A numerical prediction data efficient processing and distribution system

Abstract

The application relates to the field of data processing and discloses a numerical prediction data efficient processing and distribution system, which comprises a state monitoring module, a mode decision module and a mode execution module. The state monitoring module acquires the number of concurrent requests, the bandwidth occupancy rate and the user terminal task critical stage of the system in real time; the mode decision module instructs the terminal to adopt a first mode of active pulling when the load is low according to the dynamic change of the load; when the load continuously grows to the high position and the terminal is in the task critical stage, the mode decision module instructs the server end to switch to a second mode of active pushing and suppresses the pulling request of the terminal which is not completed; the mode execution module monitors the data receiving state in the second mode, generates a correction instruction when delay or packet loss occurs, reduces the pushing frequency and allows the terminal to initiate a limited active pulling request to compensate for the missing data. Through the adaptive mode switching and dynamic correction mechanism, the application realizes efficient and reliable distribution of numerical prediction data.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically to a system for efficient processing and distribution of numerical weather prediction data. Background Technology

[0002] Numerical forecast data (such as aviation meteorological forecast data) is characterized by large data volume, high update frequency, and strict timeliness requirements. In the aviation operation environment, user terminals include aircraft in the air, control towers, approach control centers, and airline operations control centers. These terminals have different and stringent requirements for the real-time performance and completeness of forecast data at different mission stages (such as takeoff, approach, landing, and cruise).

[0003] Currently, numerical weather prediction data distribution systems typically employ either a client-side active retrieval mode or a server-side active push mode. When system load is low, the active retrieval mode effectively meets users' on-demand data needs, offering flexible resource utilization. However, under high system load, a large number of concurrent retrieval requests can cause server load spikes and communication link congestion, potentially leading to delayed arrival or even loss of critical forecast data. Especially when multiple aircraft are simultaneously in critical phases such as takeoff or landing, or when the terminal area encounters severe weather, ground data centers face immense pressure from concurrent requests, severely impacting flight safety and operational efficiency.

[0004] On the other hand, while a simple server-driven push model can alleviate request congestion to some extent, it may lead to unnecessary bandwidth waste under low load and cannot adapt to the dynamic data acquisition needs of terminals at different mission stages. Furthermore, existing systems generally lack real-time awareness of critical phases of aircraft missions, making it difficult to prioritize data delivery during high-risk phases when resources are scarce. When delays or packet loss occur during the push process, there is also a lack of effective compensation and correction mechanisms, making it difficult to guarantee the integrity and timeliness of data distribution. Summary of the Invention

[0005] The purpose of this invention is to provide an efficient numerical weather prediction data processing and distribution system that solves the technical problem of how to adaptively switch between pull and push modes based on the dynamic load changes of the airborne data distribution system and the critical stages of the user terminal's mission, and how to effectively correct for transmission anomalies.

[0006] The objective of this invention can be achieved through the following technical solutions: A system for efficient processing and distribution of numerical weather prediction data, comprising: The status monitoring module is used to obtain the number of concurrent requests, communication link bandwidth utilization, and the current key stages and location information of user terminals in real time. The mode decision module is used to determine whether the system load is in a low or high operating range based on the dynamic changes in the number of concurrent requests and bandwidth utilization. When the system load is in the low operating range, the user terminal is instructed to adopt the first mode, that is, allow the user terminal to actively pull numerical forecast data as needed; When the system load shows a continuous upward trend and reaches a high operating range, and the user terminal is in a preset critical mission phase, the command server adopts a second mode. In this mode, the server actively pushes critical numerical forecast data to the corresponding user terminal according to priority, and at the same time generates a mode switching flag. This flag is used to suppress pull requests initiated but not yet completed by the user terminal in the first mode. The priority is based on the user terminal's mission critical phase from high to low, and critical numerical forecast data is pushed to the user terminal with the earlier critical phase. The order of the mission critical phases is predefined as follows: takeoff and landing phases are the highest priority, approach phase is the second highest priority, and cruise phase is the normal priority. The mode execution module is used to monitor the actual data reception status and current concurrency environment parameters of the user terminal after the second mode is activated. When it detects that the pushed data has a delay or packet loss that exceeds the preset tolerance range, it generates a correction instruction to temporarily reduce the push frequency on the server side and allow the user terminal to initiate a limited number of active pull requests at a limited pull rate to compensate for the missing data. At the same time, the correction instruction acts in reverse on the mode decision module to dynamically adjust the sensitivity of subsequent mode switching.

[0007] As a further technical solution, the system also includes: The cache consistency verification module is used for: After the data is actively pushed to the user terminal in the second mode, the consistency between the forecast data version in the local cache of the terminal and the latest version on the server is verified. If the versions are the same, suppress any active pull actions from the terminal in the first mode; If the versions are inconsistent, an incremental fetch request will be triggered. The processing method of this incremental fetch request depends on the currently active mode: If the current mode is 1, the response will follow the normal active fetch process. If the current mode is the second mode, the server will treat the incremental fetch request as an immediately executed supplementary push task. When an incremental fetch request fails due to communication link congestion, the cache consistency verification module sends a mode maintenance request to the mode decision module, forcing the second mode to be maintained for at least one preset time window.

[0008] As a further technical solution, the system also includes: The mode pre-switching module is used for: During the first mode operation, the response time of the user terminal's pull request and the load trend of the server are calculated in real time. When the response time deviates from the historical average and the load trend shows a continuous upward curve, if the triggering conditions of the second mode are not met, the pre-activation of the second mode is triggered. The pre-set key forecast data is sent to the user terminal cache as background task at a lower sending rate and queue order than the regular push data, and marked as pre-push data. When the actual environmental conditions reach the triggering conditions of the second mode, the mode decision module will seamlessly switch from pre-activation to full activation, and the cached pre-push data will be used directly as valid data. Meanwhile, the pre-push data serves as a correction benchmark: when the timestamp of the data retrieved by the user terminal in the first mode is later than the timestamp of the pre-push data, the retrieved lagging data is discarded, and the terminal display is updated based on the pre-push data. When the pre-push data itself is missing or erroneous, the mode pre-switching module generates a rollback instruction, temporarily disabling the second mode pre-activation function and forcibly restoring to the pure first mode until the server load returns to normal.

[0009] As a further technical solution, the mode pre-switching module is also used to perform the following process during the second mode pre-activation period: The pre-push data is divided into multiple consecutive time segments. After each segment is sent to multiple user terminals, the receipt quality vector returned by each terminal is collected. This vector contains data integrity identifier and timestamp deviation value. For the same time segment, if more than half of the receipt quality vectors indicate that the data is complete and the timestamp deviation is less than the historical average deviation, then the segment is marked as reliable; otherwise, it is marked as questionable. For questionable time segments, perform an incremental re-push loop: Repeat the push of the segment at the initial repush interval, double the interval after each repush, and collect receipts again; If more than half of the credible judgments are obtained after two consecutive re-pushings, the re-pushing will be terminated and the fragment will be adopted. If a credible determination cannot be obtained after three retries, a confidence rejection flag is generated, the rollback instruction is executed, and any pre-activation attempts in the subsequent pre-activation cycle are suppressed.

[0010] As a further technical solution, the mode pre-switching module also includes: Obtain the reliable proportion R of all time segments within each pre-activation cycle, and whether the system actually switches to full second mode after the end of the cycle. Calculate the data latency improvement rate ΔD and bandwidth usage reduction rate ΔB after the switch. Obtain the comprehensive benefit index E, expressed as: E=(ΔD+ΔB) / 2, and maintain a sliding window to store (R,E) pairs from the most recent pre-activation cycles; If the current period's R is lower than the average R within the sliding window, but E is higher than the average E within the sliding window, it is determined to be overly conservative, and the trigger threshold for pre-activation is lowered. If the current period's R is higher than the average R within the sliding window, but E is lower than the average E within the sliding window, it is determined to be excessive waste, and the trigger threshold for pre-activation is increased. If both R and E in the current period are higher than the mean or lower than the mean, the pre-activation trigger threshold remains unchanged.

[0011] As a further technical solution, the dynamic change characteristic is characterized by a switching tendency value P(t), the expression of which is: P(t)=max[-L,min(L,(1-λ)*P(t-1)+Δ(t))]; The single-step increment Δ(t) is obtained as follows: First, calculate the ratio of the current number of concurrent requests to the historical baseline concurrency, then subtract the ratio of the historical baseline bandwidth utilization to the current bandwidth utilization, and then multiply the difference by the dynamic adjustment factor. In the formula, λ is the attenuation coefficient, L is the amplitude limit of the cumulative value, and P(t-1) is the switching tendency value of the previous sampling period.

[0012] As a further technical solution, the process of determining whether the system load is in a low-level or high-level operating range is as follows: When P(t) is continuously positive and its value increases, it is determined that the system load shows a continuous growth trend and is approaching the high-level operating range. When P(t) is continuously negative and its value decreases, it is determined that the system load is in the low operating range. The specific switching execution method is as follows: When P(t) first exceeds +θ, a switch from the first mode to the second mode is executed; When P(t) first falls below -θ, a switch from the second mode to the first mode is executed; after the switch, P(t) is cleared to zero; where θ is the execution threshold.

[0013] As a further technical solution, the process of adjusting the mode switching sensitivity based on the correction command is as follows: Maintain a fixed-length M-length correction instruction time series buffer to store the sequence of received time intervals of the most recent M correction instructions and the corresponding delay overscalar sequence; Calculate the coefficient of variation of the time interval series, which is the ratio of the standard deviation to the mean; calculate the rate of change of the first difference sign of the delayed superscalar series, which is the proportion of the number of times the adjacent difference directions change out of the total number of comparisons. If the coefficient of variation is greater than the first dynamic threshold and the first-order difference sign rate of change is greater than the basic rate of change, it is determined to be a transient disturbance. If the coefficient of variation is less than the second dynamic threshold and the slope of the linear fit of the delayed superscalar sequence is positive, it is determined to be a persistent congestion type. Based on the judgment results, the execution threshold θ and the attenuation coefficient λ are adjusted as follows: When a transient disturbance is identified: the increase in the attenuation coefficient is equal to the base adjustment step size coefficient multiplied by the ratio of the current delay overshoot to the historical reference delay, and the increased attenuation coefficient is limited to a preset upper limit boundary; The increase in the execution threshold is equal to one-tenth of the increase in the attenuation coefficient; When determined to be persistent congestion: The increase in the execution threshold is equal to the base adjustment step size multiplied by the ratio of the current delay overshoot to the historical baseline delay, and the increased execution threshold is limited to the preset upper limit boundary; The increase in the attenuation coefficient is equal to one-tenth of the increase in the execution threshold.

[0014] As a further technical solution, the process of predicting future trends based on the rate of change of the delay overscalar in the correction instruction is as follows: Calculate the difference between the current delay overscalar and the delay overscalar of the previous correction instruction, and then divide it by the time interval between the two to obtain the rate of change of the delay overscalar. If the rate of change is positive, an additional lead step is added to the execution threshold or decay coefficient. This lead step is equal to the preset prediction coefficient multiplied by the rate of change. If the rate of change is negative, the increase will be halved before execution.

[0015] The beneficial effects of this invention are: (1) This invention uses a status monitoring module to obtain the number of concurrent requests, bandwidth utilization, and key stages of terminal tasks in real time. The mode decision module uses a switching tendency value with memory to comprehensively judge the load trend. When the load continues to increase and the terminal is in a high-risk stage such as takeoff or landing, the mode pre-switching module is triggered in advance to pre-push key data with low-priority background tasks. After the conditions are met, it seamlessly switches to the fully active push mode. This linkage mechanism avoids the switching breakpoint of pulling and then pushing in the traditional solution, so that the aircraft can always obtain the latest weather forecast data first during approach or landing, effectively ensuring flight safety. (2) When delays or packet loss occur in the second mode, the mode execution module generates correction instructions. On the one hand, it temporarily reduces the push frequency and opens limited pull to compensate for missing data. On the other hand, it acts in reverse on the mode decision module to dynamically adjust the switching threshold and attenuation coefficient. At the same time, the cache consistency verification module triggers incremental pull when the versions are inconsistent. If it fails due to congestion, it forces the maintenance of the second mode to avoid frequent switching back. The three work together to form a feedback chain of anomaly detection, compensation transmission, parameter adjustment and mode stability, so that the system can maintain the continuity of data distribution under transient disturbances or continuous congestion. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a system logic framework diagram of the present invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 As shown, this invention is an efficient numerical weather prediction data processing and distribution system, comprising: The status monitoring module is used to obtain the number of concurrent requests, communication link bandwidth utilization, and the current key stages and location information of user terminals in real time. The mode decision module is used to determine whether the system load is in a low or high operating range based on the dynamic changes in the number of concurrent requests and bandwidth utilization. When the system load is in the low operating range, the user terminal is instructed to adopt the first mode, that is, allow the user terminal to actively pull numerical forecast data as needed; When the system load shows a continuous growth trend and reaches a high operating range, and the user terminal is in the preset critical stage of the task, the instruction server adopts the second mode, that is, the server actively pushes the key numerical forecast data to the corresponding user terminal according to the priority, and at the same time generates a mode switching flag. This flag is used to suppress the pull request that the user terminal has initiated but has not yet completed in the first mode. The mode execution module monitors the actual data reception status and current concurrency environment parameters of the user terminal after the second mode is activated. When it detects that the pushed data has a delay or packet loss exceeding the preset tolerance range, it generates a correction instruction to temporarily reduce the push frequency on the server side and allow the user terminal to initiate a limited number of active pull requests at a limited pull rate to compensate for missing data. The limited pull rate and the limited number of requests are dynamically calculated by the server side based on the current bandwidth utilization rate and do not depend on preset fixed values. At the same time, the correction instruction acts inversely on the mode decision module to dynamically adjust the sensitivity of subsequent mode switching. For example, the dynamic calculation rules are as follows: limited pull rate = current remaining available bandwidth × 0.2 (fixed ratio, 80% bandwidth reserved for core push); limited number of requests = number of currently online key terminals × 1 (single terminal, single compensation pull to avoid secondary congestion); the calculation trigger is: when the bandwidth utilization rate is >80%, rate limiting is initiated; when it is ≤60%, the default pull permission is restored.

[0020] In the process of distributing aeronautical numerical forecast data, a single data retrieval mode can lead to server overload and link congestion when the system load is high and user terminals (such as aircraft and control towers) are in critical stages such as takeoff and landing. This can cause delays or loss of critical forecast data and threaten flight safety. On the other hand, a single push mode can waste bandwidth resources when the load is low. Therefore, this solution uses a status monitoring module to obtain real-time data on the number of concurrent requests, bandwidth utilization, and critical stages of terminal tasks. The mode decision module dynamically determines whether the load is low or high: in low load conditions, it uses a first mode where the terminal actively pulls data; in high load conditions, when the terminal is in a critical stage, it switches to a second mode where the server actively pushes data and suppresses incomplete pull requests. The mode execution module monitors data reception quality in the second mode. When delays or packet loss exceed the tolerance range, it generates a correction command, reduces the push frequency, and allows the terminal to initiate a limited number of active pulls at a limited rate to compensate for missing data. This adaptively switches between pull and push modes, balancing flexibility under low load and data timeliness during critical high-load stages. Furthermore, it suppresses incomplete pull requests through mode switching flags to avoid redundant transmission and resource waste. A correction mechanism is also introduced to compensate for missing data through limited pulls when push anomalies occur, improving transmission reliability. This solution is particularly suitable for aviation scenarios, prioritizing data push during high-risk stages such as takeoff and landing, thus enhancing flight safety.

[0021] The system also includes: The cache consistency verification module is used for: After the data is actively pushed to the user terminal in the second mode, the consistency between the forecast data version in the local cache of the terminal and the latest version on the server is verified. If the versions are the same, suppress any active pull actions from the terminal in the first mode; If the versions are inconsistent, an incremental fetch request will be triggered. The processing method of this incremental fetch request depends on the currently active mode: If the current mode is 1, the response will follow the normal active fetch process. If the current mode is the second mode, the server will treat the incremental pull request as an immediately executed supplementary push task, and the task will be sent before all regular push data in the current server's queue to be sent. When an incremental fetch request fails due to communication link congestion, the cache consistency verification module sends a mode maintenance request to the mode decision module, forcing the system to maintain the second mode for at least one preset time window to avoid data oscillations caused by frequent switchbacks. The default preset time window is 30 seconds, set according to the minimum instruction cycle of air traffic control; the dynamic adjustment rule is that when the link packet loss rate is >5%, the time window is automatically extended to 60 seconds; when the packet loss rate is <1%, it is shortened to 15 seconds.

[0022] Because in the second mode (active push), the user terminal's local cache may be inconsistent with the server version due to historical data residue or push loss, causing the terminal to use outdated forecast data. In addition, frequent back-switchovers may be triggered by link congestion during mode switching, causing system instability. Therefore, this solution adds a cache consistency verification module: after the push arrives, it verifies the consistency between the terminal cache and the server version. If they are consistent, it inhibits the terminal's active pull in the first mode. If they are inconsistent, it triggers an incremental pull request, which is handled according to the current mode (normal pull in the first mode, and a supplementary push task to be executed immediately in the second mode). If the incremental pull fails due to congestion, it is forced to remain in the second mode for at least a preset time window. This ensures that the terminal always uses the latest version of forecast data, avoids critical decision errors caused by cache inconsistency, and can incrementally pull only the differences, reducing bandwidth consumption. At the same time, incremental pull is promoted to a high-priority task in the second mode to ensure the timeliness of version synchronization. By forcibly maintaining the second mode, it prevents frequent back-switchovers and improves system stability, which is especially suitable for the data continuity requirements of critical phases of aviation missions.

[0023] The system also includes: The mode pre-switching module is used for: During the first mode operation, the response time of the user terminal's pull request and the load trend of the server are calculated in real time. When the response time deviates from the historical average and the load trend shows a continuous upward curve, if the triggering conditions of the second mode are not met, the pre-activation of the second mode is triggered. The pre-set key forecast data is sent to the user terminal cache as background task at a lower sending rate and queue order than the regular push data, and marked as pre-push data. When the actual environmental conditions reach the triggering conditions of the second mode, the mode decision module will seamlessly switch from pre-activation to full activation, and the cached pre-push data will be used directly as valid data to avoid repeated transmission. Meanwhile, the pre-push data serves as a correction benchmark: when the timestamp of the data retrieved by the user terminal in the first mode is later than the timestamp of the pre-push data, the retrieved lagging data is discarded, and the terminal display is updated based on the pre-push data. When the pre-push data itself is missing or erroneous, the mode pre-switching module generates a rollback instruction, temporarily disabling the second mode pre-activation function and forcibly restoring to the pure first mode until the server load returns to normal.

[0024] Because there is a lag when switching from the first mode to the second mode: when the load rises rapidly but the switching conditions are not yet met, the terminal is still using the pull mode, which may have already resulted in response delays or data loss; direct switching may also lead to push failure due to sudden congestion. At the same time, the pre-pushed data itself may be inaccurate or missing, lacking a correction benchmark; therefore, this solution adds a mode pre-switching module: during the operation of the first mode, when the response time deviates from the historical average and the load continues to rise but the triggering conditions of the second mode are not met, pre-activation is triggered, and key forecast data is pre-sent to the terminal cache at a low rate as background tasks and marked as pre-pushed data; when the subsequent conditions are met, it seamlessly switches to full activation, with the pre-pushed data serving as a correction benchmark (discarding pull data with later timestamps). If the pre-pushed data itself is missing or incorrect, a rollback instruction is generated, temporarily disabling the pre-activation function and forcibly restoring to pure first mode; on the one hand, pre-activation achieves a smooth mode transition, avoiding data gaps or delays at the moment of switching, and on the other hand, the pre-pushed data, as a correction benchmark, can identify and discard lagging data, ensuring that the terminal displays the latest information. Automatic rollback when pre-push quality is poor avoids polluting the terminal cache due to pre-activation, making it particularly suitable for terminal areas with rapidly increasing loads in aviation scenarios, and preparing meteorological data for critical aircraft in advance.

[0025] The mode pre-switching module is also used to perform the following process during the second mode pre-activation period: The pre-push data is divided into multiple consecutive time segments. After each segment is sent to multiple user terminals, the receipt quality vector returned by each terminal is collected. This vector contains data integrity identifier and timestamp deviation value. For the same time segment, if more than half of the receipt quality vectors indicate that the data is complete and the timestamp deviation is less than the historical average deviation (dynamically calculated from the deviation of the most recent successfully pushed segments), then the segment is marked as trustworthy; otherwise, it is marked as questionable. The historical average deviation is obtained by taking the 10 most recent successfully pushed segments, taking the arithmetic mean of the 10 deviation values, and removing outliers that deviate from the mean by more than 3 times. For questionable time segments, perform an incremental re-push loop: Repeat the push of the segment at the initial repush interval, double the interval after each repush, and collect receipts again; If more than half of the credible judgments are obtained after two consecutive re-pushings, the re-pushing will be terminated and the fragment will be adopted. If a credible determination cannot be obtained after three retries, a confidence rejection flag is generated, the rollback instruction is executed, and any pre-activation attempts in the subsequent pre-activation cycle are suppressed.

[0026] The maximum number of retransmissions and the requirement for consecutive successes are dynamically set by the system based on the historical reliability of the communication link. Historical reliability quantification index: average packet loss rate of the link over the past 10 minutes; dynamic setting rules: packet loss rate ≤ 1%, maximum of 2 retransmissions and 1 consecutive success; 1% < packet loss rate ≤ 5%, maximum of 3 retransmissions and 2 consecutive successes; packet loss rate > 5%, maximum of 4 retransmissions and 2 consecutive successes.

[0027] During the pre-activation period, pushed data may experience packet loss or delays due to network instability, making the pre-pushed data unreliable. Simple global retransmission will increase bandwidth consumption and cannot distinguish which segments need to be retransmitted. Therefore, this scheme divides the pre-push data into continuous time segments. After sending each segment to multiple terminals, it collects the receipt quality vector (integrity identifier and timestamp deviation value). For the same segment, if more than half of the terminals provide complete receipts and the deviation is less than the historical average deviation, it is marked as trustworthy; otherwise, it is marked as questionable. For questionable segments, an incremental re-push loop is executed (initial interval, each re-push interval doubles). If more than half of the terminals provide trustworthy results after two consecutive re-pushes, the segment is adopted. If it is still untrustworthy after three re-pushes, a confidence rejection flag is generated, a rollback instruction is executed, and any pre-activation attempts in the subsequent pre-activation cycle are suppressed. This utilizes multi-terminal feedback for confidence determination, avoiding single-point misjudgment. An exponential backoff re-push strategy is adopted to avoid exacerbating network burden during congestion. A re-push limit and consecutive success conditions are set to prevent infinite retries, quickly determine untrustworthy segments, and trigger a global rollback. This improves the overall quality of the pre-push data and ensures the effectiveness of the pre-activation mechanism, making it particularly suitable for environments where multiple aircraft operate simultaneously in aviation.

[0028] The mode pre-switching module further includes: Obtain the confidence ratio R (number of confidence segments / total number of segments) of all time segments within each pre-activation period, and whether the system actually switches to full second mode after the end of the period. Calculate the data latency improvement rate ΔD (difference in latency before and after switching divided by latency before switching) and the bandwidth usage reduction rate ΔB after switching. Obtain the comprehensive benefit index E, expressed as: E=(ΔD+ΔB) / 2, and maintain a sliding window. The typical length of the sliding window is 8 pre-activation cycles. Store the (R,E) pairs of the most recent pre-activation cycles. After each new cycle is completed, remove the oldest data and add the new data, keeping the window length unchanged. If the current period's R is lower than the average R within the sliding window, but E is higher than the average E within the sliding window, it is judged as overly conservative, and the trigger threshold for pre-activation (i.e., the deviation ratio requirement between response time and historical average) is lowered so that subsequent pre-activation is triggered earlier. If the current period's R is higher than the average R within the sliding window, but E is lower than the average E within the sliding window, it is determined to be excessive waste, the trigger threshold for pre-activation is increased, and subsequent pre-activation is delayed. If both R and E in the current period are higher than the mean or lower than the mean, the pre-activation trigger threshold remains unchanged.

[0029] The above adjustments are iterated after each pre-activation cycle is completed, and the updates are continuously slidable without the need for preset convergence conditions.

[0030] If the pre-activation trigger threshold (such as the degree of response time deviation) remains fixed, it may be too conservative (missing the opportunity for early pre-activation) or too wasteful (frequent pre-activation occupies a lot of bandwidth) in some scenarios, failing to adapt to dynamically changing network and load conditions. Therefore, the reliable ratio R of each pre-activation cycle, as well as the data latency improvement rate ΔD and bandwidth usage reduction rate ΔB after switching, are obtained. A comprehensive benefit index E=(ΔD+ΔB) / 2 is defined, and a sliding window is maintained to store (R,E) pairs for the most recent few cycles. If R is lower than the window mean but E is higher than the window mean in the current cycle, it is judged as overly conservative, and the pre-activation trigger threshold is lowered. If R is higher than the mean but E is lower than the mean, it is judged as overly wasteful, and the trigger threshold is raised. In other cases, the threshold remains unchanged. This allows for dynamic adjustment of the pre-activation threshold based on historical benefit feedback, achieving self-optimization. It also distinguishes between overly conservative and overly wasteful deviations, adopting different adjustment directions for each, making it more precise. It does not require preset convergence conditions, continuously slides and updates, and adapts to environmental changes. In aviation scenarios, the pre-activation strategy can be automatically adjusted according to the traffic patterns of different airports and different time periods, improving the overall resource utilization rate.

[0031] The dynamic change characteristic is characterized by the switching tendency value P(t), which is expressed as follows: P(t)=max[-L,min(L,(1-λ)*P(t-1)+Δ(t))]; The single-step increment Δ(t) is obtained as follows: First, calculate the ratio of the current number of concurrent requests to the historical baseline concurrency, then subtract the ratio of the historical baseline bandwidth utilization to the current bandwidth utilization, and then multiply the difference by the dynamic adjustment factor. In the formula, λ is the attenuation coefficient, L is the amplitude limit of the cumulative value to prevent the absolute value of P(t) from being too large, which would lead to sluggish or excessive switching, and P(t-1) is the switching tendency value of the previous sampling period.

[0032] The expression for the single-step increment Δ(t) is: ; In the formula, Q(t) represents the current number of concurrent requests, and B(t) represents the current bandwidth utilization; Qb and Bb are historical statistical baseline values ​​(sliding window averages), which are dynamically updated over time; is an adjustment factor. =(1+F) / (1+R), where F is the confidence rejection flag, F=1 when the confidence rejection flag is triggered, and F=0 when it is not triggered, and R is the confidence ratio, that is, the number of confident fragments in the pre-activation period / the total number of fragments, with a value of 0 to 1.

[0033] System load depends not only on the number of concurrent requests but also on bandwidth utilization. A single metric (request count or bandwidth only) can easily lead to misjudgment: for example, if there are few concurrent requests but the bandwidth is already full for other services, it should still be considered a high load. The term in the formula reflects the relative change in the number of requests (the more requests, the smaller this term), minus... The term reflects the relative change in bandwidth usage (the higher the bandwidth usage, the larger the subtrahend), and the difference between the two can comprehensively express the direction of system pressure.

[0034] Directly using the instantaneous ratio difference will produce severe jitter. Introducing a first-order low-pass filter (1-λ)*P(t-1)+Δ(t) gives P(t) inertia, allowing historical trends to continue and attenuating short-term spikes. The larger λ is, the faster historical forgetting occurs, and the more sensitive the system response; the smaller λ is, the more persistent the historical influence, and the more stable the system.

[0035] Without limiting, P(t) could grow indefinitely due to continuous positive increments, causing the switching threshold +θ to never be exceeded (or requiring an extremely long time to fall back down). By clamping P(t) within [-L,+L] using max[-L,min(L,.)], the tendency value is always within a predictable range, making the setting of the switching threshold θ meaningful. Limiting also prevents numerical overflow in extreme cases.

[0036] Decrease as the number of requests increases. It increases with bandwidth usage. Subtracting the two: when the number of requests surges or bandwidth usage increases, the difference Δ(t) becomes negative, causing P(t) to shift in the negative direction, indicating that the system tends towards pull mode; when the number of requests decreases or bandwidth is released, the difference becomes positive, and P(t) shifts in the positive direction, tending towards push mode. The above construction method unifies the two competing factors into a single scalar.

[0037] Because the dynamic characteristics of system load (number of concurrent requests and bandwidth utilization) are subject to instantaneous fluctuations, directly using instantaneous values ​​for mode switching will lead to frequent erroneous switching. Therefore, this solution uses the formula: P(t)=max[-L,min(L,(1-λ)*P(t-1)+Δ(t))], implements the accumulation and decay mechanism to smooth instantaneous fluctuations and avoid false switching due to brief spikes; at the same time, it considers the relative changes of two dimensions, concurrent requests and bandwidth usage, which is more comprehensive than a single indicator; the amplitude limit boundary prevents the tendency value from being too large or too small, ensuring numerical stability.

[0038] The process of determining whether the system load is in the low or high operating range is as follows: When P(t) is continuously positive and its value increases, it is determined that the system load shows a continuous growth trend and is approaching the high-level operating range. When P(t) is continuously negative and its value is decreasing, it is determined that the system load is in the low operating range; continuous means 3 consecutive sampling periods, and the default sampling period is 1 second; the increment or decrement is determined when the difference between adjacent periods of P(t) is ≥0.1 (dimensionless threshold). The specific switching execution method is as follows: When P(t) first exceeds +θ, a switch from the first mode to the second mode is executed; When P(t) first falls below -θ, a switch from the second mode to the first mode is executed; after the switch, P(t) is cleared to zero; where θ is the execution threshold.

[0039] This scheme can use positive and negative signs and increasing or decreasing trends to make judgments, avoiding misjudgments caused by single threshold comparisons; it sets positive and negative bidirectional thresholds to form hysteresis loops to prevent frequent switching near the boundary; after switching, the tendency value is cleared, so that the next switching needs to be re-accumulated, further suppressing the ping-pong effect, which is suitable for the high dynamic environment of aviation terminal areas.

[0040] The process of adjusting mode switching sensitivity based on correction instructions is as follows: Maintain a fixed-length M-length correction instruction time series buffer to store the sequence of received time intervals of the most recent M correction instructions and the corresponding delay overscalar sequence; Calculate the coefficient of variation of the time interval series, which is the ratio of the standard deviation to the mean; calculate the rate of change of the first difference sign of the delayed superscalar series, which is the proportion of the number of times the adjacent difference directions change out of the total number of comparisons. If the coefficient of variation is greater than the first dynamic threshold and the first-order difference sign rate of change is greater than the basic rate of change (typically 0.5 for the basic rate of change), it is determined to be a transient disturbance. If the coefficient of variation is less than the second dynamic threshold and the slope of the linear fit of the delayed superscalar sequence is positive, it is determined to be a persistent congestion type. The typical value of the first dynamic threshold is 1.5 times the historical average of the coefficient of variation within the current sliding window, and the typical value of the second dynamic threshold is 0.5 times that historical average. Based on the judgment results, the execution threshold θ and the attenuation coefficient λ are adjusted as follows: When a transient disturbance is identified: the increase in the attenuation coefficient is equal to the base adjustment step size coefficient multiplied by the ratio of the current delay overshoot to the historical reference delay, and the increased attenuation coefficient is limited to a preset upper limit boundary; The increase in the execution threshold is equal to one-tenth of the increase in the attenuation coefficient, and the execution threshold is not set with an independent upper limit or is only set with a loose upper limit that is much higher than the normal operating value. That is, the adjustment range of the attenuation coefficient is much greater than the adjustment range of the execution threshold. Prioritize increasing the attenuation coefficient λ (to accelerate historical forgetting and enable the system to recover quickly from brief disturbances), while only slightly increasing or keeping the execution threshold θ unchanged, in order to avoid excessively increasing the switching threshold and causing sluggish response; When determined to be persistent congestion: The increase in the execution threshold is equal to the base adjustment step size multiplied by the ratio of the current delay overshoot to the historical baseline delay, and the increased execution threshold is limited to the preset upper limit boundary; The increase in the attenuation coefficient is equal to one-tenth of the increase in the execution threshold; That is, the adjustment range of the execution threshold is much greater than the adjustment range of the attenuation coefficient.

[0041] Prioritize increasing the execution threshold θ (increase the switching threshold to avoid frequent switching in congested environments), while moderately increasing λ, with the two working in tandem according to a preset proportional coefficient; This solution can automatically distinguish between transient disturbances and persistent congestion, and adopt different parameter adjustment strategies. During transient disturbances, the attenuation coefficient is increased first, and the system recovers quickly. During persistent congestion, the execution threshold is increased first to avoid repeated switching that aggravates the congestion. The adjustment range is proportional to the degree of delay exceeding the standard, achieving smooth adaptation. It is particularly suitable for mixed scenarios in aviation communications, such as brief interruptions caused by aircraft maneuvers and long-term high loads in the terminal area.

[0042] The process of predicting future trends based on the rate of change of the delay overscalar in the correction instruction is as follows: Calculate the difference between the current delay overscalar and the delay overscalar of the previous correction instruction, and then divide it by the time interval between the two to obtain the rate of change of the delay overscalar. If the rate of change is positive, an additional lead step is added to the execution threshold or decay coefficient. This lead step is equal to the preset prediction coefficient multiplied by the rate of change. If the rate of change is negative, the increase will be halved before execution.

[0043] All adjusted execution thresholds and attenuation coefficients remain within their respective preset upper and lower boundaries; When a preset number of consecutive correction commands (typically 5) are all determined to be transient disturbances, and the delay overshoot in each command is below the preset ignore threshold, the positive and negative thresholds and the attenuation coefficient are reset to the system's preset initial values. The ignore threshold is defined as a delay overshoot of <0.3 (the ratio of historical baseline delay); parameters are not triggered if the delay exceeds this value. The upper limit boundary is defined as an upper limit of 0.9 for the attenuation coefficient and an upper limit of 2.0 for the execution threshold (both dimensionless values).

[0044] This scheme introduces the rate of change as a leading signal, making parameter adjustments predictable and enabling early responses to congestion worsening. When the rate of change is positive, the leading step size is increased to accelerate the response; when the rate of change is negative, the increase is halved to avoid over-adjustment. Linked with the judgment and adjustment in claim 8, it further enhances the adaptive capability, enabling early detection of the accelerating upward trend of delays caused by multiple aircraft simultaneously entering critical phases, timely raising the switching threshold or the forgetting speed, and achieving an upgrade from passive response to proactive prediction. This reduces the problem that parameter adjustments are usually based on current or historical delay overshoots, lacking the ability to predict future trends, and the probability of lag when adjusting only based on the current value when delay overshoots are accelerating.

[0045] Example: Airport aviation meteorological numerical forecast data is distributed to terminals including the airport tower, approach control center, and three civil aviation flights: an A320 in the critical takeoff phase, a B737 in the approach phase, and a C919 in the cruise phase.

[0046] Basic parameter presets: Historical benchmarks: Concurrent request benchmark Qb = 100 requests / second, bandwidth utilization benchmark Bb = 40%; Algorithm parameters: attenuation coefficient λ=0.2, amplitude limiting boundary L=10, mode switching execution threshold θ=5. =1; Key thresholds: Data latency tolerance limit 200ms, packet loss tolerance limit 1%; Pre-activation trigger conditions: pull response time exceeds historical average by 1.5 times, load continues to rise; Priority: Takeoff phase > Approach phase > Cruise phase.

[0047] Phase 1: Low-load operation, first mode (terminal actively pulls). The status monitoring module collects data in real time: concurrent requests Q(t) = 80 times / second, bandwidth usage B(t) = 30%; terminal mission phases: A320 takeoff, B737 approach, C919 cruise.

[0048] The pattern decision module calculates the switching tendency value P(t): Single-step increment is: =1×(100 / 80-40 / 30)=-0.25; P(t)=max[-10,min(10,(1-0.2)×0+(-0.25))]=-0.25; If P(t) remains negative, the system load is determined to be low, and the instruction terminal adopts the first mode: the tower, control center, and flight terminal actively retrieve meteorological data as needed.

[0049] During operation, terminal data retrieval is smooth with no delay or congestion, and cached data is consistent with the server version.

[0050] Phase 2: Load increases, triggering pre-activation of the second mode; Status monitoring module data collected: concurrent requests increased to Q(t) = 140 times / second, bandwidth usage B(t) = 55%; terminal pull response time increased from 50ms to 90ms (1.5 times higher than the historical average).

[0051] The mode pre-switching module determines that the conditions for the formal triggering of the second mode have not been met, but the load continues to rise and the response times out, thus triggering the pre-activation of the second mode.

[0052] Key meteorological data (wind direction, wind speed, visibility) for takeoff or approach are pre-pushed to the terminal cache at a low rate and marked as pre-push data; The pre-push data is split into 1-minute segments, and after sending, the terminal receipt quality vector (integrity + timestamp deviation) is collected.

[0053] Pre-push data verification and re-push of the third data segment: Only 40% of terminals have complete receipts, and the timestamp deviation is 80ms (exceeding the historical average). This segment is marked as suspicious, and an incremental re-push loop is started. First push: 1 second interval, 50% reliability of receipt; Second push: 2-second interval, 75% reliability of response; Third push: 4-second interval, 85% reliability of response; if the criteria are met twice consecutively, the push is terminated and the segment is adopted.

[0054] The pre-activation benefit assessment showed a confidence level of 90% (R=90%), a latency improvement rate of ΔD=35% after the switchover, and a bandwidth utilization reduction rate of ΔB=20%. The overall benefit E = (35% + 20%) / 2 = 27.5%, which is higher than the average of the sliding window, keeping the pre-activation threshold unchanged.

[0055] Phase 3: High load and critical phase, switch to the second mode (server proactive push). The pattern decision module recalculates the switching tendency value P(t): Δ(t)=1×(100 / 160-40 / 65)=0.22; P(t)=max[-10,min(10,(1-0.2)×(-0.25)+0.22)]=0.02; Calculation of P(t) by continuous sampling: Sample 1: Q(t) = 160 times / second, B(t) = 65%; Δ(t) = 100 / 160 - 40 / 65 = +0.01; P(t) = 0.8 × 0.02 + 0.01 = 0.026; Sample 2: Q(t) = 170 times / second, B(t) = 68%; Δ(t) = 100 / 170 - 40 / 68 = +0.022; P(t) = 0.8 × 0.026 + 0.022 = 0.0428; Sample 3: Q(t) = 180 times / second, B(t) = 70%; Δ(t) = 100 / 180 - 40 / 70 = +0.031; P(t) = 0.8 × 0.0428 + 0.031 = 0.0652; Sample 4: Q(t) = 190 times / second, B(t) = 72%; Δ(t) = 100 / 190 - 40 / 72 = +0.045; P(t) = 0.8 × 0.0652 + 0.045 = 0.0972; Sample 5: Q(t) = 200 times / second, B(t) = 74%; Δ(t) = 100 / 200 - 40 / 74 = +0.062; P(t) = 0.8 × 0.0972 + 0.062 = 0.1398; Samples 6 to N (continuous high-level accumulation) show a continuously increasing load, with Δ(t) remaining positive and P(t) accumulating rapidly: 0.1398, 0.45, 1.2, 2.8, 4.2, 5.1, 6.0; After continuous sampling, P(t) rises to +6, exceeding θ=5 for the first time, satisfying the conditions of high load and the terminal being in the critical stage of takeoff or approach, and executing the switch from the first mode to the second mode: Generate mode switching flag to suppress unfinished fetch requests from the terminal; The server pushes data according to priority: first pushes key meteorological data to A320 (takeoff), then to B737 (approach), and finally to C919 (cruise).

[0056] In operation, the server centrally pushes data, and the terminal does not need to actively pull data. Concurrent requests drop sharply, and bandwidth usage falls back to 45%.

[0057] Phase 4: Push anomaly, the mode execution module triggers correction; The mode execution module detected that the A320 terminal had a data latency of 250ms (exceeding the tolerance limit) and a packet loss rate of 1.2%, and generated a correction instruction: the server reduced the push frequency, for example, from once / 10s to once / 15s; the terminal was allowed to initiate up to 3 active pulls at a limited rate to compensate for missing data; the correction instruction was fed back to the mode decision module to adjust the switching sensitivity.

[0058] Congestion type determination: Collect the most recent 10 correction instructions and calculate: the time interval variation coefficient is 0.3, which is less than the second dynamic threshold of 0.4, and the linear slope of the delay overscalar is positive, which is determined to be persistent congestion: the execution threshold θ is increased from 5 to 7; the attenuation coefficient λ is slightly increased to avoid frequent mode switching.

[0059] Phase 5: Data synchronization and cache consistency verification module; After the cache consistency verification module pushes data to the terminal, it verifies the local cache version against the server version: if the B737 terminal cache version lags behind by 1 time slice, an incremental fetch request is triggered; the system is in the second mode, and the server converts the incremental request to priority supplementary push, sending the missing data in the queue; if the supplementary push is successful, the versions are consistent, and the terminal is prevented from actively fetching.

[0060] Exception handling: If incremental push fails due to congestion, the module sends a hold request and forcibly maintains the second mode for 5 minutes to avoid data fluctuations.

[0061] Phase 6: Load decreases, switch back to first mode; Status monitoring module data collected: concurrent requests dropped to 90 times / second, bandwidth usage was 35%, and all terminals had exited the critical takeoff or approach phase.

[0062] The pattern decision module calculates P(t): Δ(t)=1×(100 / 90-40 / 35)=-0.15; P(t)=max[-10,min(10,(1-0.2)×6+(-0.15))]=-0.15; When the cumulative calculation of P(t) drops to -6, that is, when it first falls below -θ=-5, the second mode is switched back to the first mode and P(t) is cleared to zero.

[0063] In the final state, the system resumes the terminal-driven data retrieval mode, with stable load and efficient and reliable data distribution.

[0064] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A system for efficient processing and distribution of numerical weather prediction data, characterized in that, include: The status monitoring module is used to obtain the number of concurrent requests, communication link bandwidth utilization, and the current key stages and location information of user terminals in real time. The mode decision module determines whether the system load is in a low or high operating range based on the dynamic changes in the number of concurrent requests and bandwidth utilization. When the load is low, the user terminal is instructed to use the first mode, which allows the terminal to actively retrieve numerical forecast data. When the load shows a continuous upward trend and reaches a high level, and the terminal is in the preset critical stage of the task, the instruction server adopts the second mode, that is, the server actively pushes the key numerical forecast data to the corresponding user terminal according to the priority, and at the same time generates a mode switching flag. This flag is used to suppress the pull request that the user terminal has initiated but not completed in the first mode. The mode execution module is used to monitor the actual data reception status and current concurrency environment parameters of the user terminal after the second mode is activated. When the pushed data has a delay or packet loss that exceeds the preset tolerance range, a correction instruction is generated to reduce the push frequency of the server and allow the user terminal to initiate a limited number of active pull requests at a limited pull rate to compensate for missing data.

2. The efficient numerical weather prediction data processing and distribution system according to claim 1, characterized in that, The system also includes: The cache consistency verification module is used for: After the data is actively pushed to the user terminal in the second mode, the consistency between the forecast data version in the local cache of the terminal and the latest version on the server is verified. If the versions are the same, suppress any active pull actions from the terminal in the first mode; If the versions are inconsistent, an incremental fetch request will be triggered. The processing method of this incremental fetch request depends on the currently active mode: If the current mode is 1, the response will follow the normal active fetch process. If the current mode is the second mode, the server will treat the incremental fetch request as an immediately executed supplementary push task. When an incremental fetch request fails due to communication link congestion, the cache consistency verification module sends a mode maintenance request to the mode decision module, forcing the second mode to be maintained for at least one preset time window.

3. The efficient numerical weather prediction data processing and distribution system according to claim 2, characterized in that, The system also includes: The mode pre-switching module is used for: During the first mode operation, the response time of the user terminal's pull request and the load trend of the server are calculated in real time. When the response time deviates from the historical average and the load trend shows a continuous upward curve, if the triggering conditions of the second mode are not met, the pre-activation of the second mode is triggered. The pre-set key forecast data is sent to the user terminal cache as background task at a lower sending rate and queue order than the regular push data, and marked as pre-push data. When the actual environmental conditions reach the triggering conditions of the second mode, the mode decision module will seamlessly switch from pre-activation to full activation, and the cached pre-push data will be used directly as valid data. Meanwhile, the pre-push data serves as a correction benchmark: when the timestamp of the data retrieved by the user terminal in the first mode is later than the timestamp of the pre-push data, the retrieved lagging data is discarded, and the terminal display is updated based on the pre-push data. When the pre-push data itself is missing or erroneous, the mode pre-switching module generates a rollback instruction, temporarily disabling the second mode pre-activation function and forcibly restoring to the pure first mode until the server load returns to normal.

4. The efficient numerical weather prediction data processing and distribution system according to claim 3, characterized in that, The mode pre-switching module is also used to perform the following process during the second mode pre-activation period: The pre-push data is divided into multiple consecutive time segments. After each segment is sent to multiple user terminals, the receipt quality vector returned by each terminal is collected. This vector contains data integrity identifier and timestamp deviation value. For the same time segment, if more than half of the receipt quality vectors indicate that the data is complete and the timestamp deviation is less than the historical average deviation, then the segment is marked as reliable; otherwise, it is marked as questionable. For questionable time segments, perform an incremental re-push loop: Repeat the push of the segment at the initial repush interval, double the interval after each repush, and collect receipts again; If more than half of the credible judgments are obtained after two consecutive re-pushings, the re-pushing will be terminated and the fragment will be adopted. If a credible determination cannot be obtained after three retries, a confidence rejection flag is generated, the rollback instruction is executed, and any pre-activation attempts in the subsequent pre-activation cycle are suppressed.

5. The efficient numerical weather prediction data processing and distribution system according to claim 4, characterized in that, The mode pre-switching module further includes: Obtain the reliable proportion R of all time segments within each pre-activation cycle, and whether the system actually switches to full second mode after the end of the cycle. Calculate the data latency improvement rate ΔD and bandwidth usage reduction rate ΔB after the switch. Obtain the comprehensive benefit index E, expressed as: E=(ΔD+ΔB) / 2, and maintain a sliding window to store (R,E) pairs from the most recent pre-activation cycles; If the current period's R is lower than the average R within the sliding window, but E is higher than the average E within the sliding window, it is determined to be overly conservative, and the trigger threshold for pre-activation is lowered. If the current period's R is higher than the average R within the sliding window, but E is lower than the average E within the sliding window, it is determined to be excessive waste, and the trigger threshold for pre-activation is increased. If both R and E in the current period are higher than the mean or lower than the mean, the pre-activation trigger threshold remains unchanged.

6. The efficient numerical weather prediction data processing and distribution system according to claim 1 or 5, characterized in that, The dynamic change characteristic is characterized by the switching tendency value P(t), which is expressed as follows: P(t)=max[-L,min(L,(1-λ)*P(t-1)+Δ(t))]; The single-step increment Δ(t) is obtained as follows: First, calculate the ratio of the current number of concurrent requests to the historical baseline concurrency, then subtract the ratio of the historical baseline bandwidth utilization to the current bandwidth utilization, and then multiply the difference by the dynamic adjustment factor. In the formula, λ is the attenuation coefficient, L is the amplitude limit of the cumulative value, and P(t-1) is the switching tendency value of the previous sampling period.

7. The efficient numerical weather prediction data processing and distribution system according to claim 6, characterized in that, The process of determining whether the system load is in a low or high operating range is as follows: When P(t) is continuously positive and its value increases, it is determined that the system load shows a continuous growth trend and is approaching the high-level operating range. When P(t) is continuously negative and its value decreases, it is determined that the system load is in the low operating range. The specific switching execution method is as follows: When P(t) first exceeds +θ, a switch from the first mode to the second mode is executed; When P(t) first falls below -θ, a switch from the second mode to the first mode is executed; after the switch, P(t) is cleared to zero; where θ is the execution threshold.

8. The efficient numerical weather prediction data processing and distribution system according to claim 6, characterized in that, The process of adjusting mode switching sensitivity based on correction instructions is as follows: Maintain a fixed-length M-length correction instruction time series buffer to store the sequence of received time intervals of the most recent M correction instructions and the corresponding delay overscalar sequence; Calculate the coefficient of variation of the time interval series, which is the ratio of the standard deviation to the mean; calculate the rate of change of the first difference sign of the delayed superscalar series, which is the proportion of the number of times the adjacent difference directions change out of the total number of comparisons. If the coefficient of variation is greater than the first dynamic threshold and the first-order difference sign rate of change is greater than the basic rate of change, it is determined to be a transient disturbance. If the coefficient of variation is less than the second dynamic threshold and the slope of the linear fit of the delayed superscalar sequence is positive, it is determined to be a persistent congestion type. Based on the judgment results, the execution threshold θ and the attenuation coefficient λ are adjusted as follows: When a transient disturbance is identified: the increase in the attenuation coefficient is equal to the base adjustment step size coefficient multiplied by the ratio of the current delay overshoot to the historical reference delay, and the increased attenuation coefficient is limited to the preset upper limit boundary; the increase in the execution threshold is equal to one-tenth of the increase in the attenuation coefficient. When persistent congestion is detected: the increase in the execution threshold is equal to the base adjustment step size coefficient multiplied by the ratio of the current delay overshoot to the historical baseline delay, and the increased execution threshold is limited to the preset upper limit boundary; the increase in the attenuation coefficient is equal to one-tenth of the increase in the execution threshold.

9. The efficient numerical weather prediction data processing and distribution system according to claim 8, characterized in that, The process of predicting future trends based on the rate of change of the delay overscalar in the correction instruction is as follows: Calculate the difference between the current delay overscalar and the delay overscalar of the previous correction instruction, and then divide it by the time interval between the two to obtain the rate of change of the delay overscalar. If the rate of change is positive, an additional lead step is added to the execution threshold or decay coefficient. This lead step is equal to the preset prediction coefficient multiplied by the rate of change. If the rate of change is negative, the increase will be halved before execution.

Images