Methods for hourly forecasting of shallow lake water levels during periods of strong winds

By separating the effects of wind and pressure fields and combining water balance and forecast forcing effects, efficient, accurate and robust hourly forecasts of water levels in large shallow lakes have been achieved, solving the problems of insufficient timeliness and spatial variation characterization in existing technologies.

CN123069231APending Publication Date: 2026-11-10TAIHU BASIN ADMINISTRATION BUREAU OF HYDROLOGY (INFORMATION CENT)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611581252.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies struggle to balance computational timeliness and spatial variation characterization in water level forecasting for large, shallow lakes under extreme weather conditions, and their forecast robustness is also insufficient.

Method used

By separating the tilting and forcing effects of wind and pressure fields, the initial static water level is obtained. The water balance driving factors are combined for hourly recursion. The forecast wind field and pressure forcing effects are superimposed to synthesize the hourly forecast water level. An anomaly diagnosis module is used to monitor data quality and dynamically adjust the forecast process under extreme conditions.

Benefits of technology

This improved the station-level accuracy and computational timeliness of lake water level forecasts during periods of strong winds, ensuring forecast robustness and data reliability under extreme conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069231A_ABST
    Figure CN123069231A_ABST
Patent Text Reader

Abstract

This invention relates to the interdisciplinary field of hydrological forecasting and lake hydrodynamics, specifically disclosing a method for hourly forecasting of shallow lake water levels during periods of strong winds. The method includes: acquiring measured hydrological and meteorological data at the current moment; removing the tilting and forcing effects of the current wind and pressure fields; restoring the initial still water levels at representative stations and extracting the initial still water level representative for the entire lake; combining water balance driving factors under windless conditions for hourly recursion to obtain a future series of representative still water levels for the entire lake; subsequently acquiring forecast meteorological data and calculating the forecast wind field and pressure forcing effects at each future moment; and using the initial still water level as a baseline, superimposing the still water amplitude of the entire lake and the corresponding forecast forcing effects to synthesize the hourly forecast water level series for each representative station. This invention balances the physical consistency of water balance with the spatial differences in wind and pressure forcing, improving the accuracy and computational timeliness of station-level forecasts during periods of strong winds.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of hydrological forecasting and lake hydrodynamics, and more specifically, to a method for hourly forecasting of water levels in shallow lakes during periods of strong winds. Background Technology

[0002] Large, shallow lakes play a crucial role in watershed flood control, water resource allocation, and aquatic ecosystem maintenance. Under extreme weather conditions such as typhoons, strong wind stress and sudden changes in air pressure drive dramatic dynamic responses in lake water bodies, leading to significant water level tilting and rise. Accurately predicting the spatiotemporal evolution of lake water levels during periods of strong winds is of significant hydrodynamic importance for assessing the risk of overflowing dikes around lakes, optimizing water conservancy project scheduling strategies, and ensuring overall flood control safety in the watershed.

[0003] Currently, water level forecasting for large, shallow lakes typically employs zero-dimensional water balance models or two-dimensional hydrodynamic models. Zero-dimensional water balance models generalize the lake as a single node, using the direct averaging of measured water levels at each station as the initial field for calculations. This method is computationally fast but cannot output the spatially differentiated water levels at various stations around the lake. Two-dimensional hydrodynamic models solve shallow water equations through grid discretization, simulating complex wind-driven flow fields and spatial water level distributions. However, at high-resolution grids, the numerical iterative solution process consumes enormous computational resources, resulting in long forecast times, which is insufficient to meet the timeliness requirements of hourly, high-frequency rolling updates in flood control scheduling. Furthermore, under extreme high wind speed conditions, the nonlinear evolution of surface wind stress drag parameters often leads to numerical divergence or deviations from physical limits in conventional extrapolation calculations.

[0004] In summary, existing methods struggle to achieve a good balance between spatial resolution and real-time computation when dealing with shallow lake water level evolution under extreme weather forcing, and their robustness under complex forcing conditions still needs improvement. Therefore, it is necessary to investigate a water level forecasting method that can balance computational timeliness with the ability to characterize spatial differences, and possesses high robustness under extreme conditions. Summary of the Invention

[0005] In view of this, this application provides a method for hourly forecasting of water levels in shallow lakes during periods of strong winds, in order to solve the aforementioned problems existing in the prior art.

[0006] According to one aspect of this application, a method for hourly forecasting of water levels in shallow lakes during periods of strong winds includes:

[0007] Obtain the measured hydrological and meteorological data at the current moment;

[0008] Based on measured hydrological and meteorological data, the tilting and forcing effects of the current wind field and pressure field on the water level are removed, the initial static water level of each representative station is restored, and the initial static water level of the entire lake is extracted.

[0009] Obtain the driving factors of water balance under windless conditions;

[0010] Starting from the initial still water level of the entire lake, and combining the water balance driving factors, hourly water balance is recursively extrapolated to obtain the sequence of the representative still water level of the entire lake in the foreseeable future period.

[0011] Obtain forecast meteorological data for the foreseeable future;

[0012] Based on forecast meteorological data, the forecast wind field forcing effect and forecast pressure forcing effect are calculated for each representative station at each future time.

[0013] Using the initial still water level as a baseline, the change in the representative still water level sequence of the entire lake relative to the initial still water level of the entire lake is superimposed, and the corresponding forecast wind field forcing effect and forecast air pressure forcing effect are superimposed respectively to synthesize the hourly forecast water level sequence of each representative station.

[0014] Optionally, when obtaining forecast meteorological data for the future forecast period, if it falls within the typhoon's impact period, specifically including:

[0015] Obtain typhoon track and intensity forecast parameters for the forecast period;

[0016] Based on the typhoon parameterized wind field model, combined with the typhoon track and intensity forecast parameters and the spatial location of each representative station, the local gradient wind speed and local gradient wind direction at the location of each representative station are inverted and calculated.

[0017] By reducing the local gradient wind speed, independent forecast wind speed and forecast wind direction are obtained for each representative station, replacing the unified forecast wind field for the entire lake, and used to calculate the forecast wind-induced water level rise for each representative station.

[0018] Optionally, the comprehensive wind-water increase coefficient required for calculating wind-water increase is pre-calibrated and updated online in the following manner:

[0019] Quasi-calm wind samples and windy samples were selected from historical hydrological and meteorological samples. The arithmetic mean of the measured water levels at representative stations in the quasi-calm wind samples was calibrated as the true value of the calm water level.

[0020] For windy samples, the static water level is restored using the candidate wind-induced water level increase coefficient to obtain the corresponding restored static water level, and the restoration error between the restored static water level and the true static water level at the corresponding time is calculated.

[0021] Within a set one-dimensional search interval, with the goal of minimizing the restoration error, the optimal comprehensive wind-water increase coefficient is searched. If the optimal value obtained by the search falls on the boundary of the one-dimensional search interval, the search interval is adaptively expanded in the direction of the boundary and the search is repeated until the optimal value leaves the boundary.

[0022] In daily forecasting operations, forecast error is calculated based on the forecast water level generated by the daily rolling forecast and the corresponding measured water level. When the forecast error exceeds the preset threshold for multiple consecutive forecast periods, the sample in the latest sliding window is extracted as a new historical hydrological and meteorological sample, and the calibration of the comprehensive wind-induced water level coefficient is re-executed.

[0023] Optionally, the root cause diagnosis logic is executed according to a preset priority order, specifically including cross-variable joint determination:

[0024] When the rate of change of the measured water level exceeds the physical maximum rate of change threshold, the wind speed change characteristics at the corresponding moment are extracted.

[0025] If the wind speed change characteristics do not show synchronous abrupt changes, the root cause of the abnormal record is diagnosed as a water level sensor malfunction.

[0026] If the wind speed change characteristics show synchronous abrupt changes, the root cause of the abnormal record is diagnosed as the real physical response process caused by extreme wind speed abrupt changes, and the validity of the measured hydrological and meteorological data is maintained.

[0027] Optionally, the lag order of the historically lagged dimensionless water level difference index and the historically lagged dimensionless net inflow index is determined in advance by the following method:

[0028] Based on the Akaike Information Criterion, the goodness of fit of the multiple linear regression model containing different candidate lag orders was evaluated, and the candidate lag order with the smallest evaluation value was selected as the optimal lag order.

[0029] The residual autocorrelation function is used to diagnose the model residual sequence at the optimal lag order, and to verify that the model residual sequence falls within the confidence interval at each lag order, so as to eliminate residual autocorrelation.

[0030] According to another aspect of this application, a system for hourly forecasting of water levels in shallow lakes during periods of strong winds is also provided, comprising:

[0031] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement the hourly water level forecasting method for shallow lakes during strong winds as described in this application.

[0032] Based on the above technical solutions, this invention takes into account both the physical consistency of water balance and the spatial differences of wind pressure forcing, thereby improving the accuracy and timeliness of site-level forecasts during strong wind periods. Attached Figure Description

[0033] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0034] Figure 1 This is a schematic diagram of a method for hourly forecasting of water levels in shallow lakes during periods of strong winds, provided in this application.

[0035] Figure 2 This is a schematic diagram of the data quality monitoring process through the anomaly diagnosis module provided in this application.

[0036] Figure 3 This is a schematic diagram of the process for calculating or predicting wind-induced water level increase provided in this application.

[0037] Figure 4 This application provides a schematic diagram of the process for obtaining forecast meteorological data for the foreseeable future period during a typhoon-affected period.

[0038] Figure 5 This is a schematic diagram of the process for obtaining the net inflow of water around the lake, provided in this application. Detailed Implementation

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Example 1

[0041] This embodiment provides a method for hourly forecasting of water levels in shallow lakes during periods of strong winds, such as... Figure 1 As shown, the method may include, but is not limited to, the following steps:

[0042] Step 101: Obtain the measured hydrological and meteorological data at the current moment.

[0043] Specifically, based on the preset forecast start time, the measured water level, lake surface wind speed, wind direction and azimuth, and lake surface air pressure at each representative station at that time are collected. In practical applications, various types of data usually have different observation frequencies. If the air pressure data is observed daily and the air pressure field is relatively stable on that day, the daily air pressure observation value can be used as the air pressure data for all hourly times within that day for unified processing. If the air pressure data is completely missing, the air pressure forcing term can be temporarily set to zero, and subsequent processing can be performed only based on the wind field data, with independent air pressure correction performed afterward. This ensures the robustness of the forecast process under conditions of inconsistent data frequencies.

[0044] Step 102: Based on measured hydrological and meteorological data, the tilting and forcing effects of the current wind field and pressure field on the water level are removed, the initial static water level of each representative station is restored, and the initial static water level of the entire lake is extracted.

[0045] In this embodiment, the measured water level at any given time can be physically decomposed into a linear superposition of a static water level component, a wind-induced water level component, and a pressure-induced water level component. To obtain the initial state representing the true water storage capacity of the lake, it is necessary to remove water level tilt disturbances caused by instantaneous meteorological forcing from the measured water level. For different meteorological conditions, this invention designs multiple execution paths.

[0046] Under the first execution path, i.e. under normal wind conditions and pressure disturbance conditions, the wind-induced water level is calculated at the current moment using the measured wind speed and direction, and the pressure-induced water level is calculated at the current moment using the measured air pressure. The wind-induced water level and the pressure-induced water level are simultaneously subtracted from the measured water level to obtain the initial static water level of the representative station.

[0047] The second method is to take calm wind conditions. When the measured wind speed is not greater than the preset calm wind threshold, the wind-induced water level can be ignored. In this case, the initial calm water level can be obtained by subtracting the air pressure-induced water level from the measured water level.

[0048] The third method is that under standard air pressure conditions, when the measured air pressure is equal to the standard atmospheric pressure, the air pressure increase is zero. In this case, the wind increase is subtracted from the measured water level to obtain the initial static water level.

[0049] The fourth type, under ideal conditions of calm wind and standard atmospheric pressure, the measured water level is directly equivalent to the initial static water level.

[0050] Under ideal conditions, the static water levels at all representative stations across the lake should be relatively uniform. However, due to the simplification of the wind-induced water level parameter model and inherent errors in the observation data, the actual reconstructed initial static water levels at each station may exhibit slight differences. To characterize the average water storage state of the entire lake and suppress the aforementioned reconstruction residuals, the initial static water levels at each representative station are arithmetically averaged, and the resulting average is used as the representative initial static water level for the entire lake. In this embodiment, an arithmetic mean is used; in other embodiments, a weighted average or other methods may also be employed.

[0051] Step 103: Obtain the water balance driving factors under windless conditions.

[0052] Specifically, the driving factors of water balance can include precipitation over the lake surface, evaporation, water intake around the lake, and net inflow into the lake. These factors together determine the natural increase or decrease in the total volume of the lake, without considering disturbances from wind and pressure fields.

[0053] Step 104: Starting from the initial representative still water level of the entire lake, and combining the water balance driving factors, hourly water balance is recursively calculated to obtain the sequence of representative still water levels of the entire lake in the foreseeable future period.

[0054] In this embodiment, the extracted initial still water level of the entire lake is used as the initial condition for numerical recursion. The changes in water volume caused by various water balance driving factors are gradually accumulated according to a preset time step. This recursion process is carried out under the assumption of no wind, which suppresses the interference of wind-induced tilt on the water balance calculation. The final output sequence reflects the evolution trend of the lake's actual water storage at each future time.

[0055] Step 105: Obtain forecast meteorological data for the future forecast period.

[0056] Specifically, forecast wind speed, forecast wind direction, and forecast air pressure are extracted from numerical weather prediction products at various time steps within the future forecast period.

[0057] Step 106: Based on the forecast meteorological data, calculate the forecast wind field forcing effect and forecast pressure forcing effect for each representative station at each future time.

[0058] To ensure the consistency of the calculation logic and the physical reversibility, the same calculation model as in step 102 is used to independently calculate the wind-induced water increase and the pressure-induced water increase of each representative station under the forecast wind field and the forecast pressure field for each moment in the future forecast period.

[0059] Step 107: Using the initial still water level as a reference, the change of the representative still water level sequence of the whole lake relative to the initial still water level of the whole lake is superimposed, and the corresponding forecast wind field forcing effect and forecast air pressure forcing effect are superimposed respectively to synthesize the hourly forecast water level sequence of each representative station.

[0060] Specifically, water balance controls the change in the average water storage of the entire lake, which is spatially uniform; however, the forcing effects of wind and pressure fields vary significantly spatially. Therefore, the composite logic for the predicted water levels at each representative station is implemented through the following formula:

[0061] ;

[0062] In the formula:

[0063] This represents the predicted water level at the i-th representative station at a future time t, in meters (m).

[0064] This indicates that the i-th representative obtained from the first stage of restoration stands at the initial time. The initial static water level, in meters, represents the station's own static water reference level;

[0065] The water level at rest in the entire lake at time t, obtained from the second stage water balance recursion, is expressed in meters.

[0066] Indicates the initial time. The initial still water level of the entire lake, in meters;

[0067] This represents the uniform water level variation caused by changes in the total lake storage capacity. This variation is uniformly applicable to all representative stations, and the unit is meters (m).

[0068] The forecast wind-water rise at time t represents the i-th representative station, characterizing the wind field forcing effect of station differences, and the unit is m;

[0069] This represents the predicted atmospheric pressure rise at time t for the i-th representative station, in meters (m).

[0070] Through the above process, step 102 has already removed the wind pressure forcing component before the recursion, and step 104's water balance recursion only involves the natural increase or decrease of pure water storage, unaffected by wind-induced water surface tilt, thus ensuring the physical consistency of water balance. Furthermore, step 106 independently calculates the station-level wind-induced water increase and pressure-induced water increase for each representative station, and step 107 superimposes differentiated forcing effects on the basis of a unified overall lake water storage amplitude, thereby enabling the output forecast water levels for each station to reflect the spatial tilt distribution driven by the wind field.

[0071] It should be noted that the above embodiments use large, shallow lakes as a typical application scenario. Those skilled in the art will understand that the method is also applicable to other shallow water bodies with similar physical characteristics, namely, water depths smaller than the lake surface and significant wind stress affecting water level tilt, including but not limited to medium-sized lakes, coastal lagoons, or large reservoirs. In specific implementation, the spatial geometric parameters of the target water body, such as the location of a representative station and the length of wind zones in each direction, need to be configured, and the comprehensive wind-induced water level increase coefficient needs to be calibrated using historical hydrological and meteorological data of the water body.

[0072] According to another aspect of this application, a system for hourly forecasting of water levels in shallow lakes during periods of strong winds is also provided, comprising:

[0073] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement the hourly water level forecasting method for shallow lakes during strong winds as described in this application.

[0074] Example 2

[0075] The process of acquiring measured hydrological and meteorological data at the current moment in Example 1 is further refined. In actual environmental monitoring, sensor failures, communication interruptions, or sudden changes in real data caused by extreme weather are often intertwined. To ensure the reliability of the input data to the forecasting system, an anomaly diagnosis module is introduced for data quality monitoring. The anomaly diagnosis module is an automated data diagnosis module based on preset rules and decision tree logic, which internally constructs a three-layer data quality monitoring logic.

[0076] One alternative approach is to monitor data quality through an anomaly diagnosis module when acquiring measured hydrological and meteorological data for the current moment, such as... Figure 2 As shown, the specific steps may include, but are not limited to, the following:

[0077] Step 201: Perform a data validity check on each measured hydrological and meteorological data item by item and generate anomaly records.

[0078] In the first-layer architecture, the anomaly diagnosis module monitors the validity of various types of input data in parallel and assigns independent code identifiers to different types of anomalies to form an initial anomaly log. Item-by-item data validity checks include:

[0079] 1) Detection of the integrity and range of measured water levels. When the current water level data of any representative station is missing, a communication interruption exception code is marked; when the measured water level exceeds the historical extreme value range of the station (including the set safety margin), a water level exceeding the limit exception code is marked.

[0080] 2) Detection of abrupt changes in measured water level. The rate of change of measured water level between two adjacent time points is calculated. When this rate of change exceeds a preset physical maximum rate of change threshold, a water level abrupt change anomaly code is generated. The physical maximum rate of change threshold can be determined by those skilled in the art based on the water depth conditions and shallow water wave propagation characteristics of the target lake, combined with the statistical distribution characteristics of historical measured water level change rates. In an optional implementation, this threshold can be set as the 99.9th percentile of the absolute value sequence of historical hourly water level change rates.

[0081] 3) Meteorological parameter range and abrupt change detection. When the measured wind speed is negative or exceeds the set wind level upper limit, a wind speed exceeding limit anomaly code is marked; when the wind speed change between adjacent moments is greater than the set wind speed abrupt change threshold, a wind speed abrupt change anomaly code is marked; when the wind direction azimuth is not within the set circumferential angle range, a wind direction anomaly code is marked; when the measured air pressure is lower than the set low pressure extreme value or higher than the set high pressure extreme value, an air pressure exceeding limit anomaly code is marked.

[0082] 4) Multi-station consistency detection. Under low wind speed conditions, calculate the standard deviation of the measured water level at each representative station in the entire lake. When the standard deviation is greater than the set spatial difference threshold, mark the water level consistency anomaly code.

[0083] Step 202: Perform spatiotemporal correlation analysis on the abnormal records, and execute the root cause diagnosis logic according to the preset priority order to determine the root cause of the abnormal records.

[0084] In the second-layer architecture, the anomaly diagnosis module receives the set of anomaly records output from the first layer and performs joint cross-variable judgment using a decision tree model. This process distinguishes between instrument malfunctions and actual physical processes, avoiding misjudging severe hydrological responses under extreme weather conditions as data errors. The diagnostic results are output as a triple containing the anomaly phenomenon, possible cause, and confidence level.

[0085] Optionally, the root cause diagnosis logic is executed according to a preset priority order, specifically including cross-variable joint determination:

[0086] When the rate of change of the measured water level exceeds the preset physical maximum rate of change threshold, the wind speed change characteristics at the corresponding moment are extracted.

[0087] If the wind speed change characteristics do not show synchronous abrupt changes, the root cause of the abnormal record is diagnosed as a water level sensor malfunction, and the measured water level at the corresponding time is marked as invalid.

[0088] If the wind speed change characteristics show synchronous abrupt changes, the root cause of the abnormal record is diagnosed as the real physical response process caused by extreme wind speed abrupt changes, and the validity of the measured hydrological and meteorological data is maintained.

[0089] Specifically, the following cross-validation branches are executed in a preset priority order:

[0090] Branch 1 addresses communication interruption anomalies. If data is missing at multiple consecutive times at the same site, the root cause is diagnosed as sensor communication interruption, and a high confidence level is assigned.

[0091] Branch 2: For abnormal water level changes, extract the wind speed change characteristics at the corresponding time. If the wind speed change characteristics do not show synchronous abrupt changes, i.e., no wind speed abrupt change anomaly code is triggered, then the root cause of the abnormal record is diagnosed as a water level sensor jump or icing failure, and a medium confidence level is assigned.

[0092] Branch 3 addresses abnormal water level changes. If wind speed changes exhibit synchronous abrupt changes, i.e., simultaneously triggering wind speed abrupt change anomaly codes, then the water level abrupt change is determined to have meteorological dynamic support, and the root cause is diagnosed as a real physical response process caused by extreme wind speed abrupt changes. Under this branch, the validity of measured hydrological and meteorological data is maintained, and no data is discarded.

[0093] Branch 4: For water level exceeding the limit anomaly, jointly analyze the corresponding air pressure records. If the air pressure data does not trigger the air pressure exceeding the limit anomaly code, then extreme air pressure disturbance factors are ruled out, and the root cause is diagnosed as water level sensor range overflow or reference drift.

[0094] Branch 5 involves spatial correlation analysis for anomalies in consistency across multiple stations. If only a single station's water level deviates from the group distribution, the root cause is diagnosed as an individual sensor anomaly at that station. If multiple stations simultaneously exhibit deviation characteristics, the root cause is diagnosed as real spatial differences caused by extreme weather or systemic transmission failures, and a lower confidence level is assigned, indicating the need for manual verification.

[0095] Step 203: Based on the root cause and the continuous characteristics and multi-station synchronization characteristics of the abnormal records, output graded early warning information, and execute corresponding forecasting and intervention strategies for different graded early warning information.

[0096] In the third-layer architecture, based on the root cause diagnosis results mentioned above, the potential impact of data anomalies on the forecasting process is comprehensively assessed, and an automated, tiered response is implemented:

[0097] When an anomaly manifests as an isolated event at a single site, and its duration does not exceed the set short-term threshold, a Level 3 warning message is output. The corresponding forecast intervention strategy is: prompt to check the sensor status of the corresponding site, and the current forecast process continues.

[0098] When the same type of anomaly occurs at the same station multiple times in a row, or when multiple stations simultaneously experience anomalies, or when the confidence level of the root cause diagnosis is lower than the set confidence threshold, a level-two warning message is output. The corresponding forecast intervention strategy is to add a data quality degradation flag to the output forecast water level sequence to prompt forecasters to review the rationality of the meteorological input data.

[0099] When multiple physical constraint checks fail continuously, or the number of stations affected by systemic communication interruption reaches a set proportion, or the water level change rate exceeds a set multiple of the physical maximum change rate threshold, a Level 1 warning message is output. The corresponding forecasting intervention strategy is to automatically suspend the current static water level restoration and forecasting process, record the interruption point, and wait for manual intervention or data recovery.

[0100] Example 3

[0101] A dynamic, sequential fault-tolerance mechanism based on data interruption backtracking is provided as an alternative to Implementation 1 when faced with continuous data loss and the inability to interpolate and fill in the gaps. In actual business operations, the data acquisition link may be interrupted for a long period of time due to force majeure. This implementation ensures the continuous execution of the forecasting process by dynamically adjusting the time window.

[0102] An alternative approach, when acquiring measured hydrological and meteorological data at the current moment, also includes a data interruption tolerance strategy:

[0103] Step 301: If the measured hydrological and meteorological data at the current moment is missing and cannot be filled in by interpolation of data from adjacent moments, then backtrack to the historical moment to obtain the last continuous and valid measured hydrological and meteorological data.

[0104] In routine forecasting operations, if a sensor communication link failure results in missing data for the current moment, interpolation using valid observations from adjacent moments is prioritized to fill the gap. When the data gap spans a long period, causing the interpolation preconditions to be unmet, a backtracking mechanism is activated. This mechanism searches the historical data queue in reverse chronological order for the most recent complete moment that simultaneously contains all four elements: measured water level, wind speed, wind direction, and air pressure. The data at that moment is locked as the last continuous and valid measured hydrological and meteorological data, and the corresponding timestamp is extracted.

[0105] Step 302: Based on the time difference between the timestamp of the last continuous and valid measured hydrological and meteorological data and the target forecast time, dynamically extend the forecast start time and shorten the forecast duration to ensure the continuous execution of the forecast process.

[0106] Because the response of a lake hydrodynamic system has a certain time inertia, within a finite time span, using historical valid states as initial conditions for recursion can still provide forecast results with engineering reference value. The time difference between the timestamp of the last consecutive valid data point and the set target forecast time is calculated, and the forecast window is dynamically reconstructed based on this time difference.

[0107] Specifically, assuming the set standard forecast duration is... The target forecast time is When a data interruption occurs, the timestamp of the last consecutive valid data obtained is... .calculate and Time difference between Based on this time difference, execute the following conditional branch logic:

[0108] like The minimum forecast lead time constraint set by the system is met, i.e. , To ensure the minimum time span for the availability of forecast results, the forecast start time will be dynamically adjusted from the originally planned current time to... And the forecast duration will be dynamically shortened to .by The measured hydrological and meteorological data at any given time are used as input to initiate the static water level restoration and subsequent recursive process described in Example 1.

[0109] like If the forecast window is less than the minimum forecast lead time constraint, it indicates that the effective initial state is too close to the target forecast time, and the forecast results lose their forward-looking guidance significance. At this time, an output data interruption warning is issued, and the forecast task for the current period is suspended, waiting for the data link to be restored before the forecast window is reconstructed starting from the new complete data time.

[0110] Example 4

[0111] This document details the specific process of separating the current wind field effect, pressure field effect, and calculating the forecast forcing effect in Example 1. This example covers the computational logic across three dimensions: spatial parameter lookup, nonlinear physical constraints, and time-based dynamic response.

[0112] One possible implementation method, the tilt forcing effect includes pressure-induced water increase and wind-induced water increase;

[0113] The forecast wind field forcing effect is the forecast wind-induced water increase, and the forecast pressure forcing effect is the forecast pressure-induced water increase;

[0114] Both the pressure increase and the predicted pressure increase are determined based on the relative change of pressure data with respect to a preset reference pressure.

[0115] Both wind-induced water level rise and forecast wind-induced water level rise are determined based on the preset comprehensive wind-induced water level rise coefficient, the effective wind zone length pre-configured for each representative station, the wind direction factor, and the wind speed. The wind direction factor is the cosine of the angle between the wind direction and the pre-configured maximum water level rise azimuth angle corresponding to the representative station.

[0116] When calculating the pressure increase, for the stress characteristics of closed shallow lakes, the pressure correction is mainly used to capture the lake surface tilting effect caused by the spatial gradient of the pressure field; under the condition of uniform pressure change throughout the lake, according to the reversible chain design of the present invention of first peeling and then superimposing, the influence of uniform amplitude on the predicted final value can be offset by itself.

[0117] The calculation of pressure-induced water level increase involves obtaining the difference between the reference pressure and the current measured or predicted pressure, calculating the ratio of this difference to the reference pressure, and then multiplying it by the dimensionless pressure-induced water level increase coefficient and the lake's multi-year average water level. The formula is as follows:

[0118] ;

[0119] In the formula:

[0120] This represents the pressure increase at time t for the i-th station, in meters (m).

[0121] β represents the dimensionless pressure-induced water level increase coefficient, the value of which can be derived by those skilled in the art based on the hydrostatic equilibrium theory, i.e., the relationship between the water level change Δh caused by a unit pressure change ΔP and the relationship Δh = ΔP / (ρg), where ρ is the water density and g is the acceleration due to gravity. This coefficient is related to the lake's multi-year average water level. The product of these factors constitutes the pressure response factor, which has the dimension of length (in meters).

[0122] This indicates the multi-year average water level of the target lake, in meters (m).

[0123] This indicates the set reference pressure, in hPa.

[0124] This represents the measured or predicted air pressure on the lake surface at time t for the i-th representative station, in hPa.

[0125] To calculate the fundamental parameters required for wind-induced water level rise, the difference between the current wind azimuth and the maximum water level rise azimuth at the representative station is calculated, and the cosine of this difference is taken as the wind direction factor. This wind direction factor represents the projected component of wind stress in the direction of maximum water level rise. The maximum water level rise azimuth at each representative station is the azimuth angle of the maximum effective wind zone corresponding to that station. This angle can be determined by those skilled in the art based on the digitized shoreline data of the target lake and the spatial coordinates of each station, by calculating the effective wind zone length at each azimuth angle and taking the azimuth angle corresponding to the maximum wind zone length.

[0126] One possible implementation involves using a gradient adaptive interpolation strategy when obtaining the effective wind zone length, which may include the following process:

[0127] Based on the adjacent azimuth of the wind direction, obtain the pre-stored length of the corresponding azimuth wind zone at the azimuth of the azimuth;

[0128] Calculate the ratio of the lengths of the wind zones at adjacent azimuth points as the gradient ratio;

[0129] If the gradient ratio is not greater than the set threshold, then linear interpolation is used to obtain the effective wind zone length;

[0130] If the gradient ratio is greater than the set threshold, cosine weighted interpolation is used to obtain the effective wind zone length.

[0131] The wind zone length of each representative station at each azimuth angle, such as setting an azimuth angle every 10° or 22.5°, can be calculated in advance based on the digital shoreline data of the target lake and the spatial coordinates of each representative station, by the distance between the intersection of the azimuth ray and the shoreline, and stored in the form of a lookup table for later use.

[0132] When obtaining the effective wind zone length, the following measurement tools and operating procedures can be used:

[0133] Ideally, a Geographic Information System (GIS) platform or remote sensing image interpretation tool should be used to load high-resolution digital shoreline data of the target lake, such as the lake boundary line in vector format. In the GIS environment, with the coordinates of each representative station as the origin, rays are generated along the azimuth of each point, and the intersection of the rays with the shoreline boundary is automatically obtained and the Euclidean distance is calculated. In the absence of GIS, a large-scale topographic map can be used in conjunction with a digital protractor and distance measuring tools for manual measurement, or a laser rangefinder, total station and other surveying instruments can be used to measure the straight-line distance to the opposite shoreline along the azimuth direction on the ground.

[0134] In practice, first, determine the calculation points for each representative station as the starting point for ray calculation. These points should ideally be located on the shoreline of the representative station or at a representative location on the lake surface. Using the north-facing wind as a reference, determine each azimuth angle sequentially at preset whole-point intervals, such as 10° or 22.5°. Draw a ray from the calculation point along each azimuth angle in the upwind direction (or downwind direction, as agreed upon in the calculation). Measure the distance between the ray and the first intersection point with the lake shoreline; this distance is recorded as the wind zone length for that azimuth. When the water area is irregularly shaped or has obstructions such as islands or peninsulas, causing the ray to be interrupted, take the distance from the calculation point to the edge of the obstruction as the wind zone length for that ray. After measurement, verify and archive the wind zone lengths for each azimuth, creating a wind zone length lookup table for each representative station for subsequent interpolation calculations.

[0135] To address the issue of varying wind zone lengths between adjacent azimuths caused by irregular lake shorelines, this embodiment constructs a conditional branch interpolation logic:

[0136] When the gradient ratio is not greater than the set threshold (e.g., set to 2.0), it indicates that the length of the wind zone in adjacent directions changes gradually. The effective wind zone length is calculated directly using the conventional linear interpolation formula based on the proportional relationship between the wind direction angles in adjacent hourly directions.

[0137] When the gradient ratio is greater than the set threshold, it indicates that there is island shading or abrupt change in the coastline, and the operation is switched to cosine weighted interpolation.

[0138] Cosine-weighted interpolation is calculated as follows:

[0139] ;

[0140] Among them, the weight term The calculation formula is:

[0141] ;

[0142] In the formula:

[0143] This represents the effective wind zone length of the i-th representative station in the direction of the actual wind direction θ, in meters;

[0144] and This represents the azimuth angle between two adjacent integer points within the interval where the actual wind direction θ falls, in degrees.

[0145] and This indicates the length of the known effective wind zone at the corresponding azimuth at the whole hour, in meters.

[0146] This represents the cosine weight corresponding to the first integer point, and is dimensionless.

[0147] Let represent the cosine weight corresponding to the second integer point, which is dimensionless and satisfies ... .

[0148] This weighting mechanism makes the first derivative of the interpolation result approach zero at the azimuth boundary, thereby eliminating the slope abrupt error at the connection between adjacent intervals.

[0149] One possible implementation is to calculate or predict wind-increase, such as Figure 3 As shown, the process may include the following:

[0150] Step 401: Based on the drag coefficient saturation mechanism, the actual wind speed is converted into a saturated equivalent wind speed. The saturated equivalent wind speed increases linearly in the low wind speed range and converges asymptotically in the high wind speed range.

[0151] Step 402: Calculate the theoretical wind-induced water increase based on the saturated equivalent wind speed, effective wind zone length, and wind direction factor;

[0152] Specifically, the theoretical wind-induced water increase is calculated based on the product of the actual wind speed and the saturated equivalent wind speed, the effective wind zone length, and the wind direction factor, combined with the comprehensive wind-induced water increase coefficient.

[0153] Step 403: Obtain the average water depth of the lake and set the physical constraint boundary of the absolute water depth in combination with the preset safety factor;

[0154] Step 404: Cut off the theoretical wind-water increase to the physical constraint boundary of absolute water depth to obtain the final wind-water increase or the predicted wind-water increase.

[0155] To address the divergence of traditional linear formulas under extreme wind speeds, a hyperbolic tangent function is introduced to construct the saturation calculation logic for the drag coefficient:

[0156] ;

[0157] Based on this, the theoretical formula for calculating wind-water increase in equilibrium is:

[0158] ;

[0159] In the formula:

[0160] This represents the converted saturated equivalent wind speed, in m / s.

[0161] This represents a pre-set saturation characteristic wind speed, expressed in m / s. Its value should satisfy the following conditions: greater than the upper limit of the historical common wind speed for the lake, and less than the lower limit of the extreme typhoon wind speed, ensuring that wind-induced water rise maintains an approximately quadratic increase within the common wind speed range, and that wind-induced water rise smoothly converges within the extreme wind speed range. The specific value can be determined by those skilled in the art based on meteorological observation data of the target lake area and existing research results on wind stress drag coefficient saturation. In one optional implementation, the value is 30 m / s.

[0162] This indicates the actual wind speed at the input time, in m / s.

[0163] Represents the hyperbolic tangent function;

[0164] This represents the calculated theoretical wind-induced water increase value, in meters (m).

[0165] k represents the comprehensive wind-induced water increase coefficient, which comprehensively incorporates physical factors such as wind stress drag coefficient, gravitational acceleration, and average lake depth. Its dimensions are implicitly determined by the dimensional balance relationship between the two sides of the formula equation, and the specific value is determined by the parameter calibration method described in Example 7.

[0166] Indicates the effective wind zone length, in meters (m).

[0167] This indicates the wind direction factor.

[0168] For example, suppose the input wind speed is much greater than the saturation characteristic wind speed, making the ratio At this point, the hyperbolic tangent term The asymptotic convergence to 1.0 causes the velocity term in the calculation to converge to... This limits the unbounded growth of wind-induced water volume as a function of the square of wind speed.

[0169] To ensure that the calculation results do not deviate from the constraints of fluid dynamics, the product of the lake's average water depth and the preset safety factor is obtained, and a positive-negative symmetrical physical constraint boundary for absolute water depth is established. , The maximum permissible negative water reduction (water level drop) limit, This represents the maximum permissible positive water level rise (water level increase) limit. When the theoretical wind-induced water level rise exceeds this limit, it is forcibly truncated to the corresponding boundary extreme value; when it is within the boundary, it is retained. In an optional implementation, a preset safety factor is set to 0.8. Those skilled in the art can adaptively adjust this factor based on the actual topography of the target lake and historical extreme wind-induced water level rise observation records, provided that the constraint boundary does not exceed the actual achievable water depth.

[0170] One possible implementation is to obtain the wind-driven water level rise or forecast using a dynamic response recursive model, which may include the following processing:

[0171] Based on the shallow water gravity wave velocity and effective wind zone length determined by the preset average lake water depth, the response time constant for the corresponding wind direction is determined.

[0172] Based on the relationship between the response time constant and the preset time step, the dimensionless relaxation coefficient is determined. Specifically, the smaller value between the ratio of the preset time step to the response time constant and the constant 1.0 is taken as the dimensionless relaxation coefficient.

[0173] Based on the quasi-static equilibrium assumption, the equilibrium wind increase at the current moment is calculated according to the wind speed, effective wind zone length, wind direction factor, and comprehensive wind increase coefficient.

[0174] The dynamic wind-water increase at the previous moment and the equilibrium wind-water increase at the current moment are weighted and recursively calculated using a dimensionless relaxation coefficient to obtain the dynamic wind-water increase at the current moment, which serves as the final or predicted wind-water increase. The dynamic wind-water increase at the initial moment of the recursion is taken from the equilibrium wind-water increase at the corresponding moment.

[0175] When acquiring dynamic wind-induced water rise, the standard shallow water wave equation is used, which states that wave velocity equals the square root of the product of gravitational acceleration and the average lake depth. The wave velocity is then obtained, and the ratio of the effective wind zone length to the wave velocity is calculated to obtain the response time constant. Compare with preset time steps. The dimensionless relaxation coefficient is obtained by taking the ratio of the relaxation coefficient to the response time constant and finding the minimum value between this ratio and the constant 1.0. .

[0176] At the start time of the forecast The wind-induced water level increase value at the representative station, calculated using measured wind speed and direction in step 102, is used as the initial value for the dynamic wind-induced water level increase sequence. .

[0177] Dynamic recursion is performed using first-order exponential relaxation logic:

[0178] ;

[0179] In the formula:

[0180] This represents the dynamic wind-water increase value of the i-th representative station at the current time t, in meters (m).

[0181] This represents the dimensionless relaxation coefficient of the i-th element at the current time t.

[0182] This indicates that the i-th representative stands in front of the previous calculation step. The dynamic wind-induced water level increase, in meters (m);

[0183] The value represents the balanced wind-water increase obtained by the i-th representative station at the current time t after truncation during the calculation or forecasting of wind-water increase, in meters.

[0184] The dynamic wind and water increase at the current moment is obtained after the above dynamic response recursion. This refers to the forecast wind and water level increases for each representative station in the main formula of step 107 of Example 1. This recursive mechanism characterizes the exponential decay asymptotic trend of water level adjustment towards a new equilibrium state, which is limited by shallow water wave propagation; when the effective wind zone is extremely short, resulting in a relaxation coefficient of 1.0, this weighted mechanism automatically degenerates into a quasi-static instantaneous equilibrium.

[0185] The following simplified scenario illustrates the working process of the dynamic response recursive model. Assume a representative station is under stable northeasterly wind conditions at time t-1, and the dynamic wind-water gain has converged to near equilibrium. At time t, the wind direction suddenly shifts to southwesterly. At this moment, the sign of the calculated equilibrium wind-water gain at the current time reverses, changing from positive to negative. However, exponential relaxation recursion causes the dynamic wind-water gain to gradually adjust from the previous positive value to the new negative equilibrium state, rather than abruptly changing. When the effective wind area of ​​the station is long and the response time constant is large, the relaxation coefficient is small, and the adjustment process is slow; when the effective wind area is short, the relaxation coefficient approaches 1.0, and the adjustment process approaches instantaneous equilibrium. This mechanism conforms to the physical law of gradual water level adjustment under conditions of limited wave propagation in shallow water.

[0186] Example 5

[0187] This paper provides a site-level wind field inversion mechanism for the typhoon impact period, as an enhanced alternative to the acquisition of forecast meteorological data for the future forecast period in Example 1 under extreme weather conditions. Conventional zero-dimensional forecast models typically use a uniform forecast wind field input across the entire lake, which is insufficient to reflect the spatially uneven distribution of the lake surface wind field caused by typhoon circulation. When acquiring forecast meteorological data for the future forecast period, if it is during the typhoon impact period, a reference should be made... Figure 4 Specifically, this may include, but is not limited to, the following steps:

[0188] Step 501: Obtain the typhoon track and intensity forecast parameters for the forecast period.

[0189] Specifically, typhoon characteristic information for each moment within the forecast period is extracted from meteorological forecast products. This characteristic information may include the typhoon's center coordinates, minimum central pressure, and maximum wind speed radius. To address the discrepancy between the frequency of meteorological forecast product releases and the hourly step size of the forecast system, a linear interpolation algorithm is used to smooth the typhoon parameters at adjacent forecast moments over time, generating a continuous sequence of hourly track and intensity forecast parameters. This provides high-temporal-resolution input for subsequent dynamic inversion.

[0190] Step 502: Based on the typhoon parameterized wind field model, combined with the typhoon path and intensity forecast parameters and the spatial location of each representative station, the local gradient wind speed and local gradient wind direction at the location of each representative station are calculated by inversion.

[0191] In this step, the known fixed spatial coordinates of each representative station and the typhoon center coordinates at the current moment are extracted. The great circle distance formula is used to calculate the actual distance of each representative station from the typhoon center at the current moment. To ensure the dimensional consistency of the dynamic calculations, the extracted minimum central pressure and the surrounding environmental pressure are converted into standard pressure units.

[0192] The Holland parametric wind field model was used to process the above parameters. Based on the transformed pressure difference, maximum wind speed radius, calculated spherical distance, and Coriolis parameters corresponding to the target latitude, the local gradient wind speed at the top of the atmospheric boundary layer at each representative station location was calculated. This inversion mechanism transforms macroscopic typhoon forecast parameters into specific local wind field parameters, effectively reflecting the spatial dynamic evolution of the asymmetric pressure gradient force and wind field structure near the typhoon center as the typhoon moves along its path. An example of the Holland parametric gradient wind speed formula is shown below:

[0193] The i-th element represents the gradient wind speed at time t. Calculated according to Holland (1980) model:

[0194] ;

[0195] In the formula:

[0196] B is Holland's shape parameter, which is dimensionless and taken empirically. , where 60 is an empirical denominator constant, with the unit hPa (consistent with the dimension of the pressure difference, so that B remains dimensionless).

[0197] kg / m 3 This represents the air density in a typhoon environment (slightly lower than the standard value due to low air pressure and high humidity).

[0198] This represents the distance between the i-th representative station and the center of the typhoon;

[0199] The radius of maximum wind speed;

[0200] hPa is the ambient air pressure.

[0201] This is the typhoon's central pressure, in hPa. It needs to be converted to Pa for calculation: multiply by 100.

[0202] For Coriolis parameters, Earth's rotational angular velocity rad / s;

[0203] and The units are all in meters (m), and need to be converted from kilometers (km) when used in calculations.

[0204] Step 503: Perform height reduction on the local gradient wind speed to obtain the independent forecast wind speed and forecast wind direction for each representative station, which replaces the unified forecast wind field for the entire lake and is used to calculate the forecast wind-induced water level rise for each representative station.

[0205] The local gradient wind speed is reduced by height to obtain the independent forecast wind speed for each representative station; based on the local gradient wind direction, the azimuth deflection is corrected by combining the cyclone rotation characteristics and the inflow angle caused by the bottom friction effect to obtain the independent forecast wind direction for each representative station; the forecast wind speed and forecast wind direction are used to replace the unified forecast wind field for the whole lake to calculate the forecast wind-induced water rise for each representative station.

[0206] Since the local gradient wind speed obtained through inversion corresponds to the top of the upper boundary layer, a preset water surface reduction factor is used to reduce its vertical height, mapping it to obtain the forecast wind speed at a standard height above the water surface, such as 10m. Simultaneously, the radial azimuth angle from the typhoon center to each representative station is calculated, and combined with the cyclone rotation physical characteristics and the inflow angle caused by the lower-level friction effect, this radial azimuth angle is deflected counterclockwise to determine the standard height forecast wind direction for each station.

[0207] The surface reduction factor can be determined by those skilled in the art based on empirical relationships of typhoon boundary layers. Typical values ​​can be referenced from meteorological literature on surface wind reduction in tropical cyclones. The inflow angle is the angle at which the typhoon's bottom-level airflow deviates from the gradient wind direction due to surface friction, and can be determined based on empirical values ​​of the well-known bottom-level structure of tropical cyclones. In one optional embodiment, the surface reduction factor is taken as 0.75, and the inflow angle is taken as 20°. The above parameters are commonly used empirical values ​​in tropical cyclone boundary layer studies, and those skilled in the art can make adaptive adjustments based on the actual surface roughness and atmospheric stratification conditions of the target lake.

[0208] Through the above processing, spatially differentiated independent forecast wind speeds and directions were generated for each representative station within the lake area. This station-level meteorological forcing data was then used to replace the default unified forecast wind field for the entire lake and input into the dynamic wind-water increase calculation module of Example 4, enabling the zero-dimensional water balance model to possess station-level wind field spatial resolution capabilities.

[0209] As an adaptive degradation mechanism, when the distance between the typhoon center and the target lake exceeds the set influence radius threshold, or when the typhoon center pressure rises above the preset safety threshold, the inversion process of this embodiment is automatically terminated, and the unified forecast wind field of the entire lake is restored to ensure the computational efficiency and logical consistency of daily forecasting operations.

[0210] Example 6

[0211] This paper provides a method for windless water balance recursion and precise estimation of net inflow to lakes. It details the steps of hourly water balance recursion in combination with water balance driving factors, including the acquisition of multi-source driving factors, control of numerical recursion stability, and downscaling allocation of core driving factors.

[0212] One possible approach is that the water balance driving factors include lake surface precipitation, water surface evaporation, water intake around the lake, and net inflow into the lake around the lake; the hourly water balance recursion is achieved by accumulating the water volume changes of each water balance driving factor hourly.

[0213] Under the assumption of no wind, the change in the water level of a lake under still water conditions is determined by the four water balance driving factors mentioned above. Numerical recursion calculations are performed with a preset step size. The specific formula is as follows:

[0214] ;

[0215] In the formula:

[0216] This represents the representative still water level of the entire lake obtained recursively at the next calculation time t+1, in meters;

[0217] This represents the representative still water level of the entire lake at the current calculation time t, in meters.

[0218] This represents the water depth calculated from the cumulative precipitation over the lake surface during the period from t to t+1, in meters.

[0219] This represents the water depth calculated from the cumulative evaporation from the lake surface during the period from t to t+1, in meters.

[0220] A constant conversion factor is used to represent the volume from 10,000 cubic meters to cubic meters, ensuring that the water volume is divided by the area and the units are uniformly converted to water depth in meters.

[0221] This represents the average surface area of ​​the target lake, in m². 2 ;

[0222] This represents the net inflow of water around the lake during the period from t to t+1, expressed in tens of thousands of cubic meters. Positive values ​​indicate net inflow, while negative values ​​indicate net outflow.

[0223] This represents the amount of water used around the lake during the period from t to t+1, expressed in tens of thousands of cubic meters.

[0224] Regarding the acquisition of various driving factors, this embodiment sets out specific processing mechanisms and alternative solutions:

[0225] For lake precipitation, the measured precipitation data or the arithmetic mean of precipitation in the numerical weather prediction grid are prioritized as hourly inputs; when hourly precipitation data is missing, an alternative strategy is adopted to evenly distribute the total daily precipitation to each calculation step.

[0226] For the water surface evaporation term, a daily evaporation characteristic curve based on multi-year observation data is extracted, and a temperature-evaporation correction coefficient is introduced based on the forecast daily temperature anomaly to dynamically correct the baseline evaporation extracted from the characteristic curve.

[0227] For water intake around the lake, a benchmark water intake is set based on the preset engineering design water supply capacity, and a comprehensive water intake correction coefficient is introduced to scale it up, taking into account operating conditions such as peak agricultural irrigation periods or dry seasons.

[0228] When performing the above recursive calculation, if the water level obtained through recursion exceeds a preset reasonable range, the water level value is truncated to the boundary of the corresponding range. One possible implementation method is to refer to... Figure 5 The net inflow of water around the lake can be obtained through the following steps:

[0229] Step 601: Calculate the water level difference between the representative station of each zone and the lake, and normalize the water level difference using the independent historical extreme values ​​of each zone to obtain the dimensionless water level difference index.

[0230] The measured water levels of representative stations in each lake area were collected, the water level difference between each representative station and the lake was calculated, and the water level difference was normalized using the pre-stored historical extreme values ​​of each area to obtain the dimensionless water level difference index.

[0231] In this embodiment, based on the spatial distribution of the main inflow and outflow waterways around the lake, the area surrounding the lake is divided into several zones. Each zone has a representative station to monitor the water level potential energy difference between that zone and the lake. The representative stations are hydrological monitoring stations set up in the inflow and outflow areas around the lake, which are spatially different from the representative stations set up inside the lake area (lake-side representative stations) described in Embodiment 1.

[0232] To address the issue of discontinuous data due to artificial scheduling affecting the flow at the lake's inlets, this embodiment uses the water level potential energy difference as a driving factor to estimate the net inflow. The measured water levels at representative stations in each zone are obtained, along with the measured lake water level, and the difference between the two is calculated. The maximum absolute value of the water level difference in the historical long-series measured data for each zone is extracted and set as a fixed normalized denominator. The ratio of the current zone's water level difference to the fixed normalized denominator is calculated to obtain a dimensionless water level difference index whose value is constrained within a fixed interval. This mechanism ensures that the normalized feature scale does not depend on predetermined calibration samples, guaranteeing the determinism of subsequent model extrapolation.

[0233] The same normalization strategy was used to process the historical series of net inflow into the lake. Specifically, the maximum absolute value of the daily total net inflow into the lake was extracted from the long historical series and set as the fixed normalized denominator for the net inflow. The ratio of the daily total net inflow to the fixed normalized denominator for each historical day was calculated to obtain the dimensionless net inflow index. This index was used as the autoregressive input for the subsequent multiple linear regression model.

[0234] Optionally, the lag order of the historically lagged dimensionless water level difference index and the historically lagged dimensionless net inflow index is determined in advance by the following method:

[0235] Based on the Akaike Information Criterion, the goodness of fit of the multiple linear regression model containing different candidate lag orders was evaluated, and the candidate lag order with the smallest evaluation value was selected as the optimal lag order.

[0236] The residual autocorrelation function is used to diagnose the model residual sequence at the optimal lag order, and to verify that the model residual sequence falls within the confidence interval at each lag order, so as to eliminate residual autocorrelation.

[0237] The optimal lag order is used to fit the multiple linear regression model to obtain the model residual sequence. The residual autocorrelation function is used to diagnose the model residual sequence and verify that the model residual sequence falls within the confidence interval at each lag order, so as to eliminate residual autocorrelation. If the model residual sequence exceeds the confidence interval at some lag orders, the candidate lag order is gradually increased and the goodness of fit assessment and residual diagnosis are re-executed until the model residual sequence meets the condition of falling within the confidence interval.

[0238] Considering the inertial effect of water volume changes in large shallow lakes, a lag term for previous water potential conditions is introduced. A set of continuous candidate lag orders is pre-defined, and the Akaike Information Criterion (AIC) is used to calculate the model's evaluation value for each candidate order. The order that minimizes the evaluation value is extracted as the optimal choice. The autocorrelation function (ACF) of the residuals is used to calculate the autocorrelation coefficient of the residual sequence at this order, and it is determined whether it falls within the set statistical confidence interval. This confirms that the model structure has filtered out the time correlation characteristics of the residuals.

[0239] Step 602: Input the current dimensionless water level difference index, the historically lagged dimensionless water level difference index, and the historically lagged dimensionless net inflow index into the pre-constructed multiple linear regression model to estimate the total daily net inflow of water around the lake; wherein, the dimensionless net inflow index is the normalized value obtained by dividing the total daily net inflow of water around the lake by the maximum absolute value of the pre-stored historical daily net inflow.

[0240] Based on the aforementioned optimal lag order, the historical lag dimensionless water level difference index sequence and the historical lag dimensionless net inflow index sequence for the corresponding period are extracted. Combined with the current dimensionless water level difference index, they are input into a multiple linear regression model pre-constructed using the least squares method to calculate and output the daily total net inflow of water around the lake.

[0241] Step 603: Based on the uniform distribution of the daily net inflow of water around the lake, and taking into account the difference between the hourly rainfall intensity and the daily average rainfall intensity in the foreseeable future period, the daily net inflow of water around the lake is dynamically distributed through a preset rainfall weighting coefficient, and after total amount conservation correction, the hourly net inflow sequence of water around the lake is obtained.

[0242] Based on the uniform distribution of the daily net inflow of water around the lake, the hourly rainfall intensity within the forecast period is extracted from the forecast meteorological data, and its daily average rainfall intensity is calculated. According to the difference between the hourly rainfall intensity and the daily average rainfall intensity, the daily net inflow of water around the lake is dynamically distributed through a preset rainfall weighting coefficient. After total amount conservation correction, the hourly net inflow sequence of water around the lake is obtained.

[0243] Since daily totals cannot be directly used for hourly recursion, this embodiment designs a downscaling allocation algorithm based on rainfall intensity. The specific calculation formula is as follows:

[0244] ;

[0245] In the formula:

[0246] This represents the net inflow of water around the lake obtained during the τth calculation period, expressed in tens of thousands of cubic meters.

[0247] This represents the baseline value for uniform distribution obtained by dividing the total daily net inflow of water around the lake by the total number of time periods within the day, expressed in tens of thousands of cubic meters.

[0248] λ represents a preset dimensionless rainfall weighting coefficient, used to adjust the concentration effect of rainfall on water distribution. Its value controls the intensity of the concentration effect of rainfall on the distribution of net inflow to the lake. When λ=0, it degenerates into a uniform distribution; the larger λ is, the higher the distribution ratio during rainfall periods. In an optional implementation, the value is 0.5. Those skilled in the art can determine a suitable value by performing regression analysis or cross-validation on historical rainfall-inflow control samples based on the runoff response characteristics of the target lake basin.

[0249] This represents the predicted rainfall intensity during the τth calculation period, in mm / h.

[0250] This indicates the average daily rainfall intensity for the forecast day, expressed in mm / h.

[0251] When rainfall is forecast for the day, that is When the value is greater than 0, this formula ensures that periods with rainfall intensity higher than the daily average receive more inflow into the lake, while periods without rainfall retain a stable baseflow constrained by λ; when no rainfall is forecast for the day, i.e. When =0, skip the above weighted calculation steps and directly set the time intervals. This involves uniform allocation. After the initial allocation for each time period is completed, the sum of the allocated value sequence is calculated, and the ratio of this sum to the daily net inflow of water around the lake is extracted. This ratio is then used to scale the allocated values ​​for each time period to ensure that the downscaled hourly sequence satisfies the strict total amount conservation condition on the daily scale.

[0252] Example 7

[0253] An adaptive calibration and rolling update closed-loop mechanism for core parameters is provided to solve the problem that the wind stress drag coefficient in the forecast model is difficult to obtain and is prone to drift with seasonal environment. By combining offline calibration and online closed-loop optimization, the long-term stable operation of the forecast system is guaranteed.

[0254] An alternative approach is to pre-calibrate and update the comprehensive wind-water increase coefficient required for calculating wind-water increase in the following manner:

[0255] Step 701: Select quasi-calm wind samples and windy samples from historical hydrological and meteorological samples, and calibrate the arithmetic mean of the measured water levels of each representative station in the quasi-calm wind samples as the true value of the calm water level.

[0256] During the offline parameter calibration phase, samples were divided based on wind speed conditions within the historical time series. When the lake surface wind speed during a historical period did not exceed a set calm wind threshold, it was classified as a quasi-calm wind sample. Under quasi-calm wind conditions, the wind stress forcing effect on the water surface is at a low level, and the oscillation energy of the lake surface converges. At this time, the water levels at representative stations across the entire lake tend to be horizontal. Therefore, the arithmetic mean of the measured water levels at each station in the quasi-calm wind sample was extracted and physically equivalent to the true calm water level corresponding to the actual water storage capacity of the lake at that moment, eliminating the cumbersome process of relying on expensive field observation equipment to obtain calibration benchmarks.

[0257] For periods when the wind speed exceeds a set wind threshold, these periods are classified as windy samples and used as the test set for subsequent calibration. The wind speed range between the wind threshold and the calm wind threshold serves as a transition buffer zone; samples within this buffer zone are not included in the calibration process to avoid interference from weak, difficult-to-identify wind-induced water level rise signals affecting the calibration accuracy. In one optional implementation, the calm wind threshold is set to 2.0 m / s, and the windy threshold is set to 4.0 m / s.

[0258] Step 702: For windy samples, use the candidate wind-induced water level increase coefficient to restore the static water level, obtain the corresponding restored static water level, and calculate the restoration error between the restored static water level and the true static water level at the corresponding time.

[0259] After obtaining the test set and calibration benchmark, candidate wind-induced water level coefficients are extracted within a preset initial parameter space. For each windy sample, the candidate wind-induced water level coefficient is substituted into the static water level restoration process described in Example 1 to remove the wind field tilt effect at that moment and calculate the corresponding restored static water level. In the time dimension, the quasi-calm wind sample with the closest timestamp to the windy sample is matched, the corresponding true static water level is extracted, and the root mean square error between the restored static water level and the true static water level is calculated. This error is used as a restoration error index to evaluate the fitting accuracy of the candidate wind-induced water level coefficient.

[0260] Step 703: Within the set one-dimensional search interval, with the goal of minimizing the restoration error, search for the optimal comprehensive wind-water increase coefficient; if the optimal value obtained by the search falls on the boundary of the one-dimensional search interval, then adaptively expand the search interval in the direction of the boundary and search again until the optimal value leaves the boundary.

[0261] For the parameter optimization process, a one-dimensional optimization algorithm is used to perform an iterative search within a given parameter search interval. To prevent the algorithm from getting trapped in local optima or causing parameter truncation due to improper initial interval settings, conditional branch logic is executed before outputting the optimization results:

[0262] In the first case, if the optimal value output by the one-dimensional search algorithm is inside the search interval, it indicates that the current interval has covered the true parameter extreme point. The optimal value is directly extracted and locked as the comprehensive wind-water increase coefficient used in the current forecast period.

[0263] In the second scenario, if the optimal value output by the one-dimensional search algorithm is exactly equal to the upper or lower bound of the search interval, it indicates that the true extreme point of the parameter may be located outside the current interval. In this case, the search interval is expanded towards the boundary according to the set step size, triggering a new round of iterative search. This expansion and re-search process is repeated until the optimal value output falls inside the new interval and is no longer located at the boundary, thus ensuring the global optimality of the calibration result.

[0264] Step 704: In daily forecasting operations, the forecast error dual index is calculated based on the forecast water level generated from the daily rolling forecast and the corresponding measured water level. When the forecast error dual index exceeds the preset threshold for multiple consecutive forecast periods, the sample in the latest sliding window is extracted as the new historical hydrological and meteorological sample, and the calibration of the comprehensive wind-induced water level coefficient is re-executed. The forecast error dual index includes the mean absolute error and the root mean square error.

[0265] During the online operation phase of the system, in order to address parameter applicability drift caused by changes in water surface roughness or seasonal succession, a closed-loop optimization mechanism driven by forecast error is established. After the daily forecast task is completed, the measured water levels of each representative station are collected and compared with the corresponding forecast water levels. Conventional statistical methods are used to calculate the mean absolute error and root mean square error, which are used as dual indicators of forecast error.

[0266] The system performs time-series monitoring of the two forecast error indicators. When the two forecast error indicators exceed the preset threshold within a single forecast period, a data quality check prompt is output, and the current parameters remain unchanged. When the two forecast error indicators continuously exceed the preset threshold for multiple consecutive forecast periods, it is determined that the current comprehensive wind-water increase coefficient has systematically failed. At this time, the system automatically extracts the data from the sliding time window closest to the current time according to the set time length, uses it as a new historical hydrological and meteorological sample, and automatically triggers steps 701 to 703 to recalibrate and update the comprehensive wind-water increase coefficient, thereby giving the forecast system the ability to self-correct and adaptively evolve.

[0267] As an example, the method of this invention was applied to historical typhoon events for verification of a large, shallow lake, using hourly water levels at several representative stations around the lake as verification targets. During the strong wind period affected by the typhoon, the average absolute error of the 24-hour forecasts at each representative station was at the centimeter level, and the calculation time for a single complete forecast was at the second level. These results demonstrate that the method of this invention can achieve station-level water level forecasts during strong wind periods while maintaining high computational timeliness. Those skilled in the art will understand that the specific forecast accuracy may vary depending on the lake morphology, meteorological forcing intensity, and the quality of the input data.

[0268] This invention constructs a decoupled-recursive-superposition reversible computational architecture, which effectively reduces the pollution of wind-induced tilt to the initial water volume calculation. By independently superimposing station-level meteorological forcing, it achieves refined differential output of water levels at each station while maintaining high timeliness of algebraic operations.

[0269] By introducing a drag coefficient saturation mechanism based on the hyperbolic tangent function and an absolute water depth constraint boundary, the theoretical wind-induced water increase converges smoothly in the high wind speed range, and the constraint prediction results do not exceed the fluid dynamics limit boundary.

[0270] An exponential relaxation recursive model considering shallow water gravity wave velocity was constructed and a parameterized typhoon wind field downscaling inversion mechanism was integrated to reasonably characterize the time memory effect of lake water and the local spatial differences of eccentric wind field forcing.

[0271] A three-layer intelligent diagnosis and dynamic continuation fault tolerance mechanism with cross-variable joint judgment is established, supplemented by smooth interpolation in high gradient areas, weighted fine estimation of rainfall without wind and water volume, and closed-loop adaptive rolling update of parameters. This enhances the forecast system's anti-interference capability and long-term adaptive evolution capability under complex meteorological conditions to the level of operational reliability.

[0272] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for hourly forecasting of water levels in shallow lakes during periods of strong winds, characterized in that, include: Obtain the measured hydrological and meteorological data at the current moment; Based on measured hydrological and meteorological data, the tilting and forcing effects of the current wind field and pressure field on the water level are removed, the initial static water level of each representative station is restored, and the initial static water level of the entire lake is extracted. Obtain the driving factors of water balance under windless conditions; Starting from the initial still water level of the entire lake, and combining the water balance driving factors, hourly water balance is recursively extrapolated to obtain the sequence of the representative still water level of the entire lake in the foreseeable future period. Obtain forecast meteorological data for the foreseeable future; Based on forecast meteorological data, the forecast wind field forcing effect and forecast pressure forcing effect are calculated for each representative station at each future time. Using the initial still water level as a baseline, the change in the representative still water level sequence of the entire lake relative to the initial still water level of the entire lake is superimposed, and the corresponding forecast wind field forcing effect and forecast air pressure forcing effect are superimposed respectively to synthesize the hourly forecast water level sequence of each representative station.

2. The method according to claim 1, characterized in that, When acquiring measured hydrological and meteorological data at the current moment, data quality is monitored through an anomaly diagnosis module, including: Perform item-by-item validity checks on the measured hydrological and meteorological data and generate anomaly records; Perform spatiotemporal correlation analysis on abnormal records and execute root cause diagnosis logic according to a preset priority order to determine the root cause of the abnormal records; Based on the root causes, the persistent characteristics of abnormal records, and the multi-station synchronization characteristics, hierarchical early warning information is output, and corresponding forecasting and intervention strategies are implemented for different hierarchical early warning information.

3. The method according to claim 1, characterized in that, The tilt forcing effect includes barometric pressure increase and wind-induced water increase; the forecast wind field forcing effect is the forecast wind-induced water increase, and the forecast barometric pressure forcing effect is the forecast barometric pressure increase. Both barometric pressure rise and forecast barometric pressure rise are determined based on the relative change of pressure data with respect to the baseline pressure. Both wind-induced water level rise and forecast wind-induced water level rise are determined based on the effective wind zone length, wind direction factor, and wind speed. The wind direction factor is the cosine of the angle between the wind direction and the maximum water level rise azimuth corresponding to the representative station.

4. The method according to claim 3, characterized in that, When obtaining the effective wind zone length, a gradient adaptive interpolation strategy is adopted, including: Based on the adjacent azimuth of the wind direction, obtain the length of the corresponding azimuth wind zone. Calculate the ratio of the lengths of the wind zones at adjacent azimuth points as the gradient ratio; If the gradient ratio is not greater than the set threshold, then linear interpolation is used to obtain the effective wind zone length. If the gradient ratio is greater than the set threshold, cosine weighted interpolation is used to obtain the effective wind zone length.

5. The method according to claim 3, characterized in that, When calculating or forecasting wind-induced water level rise, the following should be included: Based on the drag coefficient saturation mechanism, the actual wind speed is converted into the saturated equivalent wind speed. The saturated equivalent wind speed increases linearly in the low wind speed range and converges asymptotically in the high wind speed range. The theoretical wind-induced water increase is calculated based on saturated equivalent wind speed, effective wind zone length, and wind direction factor. Obtain the average water depth of the lake and set the physical constraint boundary of the absolute water depth in combination with the preset safety factor; The theoretical wind-water increase is cut off to the physical constraint boundary of absolute water depth to obtain the final or predicted wind-water increase.

6. The method according to claim 5, characterized in that, When calculating or forecasting wind-induced water rise, it is obtained through a dynamic response recursive model, including: The response time constant for the corresponding wind direction is determined based on the shallow water gravity wave velocity and the effective wind zone length. The dimensionless relaxation coefficient is determined based on the relationship between the response time constant and the preset time step. The equilibrium wind-water increase at the current moment is calculated based on the quasi-static equilibrium assumption. The dynamic wind-water increase at the previous moment and the equilibrium wind-water increase at the current moment are weighted and recursively calculated using a dimensionless relaxation coefficient to obtain the dynamic wind-water increase at the current moment, which serves as the final or predicted wind-water increase.

7. The method according to claim 1, characterized in that, The driving factors of water balance include precipitation on the lake surface, evaporation on the lake surface, water intake around the lake, and net inflow of water around the lake; hourly water balance is calculated by accumulating the water changes of each driving factor of water balance hourly.

8. The method according to claim 7, characterized in that, The net inflow of water around the lake was obtained through the following steps: The water level difference between the representative station of each zone and the lake is calculated, and the water level difference is normalized by using the independent historical extreme values ​​of each zone to obtain the dimensionless water level difference index. The current dimensionless water level difference index, the historically lagged dimensionless water level difference index, and the historically lagged dimensionless net inflow index are input into a pre-constructed multiple linear regression model to estimate the daily total net inflow of water around the lake. Based on the uniform distribution of the daily net inflow of water around the lake, and taking into account the difference between the hourly rainfall intensity and the daily average rainfall intensity in the foreseeable future period, the daily net inflow of water around the lake is dynamically distributed through a preset rainfall weighting coefficient. After total amount conservation correction, the hourly net inflow sequence around the lake is obtained.

9. The method according to claim 1, characterized in that, When acquiring the measured hydrological and meteorological data at the current moment, a data interruption fault tolerance strategy is also included: If the measured hydrological and meteorological data at the current moment is missing and cannot be supplemented by interpolation of data from adjacent moments, then backtrack to the historical moment to obtain the last continuous and valid measured hydrological and meteorological data. Based on the time difference between the timestamp of the last consecutive valid measured hydrological and meteorological data and the target forecast time, the forecast start time is dynamically extended and the forecast duration is shortened to ensure the continuous execution of the forecast process.

10. A system for hourly forecasting of water levels in shallow lakes during periods of strong winds, characterized in that, include: At least one processor; as well as, A memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by the processor to implement the method according to any one of claims 1 to 9.