A method and system for generating typhoon real-time data based on a high time-frequency update mechanism

Abstract

This invention discloses a method and system for generating typhoon real-time data based on a high-frequency update mechanism, specifically relating to the field of meteorological information processing. The method includes edge computing nodes collecting local typhoon observation data within their respective regions, performing edge real-time data generation processing on the local typhoon observation data, and associating and encapsulating the corresponding collection period, edge generation period, and impact period with the edge real-time data generation processing results to obtain an edge result package. A central processing node receives the edge result package and central observation data that arrive on time within the target release period, and performs real-time data fusion generation on the timely-arriving edge result package and central observation data to obtain the current round of real-time data results. By performing rewriting eligibility determination, allowed rewriting range constraints, local write-back updates, and cross-round inheritance calls on late-arriving edge observation data, late-arriving observation data can be continuously incorporated into subsequent typhoon real-time data generation processes without rewriting already closed areas.

Description

Technical Field

[0001] This invention relates to the field of meteorological information processing technology, and more specifically, to a method and system for generating typhoon real-time data based on a high-frequency update mechanism. Background Technology

[0002] In the field of typhoon real-time data fusion generation, the mainstream practice in the industry is to solve the problem of unified access and rapid mapping of multi-source observation results within the same release cycle. Usually, the central side collects data uploaded by satellites, weather radars, ground automatic stations, buoys and coastal stations according to a preset update cycle. Time alignment, spatial registration, missing data filling and weighted fusion are performed on the data from different sources to generate the typhoon real-time results at the corresponding time. For example, in application scenarios that release information minute by minute in coastal areas, the system needs to continuously receive local observation results uploaded by edge computing nodes and continuously output typhoon real-time maps that can be directly used for early warning linkage. At the same time, it must also meet the hard constraints that the released results cannot be frequently bounced back, the delayed data transmission from the edge side cannot be invalid for a long time, and the overall generation link cannot be repeatedly recalculated globally due to local retransmission. However, under this constraint, the mainstream approach will expose a type of defect. When local observations arrive late due to link jitter, short-term offline, or batch retransmission at coastal edge nodes, if such late data is allowed to directly participate in the rewriting of the results of the published rounds, phenomena such as local structural reversals, typhoon center position pullback, sudden changes in the boundary of the heavy rainfall belt, or repeated oscillations of the outer edge of the wind circle will occur between two or more consecutive rounds of published results. On the other hand, if late data is not accepted at all, it will lead to the inability of real observations on the edge side to enter the continuous published results for a long time, resulting in continuous distortion of local realities. The fundamental reason is that existing technologies generally assume that newly arrived data naturally has the right to rewrite results, focusing on how to integrate data values, but not on whether late data still has the qualification to rewrite in the continuous high-frequency publication chain, which round of results can be rewritten, and to what extent rewriting is allowed. The technical problem this application aims to solve is: how to establish an effective result rewriting qualification control mechanism for late-arriving local observation data during the high-frequency continuous release of typhoon real-time data based on edge computing, so as to avoid local rebound of the released real-time results, while ensuring that real observations on the edge side can be continuously incorporated into the subsequent real-time generation process. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a typhoon real-time data fusion generation method and system based on a high-frequency update mechanism. By performing rewriting eligibility determination, allowable rewriting range constraints, local write-back updates, and cross-round inheritance calls on late-arriving edge observation data, late-arriving observation data can be continuously incorporated into the subsequent typhoon real-time data generation process without rewriting the closed area, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for generating typhoon real-time data based on a high-frequency update mechanism, comprising: S1. Edge computing nodes collect local typhoon observation data within their respective areas, perform edge real-time generation processing on the local typhoon observation data, and associate and encapsulate the corresponding collection time period, edge generation time period, and impact time period with the edge real-time generation processing results to obtain an edge result package. S2. The central processing node receives the edge result packets and central observation data that arrive on time within the target release cycle. It performs current round real-time fusion generation on the timely edge result packets and central observation data to obtain the current round real-time results. Based on the data cutoff time and the time period coverage of the data included in the current round real-time results, it determines the closed boundary and then divides the current round real-time results into closed and unclosed areas. S3. The central processing node receives late edge result packets that do not arrive on time within the target release period, compares the impact period of each late edge result packet with the closed boundary, and determines the corresponding impact area of ​​each late edge result packet with the closed area and the unclosed area to obtain the rewriting qualification mark and allowed rewriting range of the late edge result packet. S4. The central processing node performs a local write-back update on the current round's live results only within the corresponding allowed rewrite range for late edge result packets that are marked as allowing rewrite eligibility, in order to obtain the current round's published live results. S5. The central processing node writes the late edge result package, which is marked as prohibited from being rewritten, into the next round candidate input set according to its collection time period, edge generation time period and impact time period, and adds a cross-round inheritance identifier to the late edge result package to obtain the next round inheritance input data. S6. The central processing node outputs the current round of real-time results and calls the next round of inherited input data in the next release cycle to participate in the next round of real-time fusion generation, so that late-arriving edge observation data can be continuously incorporated into the subsequent typhoon real-time generation process without rewriting the closed area.

[0005] In a preferred embodiment, S1 includes: S1-1. Obtain the local typhoon observation data output by each observation source, extract the observation source identifier, collection timestamp, spatial location identifier and observation value, map the collection timestamp to a unified edge time base, and arrange them in time order and spatial location order to obtain the local observation sequence. S1-2. Perform linear interpolation on the missing observations in the local observation sequence before and after the time interval, and take the arithmetic mean of multiple observations at the same time and the same spatial location to generate the edge real-time generation processing result. At the same time, determine the earliest and latest acquisition timestamps corresponding to the edge real-time generation processing result as the acquisition period, and determine the generation timestamp of the edge real-time generation processing result as the edge generation period. S1-3. Calculate the duration of the data collection period, and determine the period of influence by taking the start time of the edge generation period as the start time of the influence and the time obtained by adding the duration to the start time of the influence as the end time of the influence. Then, associate and encapsulate the edge real-time generation processing result, the data collection period, the edge generation period, and the period of influence to obtain the edge result package.

[0006] In a preferred embodiment, in S2, the generation of the current round's live results and the process of dividing the closed and unclosed regions include: S2-1. Obtain the edge result packages and central side observation data that arrive on time within the target release period, extract the edge real-time generation and processing results, collection time period, edge generation time period and impact time period corresponding to each edge result package, and map the central side observation data to the target release period according to the collection timestamp to obtain the fusion input set for this round. S2-2. Map the edge real-time generation and processing results of each edge in the current round of fusion input set to a unified real-time grid according to their spatial location. Then, overlay the mapped edge real-time generation and processing results with the central observation data in the same real-time grid according to the order of collection time to generate the current round of real-time results. At the same time, determine the latest collection timestamp in the current round of fusion input set as the data cutoff time. S2-3. Calculate the end time of the influence of the input data corresponding to each real-time grid in the current round of real-time results. Determine the real-time grids whose influence end time is earlier than or equal to the data cutoff time as closed regions, and determine the real-time grids whose influence end time is later than the data cutoff time as unclosed regions.

[0007] In a preferred embodiment, S3 includes: S3-1. Map the affected area in the late edge result package to the unified real-time grid corresponding to the current round of real-time results, write the affected time period into the corresponding real-time grid node according to the grid, and construct the late rewriting spatiotemporal graph according to the adjacency relationship of the real-time grid to obtain the candidate rewriting node set. S3-2. For each real-world grid node in the candidate rewrite node set, calculate the node rewrite confidence based on the temporal relationship between its influence period and the closed boundary. Real-world grid nodes whose influence ends before the closed boundary are marked as closed conflict nodes, real-world grid nodes whose influence begins after the closed boundary are marked as directly rewriteable nodes, and real-world grid nodes whose influence period crosses the closed boundary are marked as nodes to be solved, thus obtaining the node state set. S3-3. Using the maximum sum of rewritten confidence of retained nodes, the minimum number of piercing edges of closed conflict nodes, and the retention of the node set and the unclosed region as joint constraints, perform a connected subgraph search on the node state set, remove the real grid nodes that form piercing paths with the closed region, and retain the real grid nodes that are connected to the unclosed region to obtain the initial rewritten subgraph.

[0008] In a preferred embodiment, S3 further includes: S3-4. Using the initial rewritten subgraph as input, iteratively perform boundary consistency verification and node state backhaul update. In each iteration, delete the real grid nodes that cause new conflicts in the boundary relationship of the current round's real results and recalculate the rewrite confidence of the remaining real grid nodes until the set of retained nodes obtained from two consecutive iterations is consistent, thus obtaining the allowable rewrite range. S3-5. Determine late edge result packets with an empty allowed rewrite range as prohibited from rewriting, and determine late edge result packets with a non-empty allowed rewrite range as allowed to rewrite, and output the corresponding rewrite qualification flag and allowed rewrite range.

[0009] In a preferred embodiment, S4 includes: S4-1. Extract the edge real-world generation and processing results from the late edge result package that is allowed to be rewritten, and map the edge real-world generation and processing results and the current round real-world results to the allowed rewrite range according to a unified real-world grid, so as to obtain the late grid value and the current grid value corresponding to each real-world grid within the allowed rewrite range. S4-2. Within the allowed rewriting range, the real grids that share a common edge with the real grids outside the allowed rewriting range are identified as boundary grids. Starting from the boundary grids, the rewriting range is expanded layer by layer. The level number corresponding to each real grid is determined according to the shortest grid step from each real grid to the nearest boundary grid. S4-3. Using the maximum level number within the allowed rewrite range as the normalization base, determine the write weight of the late grid value for each real grid according to the ratio of its level number to the maximum level number, and calculate the updated grid value according to the write weight and the remaining weight of the corresponding current grid value. Boundary grids with a level number of zero retain the current grid value, and internal grids with a level number equal to the maximum level number write the late grid value. S4-4. Write back the updated grid values ​​corresponding to each live grid to the live grids corresponding to the current round's live results within the allowed rewrite range, and keep the values ​​of each live grid outside the allowed rewrite range unchanged, to obtain the current round's published live results.

[0010] In a preferred embodiment, S5 includes: S5-1. Extract the edge real-time generation processing result, collection period, edge generation period and impact period from the late edge result package characterized by the rewriting qualification mark as prohibiting rewriting. Generate a temporal inheritance index according to the start time of the collection period, the end time of the collection period, the edge generation period and the end time of the impact period to obtain candidate inheritance entries. S5-2. Write the candidate inheritance entries into the candidate input set for the next round, and associate the temporal inheritance index with the corresponding edge real-time generation processing result. Add cross-round inheritance identifiers to candidate inheritance entries whose collection period ends later than the current round's data cutoff time and whose influence period ends later than the current round's data cutoff time, to obtain the next round's inheritance input data. S5-3. Arrange the next round of inheritance input data in sequence according to the time-sequence inheritance index, and output the next round of inheritance input data with cross-round inheritance identifier to the next release cycle call.

[0011] In a preferred embodiment, S6 includes: S6-1. Output the current round of release results, and extract the data cutoff time, closed area and unclosed area corresponding to the current round of release results. Write the data cutoff time, closed area and unclosed area into the fusion control information of the next release cycle to obtain the fusion constraint data of the next round. S6-2. Obtain the next round of inheritance input data, extract the edge real-time generation processing results, collection time period, edge generation time period, impact time period and cross-round inheritance identifier corresponding to each next round of inheritance input data, compare the impact time period with the data cutoff time in the next round of fusion constraint data, and map the impact area with the closed area and the unclosed area to obtain the next round of inheritance candidate set; S6-3. Merge the next round of inherited candidate sets with the newly arrived edge result packages and center-side observation data in the next release cycle, and use them as input data for the next round of real-time fusion generation, so that the late-arriving edge observation data can be continuously incorporated into the subsequent typhoon real-time generation process without rewriting the closed area.

[0012] In a preferred embodiment, in S6-2, the process of generating the next round of inheritance candidate sets includes: S6-2-1. Obtain the next round of inheritance input data, extract the edge real-world generation processing results, collection time period, edge generation time period, influence time period, influence area and cross-round inheritance identifier corresponding to each next round of inheritance input data, and map the influence area to the unified real-world grid corresponding to the next round of fusion constraint data to obtain the inheritance mapping result; S6-2-2. Compare the end time of the influence period of each next round of inheritance input data in the inheritance mapping result with the data cutoff time. For the next round of inheritance input data whose end time of influence period is later than the data cutoff time, retain the cross-round inheritance identifier. For the next round of inheritance input data whose end time of influence period is earlier than or equal to the data cutoff time, delete the cross-round inheritance identifier to obtain the time series filtering result. S6-2-3. Perform grid correspondence statistics on the affected area corresponding to the next round inheritance input data with the cross-round inheritance identifier retained in the time series filtering results, and the closed and unclosed areas. Extract the next round inheritance input data in which the number of actual grids falling into the unclosed area is greater than the number of actual grids falling into the closed area, and obtain the next round inheritance candidate set.

[0013] A typhoon real-time data fusion generation system based on a high time-frequency update mechanism includes: The encapsulation and generation module collects local typhoon observation data within its area through edge computing nodes, performs edge real-time generation processing on the local typhoon observation data, and associates and encapsulates the corresponding collection time period, edge generation time period, and impact time period with the edge real-time generation processing results to obtain an edge result package. The fusion closure module receives edge result packets and central observation data that arrive on time within the target release cycle through the central processing node. It performs current round of real-time fusion generation on the timely edge result packets and central observation data to obtain the current round of real-time results. Based on the data cutoff time corresponding to the current round of real-time results and the time period coverage of the data already included, it determines the closure boundary and then divides the current round of real-time results into closed and unclosed areas. The qualification determination module receives late edge result packets that do not arrive on time within the target release cycle through the central processing node. It compares the impact period of each late edge result packet with the closed boundary, and determines the corresponding impact area of ​​each late edge result packet with the closed area and the unclosed area to obtain the rewriting qualification mark and allowed rewriting range of the late edge result packet. The local write-back module, through the central processing node, identifies late-arriving edge result packets that are marked as allowing rewrite eligibility. It then performs local write-back updates on the current round's live results only within the corresponding allowed rewrite range to obtain the current round's published live results. The cross-round inheritance module, through the central processing node, writes the late edge result package, which is characterized by the rewriting qualification mark as prohibited from rewriting, into the next round candidate input set according to its collection time period, edge generation time period, and impact time period, and adds a cross-round inheritance identifier to the late edge result package to obtain the next round inheritance input data; The continuous inclusion module outputs the current round of real-time data release results through the central processing node, and calls the next round of inherited input data in the next release cycle to participate in the next round of real-time data fusion generation, so that late-arriving edge observation data can be continuously included in the subsequent typhoon real-time data generation process without rewriting the closed area.

[0014] The technical effects and advantages of this invention are as follows: By setting rewrite eligibility flags and allowed rewrite ranges for late edge result packages, late observations no longer automatically acquire global write-back rights, thus reducing local bounces of published live results and ensuring that valid late observations are continuously incorporated into subsequent generation processes. By dividing closed and unclosed areas based on the data cutoff time and the period of impact, it is possible to distinguish between areas that can be stably retained and areas that can be further adjusted in the current round, thereby improving the update boundary control in high-frequency continuous releases; By generating edge result packages for associated acquisition periods, edge generation periods, and impact periods at the edge side, a unified time basis can be provided for subsequent fusion, judgment, and inheritance at the center side, thereby improving the consistency of connection under multi-source asynchronous input; By performing local write-back updates only within the allowed rewrite range and assigning late grid value write weights according to hierarchical sequence numbers, local corrections can be completed while keeping the real-world grid values ​​outside the range unchanged, thereby reducing the overall recalculation pressure caused by local retransmissions. By transferring late edge results packets that are prohibited from being rewritten into the next round of candidate input sets and attaching a cross-round inheritance flag, we can avoid the direct discarding of late observations, thereby alleviating the problem that real observations on the edge side cannot be included in the continuous publication of results for a long time. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention.

[0016] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0017] 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.

[0018] Refer to the instruction manual appendix Figure 1-2 The present invention provides a method for generating typhoon real-time data based on a high-frequency update mechanism, comprising: S1. Edge computing nodes collect local typhoon observation data within their respective areas, perform edge real-time generation processing on the local typhoon observation data, and associate and encapsulate the corresponding collection time period, edge generation time period, and impact time period with the edge real-time generation processing results to obtain an edge result package. S1 is used to generate edge result packages on the edge computing node side that can be directly called by the central processing node. Its overall purpose is to first unify the local typhoon observation data output from different observation sources under the same time reference and the same spatial arrangement rules, and then perform missing completion and isotope synthesis on the local observation sequence to form an edge real-time generation and processing result that can characterize the current local typhoon situation in the region. Subsequently, the edge real-time generation and processing result is further supplemented with the acquisition period, edge generation period, and impact period that can participate in the subsequent closure boundary determination and rewriting eligibility determination, so that the output result not only includes the observation value itself, but also includes time attributes and encapsulation structure that can be reused in subsequent steps. The implementation process includes the following steps: The purpose of S1-1 is to organize raw typhoon local observation data from different observation sources into a continuously processable local observation sequence. Its mechanism involves first standardizing the time representation and then standardizing the arrangement, thus providing consistent input for subsequent missing observation completion and co-location observation synthesis. In practice, edge computing nodes receive typhoon local observation data from various observation sources within their respective regions. This data may include one or more of the following: wind speed, wind direction, air pressure, rainfall, radar echo intensity, or local cloud image features. The edge computing nodes extract the observation source identifier, acquisition timestamp, spatial location identifier, and observation value from each typhoon local observation data. The spatial location identifier can be latitude and longitude coordinates, grid number, or station number. Subsequently, each acquisition timestamp is mapped to a unified edge time base. This unified edge time base can be a standard time axis maintained locally by the edge computing nodes. The edge computing nodes merge the mapped acquisition timestamps according to a preset time granularity, which can be determined by the edge nodes. The configuration file is set to 30 seconds, 1 minute, or 2 minutes, and the actual deployment can be selected according to the observation density of the region and the target release cycle length. When the collection timestamps output by multiple observation sources are mapped to the same time granularity unit, they are grouped into the same collection time. After the time mapping is completed, the edge computing nodes arrange all typhoon local observation data in the order of collection time, and arrange them in the coding order of spatial location identifiers within the same collection time to obtain a local observation sequence. Each record in the local observation sequence includes at least the observation source identifier, collection time, spatial location identifier, and observation value, and is written to the edge node cache for S1-2 to read. When a typhoon local observation data lacks an observation source identifier or collection timestamp, the edge computing node marks the data as invalid data and removes it from the local observation sequence construction process. When a typhoon local observation data lacks a spatial location identifier but can be matched to fixed station coordinates through the observation source identifier, the station location lookup table is called to complete the spatial location identifier before participating in the arrangement. The purpose of S1-2 is to generate edge real-time processing results based on local observation sequences and simultaneously determine the acquisition period and edge generation period. Its mechanism involves obtaining continuous, unified, and taggable local real-time results on the edge side through missing observation completion and co-location observation synthesis. In specific implementation, the edge computing node sequentially reads the local observation sequence from the edge node buffer. For each spatial location, it identifies the corresponding observation sequence and checks whether there is a missing acquisition time between two adjacent known acquisition times. When a missing acquisition time exists simultaneously with both a preceding and following known observation, the corresponding linear interpolation result is calculated based on the proportion of the time interval between the missing acquisition time and the preceding known acquisition time to the total time interval between the two known acquisition times. The results are then entered into the spatial location record corresponding to the missing acquisition time. When a missing acquisition time is located at the beginning or end of the sequence and it is impossible to obtain two known observations before and after it at the same time, the missing acquisition time is kept as a null value and is not included in this edge real-time generation process. After the missing observations are filled in, the edge computing node calculates the arithmetic mean of multiple observations from multiple observation sources under the same acquisition time and the same spatial location identifier to obtain the synthetic observation corresponding to the acquisition time and the spatial location identifier. After repeating the above process for all spatial location identifiers and all acquisition times, the edge real-time generation process result is obtained. The edge real-time generation process result can be represented as a set of local gridded observation results composed of multiple acquisition times and multiple spatial location identifiers. Subsequently, the edge computing node extracts the earliest and latest acquisition timestamps from all records constituting the edge real-time generation processing results, and determines these earliest and latest acquisition timestamps as the start and end times of the acquisition period. Simultaneously, the edge computing node records the system generation time when the edge real-time generation processing results are written to the result buffer, and determines this system generation timestamp as the start and end times of the edge generation period. Recording the edge generation period as a single generation time is to ensure a consistent representation of the edge result package's formation time in subsequent steps. The edge real-time generation processing results, acquisition period, and edge generation period are all written together to the edge result temporary storage area for S1-3 to read. In practical applications, such as when a coastal edge computing node simultaneously receives wind speed data from automatic weather stations, air pressure data from buoys, and local radar echo intensity data, different observation sources may generate multiple observation values ​​for the same grid location within the same minute granularity. In this case, the edge computing node first performs linear interpolation on the missing intermediate time, and then takes the arithmetic average of the multiple observation values ​​for the same minute and the same grid location to obtain the local composite observation value of the grid location within that minute. The purpose of S1-3 is to supplement the edge real-time generation and processing results with the impact time period and encapsulation structure that can be directly called by the subsequent central processing node. Its mechanism is to convert the data time span corresponding to the collection time period into the effective impact window of the edge real-time generation and processing results in the subsequent fusion link, and to make the edge real-time generation and processing results and their time attributes form a unified result package. In specific implementation, the edge computing node reads the edge real-time generation and processing results, the collection time period, and the edge generation time period from the edge result temporary storage area. First, it calculates the duration of the collection time period by subtracting the start time of the collection time period from the end time of the collection time period to obtain the time difference value. Then, it uses the start time of the edge generation time period as the impact start time, and adds the time difference value to the impact start time to obtain the impact end time, thereby determining the impact time period. The impact time period here is used to characterize the effective time range of the edge real-time generation and processing results when the subsequent central processing node performs closed boundary comparison and late edge result package rewriting qualification determination, rather than a direct measurement of the actual duration of the natural typhoon evolution. After determining the impact period, the edge computing node further forms a spatial coverage area based on the set of all spatial location identifiers involved in generating the edge real-time generation processing results, and records this spatial coverage area as the impact area, so that it can be directly called in S3 when mapping the impact area corresponding to the late edge result package to the unified real-time grid. Finally, the edge computing node writes the edge real-time generation processing results, acquisition period, edge generation period, impact period, and impact area into a unified encapsulation structure in the order of preset fields. The unified encapsulation structure includes at least the result identifier field, observation source set field, edge real-time generation processing result field, acquisition period field, edge generation period field, impact period field, and impact area field, thereby obtaining the edge result package, and outputting the edge result package to the receiving queue of the central processing node for S2 to call. When the acquisition period duration is zero, the edge computing node records both the impact start time and impact end time as the start time of the edge generation period to maintain the consistency of the subsequent closed boundary comparison caliber. When the set of spatial location identifiers corresponds to only a single site or a single grid, the single site or single grid is directly written as the impact area into the edge result package. The above implementation process unifies the time base, fills in missing local observations, synthesizes co-located observations and adds the acquisition period, edge generation period, impact period and impact area, so that the edge result package output by the edge computing node has three types of information: observation content, time span and spatial coverage. This allows the central processing node to directly use it for subsequent current-round fusion, closure boundary determination, late edge result package rewriting eligibility determination and cross-round inheritance processing, avoiding the lack of time and spatial attribute basis in subsequent steps due to simply uploading observations. In practical applications: When an edge computing node in a coastal area continuously receives typhoon local observation data corresponding to three automatic weather stations, one buoy, and a set of local radar submaps within a target release cycle, the edge computing node first maps each acquisition timestamp to a unified edge time base with a 1-minute granularity, and then forms a local observation sequence according to spatial location. For a missing buoy pressure value in a certain minute, if there are valid pressure values ​​in the two minutes before and after it, the pressure value of the missing minute is obtained by linear interpolation based on the pressure values ​​of the two minutes before and after. If there are both automatic weather station wind speed values ​​and radar-converted wind speed values ​​at the same grid location in the same minute, the arithmetic mean of the two is taken to generate the composite observation value of that grid location. Then, the acquisition period is determined by the earliest and latest acquisition timestamps of all the results involved in the generation, the edge generation period is determined by the system time when the edge computing node completes the writing of the results, and the affected period is determined by adding the start time of the edge generation period to the acquisition period duration. At the same time, the affected area is determined based on the set of spatial location identifiers of the results involved in the generation, and finally, an edge result packet that can be directly sent to the central processing node is formed.

[0019] S2. The central processing node receives the edge result packets and central observation data that arrive on time within the target release cycle. It performs current round real-time fusion generation on the timely edge result packets and central observation data to obtain the current round real-time results. Based on the data cutoff time and the time period coverage of the data included in the current round real-time results, it determines the closed boundary and then divides the current round real-time results into closed and unclosed areas. The above-mentioned S2 implementation process is used to organize the timely arriving edge result packets within the target release cycle and the central observation data into the current round fusion input set at the central processing node, and generate the current round's real-time results on a unified real-time grid. Then, based on the relationship between the end time of the influence of the input data corresponding to each real-time grid and the data cutoff time, the current round's real-time results are divided into closed and open regions for use in the qualification determination of subsequent late edge result packets. This implementation process includes the following steps: The purpose of S2-1 is to form the current round's fusion input set that can participate in the fusion calculation. Its mechanism involves first determining whether the edge result packets belong to the on-time arrival input of the current round according to the target release cycle, and then unifying the edge result packets with the central observation data into a fusion object under the same round. In specific implementation, the central processing node reads the start and end times of the current target release cycle and receives the edge result packet arrival records and the central observation data acquisition records. For each edge result packet, its reception completion time is extracted. Edge result packets with reception completion times earlier than or equal to the end time of the current target release cycle are identified as on-time arrival edge result packets, while edge result packets with reception completion times later than the end time of the current target release cycle are reserved for subsequent late-arriving edge result packet processing. Subsequently, edge real-time generation processing results, acquisition time periods, and edge data are extracted from each on-time arrival edge result packet. The edge generation period, impact period, and impact area are recorded and written into the edge input buffer for this round. Simultaneously, the acquisition timestamp, spatial location identifier, and observation value are read from the central observation data. Central observation data whose acquisition timestamp falls between the start and end times of the current target release cycle are retained, while other central observation data are removed from this round of processing. Finally, the timely-arriving edge result packets in the edge input buffer for this round are merged with the retained central observation data to form the fusion input set for this round, and written into the fusion calculation buffer for S2-2 to read. When no timely-arriving edge result packets are received within the current target release cycle, the fusion input set for this round is constructed solely from the central observation data. When the central observation data is empty, the fusion input set for this round is constructed solely from the timely-arriving edge result packets. When both types of inputs are empty, an empty round marker is output, and the fusion process is retried in the next target release cycle. The purpose of S2-2 is to generate the current round of real-time results and determine the data cutoff time on a unified real-time grid. Its mechanism is to map the edge real-time generation and processing results and the central observation data to the same spatial unit, and then perform sequential writing on the same real-time grid according to the order of collection time. This ensures that each real-time grid retains the last valid grid value in the current round, as well as the corresponding latest collection time and the end time of the impact. In specific implementation, the central processing node first reads the pre-established unified real-time grid, which covers the current typhoon monitoring area. Each real-time grid has a unique grid number and a fixed spatial range, and the spatial location identifier is mapped to the unique real-time grid through a geographic mapping relationship. Subsequently, it reads the current round of fusion input set from the fusion computing cache, maps each edge real-time generation and processing result to the corresponding real-time grid according to the impact area or spatial location identifier, maps the central observation data to the corresponding real-time grid according to the spatial location identifier, and retains the collection timestamp, observation value or edge real-time generation and processing result value, the end time of the impact, and the source identifier for each mapped input item. After mapping is completed, all mapped input items within the same real-time grid are sorted from earliest to latest according to their collection timestamps, and then written sequentially. Specifically, the first sorted input item is written to the current record value field of the real-time grid, and then subsequent input items are read one by one. When a subsequent input item is mapped to the same real-time grid, the current record value field of the real-time grid is updated with the subsequent input item, and the collection timestamp of the subsequent input item is written to the latest collection time field of the real-time grid. The end time of the impact corresponding to the subsequent input item is written to the current end time of the impact field of the real-time grid. The grid record set of all real-time grids together constitutes the real-time result of the current round. Based on this, the collection timestamps of all input items in the current round of fusion input set are counted, and the latest collection timestamp is taken as the data cutoff time. The data cutoff time is written into the metadata area of ​​the current round's real-time results for S2-3 to read. When there are two input items with the same collection timestamp in the same real-time grid, the writing is performed according to the source priority preset in the system configuration file. When an input item lacks an impact end time but has an impact period, the impact end time is extracted from the impact period and then participated in the writing. The purpose of S2-3 is to divide the current round of real-time results into closed and open regions based on the comparison between the end time of the influence of the input data corresponding to each real-time grid and the data cutoff time. Its mechanism is to separate real-time grids that have completed the current round's time coverage from those that may still be affected by subsequent late edge result packets. In practice, the central processing node reads the data cutoff time from the metadata area of ​​the current round's real-time results and the current influence end time from the grid record set of each real-time grid. Then, a comparison operation is performed on each of the real-time grids. If the current influence end time of a real-time grid is earlier than or equal to the data cutoff time, that real-time grid is marked as a closed grid; if the current influence end time of a real-time grid is later than the data cutoff time, that real-time grid is marked as an open grid. The set of all closed grids is determined as the closed region, and the set of all open grids is determined as the open region. Based on this, the grid boundaries that have common edges between closed and unclosed grids are extracted according to the adjacency relationship of the unified real-world grid, and the boundary relationships corresponding to these common edges are recorded as closed boundaries for subsequent comparison of the impact time period and determination of the impact area of ​​late edge result packets. Finally, the current round's real-world results, data cutoff time, closed area, unclosed area, and closed boundary are written together into the current round's result storage area for S3 to read. When a real-world grid does not receive any mapped input items in the current round, the real-world grid is marked as an unassigned grid and excluded from the division of closed and unclosed areas. When the current impact end time of a real-world grid is missing, the impact time period corresponding to the last input item written to the real-world grid is retrieved and the impact end time is extracted again before the comparison is performed. Through the above implementation process, the central processing node first organizes the edge result packets that arrive on time within the target release cycle and the central observation data into a unified fusion input set for this round. Then, it completes the writing on the unified real-time grid in the order of collection time, and obtains the current round real-time result containing the current record value, the latest collection time field, and the current impact end time field. Furthermore, it uses the grid-by-grid comparison results of the current impact end time and the data cutoff time to form closed areas, unclosed areas, and closed boundaries, so that the rewriting qualification judgment of subsequent late edge result packets has clear time and space basis. In practical applications: When the release cycle of a target is set to 10 minutes, the central processing node first identifies the coastal station edge result packets received before the end of the 10-minute cycle as timely arriving edge result packets, and includes the observation data from the central meteorological platform received at the same time into the current round of fusion input set; then, the local wind field results from the edge result packets and the point pressure observations from the central meteorological platform are mapped to the same unified real-time grid. For multiple input items arriving successively within the same real-time grid, the current recorded value and the current impact end time of the real-time grid are updated sequentially from earliest to latest according to the collection timestamp; after all real-time grids are written, the latest collection timestamp among all input items is taken as the data cutoff time, and then the current impact end time of each real-time grid is compared with the data cutoff time one by one. Real-time grids with impact end times earlier than or equal to the data cutoff time are classified into closed areas, and real-time grids with impact end times later than the data cutoff time are classified into unclosed areas. The common edge between the two types of real-time grids is recorded as the closed boundary, thus providing a basis for determining whether subsequent late edge result packets are eligible for rewriting.

[0020] S3. The central processing node receives late edge result packets that do not arrive on time within the target release period, compares the impact period of each late edge result packet with the closed boundary, and determines the corresponding impact area of ​​each late edge result packet with the closed area and the unclosed area to obtain the rewriting qualification mark and allowed rewriting range of the late edge result packet. The above-described S3 implementation process is used to solve the rewriteability of late edge result packets on the central processing node side. First, the affected area and affected time period of the late edge result packets are mapped to the unified real-world grid corresponding to the current round's real-world results. Then, based on closed boundaries, candidate rewrite nodes are divided into different node states. The allowed rewrite range is obtained through connected subgraph search and iterative consistency verification, and finally, a rewrite eligibility flag is output. The principle is that instead of directly writing the late edge result packets back to the current round's real-world results, the late edge result packets are first transformed into spatiotemporal graph solution objects constrained by closed boundaries. Then, the rewriteable parts are determined through temporal comparison, spatial connectivity constraints, and boundary conflict resolution. This implementation process includes the following steps: The purpose of S3-1 is to transform late edge result packets into a computable set of candidate rewriting nodes. Its mechanism is to first unify the spatial and temporal carriers, and then construct a late rewriting spatiotemporal map based on a unified real-world grid. In specific implementation, the central processing node receives the late edge result packets and reads the edge real-world generation processing results, the affected time period, and the affected area. The affected area is the spatial coverage range formed by the set of spatial location identifiers of the participants in generating the edge real-world processing results in S1. Then, the affected area is mapped to the unified real-world grid corresponding to the current round of real-world results. Specifically, according to the geographical mapping relationship between spatial location identifiers and real-world grid numbers, each real-world grid covered by the affected area is determined as a candidate real-world grid node. For each candidate real-world grid node, the start time of the impact, the end time of the impact, and the late edge result packet identifier corresponding to the late edge result packet are written. After the node writing is completed, connecting edges are established for adjacent candidate real-world grid nodes based on the adjacency relationship of the unified real-world grid. The adjacency relationship adopts the same four-adjacency relationship as S2, that is, connecting edges are established between two real-world grid nodes that have common edges in the top, bottom, left, and right directions, thereby constructing a late rewrite spatiotemporal graph. Finally, all candidate real-world grid nodes in the late rewrite spatiotemporal graph are output as a candidate rewrite node set and written to the rewrite qualification determination buffer for S3-2 to read. When the affected area mapping corresponds to only a single real-world grid, the single real-world grid is directly used as the candidate rewrite node set. When the late edge result package lacks an affected area but retains the set of spatial location identifiers that participated in the generation of edge real-world generation processing results, the central processing node regenerates the affected area based on the set of spatial location identifiers and then performs the mapping. The purpose of S3-2 is to calculate the node rewrite confidence and determine the node state for each real-world grid node in the candidate rewrite node set. Its mechanism involves transforming the temporal relationship between the influencing time period and the closed boundary into a unified metric for node rewriteability, providing a comparable solution basis for subsequent connected subgraph searches. In practice, the central processing node reads the candidate rewrite node set from the rewrite eligibility cache and the closed boundary from the current round's result storage area. The closed boundary corresponds to the grid boundary relationship between closed and unclosed regions, using the current round's data cutoff time as the temporal judgment benchmark. Subsequently, a temporal comparison operation is performed on each candidate real-world grid node. When the end time of the influence of a candidate real-world grid node is earlier than or equal to the data cutoff time, the candidate real-world grid node is marked as a closed conflict node, and the node rewrite confidence is set to 0; when the start time of the influence of a candidate real-world grid node is later than the data cutoff time, the candidate real-world grid node is marked as a directly rewritable node, and the node rewrite confidence is set to 1; when the start time of the influence of a candidate real-world grid node is earlier than the data cutoff time and the end time of the influence is later than the data cutoff time, the candidate real-world grid node is marked as a node to be solved, and the influence is set to 0. The node rewrite confidence of a candidate real-world grid node is obtained by subtracting the data cutoff time from the end time and dividing the total duration by subtracting the start time from the end time of the influence. This forms a node state set including node state and node rewrite confidence, which is written into the subgraph search cache for S3-3 to read. When the end time of the influence of a candidate real-world grid node is equal to the start time of the influence, the node rewrite confidence of the candidate real-world grid node is directly recorded as 0 or 1. The start time of the influence is recorded as 1 when it is later than the data cutoff time, and as 0 in other cases to avoid the denominator being zero. The purpose of S3-3 is to find the initial rewritten subgraph that satisfies the rewriting objective from the node state set. Its working mechanism is to prioritize retaining candidate real grid nodes with higher node rewriting confidence while maintaining connectivity with the unclosed region, and to exclude candidate real grid nodes that will penetrate the closed region. In specific implementation, the central processing node reads the node state set from the subgraph search buffer and reads the unclosed region and the closed region from the current round result storage area. First, all directly rewritable nodes are used as the initial search nodes, and the nodes to be solved connected to the initial search nodes through four-adjacency relationships are used as expansion nodes. A breadth-first search of the connected subgraph is performed. During the search, if a candidate real-world grid node shares a common edge with any real-world grid in the closed region, this common edge is recorded as a penetrating edge. A penetrating edge characterizes the possibility that the candidate real-world grid node, after being written, may directly affect the boundary of the closed region. Subsequently, three quantities are statistically analyzed for each connected candidate subgraph: the sum of the node rewrite confidence of all candidate real-world grid nodes in the connected candidate subgraph, the relationship between the connected candidate subgraph and the closed region, and the relationship between the candidate subgraph and the closed region. The number of penetrating edges is considered, and the third is whether the candidate connected subgraph is adjacent to at least one real grid in the unclosed region. After the statistics are completed, the candidate connected subgraphs that are not adjacent to the unclosed region are removed. Among the remaining candidate connected subgraphs, the one with the larger sum of node rewrite confidence is selected first. When the sum of node rewrite confidence of two candidate connected subgraphs is the same, the one with fewer penetrating edges is selected. The initial rewrite subgraph is thus determined and written into the consistency check buffer for S3-4 to read. When there are no directly rewriteable nodes in the candidate rewrite node set, the node to be solved with a node rewrite confidence greater than 0 is used as the initial search node to perform the same connected subgraph search. The purpose of S3-4 is to perform boundary consistency checks and node state backhaul updates on the initial rewritten subgraph to obtain a stable allowable rewrite range. Its mechanism involves iteratively deleting newly added conflicting nodes and recalculating the node rewrite confidence of the remaining nodes, gradually eliminating the disruption of the boundary relationships of the current round's live results by the initial rewritten subgraph. Specifically, the central processing node reads the initial rewritten subgraph from the consistency check cache and reads the closed region, unclosed region, and closed boundary from the current round's result storage area. Then, it performs boundary consistency checks on all candidate live grid nodes in the initial rewritten subgraph one by one. Specifically, it checks whether retaining a candidate live grid node causes it to form a new direct adjacency with the closed region through a four-adjacency relationship, or whether nodes originally only adjacent to unclosed regions gain a new direct adjacency with the closed region after retention. If any of these situations occur, the node is then... Candidate real-world grid nodes are identified as newly added conflict nodes and removed from the initial rewritten subgraph. After this round of deletion, the node rewriting confidence of the remaining candidate real-world grid nodes is recalculated. The recalculation method is as follows: directly rewritable nodes are still recorded as 1, closed conflict nodes are still recorded as 0, and the node to be solved is still calculated by dividing the duration obtained by subtracting the data cutoff time from the end time of its influence by the total duration obtained by subtracting the start time of its influence from the end time of its influence. At the same time, the node rewriting confidence of candidate real-world grid nodes that have lost their connectivity with the unclosed region is updated to 0. Based on this, a new set of retained nodes is generated, and the next round of boundary consistency verification is performed. When the set of retained nodes obtained from two consecutive iterations is completely consistent, the iteration is stopped, and the real-world grid range corresponding to the set of retained nodes is determined as the allowable rewriting range. When the set of retained nodes is empty after the first iteration, the allowable rewriting range is directly recorded as an empty set. The purpose of S3-5 is to output a rewrite qualification flag based on the allowed rewrite range. Its mechanism is to convert the aforementioned spatiotemporal graph solution results into control results that can be directly called by subsequent local write-back updates or cross-round inheritance processing. In specific implementation, the central processing node reads the allowed rewrite range and counts the number of real-world grids contained in the allowed rewrite range. When the number of real-world grids in the allowed rewrite range is equal to zero, the late edge result package is determined to be prohibited from rewriting, and a prohibited rewrite flag corresponding to the late edge result package identifier is generated. When the number of real-world grids in the allowed rewrite range is greater than zero, the late edge result package is determined to be allowed to rewrite, and an allowed rewrite flag corresponding to the late edge result package identifier is generated. Subsequently, the central processing node writes the late edge result package identifier, rewrite qualification flag, and allowed rewrite range together into the rewrite qualification result table for S4 to read. When there are multiple late edge result packages within the same target release cycle, S3-1 to S3-5 are repeated for each late edge result package, and their respective rewrite qualification flags and allowed rewrite ranges are written into the rewrite qualification result table. Through the above implementation process, the central processing node first transforms the late edge result packet into a late rewrite spatiotemporal graph, then calculates the node rewrite confidence using the temporal relationship between the period of influence and the data cutoff time, and obtains the allowed rewrite range through connected subgraph search and boundary consistency iteration, and finally outputs the rewrite qualification flag, thereby avoiding the late edge result packet from directly writing back to the closed area, while retaining the effective part that can be applied to the unclosed area. In practical applications: When a coastal edge computing node retransmits a late edge result packet after the current target release cycle ends, the central processing node first maps the affected area of ​​the late edge result packet onto a unified real-world grid, forming multiple candidate real-world grid nodes. Then, the start and end times of the effects of these candidate real-world grid nodes are compared with the current round's data cutoff time. Candidate real-world grid nodes that are significantly later than the data cutoff time are designated as directly rewritable nodes; those that span the data cutoff time are designated as nodes to be solved; and those that end earlier than or equal to the data cutoff time are designated as closed-loop nodes. The process begins with identifying bridging nodes. Next, starting with directly rewritable nodes, a connected subgraph search is performed. The sum of node rewrite confidence and the number of penetrating edges for each connected candidate subgraph are calculated. Connected candidate subgraphs connected to unclosed regions and having a large sum of node rewrite confidence are selected as the initial rewrite subgraphs. Subsequently, candidate real-world grid nodes that would create direct adjacencies to closed regions are deleted until the set of retained nodes is consistent across two consecutive iterations. If multiple candidate real-world grid nodes are still retained, the late-arriving edge result packet is determined as rewritable, and the real-world grid range formed by these candidate real-world grid nodes is output as the rewritable range for subsequent local write-back updates.

[0021] S4. The central processing node performs a local write-back update on the current round's live results only within the corresponding allowed rewrite range for late edge result packets that are marked as allowing rewrite eligibility, in order to obtain the current round's published live results. S4 is used to perform local write-back updates on late-edge result packets that are marked as allowing rewrite. Its core is to first extract late-edge grid values ​​and current grid values ​​within the allowed rewrite range, then construct a hierarchical sequence from the outside in based on the positional relationship between each real-world grid and the boundary within the allowed rewrite range, and use the hierarchical sequence to generate write weights that gradually increase from the boundary inwards, so that late-edge result packets are smoothly written to the current round's real-world results only within the allowed rewrite range. This implementation process includes the following steps: The purpose of S4-1 is to prepare the corresponding late grid values ​​and current grid values ​​for local write-back updates within the allowed rewrite range. Its mechanism involves first unifying the late edge result package and the current round's real-world results to the same real-world grid, and then forming directly computable dual-value inputs grid-by-grid within the allowed rewrite range. In specific implementation, the central processing node reads the late edge result package, marked as allowed for rewrite, and its allowed rewrite range from the rewrite eligibility result table, extracts the edge real-world generation processing results from the late edge result package, and reads the current round's real-world results from the current round's result storage area. Subsequently, the edge real-world generation processing results are mapped to the real-world grid corresponding to the allowed rewrite range according to the unified real-world grid, and the current round's real-world results are updated accordingly. The current recorded value of the corresponding real-time grid in the next real-time result is extracted synchronously. For each real-time grid within the allowed rewrite range, the late grid value obtained by mapping from the late edge result package and the current grid value read from the current round of real-time results are recorded respectively to form a local write-back calculation table. This local write-back calculation table is written into the write-back calculation buffer for S4-2 to read. When a real-time grid does not have a corresponding late grid value in the edge real-time generation processing result, the real-time grid is deleted from the allowed rewrite range and then participates in subsequent calculations. When the corresponding real-time grid in the current round of real-time results does not have a current recorded value, the current grid value is made up by the arithmetic mean of the adjacent assigned real-time grids in the current round of real-time results and then written into the local write-back calculation table. The purpose of S4-2 is to determine the hierarchical sequence number of each real-world grid within the permissible rewrite range. Its mechanism involves using the boundary grid as the zero-layer starting point and expanding layer by layer into the permissible rewrite range according to the adjacency relationship of the unified real-world grids, giving each real-world grid a hierarchical position reflecting its distance from the boundary. In practice, the central processing node reads the local write-back calculation table from the write-back calculation buffer and determines all real-world grids within it based on the permissible rewrite range. For each real-world grid, it checks whether there are any real-world grids outside the permissible rewrite range in its four adjacent directions. If so, that real-world grid is designated as the boundary. The grid is first constructed, and its level number is recorded as 0. Then, starting with all boundary grids, a layer-by-layer expansion is performed. The actual grids that are directly adjacent to the boundary grids and are within the allowed rewrite range are assigned level number 1. The actual grids that are directly adjacent to the actual grids with level number 1 and have not yet been assigned a level number are assigned level number 2, and so on, until all actual grids within the allowed rewrite range have been assigned level numbers. Finally, the level numbers corresponding to each actual grid are written back to the local write-back calculation table for S4-3 to read. When all actual grids within the allowed rewrite range are boundary grids, the level numbers of all actual grids are recorded as 0. The purpose of S4-3 is to calculate the updated grid value based on the hierarchical index. Its mechanism involves constructing a write weight for late-arriving grid values ​​using the ratio between the hierarchical index and the maximum hierarchical index. This ensures that boundary grids maintain their current grid values ​​while internal grids gradually increase the write ratio of late-arriving grid values. In practice, the central processing node reads the hierarchical index corresponding to each real-world grid from the local write-back calculation table, taking the maximum hierarchical index as the normalization base. For boundary grids with a hierarchical index of 0, the current grid value is directly determined as the updated grid value. For other real-world grids with a maximum hierarchical index greater than 0, the hierarchical index of that real-world grid is first divided by the maximum hierarchical index. The system calculates the write weight of the late grid value based on the level number, then subtracts the write weight from 1 to get the remaining weight of the current grid value. Finally, it calculates the updated grid value by multiplying the late grid value by the write weight and the current grid value by the remaining weight. For internal grids with a level number equal to the maximum level number, the write weight is 1 and the remaining weight is 0, so the updated grid value is directly taken from the late grid value. After the updated grid value calculation for all real grids is completed, the updated grid value is written to the local write-back calculation table for S4-4 to read. When the maximum level number is equal to 0, all real grids within the rewrite range are allowed to retain their current grid value, and the late grid value is not written. The purpose of S4-4 is to generate the current round's published live results based on the updated grid values. Its mechanism involves replacing only the current record value of the corresponding live grid within the allowed rewrite range, while keeping the values ​​of live grids outside the allowed rewrite range unchanged. This limits the late-arriving edge result packets to being written into the current round's live results within the permitted range. In practice, the central processing node reads the updated grid values ​​corresponding to all live grids from the local write-back calculation table and writes them back one by one to the live grids within the allowed rewrite range in the current round's live results according to a unified live grid number. For live grids outside the allowed rewrite range, the original current record value in the current round's live results remains unchanged. After all write-backs are completed, the current round's published live results are obtained and written to the current round's published results storage area for S5 and S6 to read. When a live grid is missing an updated grid value during the write-back process, the original current record value of that live grid is retained. Through the above implementation process, the central processing node first extracts the late-arriving grid value and the current grid value within the allowed rewrite range. Then, it expands inward layer by layer according to the boundary grid to form a hierarchical sequence number, and calculates and updates the grid value based on the hierarchical sequence number. Finally, it completes the local write-back update only within the allowed rewrite range, thereby avoiding the impact of late edge result packets on the actual grid outside the allowed rewrite range. In practical applications: when the allowed rewrite range corresponding to a certain late edge result packet covers nine adjacent actual grids, the central processing node first extracts the late-arriving grid value and the current grid value corresponding to these nine actual grids. Next, the outer real-time grids that share a common edge with the real-time grids outside the allowed rewrite range are identified as boundary grids and assigned a level number of 0. The inner real-time grids at the center are assigned a larger level number. Then, the write weight of each real-time grid is calculated using the maximum level number as the normalization base. This ensures that the outer boundary grids maintain their current grid values, the middle layer real-time grids mix late grid values ​​and current grid values ​​proportionally, and the innermost real-time grids directly write late grid values. Finally, the updated grid values ​​are written back to the nine real-time grids corresponding to the current round's real-time results to obtain the current round's published real-time results.

[0022] S5. The central processing node writes the late edge result package, which is marked as prohibited from being rewritten, into the next round candidate input set according to its collection time period, edge generation time period and impact time period, and adds a cross-round inheritance identifier to the late edge result package to obtain the next round inheritance input data. The implementation process of S5 described above is used to transform late-end result packages, which are marked as prohibited from being rewritten, into next-round inheritance input data that can be called in the next release cycle. Its core lies in first generating a temporal inheritance index that can represent the time source relationship for the late-end result package, then determining whether the late-end result package meets the cross-round inheritance conditions based on the current round's data cutoff time, and finally arranging and outputting it to the next release cycle according to the completed temporal inheritance index. This implementation process includes the following steps: The purpose of S5-1 is to organize late-arriving edge result packets that are prohibited from being rewritten into candidate inheritance entries that can enter the inheritance process. Its mechanism is to extract the time field uniformly and generate a temporal inheritance index, so that each late-arriving edge result packet has a comparable and sortable inheritance time identifier. In specific implementation, the central processing node reads late-arriving edge result packets marked as prohibited from being rewritten from the rewriting qualification result table, extracts the edge situation generation processing result, collection period, edge generation period, and impact period from them; then it reads the start time of the collection period, the end time of the collection period, the start time of the edge generation period, and the end time of the impact period, and combines them according to a preset field order to generate a temporal inheritance index. The preset field order is fixed as the start time of the collection period, the end time of the collection period, the start time of the edge generation period, and the end time of the impact period, to ensure that the temporal inheritance index of different late edge result packages has a unified comparison caliber. After the generation is completed, the temporal inheritance index and the corresponding edge real-time generation processing result are written together into the candidate inheritance entry table to obtain candidate inheritance entries, and the candidate inheritance entries are written into the inheritance processing cache for S5-2 to read. When a late edge result package lacks an edge generation period but retains the generation timestamp, the generation timestamp is used to make up the start time of the edge generation period. When a late edge result package lacks the end time of the impact period, the impact period field is retrieved and extracted again. The purpose of S5-2 is to filter out entries that meet the cross-round inheritance conditions from candidate inheritance entries and form the input data for the next round of inheritance. Its mechanism involves a dual comparison of the collection period and the impact period of the candidate inheritance entries using the current round's data cutoff time, retaining late-arriving edge result packets that have not yet completed time coverage within the current round. In practice, the central processing node reads candidate inheritance entries from the inheritance processing cache and the current round's data cutoff time from the current round's result storage area. Then, the candidate inheritance entries are written into the next round's candidate input set, and the temporal inheritance index of each candidate inheritance entry is compared with the corresponding edge situation generation processing result. Associate the records; after writing, perform timing judgment on each candidate inheritance entry: when the end time of the collection period is later than the data cutoff time of the current round, and the end time of the influence period is later than the data cutoff time of the current round, add a cross-round inheritance identifier to the candidate inheritance entry; when either of the above two conditions is not met, do not add a cross-round inheritance identifier to the candidate inheritance entry; thus forming the next round of inheritance input data, and writing it to the inheritance result cache for S5-3 to read; when the edge real-time generation processing results of multiple candidate inheritance entries correspond to the same late edge result packet identifier, only the candidate inheritance entry with the later start time of the edge generation period is retained as a valid entry; The purpose of S5-3 is to provide a clear order of inheritance inputs for the next release cycle. Its mechanism involves using a temporal inheritance index to uniformly sort the next-round inheritance input data with cross-round inheritance identifiers, ensuring a stable temporal order when the next release cycle is called. In practice, the central processing node reads the next-round inheritance input data from the inheritance result cache, extracts the entries with cross-round inheritance identifiers, and sorts them according to the field order corresponding to the temporal inheritance index. Specifically, it sorts them first by the start time of the collection period in ascending order, then by the end time of the collection period in ascending order. If the first two items are the same, they are sorted by the start time of the edge generation period in ascending order. If the first three items are still the same, they are sorted by the end time of the affected period in ascending order. After sorting, the next-round inheritance input data with cross-round inheritance identifiers is written into the next release cycle call queue according to the sorting result and output to the next release cycle call. When there is no next-round inheritance input data with cross-round inheritance identifiers, an empty inheritance queue marker is output for S6 to identify. Through the above implementation process, the central processing node first generates a temporal inheritance index for the late edge result package that is prohibited from being rewritten, then determines whether it has the conditions for cross-round inheritance based on the data cutoff time of the current round, and finally outputs the entries that have the conditions for cross-round inheritance to the next release cycle in a unified order, so that the late edge result package that is prohibited from being rewritten in the current round can continue to participate in the live fusion generation in subsequent rounds. In practical applications: When a late edge result packet is determined to be unrewriteable in the current round, the central processing node first extracts the collection period, edge generation period, and impact period of the late edge result packet, and generates a temporal inheritance index according to the start time of the collection period, the end time of the collection period, the start time of the edge generation period, and the end time of the impact period. Then, the temporal inheritance index and the edge real-time generation processing result are written into the candidate inheritance entry, and the end time of the collection period and the end time of the impact period are compared with the data cutoff time of the current round. If both are later than the data cutoff time of the current round, a cross-round inheritance identifier is added to the candidate inheritance entry. Finally, the candidate inheritance entries with the cross-round inheritance identifier are written into the call queue of the next release cycle according to the temporal inheritance index order, so that they can participate in the subsequent real-time fusion generation in the next release cycle in a priority manner.

[0023] S6. The central processing node outputs the current round of real-time results and calls the next round of inherited input data in the next release cycle to participate in the next round of real-time fusion generation, so that late-arriving edge observation data can be continuously incorporated into the subsequent typhoon real-time generation process without rewriting the closed area. Furthermore, S6 is used to connect the current round of live data release results with the inherited data retained from the previous round into the next release cycle. Its core lies in first extracting closure information that can constrain the next round of fusion from the current round of live data release results, then performing temporal and spatial filtering on the inherited input data for the next round, and finally merging it with the newly arriving edge result packets and center-side observation data in the next release cycle to form the input data generated for the next round of live data fusion. This implementation process includes the following steps: The purpose of S6-1 is to establish fusion constraint data consistent with the current round's release results for the next release cycle. Its mechanism involves converting the time boundaries and spatial closure states already formed in the current round into control inputs that can be directly accessed in the next round. Specifically, the central processing node outputs the current round's release results and extracts the corresponding data cutoff time, closed area, and unclosed area from the current round's release result storage area. Then, according to a preset field order, the data cutoff time, closed area, and unclosed area are written into the fusion control information for the next release cycle. The preset field order is fixed as data cutoff time, closed area, and unclosed area to ensure that the next release cycle can read the data using a unified standard. After writing, the next round's fusion constraint data is obtained and written to the fusion constraint cache for S6-2 to read. When no valid unclosed area is formed in the current round's release results, the unclosed area is recorded as an empty set; when no valid closed area is formed in the current round's release results, the closed area is recorded as an empty set. The purpose of S6-2 is to filter out the next round of inheritance candidate sets from the next round of inheritance input data that can participate in the next release cycle fusion. Its working mechanism is to use the data cutoff time, closed area and unclosed area in the next round of fusion constraint data to perform time comparison and spatial mapping on the inheritance data, and retain the inheritance entries that still mainly affect the unclosed area. In specific implementation, the central processing node obtains the next round of inheritance input data from the inheritance result cache, and extracts the edge real-time generation processing result, collection time period, edge generation time period, affected time period, affected area and cross-round inheritance identifier corresponding to each next round of inheritance input data; then it reads the data cutoff time, closed area and unclosed area in the next round of fusion constraint data from the fusion constraint cache, and performs time comparison on each next round of inheritance input data, specifically by setting the affected time period... The end time is compared with the data cutoff time, and only the next round inheritance input data with the end time of the affected period being later than the data cutoff time and with the cross-round inheritance identifier is retained. Then, spatial mapping is performed on the retained next round inheritance input data to map its affected area to a unified real-world grid, and the number of real-world grids falling into closed areas and unclosed areas after mapping are counted respectively. When the number of real-world grids falling into unclosed areas is greater than the number of real-world grids falling into closed areas, the next round inheritance input data is written into the next round inheritance candidate set. After all filtering is completed, the next round inheritance candidate set is written into the next round fusion input buffer for S6-3 to read. When the affected area of ​​a certain next round inheritance input data does not fall into any closed or unclosed area after mapping, the next round inheritance input data is removed from this filtering. The purpose of S6-3 is to generate the input data for the next round of live fusion. Its mechanism involves integrating the filtered inherited data with newly arriving edge result packets and center-side observation data from the next release cycle into the same input set, allowing late-arriving edge observation data to enter subsequent rounds without rewriting already closed regions. Specifically, the central processing node reads the next round's inherited candidate set from the next round's fusion input buffer and receives newly arriving edge result packets and center-side observation data from the next release cycle. Then, the next round's inherited candidate set, newly arrived edge result packets, and center-side observation data are written into the next round's fusion input column according to a unified field structure. The table, with a unified field structure, includes at least the fields of source identifier, collection timestamp or collection period, affected period, affected area, and data content. After writing, the next round of fusion input list is sorted according to the collection time, and the sorted result is determined as the input data for the next round of live fusion generation, and written to the fusion calculation cache for the next round of S2 to call. When the next round of inheritance candidate set is empty, only the newly arrived edge result package and the central side observation data are used as the input data for the next round of live fusion generation. When both the newly arrived edge result package and the central side observation data are empty, the next round of inheritance candidate set is retained to participate in the next round of live fusion generation alone. Through the above implementation process, the central processing node first extracts the data cutoff time, closed areas, and unclosed areas from the current round of real-time data releases to form the fusion constraint data for the next round. Then, based on time and space conditions, it filters out the next round of inheritance candidate sets from the next round of inheritance input data. Finally, it merges this data with the newly arriving edge result packets and central observation data in the next release cycle to form the input data for the next round of real-time data fusion generation. This allows late-arriving edge observation data to be continuously incorporated into the subsequent typhoon real-time data generation process without rewriting the closed areas. In practical applications: after the current round of real-time data releases has already formed the data cutoff time, closed areas, and unclosed areas, the central processing node first writes these three types of information into the next round of data fusion. The first round of data fusion constrains the data; then, the next round of inherited input data retained from the previous round is read. For each inherited data, the end time of the influence period is compared with the data cutoff time. Only inherited data whose end time of the influence period is later than the data cutoff time and which has a cross-round inheritance identifier is retained. Then, its influence area is mapped to a unified real-world grid. The number of real-world grids that fall into unclosed and closed areas is counted. Only inherited data that mainly fall into unclosed areas are retained in the next round of inheritance candidate set. Finally, the next round of inheritance candidate set, together with the newly arrived edge result packages and center-side observation data in the next release cycle, are written into the next round of fusion input list and sorted according to the acquisition time order as the input data generated by the next round of real-world fusion.

[0024] Furthermore, the implementation process of S6-2 is used to screen out the next round of inheritance candidates that can continue to participate in the fusion from the next round of inheritance input data before the start of the next release cycle. Its core lies in first mapping the influence area of ​​the next round of inheritance input data to a unified real-world grid, then performing time-based filtering on inheritance entries based on the data cutoff time, and then performing spatial filtering based on the grid distribution of closed and unclosed regions, thereby retaining inheritance entries that mainly affect unclosed regions. This implementation process includes the following steps: The purpose of S6-2-1 is to organize the next round of inheritance input data into inheritance mapping results that can be directly subjected to time and space filtering. Its mechanism involves first uniformly reading the time and space fields from the next round of inheritance input data, and then mapping the affected area to the unified real-world grid corresponding to the next round of fusion constraint data. In specific implementation, the central processing node retrieves the next round of inheritance input data from the inheritance result cache, extracting the edge real-world generation processing results, collection time period, edge generation time period, affected time period, affected area, and cross-round inheritance identifier corresponding to each next round of inheritance input data. Subsequently, it reads the unified real-world grid corresponding to the next round of fusion constraint data and, according to the relationship between the affected area and the real-world grid number... The geographic mapping relationship maps the influence area of ​​each next-round inherited input data to several corresponding real-world grids. After mapping, for each next-round inherited input data, the corresponding real-world grid set, the end time of the influence period, and the cross-round inheritance identifier are recorded to form the inheritance mapping result. The inheritance mapping result is written to the inheritance filtering cache for S6-2-2 to read. When a next-round inherited input data lacks an influence area but retains a set of spatial location identifiers, the central processing node regenerates the influence area based on the set of spatial location identifiers and then performs the mapping. When a next-round inherited input data does not correspond to any real-world grids after mapping, the next-round inherited input data is removed from the inheritance mapping result. The purpose of S6-2-2 is to perform time filtering on the inheritance mapping results based on the data cutoff time. Its mechanism is to retain inheritance entries that are still later than the current time boundary and delete the cross-round inheritance identifiers corresponding to inheritance entries that have already fallen within the time range covered by the current round. In specific implementation, the central processing node reads the inheritance mapping results from the inheritance filtering cache and the data cutoff time from the next round of fusion constraint data. Then, it extracts the end time of the influence period for each next round inheritance input data in the inheritance mapping results and compares the end time of the influence period with the data cutoff time. When the end time of the influence period is later than the data cutoff time, the cross-round inheritance identifier of the next round inheritance input data is retained. When the end time of the influence period is earlier than or equal to the data cutoff time, the cross-round inheritance identifier of the next round inheritance input data is deleted. After all comparisons are completed, the time-series filtering results are obtained and written to the spatial filtering cache for S6-2-3 to read. When a next round inheritance input data lacks the end time of the influence period but retains the complete influence period, the end time is extracted from the influence period before the comparison is performed. The purpose of S6-2-3 is to extract the next round of inheritance candidate sets that are truly suitable for continued fusion in the next release cycle from the temporal screening results based on the grid distribution of closed and unclosed regions. Its mechanism is to retain inheritance entries that mainly affect unclosed regions and prevent inheritance entries from affecting closed regions again. In specific implementation, the central processing node reads the temporal screening results from the spatial screening buffer and extracts the next round of inheritance input data that retains the cross-round inheritance identifier. Then, it reads the closed and unclosed regions from the next round of fusion constraint data. For each next round of inheritance input data that retains the cross-round inheritance identifier, it counts the number of real grids falling into unclosed regions and the number of real grids falling into closed regions in the corresponding real grid set. When the number of real-world grids falling into an unclosed region is greater than the number of real-world grids falling into a closed region, the next round of inheritance input data is written into the next round of inheritance candidate set; when the number of real-world grids falling into an unclosed region is less than or equal to the number of real-world grids falling into a closed region, the next round of inheritance input data is not written into the next round of inheritance candidate set; after all statistics are completed, the next round of inheritance candidate set is obtained and output to the next round fusion input buffer for S6-3 to read; when the set of real-world grids corresponding to a certain next round of inheritance input data falls only into an unclosed region, it is directly written into the next round of inheritance candidate set; when the set of real-world grids corresponding to a certain next round of inheritance input data does not fall into either a closed region or an unclosed region, the next round of inheritance input data is removed. Through the above implementation process, the central processing node first maps the next round of inheritance input data to a unified real-time grid, then performs time filtering based on the data cutoff time, and subsequently performs spatial filtering based on the grid distribution of closed and unclosed areas, finally obtaining the next round of inheritance candidate set, thus ensuring that the inheritance entries entering the next release cycle still mainly affect the unclosed areas. In practical applications: when multiple next round inheritance input data are retained in the previous round, the central processing node first extracts the edge real-time generation processing results, collection time period, edge generation time period, impact time period, impact area, and cross-round inheritance identifier of each inheritance data, and maps the impact area to a unified real-time grid to form the inheritance mapping result; then, it compares the end time of the impact time period of each inheritance data with the data cutoff time, and only retains the cross-round inheritance identifier of inheritance data whose impact time period ends later than the data cutoff time; finally, it counts the number of real-time grids that fall into the unclosed and closed areas after the impact area mapping of the inheritance data with retained cross-round inheritance identifiers, and only retains the inheritance data with a large number of real-time grids falling into the unclosed areas in the next round of inheritance candidate set, so that it can continue to participate in real-time fusion generation in the next release cycle.

[0025] Furthermore, this solution, based on a typhoon real-time data fusion generation method using a high time-frequency update mechanism, also includes a typhoon real-time data fusion generation system based on a high time-frequency update mechanism, the system comprising: The encapsulation and generation module collects local typhoon observation data within its area through edge computing nodes, performs edge real-time generation processing on the local typhoon observation data, and associates and encapsulates the corresponding collection time period, edge generation time period, and impact time period with the edge real-time generation processing results to obtain an edge result package. The fusion closure module receives edge result packets and central observation data that arrive on time within the target release cycle through the central processing node. It performs current round of real-time fusion generation on the timely edge result packets and central observation data to obtain the current round of real-time results. Based on the data cutoff time corresponding to the current round of real-time results and the time period coverage of the data already included, it determines the closure boundary and then divides the current round of real-time results into closed and unclosed areas. The qualification determination module receives late edge result packets that do not arrive on time within the target release cycle through the central processing node. It compares the impact period of each late edge result packet with the closed boundary, and determines the corresponding impact area of ​​each late edge result packet with the closed area and the unclosed area to obtain the rewriting qualification mark and allowed rewriting range of the late edge result packet. The local write-back module, through the central processing node, identifies late-arriving edge result packets that are marked as allowing rewrite eligibility. It then performs local write-back updates on the current round's live results only within the corresponding allowed rewrite range to obtain the current round's published live results. The cross-round inheritance module, through the central processing node, writes the late edge result package, which is characterized by the rewriting qualification mark as prohibited from rewriting, into the next round candidate input set according to its collection time period, edge generation time period, and impact time period, and adds a cross-round inheritance identifier to the late edge result package to obtain the next round inheritance input data; The continuous inclusion module outputs the current round of real-time data release results through the central processing node, and calls the next round of inherited input data in the next release cycle to participate in the next round of real-time data fusion generation, so that late-arriving edge observation data can be continuously included in the subsequent typhoon real-time data generation process without rewriting the closed area.

[0026] Working Principle: The core idea of ​​this scheme is to first organize the local typhoon observation data within its jurisdiction into directly usable edge result packages at the edge computing nodes. Then, the central processing node continuously generates and updates the typhoon real-time results according to the release cycle. Specifically, data from different observation sources are first time-unified, missing data is filled in, and locally synthesized to form edge real-time generation and processing results, while simultaneously attaching the acquisition time period, edge generation time period, and impact time period. Subsequently, the central processing node merges the timely arriving edge result packages with the central observation data to generate the current round of real-time results, and divides the closed and unclosed areas according to the relationship between the impact end time and the data cutoff time of each grid data. For edge result packages that arrive late, they are not directly written back. Instead, it is first determined whether their impact range is still eligible to rewrite the current results. If rewriting is allowed, only a local write-back update is performed within the allowed rewriting range. If rewriting is not allowed, it proceeds to the next round to inherit the input data and continues to participate in the fusion in subsequent release cycles. This process can continuously absorb late data while avoiding repeated modification of already stably released areas. For example, in coastal typhoon monitoring scenarios, data from offshore buoys, shore-based automatic weather stations, local radars, and satellite sub-regions are continuously uploaded. However, the network status of different nodes is unstable, and data from some edge nodes arrives a few minutes late. According to this solution, each edge computing node first organizes the locally collected wind speed, air pressure, rainfall, or echo data into edge result packages and uploads them. Within the current 10-minute release cycle, the central processing node first uses the timely-arriving data to generate a version of the current round of typhoon status map, while marking which areas have been closed and which areas may still be affected by subsequent data. If a result package from a coastal station arrives late after the current round has ended, the system will not directly overwrite the current status map. Instead, it will first determine which unclosed areas the late data will mainly affect. Only the parts that meet the conditions will be partially written, while the parts that do not meet the conditions will be left to participate in the generation in the next round. In this way, in actual release, it is possible to update the typhoon status frequently while ensuring that the already stable areas do not change back and forth, making the entire typhoon status map both continuous and reliable.

[0027] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. 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.

Claims

1. A method for generating typhoon real-time data based on a high-frequency update mechanism, characterized in that, include: S1. Edge computing nodes collect local typhoon observation data within their respective areas, perform edge real-time generation processing on the local typhoon observation data, and associate and encapsulate the corresponding collection time period, edge generation time period, and impact time period with the edge real-time generation processing results to obtain an edge result package. S2. The central processing node receives the edge result packets and central observation data that arrive on time within the target release cycle. It performs current round real-time fusion generation on the timely edge result packets and central observation data to obtain the current round real-time results. Based on the data cutoff time and the time period coverage of the data included in the current round real-time results, it determines the closed boundary and then divides the current round real-time results into closed and unclosed areas. S3. The central processing node receives late edge result packets that do not arrive on time within the target release period, compares the impact period of each late edge result packet with the closed boundary, and determines the corresponding impact area of ​​each late edge result packet with the closed area and the unclosed area to obtain the rewriting qualification mark and allowed rewriting range of the late edge result packet. S4. The central processing node performs a local write-back update on the current round's live results only within the corresponding allowed rewrite range for late edge result packets that are marked as allowing rewrite eligibility, in order to obtain the current round's published live results. S5. The central processing node writes the late edge result package, which is marked as prohibited from being rewritten, into the next round candidate input set according to its collection time period, edge generation time period and impact time period, and adds a cross-round inheritance identifier to the late edge result package to obtain the next round inheritance input data. S6. The central processing node outputs the current round of real-time results and calls the next round of inherited input data in the next release cycle to participate in the next round of real-time fusion generation, so that late-arriving edge observation data can be continuously incorporated into the subsequent typhoon real-time generation process without rewriting the closed area.

2. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 1, characterized in that: S1 includes: S1-1. Obtain the local typhoon observation data output by each observation source, extract the observation source identifier, collection timestamp, spatial location identifier and observation value, map the collection timestamp to a unified edge time base, and arrange them in time order and spatial location order to obtain the local observation sequence. S1-2. Perform linear interpolation on the missing observations in the local observation sequence before and after the time interval, and take the arithmetic mean of multiple observations at the same time and spatial location to generate the edge real-time generation processing result. At the same time, determine the earliest and latest acquisition timestamps corresponding to the edge real-time generation processing result as the acquisition period, and determine the generation timestamp of the edge real-time generation processing result as the edge generation period. S1-3. Calculate the duration of the data collection period, and determine the period of influence by taking the start time of the edge generation period as the start time of the influence and the time obtained by adding the duration to the start time of the influence as the end time of the influence. Then, associate and encapsulate the edge real-time generation processing result, the data collection period, the edge generation period, and the period of influence to obtain the edge result package.

3. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 2, characterized in that: In S2, the generation of the current round's live results and the process of dividing the closed and open regions include: S2-1. Obtain the edge result packages and central side observation data that arrive on time within the target release period, extract the edge real-time generation and processing results, collection time period, edge generation time period and impact time period corresponding to each edge result package, and map the central side observation data to the target release period according to the collection timestamp to obtain the fusion input set for this round. S2-2. Map the edge real-time generation and processing results of each edge in the current round of fusion input set to a unified real-time grid according to their spatial location. Then, overlay the mapped edge real-time generation and processing results with the central observation data in the same real-time grid according to the order of collection time to generate the current round of real-time results. At the same time, determine the latest collection timestamp in the current round of fusion input set as the data cutoff time. S2-3. Calculate the end time of the influence of the input data corresponding to each real-time grid in the current round of real-time results. Determine the real-time grids whose influence end time is earlier than or equal to the data cutoff time as closed regions, and determine the real-time grids whose influence end time is later than the data cutoff time as unclosed regions.

4. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 3, characterized in that: S3 includes: S3-1. Map the affected area in the late edge result package to the unified real-time grid corresponding to the current round of real-time results, write the affected time period into the corresponding real-time grid node according to the grid, and construct the late rewriting spatiotemporal graph according to the adjacency relationship of the real-time grid to obtain the candidate rewriting node set. S3-2. For each real-world grid node in the candidate rewrite node set, calculate the node rewrite confidence based on the temporal relationship between its influence period and the closed boundary. Real-world grid nodes whose influence ends before the closed boundary are marked as closed conflict nodes, real-world grid nodes whose influence begins after the closed boundary are marked as directly rewriteable nodes, and real-world grid nodes whose influence period crosses the closed boundary are marked as nodes to be solved, thus obtaining the node state set. S3-3. Using the maximum sum of rewritten confidence of retained nodes, the minimum number of piercing edges of closed conflict nodes, and the retention of the node set and the unclosed region as joint constraints, perform a connected subgraph search on the node state set, remove the real grid nodes that form piercing paths with the closed region, and retain the real grid nodes that are connected to the unclosed region to obtain the initial rewritten subgraph.

5. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 4, characterized in that: S3 further includes: S3-4. Using the initial rewritten subgraph as input, iteratively perform boundary consistency verification and node state backhaul update. In each iteration, delete the real grid nodes that cause new conflicts in the boundary relationship of the current round's real results and recalculate the rewrite confidence of the remaining real grid nodes until the set of retained nodes obtained from two consecutive iterations is consistent, thus obtaining the allowable rewrite range. S3-5. Determine late edge result packets with an empty allowed rewrite range as prohibited from rewriting, and determine late edge result packets with a non-empty allowed rewrite range as allowed to rewrite, and output the corresponding rewrite qualification flag and allowed rewrite range.

6. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 5, characterized in that: S4 includes: S4-1. Extract the edge real-world generation and processing results from the late edge result package that is allowed to be rewritten, and map the edge real-world generation and processing results and the current round real-world results to the allowed rewrite range according to a unified real-world grid, so as to obtain the late grid value and the current grid value corresponding to each real-world grid within the allowed rewrite range. S4-2. Within the allowed rewriting range, the real grids that share a common edge with the real grids outside the allowed rewriting range are identified as boundary grids. Starting from the boundary grids, the rewriting range is expanded layer by layer. The level number corresponding to each real grid is determined according to the shortest grid step from each real grid to the nearest boundary grid. S4-3. Using the maximum level number within the allowed rewrite range as the normalization base, determine the write weight of the late grid value for each real grid according to the ratio of its level number to the maximum level number, and calculate the updated grid value according to the write weight and the remaining weight of the corresponding current grid value. Boundary grids with a level number of zero retain the current grid value, and internal grids with a level number equal to the maximum level number write the late grid value. S4-4. Write back the updated grid values ​​corresponding to each live grid to the live grids corresponding to the current round's live results within the allowed rewrite range, and keep the values ​​of each live grid outside the allowed rewrite range unchanged, to obtain the current round's published live results.

7. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 6, characterized in that: S5 includes: S5-1. Extract the edge real-time generation processing result, collection period, edge generation period and impact period from the late edge result package characterized by the rewriting qualification mark as prohibiting rewriting. Generate a temporal inheritance index according to the start time of the collection period, the end time of the collection period, the edge generation period and the end time of the impact period to obtain candidate inheritance entries. S5-2. Write the candidate inheritance entries into the candidate input set for the next round, and associate the temporal inheritance index with the corresponding edge real-time generation processing result. Add cross-round inheritance identifiers to candidate inheritance entries whose collection period ends later than the current round's data cutoff time and whose influence period ends later than the current round's data cutoff time, to obtain the next round's inheritance input data. S5-3. Arrange the next round of inheritance input data in sequence according to the time-sequence inheritance index, and output the next round of inheritance input data with cross-round inheritance identifier to the next release cycle call.

8. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 7, characterized in that: S6 includes: S6-1. Output the current round of release results, and extract the data cutoff time, closed area and unclosed area corresponding to the current round of release results. Write the data cutoff time, closed area and unclosed area into the fusion control information of the next release cycle to obtain the fusion constraint data of the next round. S6-2. Obtain the next round of inheritance input data, extract the edge real-time generation processing results, collection time period, edge generation time period, impact time period and cross-round inheritance identifier corresponding to each next round of inheritance input data, compare the impact time period with the data cutoff time in the next round of fusion constraint data, and map the impact area with the closed area and the unclosed area to obtain the next round of inheritance candidate set; S6-3. Merge the next round of inherited candidate sets with the newly arrived edge result packages and center-side observation data in the next release cycle, and use them as input data for the next round of real-time fusion generation, so that the late-arriving edge observation data can be continuously incorporated into the subsequent typhoon real-time generation process without rewriting the closed area.

9. The typhoon real-time data fusion generation method based on a high time-frequency update mechanism according to claim 8, characterized in that: In S6-2, the process of generating the next round of inheritance candidate sets includes: S6-2-1. Obtain the next round of inheritance input data, extract the edge real-world generation processing results, collection time period, edge generation time period, influence time period, influence area and cross-round inheritance identifier corresponding to each next round of inheritance input data, and map the influence area to the unified real-world grid corresponding to the next round of fusion constraint data to obtain the inheritance mapping result; S6-2-2. Compare the end time of the influence period of each next round of inheritance input data in the inheritance mapping result with the data cutoff time. For the next round of inheritance input data whose end time of influence period is later than the data cutoff time, retain the cross-round inheritance identifier. For the next round of inheritance input data whose end time of influence period is earlier than or equal to the data cutoff time, delete the cross-round inheritance identifier to obtain the time series filtering result. S6-2-3. Perform grid correspondence statistics on the affected area corresponding to the next round inheritance input data with the cross-round inheritance identifier retained in the time series filtering results, and the closed and unclosed areas. Extract the next round inheritance input data in which the number of actual grids falling into the unclosed area is greater than the number of actual grids falling into the closed area, and obtain the next round inheritance candidate set.

10. A typhoon real-time data fusion generation system based on a high-frequency update mechanism, characterized in that, include: The encapsulation and generation module collects local typhoon observation data within its area through edge computing nodes, performs edge real-time generation processing on the local typhoon observation data, and associates and encapsulates the corresponding collection time period, edge generation time period, and impact time period with the edge real-time generation processing results to obtain an edge result package. The fusion closure module receives edge result packets and central observation data that arrive on time within the target release cycle through the central processing node. It performs current round of real-time fusion generation on the timely edge result packets and central observation data to obtain the current round of real-time results. Based on the data cutoff time corresponding to the current round of real-time results and the time period coverage of the data already included, it determines the closure boundary and then divides the current round of real-time results into closed and unclosed areas. The qualification determination module receives late edge result packets that do not arrive on time within the target release cycle through the central processing node. It compares the impact period of each late edge result packet with the closed boundary, and determines the corresponding impact area of ​​each late edge result packet with the closed area and the unclosed area to obtain the rewriting qualification mark and allowed rewriting range of the late edge result packet. The local write-back module, through the central processing node, identifies late-arriving edge result packets that are marked as allowing rewrite eligibility. It then performs local write-back updates on the current round's live results only within the corresponding allowed rewrite range to obtain the current round's published live results. The cross-round inheritance module, through the central processing node, writes the late edge result package, which is characterized by the rewriting qualification mark as prohibited from rewriting, into the next round candidate input set according to its collection time period, edge generation time period, and impact time period, and adds a cross-round inheritance identifier to the late edge result package to obtain the next round inheritance input data; The continuous inclusion module outputs the current round of real-time data release results through the central processing node, and calls the next round of inherited input data in the next release cycle to participate in the next round of real-time data fusion generation, so that late-arriving edge observation data can be continuously included in the subsequent typhoon real-time data generation process without rewriting the closed area.

Images