Wind power cluster-oriented high concurrency real-time database synchronization method

Abstract

The application belongs to the technical field of databases, and discloses a high-concurrency real-time database synchronization method for a wind power cluster, which comprises the following steps: establishing database partitions for nodes at each level of a machine-field-cluster-network, and pre-constructing buffer time slots at the tail of the partitions; when the concurrent writing exceeds a steady-state processing threshold, increment data is temporarily stored in the corresponding time slot and an attribution record is generated; the state of a data point is determined based on the corresponding record, buffer and main partition data are sequentially merged and synchronized for a buffer-merge-waiting point, and a direct extraction synchronization is performed for a direct extraction point; a delay tolerance window is set based on load evaluation, and a degradation strategy is triggered when the delay exceeds the limit, difference data blocks are extracted through hash comparison and transmitted according to priority levels. The concurrent processing capacity, data synchronization efficiency and system adaptability are improved.

Description

Technical Field

[0001] This invention relates to the field of database technology, and more specifically, to a high-concurrency real-time database synchronization method for wind power clusters. Background Technology

[0002] With the deep penetration of new energy and IoT technologies into fields such as smart grids and wind power cluster monitoring, data synchronization scenarios are exhibiting significant characteristics of "multi-level, high concurrency, and strong timeliness." Taking the wind power cluster scenario as an example, the system needs to synchronize massive amounts of collected data generated by multiple nodes such as turbines, substations, clusters, and the power grid across different levels. This type of data is not only large in scale and has varying sampling frequencies (such as the coexistence of high-frequency data at the millisecond level and low-frequency data at the minute level), but also has significantly different tolerances for data transmission latency at different levels of links. This places extremely stringent requirements on the high-concurrency write throughput, resource utilization, and dynamic adaptive capabilities of real-time database systems.

[0003] In related technologies, continuous influx of write requests are typically handled directly based on fixed hardware resource configurations, and cross-node data synchronization is performed using an indiscriminate, periodic full-transmission method. This makes it prone to overcoming system processing bottlenecks during sudden high-concurrency writes, leading to write blockages or data loss. Furthermore, large packet transmissions during network fluctuations cause severe channel congestion, significantly impacting the timeliness of data synchronization. Moreover, these technologies cannot effectively cope with the dynamic changes in multi-level network loads and underlying link states within wind power clusters, resulting in weak high-concurrency handling capabilities, network transmission redundancy, and poor system adaptability. Therefore, this invention proposes a high-concurrency real-time database synchronization method for wind power clusters to address these problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: High-concurrency real-time database synchronization methods for wind power clusters include: The turbine nodes, farm nodes, cluster nodes and grid nodes in the wind power cluster are added to the data synchronization platform according to the hierarchical relationship of turbine-farm-cluster-grid. Data domain planning is performed when nodes are added, a corresponding real-time database partition is established for each level node, and several blank buffer time slots are pre-initialized in each database partition. Based on the buffer time slot, when the number of concurrent write requests in a certain data acquisition cycle of the wind power cluster exceeds the steady-state processing threshold of the current database partition, the incremental data packets exceeding the steady-state processing threshold are temporarily stored in the corresponding buffer time slot according to the device dimension label to which the data point belongs, and the data point affiliation relationship of each buffer time slot is recorded to obtain the affiliation record table. Based on the ownership record table, when an upper-level node initiates a data synchronization request to a lower-level node, it checks whether there are any uncommitted buffer time slot records for the data points involved in the request. If they exist, the data in the buffer time slot is merged with the data in the primary partition before synchronization is performed. If they do not exist, the data is directly extracted from the primary partition and synchronization is performed. Based on the differences in latency sensitivity of cross-level synchronization, differentiated latency tolerance windows are set for the synchronization links between the airport, the group, and the network. When the measured latency on the synchronization link exceeds the latency tolerance window of the corresponding level, the synchronization degradation strategy of the corresponding link is triggered.

[0005] Furthermore, the process of initializing several blank buffer time slots includes: The node attribute information carried by each level of the access data synchronization platform is statistically analyzed. The node attribute information includes device type identifier, sampling frequency identifier, and number of data points. Based on the sampling frequency identifier, the data points are divided into high-frequency data sampling points and low-frequency data sampling points, and then grouped and aggregated according to the device type identifier to obtain a high-frequency data point group set and a low-frequency data point group set. Calculate the rated data throughput corresponding to the high-frequency data point group and the low-frequency data point group respectively, and perform database partitioning based on their total real-time database storage capacity in the corresponding level nodes. Initialize the index structure of the database partitions, and pre-construct a fixed proportion of storage space at the end of each database partition to obtain blank buffer time slots. Write the time slot number, starting storage address, ending storage address, and writable capacity into the buffer time slot metadata table, and mark the buffer time slot status as idle.

[0006] Furthermore, the database partitioning includes establishing high-priority storage partitions and standard storage partitions based on the division of high-frequency data points and low-frequency data points.

[0007] Furthermore, the process of obtaining the attribution record table includes: Read the historical write peak records of the current database partition, and calculate the steady-state processing threshold within the current data acquisition cycle using the sliding average of the historical write peak records and the preset safety factor; Continuously monitor the real-time write request count of the database partition corresponding to each level node. When the write request count is higher than the steady-state processing threshold in three consecutive sampling intervals, the corresponding database partition is determined to enter the concurrent write peak state. For database partitions that have entered peak concurrent write state, all write requests arriving within the current data acquisition cycle are grouped according to the device dimension label corresponding to the data point, and incremental data packets exceeding the steady-state processing threshold are identified. The identified incremental data packets are matched based on the pre-established multi-dimensional index anchor points, and the incremental data packets are imported into the corresponding buffer time slots based on the matching results. At the same time, the data point number, the data acquisition cycle, the data point type identifier, and the temporary data volume are written into the buffer time slot's ownership record table. The target buffer time slot matching the corresponding device dimension label is retrieved from the buffer time slot of the current database partition according to the ownership record table. If the target buffer time slot has remaining writable capacity, the incremental data packet is written to the target buffer time slot, and the data point number and write position record of the corresponding incremental data packet are appended to the ownership record table of the buffer time slot. If the remaining capacity of the target buffer time slot is insufficient, the buffer time slot rolling expansion operation is initiated to merge adjacent free buffer time slots into an expanded buffer time slot before writing is performed.

[0008] Furthermore, the process of initiating the buffer time slot rolling expansion operation includes: Retrieve adjacent idle buffer time slots in the current database partition to obtain a list of idle adjacent time slots; The list of adjacent idle time slots is segmented according to the continuity of the time slot numbers to obtain a set of consecutive idle time slot segments; Select the shortest consecutive idle time slot from the set of consecutive idle time slots that are adjacent to the target buffer time slot to obtain the time slot segment to be merged; The start and end addresses of the time slot segments to be merged are merged into the address range of the target buffer time slot, and the storage boundary of the target buffer time slot is updated in the buffer time slot metadata table to obtain the expanded buffer time slot; and the original time slot entries of the time slot segments to be merged are marked as merged.

[0009] Furthermore, the process of checking whether there are any uncommitted buffer time slot records for the data points involved in the request includes: When an upper-level node initiates a data synchronization request to a lower-level node, extract the list of requested data point numbers and the requested time range parameter.

[0010] Based on the list of requested data point numbers, query the record status of each data point number in the buffer time slot attribution record table one by one; if a data point number has a record entry in the attribution record table with a status of uncommitted, mark the corresponding data point as a buffer-containing point to be merged and record its corresponding buffer time slot number; if a data point number does not have an uncommitted record in the attribution record table, mark the corresponding data point as a primary partition direct retrieval point. For points with buffers to be merged, extract buffer data segments from their corresponding buffer time slots according to the requested time range parameters to obtain a set of buffer data segments; at the same time, extract the primary partition data segments of the same data point within the same time range from the primary partition to obtain a set of primary partition data segments. The buffer data segment set and the primary partition data segment set are merged in order according to the timestamp field to obtain the merged data segment set; The merged data segment set and the data from the primary partition direct fetch point are encapsulated together into a synchronization response data packet, serialized according to the specified transmission format, and sent to the upper-layer node.

[0011] Furthermore, the process of orderly merging includes: The data records in the buffer data segment set and the primary partition data segment set are sorted in ascending order according to the timestamp field to obtain the buffer ordered sequence and the primary partition ordered sequence, respectively. Set two pointers to point to the current read position of the buffered ordered sequence and the main partition ordered sequence respectively, and initially point to the first record of their respective sequences; Compare the timestamps of the records currently pointed to by the two pointers, write the record with the smaller timestamp to the merge output buffer, and move the corresponding pointer one position to the right; when the timestamps of the records pointed to by the two pointers are equal, it is determined that the timestamps overlap, the corresponding record in the buffered ordered sequence is written to the merge output buffer, and the corresponding record in the main partition ordered sequence is marked as overwritten and skipped, and the two pointers move one position to the right in sync. Repeat the above comparison operation until both pointers reach the end of their respective sequences. Export all records in the merge output buffer in the order they were written to obtain the merged data segment set.

[0012] Furthermore, the process of setting the differentiated latency tolerance window includes: Obtain historical latency sampling sequences of synchronization links between each level of the airport-field-group-network, calculate the mean latency and standard deviation of latency for each synchronization link in the last N sampling periods, and obtain the latency statistical parameters of each link; Calculate the initial value of the basic tolerance window for the corresponding synchronization link based on the latency statistics parameters of each link. Obtain the total number of online data points currently carried by each level node, and use the ratio of the total number of online data points to the designed rated data point capacity of the corresponding level node to obtain the load rate of each level node; based on the load rate of each level node, linearly amplify and correct the initial value of the basic tolerance window of the corresponding synchronization link to obtain the load-corrected tolerance window value. After load correction, the tolerance window value is accumulated and verified sequentially from the machine layer to the power grid layer. The load correction tolerance window value of each link that passes the accumulated constraint verification is determined as the final delay tolerance window, written into the synchronization link configuration table, and sent to the corresponding link management process.

[0013] Furthermore, the execution process of the synchronous degradation strategy includes: When a synchronization link triggers the synchronization degradation strategy, the sending end performs block processing on the current set of data to be synchronized for the corresponding synchronization link. The set of data to be synchronized is divided into several data blocks according to the device dimension label to which the data point belongs, thus obtaining the set of data blocks to be synchronized. For each data block in the set of data blocks to be synchronized, calculate the hash digest value of its data content to obtain a hash digest set for each data block; encapsulate the hash digest set of each data block into a digest synchronization message according to the data block number, and transmit it to the receiving end through the downgraded synchronization link; The receiving end parses the digest synchronization message and extracts the hash digest set of each data block; it calculates the digest value of the corresponding data block with the same hash algorithm on the local side of the receiving end to obtain the local hash digest set of each data block. The hash digest set of each data block on the receiving end is compared block by block with the hash digest set of each data block received, and the data block numbers with inconsistent hash digest values ​​are filtered out to obtain a list of differing data block numbers; The receiving end encapsulates the list of difference data block numbers into a difference request message and sends it back to the sending end; the sending end extracts the complete data content of the corresponding data block according to the list of difference data block numbers, encapsulates it into a difference data response message and sends it to the receiving end; the receiving end replaces the corresponding local data block with the data content in the difference data response message.

[0014] Furthermore, the process by which the sending end extracts the complete data content of the corresponding data block based on the list of difference data block numbers, encapsulates it into a difference data response message, and sends it to the receiving end also includes: For each differential data block in the differential data block number list, at the sending end, extract the absolute deviation between the current value of all data points in the corresponding differential data block and the corresponding snapshot value of the corresponding data block when the receiving end last successfully synchronized, and obtain the set of deviations for each data point. Calculate the maximum value, mean value and proportion of non-zero deviation points for each set of deviations to obtain the data block deviation feature triplet; Based on the triplet of data block deviation characteristics, the difference level determination rules are set as follows: when the maximum deviation exceeds the preset high-risk threshold or the proportion of non-zero deviation points exceeds the preset high-density threshold, the difference level of the corresponding data block is set to the emergency level; when the maximum deviation is between the preset high-risk threshold and the preset medium-risk threshold or the proportion of non-zero deviation points is between the preset high-density threshold and the preset medium-density threshold, the difference level is set to the priority level; in other cases, the difference level is set to the normal level. In the difference request message, each difference data block number is appended with its corresponding difference level identifier. After receiving the difference request message, the sender prioritizes the difference data blocks according to the difference level identifier, encapsulates and sends the complete data content of the urgent difference data blocks first, then sends the priority difference data blocks, and finally sends the normal difference data blocks.

[0015] The technical effects and advantages of this invention for high-concurrency real-time database synchronization method for wind power clusters: 1. This invention effectively resolves the problem of concurrent write impact on massive data collection in wind power clusters under multi-level architecture by establishing a concurrent diversion and data dual-pointer ordered merging mechanism based on buffer time slots; it enhances the integrity of cross-node data synchronization, can accurately identify incremental data packets exceeding the steady-state threshold, and ensures seamless merging of uncommitted buffered data with historical data in the main partition, thereby providing key data support for the overall status monitoring and collaborative scheduling of wind farm clusters.

[0016] 2. This invention helps to build an adaptive control system for wind power cluster data transmission. By dynamically adjusting the latency tolerance window based on node load and historical parameters, it prevents transmission blockage caused by network fluctuations, improves the operational reliability of cross-level communication links, and enhances the operational and maintenance collaboration efficiency of wind power management departments.

[0017] 3. The hash digest difference comparison and graded priority transmission degradation synchronization mechanism of the present invention can effectively overcome the bandwidth transmission bottleneck under network latency exceeding the limit, and ensure the priority access of core emergency business data and the security of power grid access.

[0018] 4. This invention can promptly detect and properly handle abnormal write surges that continuously flood into multi-level nodes; and by constructing a real-time synchronization closed-loop mechanism for data "concurrent buffering - ordered merging - degraded transmission", it provides a solid technical foundation for building a smart, efficient, and highly available modern wind power cluster scheduling and monitoring system. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the high-concurrency real-time database synchronization method for wind power clusters according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The high-concurrency real-time database synchronization method for wind power clusters provided in this application embodiment can be used in wind power cluster data acquisition and monitoring scenarios. Based on the above application scenario, it can be understood that the execution subject of each step can be a computer device. The corresponding computer device refers to any electronic device with data computing, processing and storage capabilities, such as industrial control computers, edge computing servers, site monitoring hosts and other terminal devices, or cloud servers and other devices. This application embodiment does not limit this.

[0022] Example 1, please refer to Figure 1 As shown, the high-concurrency real-time database synchronization method for wind power clusters provided in this embodiment includes: First, the turbine nodes, farm nodes, cluster nodes, and grid nodes in the wind power cluster are registered and connected to the data synchronization platform sequentially according to the hierarchical relationship of the airport network. Upon node connection, the node attribute information carried by each node is read. Based on the node attribute information, database partitioning is performed on the local storage resources of each level of node. A corresponding real-time database partition is established for each level of node, and several blank buffer time slots are pre-initialized at the end of each database partition, marking the buffer time slots as idle. The hierarchical relationship of the airport network refers to the sequence from the single-unit wind turbine equipment layer upwards through the wind farm aggregation layer, multi-farm cluster adjustment layer, and so on. The system has a four-level tree-like hierarchical structure from the degree layer down to the power grid access layer; database partitions refer to logical data storage partitions that are independently allocated to each level node within the total storage capacity of the real-time database, with clear storage boundaries and index structures. Within each partition, high-priority storage partitions and standard storage partitions are established according to the division of high-frequency data points and low-frequency data points; buffer time slots refer to storage space segments reserved at the end of each database partition storage, equally divided according to the acquisition cycle. In the initial state, they do not carry any data points, but only maintain the slot index and ownership relationship record structure, and are used to temporarily accept incremental data packets that exceed the steady-state processing capacity of the main partition during peak concurrent write periods.

[0023] Based on the buffer time slot, when it is detected that the concurrent write request volume of a certain database partition in a certain data acquisition cycle of the wind power cluster exceeds the steady-state processing threshold of the corresponding partition, the incremental data packets exceeding the steady-state processing threshold are grouped according to the device dimension label to which the data point belongs, and each group of incremental data packets is temporarily stored in the corresponding buffer time slot. At the same time, fields such as data point number, data acquisition cycle, data point type identifier, and temporary data volume are written into the attribution record table to obtain the attribution record table.

[0024] Based on the attribution record table, when an upper-level node initiates a data synchronization request to a lower-level node, it extracts the list of requested data point numbers and the requested time range parameters, and queries the record status of each data point number in the attribution record table one by one. If a data point number has a record entry in the attribution record table with an uncommitted status, the corresponding data point is marked as a buffer-containing point to be merged, the buffer data segment is extracted from its corresponding buffer time slot, and the same data point in the same time range in the main partition is extracted from the main partition. After orderly merging according to the timestamp field, synchronization is performed. If a data point number does not have any uncommitted records in the attribution record table, the corresponding data point is marked as a main partition direct retrieval point, and data is directly retrieved from the main partition for synchronization. Here, the main partition refers to the main storage area in the real-time database partition used to store committed data, which is different from the pre-built buffer time slot area at the tail. A buffer-containing point to be merged refers to a data point with buffer data that has not yet been committed to the main partition and needs to be merged with the main partition data before synchronization. A main partition direct retrieval point refers to a data point with no uncommitted records in the attribution record table and whose data can be directly read from the main partition.

[0025] Based on the differences in latency sensitivity across different synchronization levels, differentiated latency tolerance windows are set for each synchronization link in the airport, airfield cluster, and network. When the measured latency on a synchronization link exceeds the corresponding latency tolerance window, the synchronization degradation strategy for that link is triggered. Differential data blocks are transmitted through a hash digest difference comparison mechanism and prioritized according to the difference level identifier to ensure that critical data arrives synchronously first. The latency tolerance window refers to the maximum acceptable synchronization delay time range determined by combining the historical latency statistics of each synchronization link with the current load rate correction value. The synchronization degradation strategy refers to the degradation synchronization mechanism that switches from full data transmission to hash digest comparison plus differential data transmission when the latency of a synchronization link exceeds the tolerance window.

[0026] In this embodiment of the invention, the detailed implementation steps for initializing the blank buffer time slot include: The system reads the node attribute information carried by each level of the access data synchronization platform. This node attribute information includes three core attributes: device type identifier, sampling frequency identifier, and number of data points. The device type identifier distinguishes different types of physical equipment, such as unit monitoring and control equipment, station monitoring equipment, cluster dispatching equipment, and grid access equipment. The sampling frequency identifier distinguishes the data point acquisition rate level; the sampling period for high-frequency data points is typically in the millisecond to second range, while the sampling period for low-frequency data points is typically in the minute range or higher. The number of data points refers to the total number of all data acquisition points currently registered for the corresponding node. Based on the sampling frequency identifier, the data points of each level of the node are divided into high-frequency data points and low-frequency data points. These are then grouped and aggregated according to the device type identifier. High-frequency data points are aggregated to form a high-frequency data point group set, and low-frequency data points are aggregated to form a low-frequency data point group set. The grouping and aggregation operation groups data points of the same frequency belonging to the same device type into the same group, facilitating subsequent calculation of the rated data throughput by group.

[0027] The rated data throughput for high-frequency data point groups and low-frequency data point groups is calculated separately. The rated data throughput is calculated by summing the products of the sampling frequency of each data point in the group and the size of the single-point data packet to obtain the theoretical maximum data generation rate of the corresponding group per unit time, which is then used as the rated data throughput of the corresponding group. Based on the rated data throughput of high-frequency and low-frequency data point groups, the database is partitioned within the total real-time database storage capacity of the corresponding level nodes. A high-priority storage partition is established for high-frequency data point groups, and a standard storage partition is established for low-frequency data point groups. The storage capacity of the high-priority storage partition is allocated according to the proportion of the rated data throughput of the high-frequency data point groups in the total rated data throughput, and the standard storage partition receives the remaining storage capacity. The physical address ranges of the high-priority storage partition and the standard storage partition do not overlap, and they manage their storage space independently.

[0028] Initialize the index structure of each database partition. The index structure includes a hash index that maps data point numbers to partition storage addresses and a B-tree index that supports time range queries. The hash index allows for quick and accurate location of storage points by data point number, while the B-tree index allows for efficient scanning of data records within a partition by time range. After the index structure initialization is complete, a fixed proportion of storage space is pre-constructed at the end of each database partition as a buffer time slot area. The capacity of the buffer time slot area is determined according to a preset proportion of the total storage capacity of the partition, which is pre-set by those skilled in the art based on actual peak concurrency characteristics. Within the buffer time slot area, several independent buffer time slot units are evenly divided according to a preset single-slot capacity. Each buffer time slot unit corresponds to a buffer time slot metadata table record. The buffer time slot metadata table record contains five fields: time slot number, starting storage address, ending storage address, writable capacity, and time slot status. The time slot number uses a continuous integer sequence starting from zero. After initialization, the time slot status field of all buffer time slots is marked as idle, indicating that the corresponding buffer time slot currently does not store any data and can accept write requests for incremental data packets.

[0029] In this embodiment of the invention, the detailed implementation steps for obtaining the attribution record table include: The system reads the historical write peak records of the current database partition, which store the maximum write request count for each period over several past data collection cycles. It calculates the moving average within a sliding window based on the historical write peak records in chronological order. The length of the sliding window is preset by experts in the field according to the actual scenario. The product of the moving average and a preset safety factor is used as the steady-state processing threshold for the current data collection cycle. The preset safety factor is used to reserve a certain processing margin based on the historical average to avoid frequent triggering of the buffering mechanism due to short-term increases in write volume; its typical value range is (0.8, 0.9). The system continuously monitors the real-time, second-level write request counts of the corresponding database partitions at each level. These counts are maintained in real-time by a request counter deployed on the database partition's write path, and the count value is updated once per sampling interval. When the write request counts for three consecutive sampling intervals are all higher than the steady-state processing threshold, the corresponding database partition is determined to have entered a concurrent write peak state. The judgment condition for three consecutive sampling intervals is used to filter out occasional instantaneous write spikes, ensuring that only continuous high-concurrency writes trigger the buffering mechanism.

[0030] For database partitions entering peak concurrent write states, all write requests arriving within the current data acquisition cycle are grouped according to the device dimension label corresponding to the data point. The grouping rules for the device dimension labels are consistent with the grouping rules used when partitioning the database, ensuring that the grouping results correspond to the allocation of buffer time slots. Incremental data packets exceeding the steady-state processing threshold are identified. The identification method for incremental data packets is as follows: all write requests in the current cycle are arranged in order of arrival time, the number of write requests before the steady-state processing threshold is defined as the normal processing range, and the remaining write requests constitute incremental data packets. The identified incremental data packets are matched based on pre-established multi-dimensional index anchors. The multi-dimensional index anchors are established with the device dimension label as the primary key, supporting rapid location of the target buffer time slot by dimensions such as device type and sampling frequency. Based on the matching results, the incremental data packets are imported into the corresponding buffer time slots, and four fields—data point number, data acquisition cycle, data point type identifier, and temporary data volume—are written into the buffer time slot allocation record table.

[0031] The target buffer time slot matching the corresponding device dimension label is retrieved from the buffer time slots of the current database partition based on the attribution record table. The retrieval process is achieved by querying the device dimension label field in the buffer time slot metadata table. If the target buffer time slot has remaining writable capacity, the incremental data packet is written to the target buffer time slot, and the data point number and write position record of the corresponding incremental data packet are appended to the attribution record table of the buffer time slot. The write position record contains three subfields: data point number, write start address, and write data length, which are used to accurately locate the storage location of the buffer data during subsequent synchronization. If the remaining capacity of the target buffer time slot is insufficient to store the current incremental data packet, a buffer time slot rolling expansion operation is initiated. The storage space is merged from adjacent free buffer time slots to form an expanded buffer time slot before writing is performed. After the buffer time slot rolling expansion operation is completed, the end storage address and writable capacity fields of the target buffer time slot in the buffer time slot metadata table are updated, and the merged adjacent free buffer time slot entries are marked as merged to ensure the consistency of the metadata table.

[0032] In this embodiment of the invention, the detailed implementation steps for initiating the buffer time slot rolling expansion operation include: The system retrieves the buffer time slot metadata table of the current database partition, filters out buffer time slot entries marked as idle in the time slot status field, and obtains a list of idle adjacent time slots with time slot numbers adjacent to the target buffer time slot. The adjacency relationship is determined by the continuity of time slot numbers, that is, buffer time slots whose time slot numbers differ from the target buffer time slot numbers by 1 are directly adjacent time slots. The list of idle adjacent time slots is sorted in ascending order of time slot numbers to facilitate subsequent segmentation processing according to the continuity of numbers. The retrieval process is implemented by traversing all entries in the buffer time slot metadata table, and the time complexity of each retrieval is linearly related to the total number of buffer time slots.

[0033] The list of adjacent idle time slots is segmented according to the continuity of the time slot numbers. The segmentation rule is as follows: if the difference between the time slot numbers of two adjacent records is 1, they are considered continuous and the continuous entries are grouped into the same time slot segment; if the difference is greater than 1, they are considered discontinuous and a new time slot segment is split off with the discontinuous position as the boundary. The segmentation operation is implemented by linear scanning, traversing the list of adjacent idle time slots, maintaining the start and end numbers of the current segment, judging the continuity of each entry and outputting the segmentation result to obtain a set of continuous idle time slot segments. From the set of continuous idle time slot segments, the continuous idle time slot segment that is adjacent to the target buffer time slot and has the shortest time slot segment length (i.e., the number of time slots contained in the corresponding segment) is selected to obtain the time slot segment to be merged. The purpose of selecting the shortest continuous idle time slot segment is to minimize the impact of a single expansion operation on the overall availability of the buffer time slot pool and to prioritize the consumption of the minimum necessary storage resources to complete the expansion requirements.

[0034] The starting and ending storage addresses of the time slot segments to be merged are merged into the address range of the target buffer time slot. Specifically, if the time slot numbers of the time slot segments to be merged are all greater than the target buffer time slot number, the ending storage address of the target buffer time slot is updated to the ending storage address of the time slot segment to be merged. If the time slot numbers of the time slot segments to be merged are all less than the target buffer time slot number, the starting storage address of the target buffer time slot is updated to the starting storage address of the time slot segment to be merged. The starting storage address field, ending storage address field, and writable capacity field of the target buffer time slot entry are updated in the buffer time slot metadata table. The writable capacity update value is the total number of storage bytes in the merged address range minus the amount of data already written, resulting in the extended buffer time slot. The original time slot entries of the time slot segments to be merged are marked as merged in the time slot status field of the buffer time slot metadata table. The merged status indicates that the storage space of the corresponding time slot has been integrated into the extended buffer time slot and will no longer participate in subsequent retrieval and allocation operations as an independent time slot.

[0035] In this embodiment of the invention, the detailed implementation steps for checking whether there are any unsubmitted buffer time slot records for the data points involved in the request include: When an upper-layer node initiates a data synchronization request to a lower-layer node, it parses the synchronization request message and extracts the list of requested data point numbers and the request time range parameter. The list of requested data point numbers contains the unique numbers of all data points involved in this synchronization request. The request time range parameter contains the start and end times of this synchronization request, represented in timestamp form. Based on the list of requested data point numbers, it queries the record status of each data point number in the buffer time slot attribution record table one by one. The query operation is implemented through the hash index of the attribution record table, using the data point number as the query key for exact matching. If a data point number has a record entry in the attribution record table with a time slot status field of "uncommitted", the corresponding data point is marked as a buffer-bound point to be merged, and its corresponding buffer time slot number is recorded. The "uncommitted" status indicates that some data of the corresponding data point is still temporarily stored in the buffer time slot and has not yet been written to the primary partition. If a data point number does not have a record entry in the attribution record table with a time slot status of "uncommitted", the corresponding data point is marked as a primary partition direct retrieval point. All data of the primary partition direct retrieval point has been stored in the primary partition and can be directly retrieved.

[0036] For points with buffers to be merged, the buffer time slot metadata table is queried based on the corresponding buffer time slot number to obtain the start and end storage addresses of the corresponding buffer time slot. The write position records of the corresponding data points within the buffer time slots are read from the attribution record table. Buffer data records that overlap with the time range are filtered according to the requested time range parameter, and the corresponding buffer data segments are extracted to obtain a buffer data segment set. Each record in the buffer data segment set contains three core fields: data point number, timestamp, and data value. Simultaneously, the primary partition data segments of the same data point within the same requested time range are extracted from the primary partition. The primary partition data segments are quickly located by time range using the B-tree time index of the primary partition to obtain the primary partition data segment set. The primary partition data segment set and the buffer data segment set use the same record format for easy unified processing later.

[0037] The buffer data segment set and the primary partition data segment set are merged in order according to the timestamp field to obtain a merged data segment set. During the ordered merging process, the data records in the buffer data segment set and the primary partition data segment set are sorted in ascending order according to the timestamp field to obtain a buffer ordered sequence and a primary partition ordered sequence. Two pointers are set to point to the current read position of the buffer ordered sequence and the primary partition ordered sequence respectively, initially pointing to the first record of their respective sequences. The timestamps of the records currently pointed to by the two pointers are compared, and the record with the smaller timestamp is written to the merge output buffer, and the corresponding pointer is moved forward one position. When the timestamps of the records pointed to by the two pointers are equal, it is determined that the timestamps overlap. The corresponding record in the buffered ordered sequence is written into the merge output buffer, and the corresponding record in the main partition ordered sequence is marked as overwritten and skipped. The two pointers move one position forward in sync. When the timestamps overlap, the buffered data will preferentially overwrite the main partition data. The semantic basis is that the data in the buffer time slot is the updated value written after the main partition data, which has higher timeliness. Repeat the above comparison operation until both pointers reach the end of their respective sequences. All records in the merge output buffer are exported in the order of writing to obtain the merged data segment set.

[0038] The merged data segment set and the data from the primary partition direct fetch point are jointly encapsulated into a synchronization response data packet. The data from the primary partition direct fetch point is extracted in batches by data point number using the hash index of the primary partition, and the extraction results are also filtered according to the requested time range parameter. The merged data segment set and the data from the primary partition direct fetch point are serialized according to the specified transmission format. The serialization format contains a three-level field structure including data point number, timestamp sequence, and corresponding data value sequence. After serialization, the synchronization response data packet is sent to the upper-layer node that initiated the request through the synchronization link, completing this data synchronization response.

[0039] In this embodiment of the invention, the detailed implementation steps for setting the differentiated latency tolerance window include: Historical latency sampling sequences of each synchronization link in the airport, airfield cluster, and network are obtained over the past N collection periods. These historical latency sampling sequences consist of end-to-end transmission delay values ​​recorded by the management processes of each link during each data synchronization operation, with the collection period consistent with the data synchronization period. For each synchronization link's historical latency sampling sequence, the mean latency and standard deviation of latency over the past N collection periods are calculated. The mean latency reflects the typical transmission delay level of the corresponding link under normal operating conditions, while the standard deviation reflects the fluctuation range of the corresponding link's latency. The mean latency and standard deviation are combined to form the latency statistical parameters for each link, resulting in a set of latency statistical parameters for each link. The value of N is predetermined by professionals in the field based on the actual link stability characteristics; a larger N value results in more stable statistical results but a more delayed response to changes in link status.

[0040] The initial value of the basic tolerance window for each synchronization link is calculated based on the statistical parameters of the delay of each link. The initial value of the basic tolerance window is calculated as follows: the sum of the mean delay and the standard deviation of the delay is used as the initial value of the basic tolerance window. The preset multiple is set in advance by the professionals in this field according to the link quality requirements, and is usually 2 to 3 times the standard deviation to cover the vast majority of delay samples within the normal fluctuation range. This calculation method based on statistical distribution ensures that the initial value of the basic tolerance window can adapt to the inherent transmission characteristics of different links and avoid false triggering caused by differences in the physical transmission conditions of links at different levels. The initial value of the basic tolerance window for airport links is usually the smallest, reflecting the highest real-time requirements of the data on the flight crew side. The initial value of the basic tolerance window for group network links is usually the largest, reflecting the relatively low delay sensitivity of the power grid side to the aggregated data.

[0041] The system obtains the total number of online data points currently carried by each level of node. The total number of online data points is counted in real time by the data synchronization platform access management module of each level of node. The load rate of each level of node is obtained by the ratio of the total number of online data points to the designed rated data point capacity of the corresponding level of node. The designed rated data point capacity refers to the maximum number of data points that can be accessed by the corresponding level of node during the design and planning. The initial value of the basic tolerance window of the corresponding synchronization link is linearly amplified and corrected according to the load rate of each level of node. The calculation formula for the linear amplification correction is: the tolerance window value after load correction is equal to the product of the initial value of the basic tolerance window and the linear correction coefficient of the load rate. The linear correction coefficient of the load rate is equal to 1 plus the product of the load rate and the preset amplification slope. The preset amplification slope is set in advance by those skilled in the art based on the node load characteristics. The higher the load rate, the larger the tolerance window value after correction, which means that more lenient latency tolerance is given under high node load conditions, reducing the false triggering of synchronization degradation caused by load.

[0042] The load-corrected tolerance window values ​​are cumulatively checked from the machine layer to the power grid layer. The rules for cumulative constraint checking are as follows: the latency tolerance window value of the upper-layer link must not be less than the latency tolerance window value of the lower-layer link, that is, the tolerance window value of the group network link must not be less than the tolerance window value of the field group link, and the tolerance window value of the field group link must not be less than the tolerance window value of the airport link. If the load-corrected tolerance window value of a link violates the above constraints, the tolerance window value of the corresponding link is adjusted to the sum of the tolerance window value of its lower-layer link and the preset minimum level difference to ensure that the latency tolerance window between levels is monotonically increasing. The cumulative constraint checking is performed layer by layer from the airport link upwards until the group network link is checked. The load-corrected tolerance window values ​​of each link that pass the cumulative constraint checking are determined as the final latency tolerance window, written into the synchronization link configuration table, and sent to the corresponding link management process. After receiving the final latency tolerance window configuration, the link management process updates the local latency comparison threshold. When the subsequent measured latency exceeds the corresponding threshold, the synchronization degradation strategy is triggered.

[0043] In this embodiment of the invention, the detailed implementation steps for executing the synchronous degradation strategy include: When the measured latency of a synchronization link exceeds the corresponding final latency tolerance window, the corresponding link's synchronization degradation strategy is triggered. At the sending end, the current set of data to be synchronized on the corresponding synchronization link is divided into blocks according to the device dimension label to which the data points belong. Data points with the same device dimension label are grouped into the same data block, resulting in a set of data blocks to be synchronized. Each data block in the set is assigned a unique data block number, using a consecutive integer sequence starting from zero. A hash digest value for the data content of each data block is calculated, using a fixed hash algorithm to ensure comparability between the sending and receiving ends. All data content of each data block is sequentially input into a hash calculation function in byte order to obtain a fixed-length hash digest value. The hash digest values ​​of each data block constitute a hash digest set for each data block. The hash digest sets of each data block are encapsulated into a digest synchronization message according to the data block number. The digest synchronization message is significantly smaller than the full data message, enabling transmission with lower load on the degraded link where latency exceeds the limit, and is then transmitted to the receiving end through the degraded synchronization link.

[0044] The receiving end parses the synchronization message and extracts the hash digest set of each data block. It then calculates the digest value of the corresponding data block locally using the same hash algorithm as the sending end, obtaining a local hash digest set for each data block. Hash algorithm consistency is a prerequisite for ensuring the validity of the comparison results; the sending and receiving ends must agree to use the same hash algorithm identifier in the synchronization link configuration table. The receiving end compares its local hash digest set with the received hash digest set block by block, using the data block number as the alignment key. The hash digest values ​​of the corresponding numbered data blocks at the sending and receiving ends are compared sequentially to see if they match. Data block numbers with inconsistent hash digest values ​​are filtered out, resulting in a list of differing data block numbers. Data blocks with consistent hash digest values ​​indicate that the content of the corresponding data blocks at the sending and receiving ends is completely identical and does not require transmission. Data blocks with inconsistent hash digest values ​​indicate that there are differences between the two ends' data, requiring further transmission of the complete data content to complete synchronization.

[0045] The receiving end encapsulates the list of differential data block numbers into a differential request message and sends it back to the sending end. Upon receiving the differential request message, the sending end, for each differential data block in the differential data block number list, extracts the absolute deviation between the current value of all data points within the corresponding differential data block and the corresponding snapshot value of the data block at the time of the last successful synchronization at the receiving end, obtaining a set of deviations for each data point. The snapshot value is recorded by the sending end after each successful data synchronization and stored in the sending end's synchronization snapshot cache using the data block number as the key. The maximum value, mean value, and percentage of non-zero deviations are calculated for each data point deviation set, resulting in a data block deviation feature triplet. The percentage of non-zero deviations is calculated by dividing the number of data points with non-zero deviations by the total number of data points contained in the corresponding data block. Based on the data... The difference level determination rules are set according to the block deviation feature triplet: when the maximum deviation exceeds the preset high-risk threshold or the proportion of non-zero deviation points exceeds the preset high-density threshold, the difference level identifier of the corresponding data block is set to the emergency level; when the maximum deviation is between the preset high-risk threshold and the preset medium-risk threshold or the proportion of non-zero deviation points is between the preset high-density threshold and the preset medium-density threshold, the difference level identifier is set to the priority level; in other cases, the difference level identifier is set to the normal level; the preset high-risk threshold, preset medium-risk threshold, preset high-density threshold, and preset medium-density threshold are preset by the personnel in this field according to the characteristics of the business data; in the response processing stage of the difference request message, each difference data block number is appended with its corresponding difference level identifier to form a list of difference data blocks with level identifiers.

[0046] The sending end prioritizes the difference data blocks according to their priority levels, based on a list of difference data blocks with level identifiers. The prioritization rule is as follows: urgent difference data blocks are placed at the beginning of the queue, followed by priority difference data blocks, and then ordinary difference data blocks are placed at the end. Data blocks within the same difference level are arranged in ascending order of their block numbers. The sending end extracts the complete data content of each difference data block sequentially according to the sorted queue order, prioritizing and sending the complete data content of urgent difference data blocks, followed by priority difference data blocks, and finally ordinary difference data blocks. The complete data content of each difference data block is encapsulated into a difference data response message. It includes three fields: data block number, difference level identifier, and complete data content of the data block. The receiving end parses the difference data response message in the order of reception, replaces the corresponding local data block with the data content in the difference data response message, and completes the synchronous update of the difference data block. After the replacement is completed, the receiving end updates the local synchronization snapshot cache, records the current synchronization data as the new snapshot value, and uses it for the next difference comparison. Through this degradation synchronization mechanism based on hash digest comparison and difference level hierarchical transmission, the total amount of data transmission is significantly reduced under the condition of synchronization link latency exceeding the limit, while ensuring that high-risk difference data is synchronized first, thus achieving reliable data synchronization guarantee in latency-sensitive scenarios.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention 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 embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0048] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0049] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A high-concurrency real-time database synchronization method for wind power clusters, characterized in that, include: The turbine nodes, farm nodes, cluster nodes and grid nodes in the wind power cluster are added to the data synchronization platform according to the hierarchical relationship of turbine-farm-cluster-grid. Data domain planning is performed when nodes are added, a corresponding real-time database partition is established for each level node, and several blank buffer time slots are pre-initialized in each database partition. Based on the buffer time slot, when the number of concurrent write requests in a certain data acquisition cycle of the wind power cluster exceeds the steady-state processing threshold of the current database partition, the incremental data packets exceeding the steady-state processing threshold are temporarily stored in the corresponding buffer time slot according to the device dimension label to which the data point belongs, and the data point affiliation relationship of each buffer time slot is recorded to obtain the affiliation record table. Based on the ownership record table, when an upper-level node initiates a data synchronization request to a lower-level node, it checks whether there are any uncommitted buffer time slot records for the data points involved in the request. If they exist, the data in the buffer time slot is merged with the data in the primary partition before synchronization is performed. If they do not exist, the data is directly extracted from the primary partition and synchronization is performed. Based on the differences in latency sensitivity of cross-level synchronization, differentiated latency tolerance windows are set for the synchronization links between the airport, the group, and the network. When the measured latency on the synchronization link exceeds the latency tolerance window of the corresponding level, the synchronization degradation strategy of the corresponding link is triggered.

2. The high-concurrency real-time database synchronization method for wind power clusters according to claim 1, characterized in that, The process of initializing several blank buffer time slots includes: The node attribute information carried by each level of the access data synchronization platform is statistically analyzed. The node attribute information includes device type identifier, sampling frequency identifier, and number of data points. Based on the sampling frequency identifier, the data points are divided into high-frequency data sampling points and low-frequency data sampling points, and then grouped and aggregated according to the device type identifier to obtain a high-frequency data point group set and a low-frequency data point group set. Calculate the rated data throughput corresponding to the high-frequency data point group and the low-frequency data point group respectively, and perform database partitioning based on their total real-time database storage capacity in the corresponding level nodes. Initialize the index structure of the database partitions, and pre-construct a fixed proportion of storage space at the end of each database partition to obtain blank buffer time slots. Write the time slot number, starting storage address, ending storage address, and writable capacity into the buffer time slot metadata table, and mark the buffer time slot status as idle.

3. The high-concurrency real-time database synchronization method for wind power clusters according to claim 2, characterized in that, The database partitioning includes establishing high-priority storage partitions and standard storage partitions based on the division of high-frequency data points and low-frequency data points.

4. The high-concurrency real-time database synchronization method for wind power clusters according to claim 2, characterized in that, The process of obtaining the attribution record table includes: Read the historical write peak records of the current database partition, and calculate the steady-state processing threshold within the current data acquisition cycle using the sliding average of the historical write peak records and the preset safety factor; Continuously monitor the real-time write request count of the database partition corresponding to each level node. When the write request count is higher than the steady-state processing threshold in three consecutive sampling intervals, the corresponding database partition is determined to enter the concurrent write peak state. For database partitions that have entered peak concurrent write state, all write requests arriving within the current data acquisition cycle are grouped according to the device dimension label corresponding to the data point, and incremental data packets exceeding the steady-state processing threshold are identified. The identified incremental data packets are matched based on the pre-established multi-dimensional index anchor points, and the incremental data packets are imported into the corresponding buffer time slots based on the matching results. At the same time, the data point number, the data acquisition cycle, the data point type identifier, and the temporary data volume are written into the buffer time slot's ownership record table. The target buffer time slot matching the corresponding device dimension label is retrieved from the buffer time slot of the current database partition according to the ownership record table. If the target buffer time slot has remaining writable capacity, the incremental data packet is written to the target buffer time slot, and the data point number and write position record of the corresponding incremental data packet are appended to the ownership record table of the buffer time slot. If the remaining capacity of the target buffer time slot is insufficient, the buffer time slot rolling expansion operation is initiated to merge adjacent free buffer time slots into an expanded buffer time slot before writing is performed.

5. The high-concurrency real-time database synchronization method for wind power clusters according to claim 4, characterized in that, The process of initiating the buffer time slot rolling expansion operation includes: Retrieve adjacent idle buffer time slots in the current database partition to obtain a list of idle adjacent time slots; The list of adjacent idle time slots is segmented according to the continuity of the time slot numbers to obtain a set of consecutive idle time slot segments; Select the shortest consecutive idle time slot from the set of consecutive idle time slots that are adjacent to the target buffer time slot to obtain the time slot segment to be merged; The start and end addresses of the time slot segments to be merged are merged into the address range of the target buffer time slot, and the storage boundary of the target buffer time slot is updated in the buffer time slot metadata table to obtain the expanded buffer time slot; and the original time slot entries of the time slot segments to be merged are marked as merged.

6. The high-concurrency real-time database synchronization method for wind power clusters according to claim 4, characterized in that, The process of checking whether there are any uncommitted buffer slot records for the data points involved in the request includes: When an upper-level node initiates a data synchronization request to a lower-level node, extract the list of requested data point numbers and the requested time range parameters. Based on the list of requested data point numbers, query the record status of each data point number in the buffer time slot attribution record table one by one; if a data point number has a record entry in the attribution record table with a status of uncommitted, mark the corresponding data point as a buffer-containing point to be merged and record its corresponding buffer time slot number; if a data point number does not have an uncommitted record in the attribution record table, mark the corresponding data point as a primary partition direct retrieval point. For points with buffers to be merged, extract buffer data segments from their corresponding buffer time slots according to the requested time range parameters to obtain a set of buffer data segments; at the same time, extract the primary partition data segments of the same data point within the same time range from the primary partition to obtain a set of primary partition data segments. The buffer data segment set and the primary partition data segment set are merged in order according to the timestamp field to obtain the merged data segment set; The merged data segment set and the data from the primary partition direct fetch point are encapsulated together into a synchronization response data packet, serialized according to the specified transmission format, and sent to the upper-layer node.

7. The high-concurrency real-time database synchronization method for wind power clusters according to claim 6, characterized in that, The process of orderly merging includes: The data records in the buffer data segment set and the primary partition data segment set are sorted in ascending order according to the timestamp field to obtain the buffer ordered sequence and the primary partition ordered sequence, respectively. Set two pointers to point to the current read position of the buffered ordered sequence and the main partition ordered sequence respectively, and initially point to the first record of their respective sequences; Compare the timestamps of the records currently pointed to by the two pointers, write the record with the smaller timestamp to the merge output buffer, and move the corresponding pointer one position to the right; when the timestamps of the records pointed to by the two pointers are equal, it is determined that the timestamps overlap, the corresponding record in the buffered ordered sequence is written to the merge output buffer, and the corresponding record in the main partition ordered sequence is marked as overwritten and skipped, and the two pointers move one position to the right in sync. Repeat the above comparison operation until both pointers reach the end of their respective sequences. Export all records in the merge output buffer in the order they were written to obtain the merged data segment set.

8. The high-concurrency real-time database synchronization method for wind power clusters according to claim 1, characterized in that, The process of setting the differentiated latency tolerance window includes: Obtain historical latency sampling sequences of synchronization links between each level of the airport-field-group-network, calculate the mean latency and standard deviation of latency for each synchronization link in the last N sampling periods, and obtain the latency statistical parameters of each link; Calculate the initial value of the basic tolerance window for the corresponding synchronization link based on the latency statistics parameters of each link. Obtain the total number of online data points currently carried by each level node, and use the ratio of the total number of online data points to the designed rated data point capacity of the corresponding level node to obtain the load rate of each level node; based on the load rate of each level node, linearly amplify and correct the initial value of the basic tolerance window of the corresponding synchronization link to obtain the load-corrected tolerance window value. After load correction, the tolerance window value is accumulated and verified sequentially from the machine layer to the power grid layer. The load correction tolerance window value of each link that passes the accumulated constraint verification is determined as the final delay tolerance window, written into the synchronization link configuration table, and sent to the corresponding link management process.

9. The high-concurrency real-time database synchronization method for wind power clusters according to claim 1, characterized in that, The execution process of the synchronous degradation strategy includes: When a synchronization link triggers the synchronization degradation strategy, the sending end performs block processing on the current set of data to be synchronized for the corresponding synchronization link. The set of data to be synchronized is divided into several data blocks according to the device dimension label to which the data point belongs, thus obtaining the set of data blocks to be synchronized. For each data block in the set of data blocks to be synchronized, calculate the hash digest value of its data content to obtain a hash digest set for each data block; encapsulate the hash digest set of each data block into a digest synchronization message according to the data block number, and transmit it to the receiving end through the downgraded synchronization link; The receiving end parses the digest synchronization message and extracts the hash digest set of each data block; it calculates the digest value of the corresponding data block with the same hash algorithm on the local side of the receiving end to obtain the local hash digest set of each data block. The hash digest set of each data block on the receiving end is compared block by block with the hash digest set of each data block received, and the data block numbers with inconsistent hash digest values ​​are filtered out to obtain a list of differing data block numbers; The receiving end encapsulates the list of difference data block numbers into a difference request message and sends it back to the sending end; the sending end extracts the complete data content of the corresponding data block according to the list of difference data block numbers, encapsulates it into a difference data response message and sends it to the receiving end; the receiving end replaces the corresponding local data block with the data content in the difference data response message.

10. The high-concurrency real-time database synchronization method for wind power clusters according to claim 9, characterized in that, The process by which the sending end extracts the complete data content of the corresponding data block based on the list of difference data block numbers, encapsulates it into a difference data response message, and sends it to the receiving end also includes: For each differential data block in the differential data block number list, at the sending end, extract the absolute deviation between the current value of all data points in the corresponding differential data block and the corresponding snapshot value of the corresponding data block when the receiving end last successfully synchronized, and obtain the set of deviations for each data point. Calculate the maximum value, mean value and proportion of non-zero deviation points for each set of deviations to obtain the data block deviation feature triplet; Based on the triplet of data block deviation characteristics, the difference level determination rules are set as follows: when the maximum deviation exceeds the preset high-risk threshold or the proportion of non-zero deviation points exceeds the preset high-density threshold, the difference level of the corresponding data block is set to the emergency level; when the maximum deviation is between the preset high-risk threshold and the preset medium-risk threshold or the proportion of non-zero deviation points is between the preset high-density threshold and the preset medium-density threshold, the difference level is set to the priority level; in other cases, the difference level is set to the normal level. In the difference request message, each difference data block number is appended with its corresponding difference level identifier. After receiving the difference request message, the sender prioritizes the difference data blocks according to the difference level identifier, encapsulates and sends the complete data content of the urgent difference data blocks first, then sends the priority difference data blocks, and finally sends the normal difference data blocks.

Images