary prediction results through weighted fusion and linear transformation. The fusion process takes into account the contribution of each feature at different time scales, improving the robustness and accuracy of the prediction. Finally, the output calibration parameters saved during the training phase are used to perform scale calibration and error compensation on the preliminary results. This corrects the bias caused by differences in the distribution of training data and actual online data, yielding the final predicted value of the network performance index.

[0048] By following the steps above, we can achieve high-precision prediction of complex network conditions, identify performance change trends in advance, and thus significantly improve the reliability and transmission efficiency of data migration path selection.

[0049] Furthermore, the steps of obtaining the data packet transmission sequence of the data to be migrated from the path configuration and identifying potential interruption points in the data packet transmission sequence using a decision tree algorithm specifically include: The timing features of the data packet transmission sequence are extracted based on the path configuration of the backup transmission path. The timing features include the data packet transmission timestamp, sequence number, acknowledgment time, and retransmission marker information. The temporal features are input into the feature generation node of the decision tree algorithm according to the preset feature matrix format, and the split gain of each feature is analyzed to generate candidate split nodes. Based on the gain magnitude and information entropy change of the candidate split nodes, the optimal split threshold for dividing the normal and abnormal segments of data packet transmission is selected, and a decision path for identifying the risk of interruption is generated. The decision path is used to judge the data packet transmission sequence packet by packet. When the judgment condition of abnormal segment is met, the corresponding position is marked as a potential interruption point and the set of potential interruption points is output.

[0050] In this embodiment, firstly, based on the path configuration of the backup transmission path, a complete data packet transmission sequence is extracted from the data to be migrated. This sequence includes multi-dimensional temporal features such as the transmission timestamp, sequence number, acknowledgment time, and retransmission marker for each data packet. These features comprehensively reflect the latency, packet loss, and retransmission behavior of data packets during network transmission. When these temporal features are constructed into a feature matrix and input into the decision tree algorithm, the system generates candidate split nodes by analyzing the split gain and information entropy changes of each feature, thereby determining which features and thresholds best distinguish normal transmission from potential anomalies. When selecting the optimal split threshold, the decision tree identifies nodes that maximize the splitting effect by optimizing the information gain or Gini coefficient, forming a decision path for determining interruption risk. Subsequently, the system uses the generated decision path to judge the data packet sequence packet by packet, gradually scanning the transmission sequence. When the features of a data packet meet the abnormal segment judgment conditions, the position is marked as a potential interruption point, and all potential interruption points are output centrally. The entire process enables fine-grained risk analysis at the data packet level, allowing the system to accurately locate key nodes that may cause transmission interruption during the migration process, thereby effectively reducing the failure probability of the migration task and improving overall transmission stability.

[0051] By following the steps above, potential interruption points at the data packet level can be accurately identified, enabling fine-grained risk prediction and improving the accuracy of path adjustment and system reliability during data migration.

[0052] Furthermore, based on the gain magnitude and information entropy changes of candidate split nodes, the optimal split threshold for dividing normal data packet transmission segments into abnormal segments is selected, and a decision path for identifying interruption risks is generated. This process specifically includes: For the characteristic splitting results of candidate splitting nodes, calculate at least one of the following indicators: information gain, gain ratio, or Gini index, to quantify the effectiveness of each candidate node in classifying the data packet transmission status. All candidate split nodes are sorted based on the calculated index values ​​to identify a set of high-priority split nodes that can minimize classification uncertainty. Entropy reduction analysis is performed on the set of high-priority splitting nodes. By comparing the entropy changes of abnormal and normal segments before and after splitting, the corresponding optimal splitting features and splitting thresholds are determined. The key splitting levels of the decision tree are constructed based on the optimal splitting features and splitting thresholds, and combined with the upper-level and side nodes to form a complete decision path structure for identifying interruption risks.

[0053] In this embodiment, firstly, the system calculates indicators such as information gain, gain ratio, or Gini index for the results generated by each candidate split node during the feature splitting process. This quantifies the effectiveness of the node in distinguishing between normal and abnormal data packets, ensuring that each split minimizes data uncertainty. Subsequently, based on the calculated indicator values, all candidate split nodes are ranked, prioritizing those nodes that most significantly improve classification accuracy and reduce entropy, forming a high-priority split node set. Next, the system performs entropy reduction analysis on this set, determining the optimal splitting feature and corresponding splitting threshold by comparing the entropy changes of normal and abnormal segments before and after splitting. This ensures that the decision tree is both accurate and stable in dividing data packet states. Finally, based on the determined optimal splitting feature and threshold, the system constructs the key splitting levels of the decision tree and combines them with upper-level nodes and side nodes to form a complete decision path structure. This allows each path to accurately map potential interruption risks, achieving fine-grained anomaly identification and risk localization of data packet sequences. The entire process, through scientific feature selection, entropy reduction analysis, and decision path construction, ensures that the system can efficiently and accurately identify potential interruption points even when facing complex network states and multi-dimensional features.

[0054] By following the steps above, we can efficiently construct accurate decision-making paths, accurately classify abnormal segments of data packets, and improve the accuracy of interruption risk identification and the reliability of data migration.

[0055] Furthermore, the step of activating the preset data caching mechanism to temporarily store the data to be migrated in a distributed buffer and obtaining a cached data copy specifically includes: The command to start the data caching mechanism is triggered, and a request for cache resources is sent to the node management center of the distributed buffer. Based on the list of available cache nodes returned by the node management center, the data to be migrated is divided into blocks, and each data block is assigned a corresponding target cache node address. At least one redundant data block copy is generated from the data block according to the preset replication strategy, and the data block and the redundant data block copy are written to multiple cache nodes respectively; Perform consistency checks on successfully written cached data blocks, and ensure the consistency and integrity of redundant data block replicas in each cache node through checksum comparison or hash check. After completing the data block writing and verification, index information corresponding to each cache node is generated and recorded in the cache management table.

[0056] In this embodiment, when the interruption risk level exceeds a preset standard, the system automatically activates a preset data caching mechanism to ensure the integrity and recoverability of the data to be migrated. First, the system triggers the caching mechanism's start command and sends a resource request to the node management center of the distributed buffer. The node management center returns a list of currently available cache nodes, including information such as node storage capacity, network latency, and load status. Subsequently, the data to be migrated is divided into blocks according to a preset strategy. Each data block is allocated to a corresponding cache node address according to the allocation strategy returned by the node management center to achieve load balancing and high availability. During the write process, the system generates at least one redundant copy according to a preset replication strategy and writes the original data block and the redundant copy to different cache nodes to improve fault tolerance. After the write is complete, the system performs consistency checks on the data blocks in each cache node, verifying the consistency and integrity of the data blocks and copies through checksums or hash comparisons to ensure data recovery in the event of network fluctuations or node failures. Finally, the system generates index information corresponding to each cache node, including data block location, version, checksum and expiration time, and records the index in the cache management table so as to quickly locate and restore the data to be migrated after network conditions are restored or path is switched, thus achieving reliable data migration and high availability storage management.

[0057] By following the steps above, data integrity can be ensured in high-risk network environments, highly reliable storage and rapid recovery of distributed cache can be achieved, and the fault tolerance and continuity of the data migration process can be improved.

[0058] Furthermore, after activating the preset data caching mechanism, temporarily storing the data to be migrated in a distributed buffer, and obtaining a copy of the cached data, the process also includes: Based on real-time network performance monitoring results, periodically check whether the network performance indicators of the backup or primary transmission path have recovered to the preset transmission quality range. When the network performance indicators are detected to meet the recovery conditions, a data recovery command is sent to the distributed buffer, and the data blocks that need to be resent are located according to the index information; The node with the lowest latency and highest data consistency among multiple cache nodes is selected as the data recovery source, and the data blocks are reorganized into a transferable data stream according to a preset sending sequence. The migrateable data stream is resent to the target migration database, and after successful data reception confirmation, the cache management table is updated to clear the corresponding data blocks that have been migrated.

[0059] In this embodiment, after the data to be migrated is temporarily stored in the distributed buffer, the system continuously monitors key network performance indicators of the backup or primary transmission path based on real-time network performance monitoring results. These indicators include latency, packet loss rate, bandwidth utilization, and jitter, to determine whether the network has recovered to the preset transmission quality range. Once the monitoring results show that the network performance recovery meets the quality requirements of the migration task, the system immediately issues a data recovery command to the distributed buffer and quickly locates the data blocks that need to be resent based on the index information stored in the cache management table. Subsequently, the system selects the node with the lowest latency and the highest data consistency from multiple cache nodes as the recovery source to ensure the efficiency and reliability of data transmission. During the data recovery process, the data blocks are reorganized into a migrateable data stream according to the preset data packet sending sequence to ensure that the data order and integrity are consistent with the initial migration plan. Finally, the system resends the data stream to the target migration database, and after the receiving end successfully confirms that the data has been written, it promptly updates the cache management table, deletes the data blocks that have been migrated, and releases the cache space. This process enables secure temporary storage of data in high-risk network environments, intelligent selection of the optimal recovery source, and sequential retransmission, ensuring the continuity and reliability of migration tasks and reducing transmission failures or delays caused by network fluctuations.

[0060] By following the steps above, efficient data recovery and sequential retransmission can be achieved after network recovery, ensuring the continuity and reliability of data migration and significantly improving the success rate of migration tasks and system robustness.

[0061] In this embodiment, the data migration method operates on an electronic device (e.g., Figure 1 The server shown can receive instructions or acquire data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0062] 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 instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0063] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0064] Further reference Figure 4 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a data migration apparatus, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0065] like Figure 4 As shown, the data migration device 400 described in this embodiment includes: The network prediction module 401 is used to collect real-time network transmission data, and to use a pre-trained neural network model to extract features from the network transmission data and predict network performance to obtain an estimated value of network performance indicators. The backup transmission module 402 is used to activate the backup transmission path and obtain the path configuration of the backup transmission path when the estimated value of the network performance index is lower than the preset index threshold. The interruption identification module 403 is used to obtain the data packet sending sequence of the data to be migrated from the path configuration and to identify potential interruption points in the data packet sending sequence using a decision tree algorithm; The risk assessment module 404 is used to determine the interruption risk level of each potential interruption point and compare the interruption risk level with the preset risk level standard. The data migration module 405 is used to encapsulate the data to be migrated into data packets when the interruption risk level is less than the preset risk level standard, and send the data packets to be migrated to the target migration database based on the data packet sending sequence.

[0066] The temporary cache module 406 is used to activate the preset data caching mechanism when the interruption risk level is greater than or equal to the preset risk level standard, and temporarily store the data to be migrated in the distributed buffer to obtain a cached data copy.

[0067] Further, please refer to Figure 5 The network prediction module 401 specifically includes: The acquisition unit 501 is used to acquire multi-dimensional parameters in real-time network transmission data and perform time-series processing on the multi-dimensional parameters, including bandwidth utilization, network latency, packet loss rate and jitter. The preprocessing unit 502 is used to preprocess the multi-dimensional parameters after time serialization, and input the preprocessed multi-dimensional parameters into the input layer of the neural network model to generate an input representation vector. The mapping unit 503 is used to perform nonlinear mapping on the input representation vector through the feature extraction layer of the neural network model, extract deep features across the time dimension, and obtain the network state feature vector. The prediction unit 504 is used to input the network state feature vector into the prediction layer of the neural network model, and to predict the network performance index through a combination of feature weighting and activation function. The quantization unit 505 is used to quantize the predicted output of the neural network model and generate an estimated value of the network performance index.

[0068] Furthermore, the prediction unit specifically includes: The fully connected subunit is used to input the network state feature vector into the fully connected weight matrix of the prediction layer, and to perform a linear weighting operation on the network state feature vector through the fully connected weight matrix to generate the first prediction intermediate value. The compensation subunit is used to introduce bias parameters to perform bias compensation based on the first prediction intermediate value, so as to obtain the second prediction intermediate value for nonlinear mapping. The activation subunit is used to process the second prediction intermediate value by applying the activation function of the neural network model to generate a nonlinear response value that characterizes the trend of network performance. The weighted sub-unit is used to input the nonlinear response value to the prediction output layer. By weighting and fusing the nonlinear response value and performing a linear transformation, a preliminary prediction result of the network performance index is generated. The calibration subunit is used to call the output calibration parameters saved during the model training phase based on the preliminary prediction results, perform scale calibration and error compensation on the predicted values, and obtain the final network performance index prediction results.

[0069] Furthermore, the interrupt identification module 403 specifically includes: The timing unit is used to extract the timing features of the data packet transmission sequence based on the path configuration of the backup transmission path. The timing features include the data packet transmission timestamp, sequence number, acknowledgment time, and retransmission marker information. The splitting unit is used to input temporal features into the feature generation node of the decision tree algorithm according to a preset feature matrix format, and to analyze the splitting gain of each feature to generate candidate splitting nodes. The partitioning unit is used to select the optimal partitioning threshold for dividing the normal and abnormal segments of data packet transmission based on the gain magnitude and information entropy change of the candidate partitioning nodes, and to generate a decision path for identifying the risk of interruption. The decision unit is used to judge the data packet transmission sequence packet by packet using the decision path. When the judgment condition of the abnormal segment is met, the corresponding position is marked as a potential interruption point and the set of potential interruption points is output.

[0070] Furthermore, the specific division units include: The index quantification subunit is used to calculate at least one of the following indices for the feature splitting results of candidate splitting nodes: information gain, gain ratio, or Gini index, to quantify the effectiveness of each candidate node in classifying the data packet transmission status. The sorting subunit is used to sort all candidate split nodes based on the calculated index values ​​and identify the set of high-priority split nodes that can minimize classification uncertainty. The entropy reduction subunit is used to perform entropy reduction analysis on the set of high-priority split nodes. By comparing the entropy changes of abnormal and normal segments before and after splitting, the corresponding optimal splitting features and splitting thresholds are determined. The path decision subunit is used to construct the key splitting level of the decision tree based on the optimal splitting features and splitting thresholds, and combines with the upper-level and side nodes to form a complete decision path structure for identifying interruption risks.

[0071] Furthermore, the temporary cache module 406 specifically includes: The startup unit is used to trigger the startup command of the data caching mechanism and send a cache resource request to the node management center of the distributed buffer. The block division unit is used to divide the data to be migrated into blocks based on the list of available cache nodes returned by the node management center, and to assign the corresponding target cache node address to each data block. The replica unit is used to generate at least one redundant data block replica from the data block according to a preset replication strategy, and write the data block and the redundant data block replica into multiple cache nodes respectively; The verification unit is used to perform consistency verification on successfully written cached data blocks. It ensures the consistency and integrity of redundant data block copies in each cache node through checksum comparison or hash verification. The index unit is used to generate index information corresponding to each cache node after the data block writing and verification are completed, and to record the index information in the cache management table.

[0072] Furthermore, the data migration device 400 also includes: The monitoring module is used to periodically check whether the network performance indicators of the backup transmission path or the main transmission path have recovered to the preset transmission quality range based on the monitoring results of real-time network performance. The recovery module is used to issue a data recovery command to the distributed buffer when the network performance indicators are detected to meet the recovery conditions, and to locate the data blocks that need to be resent based on the index information. The reorganization module is used to select the node with the lowest latency and the highest data consistency from multiple cache nodes as the data recovery source, and reorganize the data blocks into a transferable data stream according to a preset sending sequence; The re-migration module is used to resend the migrateable data stream to the target migration database, and after successful data reception confirmation, update the cache management table and clear the corresponding data blocks that have been migrated.

[0073] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0074] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0075] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0076] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for data migration methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0077] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, for example, to execute computer-readable instructions for the data migration method.

[0078] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0079] This application also provides an embodiment, namely, a computer device including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the data migration method described above.

[0080] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the data migration method described above.

[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0082] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0083] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A data migration method, characterized in that, include: Real-time network transmission data is collected, and a pre-trained neural network model is used to extract features and predict network performance of the network transmission data to obtain an estimated value of network performance indicators. If the estimated value of the network performance indicator is lower than the preset indicator threshold, then the backup transmission path is activated and the path configuration of the backup transmission path is obtained. Obtain the data packet sending sequence of the data to be migrated from the path configuration, and use a decision tree algorithm to identify potential interruption points in the data packet sending sequence; Determine the interruption risk level of each potential interruption point and compare the interruption risk level with a preset risk level standard; If the interruption risk level is less than the preset risk level standard, the data to be migrated is encapsulated into a data packet and sent to the target migration database in the form of a data packet based on the data packet sending sequence. If the interruption risk level is greater than or equal to the preset risk level standard, the preset data caching mechanism is activated, and the data to be migrated is temporarily stored in a distributed buffer to obtain a cached data copy.

2. The data migration method as described in claim 1, characterized in that, The steps of collecting real-time network transmission data, using a pre-trained neural network model to extract features from the network transmission data and predict network performance to obtain estimated network performance indicators specifically include: Collect multi-dimensional parameters from the real-time network transmission data and perform time-series processing on the multi-dimensional parameters, wherein the multi-dimensional parameters include bandwidth utilization, network latency, packet loss rate and jitter. The multi-dimensional parameters after time serialization are preprocessed, and the preprocessed multi-dimensional parameters are input into the input layer of the neural network model to generate an input representation vector. The input representation vector is nonlinearly mapped by the feature extraction layer of the neural network model to extract deep features across the time dimension, thereby obtaining the network state feature vector. The network state feature vector is input into the prediction layer of the neural network model, and the network performance index is predicted by combining feature weighting and activation function. The predicted output of the neural network model is quantified to generate an estimated value of the network performance index.

3. The data migration method as described in claim 2, characterized in that, The step of inputting the network state feature vector into the prediction layer of the neural network model, and predicting the network performance index through a combination of feature weighting and activation functions, specifically includes: The network state feature vector is input into the fully connected weight matrix of the prediction layer, and a linear weighting operation is performed on the network state feature vector through the fully connected weight matrix to generate the first prediction intermediate value. Based on the first predicted intermediate value, a bias parameter is introduced for bias compensation to obtain a second predicted intermediate value for nonlinear mapping. The activation function of the neural network model is applied to the second predicted intermediate value to generate a nonlinear response value that characterizes the trend of network performance. The nonlinear response value is input into the prediction output layer. By weighting and fusing the nonlinear response value and performing a linear transformation, a preliminary prediction result of the network performance index is generated. Based on the preliminary prediction results, the output calibration parameters saved during the model training phase are called to perform scale calibration and error compensation on the predicted values, so as to obtain the final network performance index prediction results.

4. The data migration method as described in claim 1, characterized in that, The step of obtaining the data packet sending sequence of the data to be migrated from the path configuration and identifying potential interruption points in the data packet sending sequence using a decision tree algorithm specifically includes: Based on the path configuration of the backup transmission path, the timing features of the data packet transmission sequence are extracted, wherein the timing features include data packet transmission timestamp, sequence number, acknowledgment time and retransmission marker information; The time-series features are input into the feature generation node of the decision tree algorithm according to a preset feature matrix format, and the split gain of each feature is analyzed to generate candidate split nodes. Based on the gain magnitude and information entropy change of the candidate split nodes, the optimal split threshold for dividing the normal and abnormal segments of data packet transmission is selected, and a decision path for identifying interruption risks is generated. The decision path is used to judge the data packet transmission sequence packet by packet. When the judgment condition of the abnormal segment is met, the corresponding position is marked as a potential interruption point, and the set of potential interruption points is output.

5. The data migration method as described in claim 4, characterized in that, The step of selecting the optimal splitting threshold for dividing normal data packet transmission segments and abnormal segments based on the gain magnitude and information entropy change of the candidate splitting nodes, and generating a decision path for identifying interruption risks, specifically includes: For the characteristic splitting results of the candidate splitting nodes, at least one of the following indicators is calculated: information gain, gain ratio, or Gini index, to quantify the effectiveness of each candidate node in classifying the data packet transmission status. All candidate split nodes are sorted based on the calculated index values ​​to identify a set of high-priority split nodes that can minimize classification uncertainty. Entropy reduction analysis is performed on the set of high-priority splitting nodes. By comparing the changes in entropy values ​​of abnormal and normal segments before and after splitting, the corresponding optimal splitting features and splitting thresholds are determined. The key splitting levels of the decision tree are constructed based on the optimal splitting features and the splitting threshold, and combined with the upper-level and side nodes to form a complete decision path structure for identifying interruption risks.

6. The data migration method as described in claim 1, characterized in that, The step of activating the preset data caching mechanism to temporarily store the data to be migrated in a distributed buffer and obtain a cached data copy specifically includes: The command to start the data caching mechanism is triggered, and a request for cache resources is sent to the node management center of the distributed buffer. Based on the list of available cache nodes returned by the node management center, the data to be migrated is divided into blocks, and a corresponding target cache node address is assigned to each data block. At least one redundant data block copy is generated from the data block according to a preset replication strategy, and the data block and the redundant data block copy are written into multiple cache nodes respectively; Perform consistency checks on successfully written cached data blocks, and ensure the consistency and integrity of redundant data block replicas in each cache node through checksum comparison or hash check. After completing the data block writing and verification, index information corresponding to each cache node is generated and recorded in the cache management table.

7. The data migration method as described in claim 6, characterized in that, After the step of activating the preset data caching mechanism, temporarily storing the data to be migrated in a distributed buffer, and obtaining a cached data copy, the method further includes: Based on real-time network performance monitoring results, periodically check whether the network performance indicators of the backup or primary transmission path have recovered to the preset transmission quality range. When the network performance indicators are detected to meet the recovery conditions, a data recovery command is sent to the distributed buffer, and the data blocks that need to be retransmitted are located according to the index information; The node with the lowest latency and highest data consistency among multiple cache nodes is selected as the data recovery source, and the data blocks are reorganized into a transferable data stream according to a preset sending sequence. The migrateable data stream is resent to the target migration database, and after successful data reception confirmation, the cache management table is updated to clear the corresponding data blocks that have been migrated.

8. A data migration device, characterized in that, include: The network prediction module is used to collect real-time network transmission data, and use a pre-trained neural network model to extract features and predict network performance of the network transmission data to obtain an estimated value of network performance indicators. A backup transmission module is used to activate a backup transmission path and obtain the path configuration of the backup transmission path when the estimated value of the network performance indicator is lower than a preset indicator threshold. The interruption identification module is used to obtain the data packet sending sequence of the data to be migrated from the path configuration, and to identify potential interruption points in the data packet sending sequence using a decision tree algorithm; The risk assessment module is used to determine the interruption risk level of each potential interruption point and compare the interruption risk level with a preset risk level standard. The data migration module is used to encapsulate the data to be migrated into a data packet when the interruption risk level is less than a preset risk level standard, and send the data packet to be migrated to the target migration database based on the data packet sending sequence. The temporary caching module is used to activate a preset data caching mechanism when the interruption risk level is greater than or equal to a preset risk level standard, and temporarily store the data to be migrated in a distributed buffer to obtain a cached data copy.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the data migration method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the data migration method as described in any one of claims 1 to 7.

Images