Plain tidal river network internal gate backflow time period determination method

Abstract

The invention discloses a method for judging a backflow time period of a gate in a plain tidal river network. The method comprises the following steps: acquiring water gate upstream and downstream long-series observation water level, drainage basin rainfall and wind field data; astronomical tide and rainfall runoff decoupling is carried out on the downstream section water level, dynamic deformation characteristic parameters are extracted and corrected, and a downstream water level forecasting curve set is obtained; physically decomposing the observation water level of the upstream shallow lake, extracting the average water level and the wind-generated water level deviation, and calculating an upstream forecast curve set in combination with rainfall; combining the forecast curve set to construct a predicted water level difference sequence set; extracting a prediction error quantile based on a pre-constructed error condition distribution library, and performing lower bound calibration and minimum extraction on the difference sequence to obtain a partial safety fusion water level difference sequence; a backflow risk period is identified based on the fused sequence and a decision threshold. According to the method, time consumption of complex numerical solution is avoided, and timeliness and robustness of advanced early warning in the backflow risk time period are improved.

Description

Technical Field

[0001] This invention relates to the fields of water conservancy project scheduling forecasting and data mining technology, and in particular to a method for determining the backflow period of gates in plain tidal river networks. Background Technology

[0002] In plain tidal river networks, sluice gate operation primarily serves flood control, drainage, and water resource allocation within the basin and region. Sluice gate backflow is a critical anomaly requiring close monitoring during operation, as it can easily lead to engineering safety risks and the failure of operational objectives. Achieving high-fidelity simulation of upstream and downstream water level evolution within a predictable timeframe and providing early warning of water level gradient reversal intervals can enhance the river network's ability to securely allocate water resources.

[0003] Existing sluice gate backflow warning and water level prediction technologies mainly fall into two categories: one is river network hydrodynamic model simulation, and the other is traditional experience-based judgment based on real-time monitoring. Hydrodynamic models require numerical simulation to solve partial differential equations to deduce the water level process at each cross-section. While the physical mechanism of this method is clear, it suffers from high computational complexity, is time-consuming, and relies heavily on extensive measured hydrological and meteorological data for model calibration and verification. On the other hand, traditional methods based on real-time monitoring and human experience primarily depend on upstream and downstream real-time water level data, combined with the experience of dispatchers, resulting in problems such as delayed response and insufficient predictability.

[0004] In summary, existing methods struggle to balance computational speed and forecast robustness when facing dynamic disturbances from multiple meteorological and environmental factors, leading to significant uncertainties in extreme risk identification during real-time operational scheduling. Therefore, it is necessary to research methods that can improve the response speed of hydrological forecasts in plain tidal river networks and enhance the robustness of extreme risk decision-making. Summary of the Invention

[0005] Purpose of the invention: In view of the above-mentioned problems of the prior art, this application provides a method for determining the backflow period of sluice gates in plain tidal river networks.

[0006] Technical solution: According to one aspect of this application, a method for determining the backflow period of a sluice gate in a plain tidal river network includes:

[0007] Acquire long-term observation water levels at the downstream section of the sluice gate, multi-station observation water levels in upstream shallow lakes, watershed rainfall data, and wind field data;

[0008] Based on long-term observation of water level and basin rainfall data, the water level at the downstream section is decoupled from astronomical tides and rainfall runoff. Dynamic deformation characteristic parameters are extracted to correct the decoupled waveform, resulting in a set of downstream water level forecast curves.

[0009] Physical decomposition was performed based on multi-station observed water level and wind field data to extract the true average water level and wind-induced water level deviation, and combined with watershed rainfall data to obtain a set of upstream water level forecast curves;

[0010] Combine it with the upstream water level forecast curve set and the downstream water level forecast curve set to construct a predicted water level difference sequence set;

[0011] The error condition distribution library is called to extract the prediction error quantiles corresponding to the current meteorological and hydrological scenario. The prediction error quantiles are then used to perform lower bound calibration and minimum value extraction on the predicted water level difference sequence set to obtain a partially safe fused water level difference sequence.

[0012] Based on this (the partial safety fusion water level difference sequence) and the preset backflow judgment threshold, the backflow risk period that meets the preset duration is identified.

[0013] Beneficial effects: This invention avoids the time-consuming complex numerical solutions and improves the timeliness and robustness of early warning for backflow risk periods. The relevant technical effects will be described in detail below with reference to specific embodiments. Attached Figure Description

[0014] Figure 1 A flowchart illustrating a method for determining the backflow period of a sluice gate in a plain tidal river network, provided as an embodiment of this application.

[0015] Figure 2 This is a flowchart for decoupling astronomical tides and rainfall runoff from downstream cross-sectional water levels, provided as an embodiment of this application.

[0016] Figure 3 The flowchart for extracting dynamic deformation feature parameters to correct the decoupled waveform and obtain a set of downstream water level forecast curves is provided in the embodiments of this application.

[0017] Figure 4 The flowchart describes the construction of a predicted water level difference sequence set based on the combined upstream forecast curve and downstream water level forecast curve set provided in the embodiments of this application.

[0018] Figure 5 The flowchart provided in this application embodiment is for identifying backflow risk periods that meet the preset duration by fusing the water level difference sequence and the backflow determination threshold. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a predetermined order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] To resolve the issue, the applicant conducted in-depth research and analysis, and discovered:

[0022] Specifically, existing methods lack accuracy and timeliness. Traditional methods based on hydrodynamic models have high computational complexity and are time-consuming. They require a large amount of measured data for model calibration and verification in the early stages, and the water level simulation results are uncertain, making it difficult to meet the requirements of flood control scheduling for rapid identification and real-time response to backflow periods.

[0023] On the other hand, the models are not very targeted. Existing hydrodynamic models usually do not incorporate wind field influence modules in the water level forecasting of upstream and downstream of sluice gates. However, in plain river network areas, the terrain is low-lying and the wind-driven flow has a significant impact on water level changes. In particular, during the formation of backflow, ignoring the quantitative effect of wind force can easily lead to deviations in the judgment of backflow situation.

[0024] On the other hand, the lack of data-driven approaches has prevented the systematic exploration of the correlation patterns and empirical relationships between multiple factors such as astronomical tides, rainfall runoff, and wind fields and changes in sluice gate water levels. It has also prevented the construction of prediction models that balance lightweight structure and rapid response, thus limiting the improvement of forecast efficiency and practicality.

[0025] To solve these problems, combined with Figures 1 to 5 The present invention will be specifically described through the following embodiments.

[0026] An example is provided, offering an optional implementation of a method for determining the backflow period of sluice gates within a plain tidal river network, particularly the overall data foundation and decoupled fusion execution architecture. Specifically, this includes:

[0027] Step 101: Obtain long-series observation water levels at the downstream section of the sluice gate, observation water levels at multiple stations in the upstream shallow lakes, watershed rainfall data, and wind field data.

[0028] Among them, the time span of the long series of water level observations at the downstream section of the sluice gate is no less than thirty days, in order to meet the separation requirements of the main astronomical tidal sequence; the time span of the basin rainfall data and wind field data is no less than one hydrological year, in order to cover synchronous observation samples of different typical hydrological periods.

[0029] In plain tidal river network areas, the downstream section of a sluice gate is affected by both the periodic fluctuations of astronomical tides and the runoff generated by rainfall within the basin, while large, shallow lakes upstream, such as Taihu Lake, are prone to water surface tilting driven by local wind fields. In order to capture the hydrodynamic response patterns, specific requirements were put forward regarding the quality and time span of the input data source.

[0030] Specifically, the long series of observed water levels at the downstream section of the sluice gate should be a continuous hourly water level observation sequence with a time resolution of not less than 1 hour, and its continuous time span should not be less than 30 days. Setting a hard data length constraint of more than 30 days is necessary because it is necessary to separate the tidal signals according to the Rayleigh criterion, especially to decouple the semi-diurnal primary lunar equinox and the semi-diurnal primary solar equinox, whose period difference is only about 0.42 hours, to avoid distortion in the harmonic constant calculation due to an excessively short data sequence.

[0031] Furthermore, the upstream shallow lakes are monitored at multiple stations, specifically including synchronous observation data from several representative water level stations distributed in different locations along the lakes. For example, hourly water levels from stations such as Dapukou, Xiaomeikou, Jiapu, Wangtingtai, and Xishan in Taihu Lake and Dongting Lake can be selected. The basin rainfall data consists of daily areal rainfall sequences from representative rain gauge stations covering the entire sluice gate-controlled basin. The wind field data includes hourly wind speed and direction sequences obtained from meteorological observation points near the sluice gate at a standard height of 10 meters.

[0032] Furthermore, the time span of watershed rainfall and wind field data should not be less than one hydrological year, so that the collected samples can comprehensively cover different typical hydrological conditions during the wet, normal, and dry seasons. After obtaining long series of raw data, linear interpolation can be used to fill in missing values, and thresholding or Z-score standardization can be used to remove anomalous jump values ​​that obviously violate physical common sense, ensuring the spatiotemporal continuity and consistency of the input data.

[0033] The Z-score standardization method can also be called the Z-score standardization method.

[0034] Step 102: Based on long-term observation of water level and basin rainfall data, the astronomical tide and rainfall runoff of the downstream section water level are decoupled, and dynamic deformation characteristic parameters are extracted to correct the decoupled waveform, so as to obtain a set of downstream water level forecast curves.

[0035] In this step, the system abandons the traditional hydrodynamic method of superimposing astronomical tides with rainfall runoff for simulation, and instead implements physical deterministic decoupling. Specifically, using a long series of quality-controlled water level observations, a least-squares harmonic analysis algorithm is employed to separate the deterministic astronomical tide level fluctuation sequence.

[0036] Simultaneously, based on the statistical mapping relationship between the collected long-term basin rainfall data and the corresponding historical water level fluctuations, the non-periodic water level response sequence generated by rainfall runoff is estimated. On this basis, using the latest measured hydrological feedback data acquired before the forecast period, dynamic deformation characteristic parameters reflecting the current river network regulation and hydrodynamic environment are extracted. These dynamic deformation characteristic parameters are used to quantify the tidal wave propagation phase shift and energy attenuation amplitude caused by previous rainfall accumulation or regulation.

[0037] Next, the extracted dynamic deformation feature parameters are used to perform time-series translation and amplitude scaling on the decoupled astronomical tide prediction waveform, and then combined and superimposed with different modes of the rainfall-runoff response sequence to generate a set of downstream water level forecast curves containing multiple differentiated correction modes. The construction of the multi-process line set preserves the uncertainty boundary characteristics of the downstream forecast water level.

[0038] Step 103 involves physical decomposition based on multi-station observed water level and wind field data to extract the true average water level and wind-induced water level deviation. This, combined with basin rainfall data, yields a set of upstream water level forecast curves. This addresses the issue that single-station water levels in large, shallow lakes upstream of sluice gates are insufficient to represent the overall storage capacity under wind conditions.

[0039] In this embodiment, a physical decomposition mechanism for the observed water levels at the monitoring stations is implemented. Specifically, based on the steady-state wind-driven current theory of shallow lakes, the observed water levels at each station are decomposed into the true average water level reflecting the overall water resource changes of the lake, and the wind-driven water level deviation influenced by the local wind field at each station. By analyzing the wind speed and direction components in the wind field data and constructing parameterized equations based on the geometric spatial coordinates of each station, the true average water level benchmark, which eliminates wind field noise, can be jointly derived from the observed water levels at multiple stations.

[0040] During the forecast period, the dynamic rise in the lake's true average water level is estimated by combining basin rainfall data, and the predetermined wind-induced water level deviation caused by forecast wind conditions at the target control section is superimposed, thereby generating a set of upstream water level forecast curves for the sluice gate. This process avoids the need for a global gridded hydrodynamic solution, which can improve the computational efficiency of the upstream control section water level.

[0041] Step 104: Combine it with the (upstream water level forecast curve set) and the downstream water level forecast curve set to construct a predicted water level difference sequence set.

[0042] After obtaining the upstream and downstream water level forecasts separately, the system performs a curve combination operation. Each candidate curve in the downstream water level forecast curve set is paired with a forecast curve in the upstream water level forecast curve set, and the difference is calculated to obtain multiple sets of hourly evolving upstream and downstream water level difference sequences within the forecast period. These sequences are then compiled into a predicted water level difference sequence set. Each curve in this sequence set represents the evolution trajectory of the upstream and downstream water pressure gradient under different tidal correction assumptions.

[0043] Step 105: Call the error condition distribution library to extract the prediction error quantiles corresponding to the current meteorological and hydrological scenario. Use the prediction error quantiles to perform lower bound calibration and minimum value extraction on the predicted water level difference sequence set to obtain the partially safe fused water level difference sequence.

[0044] In other words, the prediction error quantiles corresponding to the current meteorological and hydrological scenario are extracted based on the pre-constructed error condition distribution library. The prediction error quantiles are then used to perform lower bound calibration and minimum value extraction on the predicted water level difference sequence set to obtain a partially safe fused water level difference sequence.

[0045] Alternatively, based on the hydrological and meteorological conditions within the forecast period corresponding to the predicted water level difference sequence set, the corresponding prediction error quantiles are matched and extracted from the pre-constructed error condition distribution library. The prediction error quantiles are then used to perform lower bound calibration and minimum value extraction on the predicted water level difference sequence set to obtain a partially safe fused water level difference sequence.

[0046] This step abandons the arithmetic mean fusion criterion and instead transforms a multi-curve set into a single scheduling decision criterion. Specifically, the system extracts tidal phase, rainfall level, and wind field quadrant features for the current forecast period to form meteorological and hydrological scenario labels. Next, it matches and retrieves the corresponding statistical prediction error quantiles under each scenario label from an error condition distribution library pre-built offline using long-series historical backcalculation data.

[0047] Since the risk of backflow occurs under unfavorable conditions such as small or negative water level differences, the selected prediction error quantile is biased towards the lower bound of the negative tail of the error distribution, for example, a quantile level of 0.05 is chosen. This quantile is superimposed on each sequence in the predicted water level difference sequence set to perform lower bound calibration. At the same time, the minimum values ​​of all calibrated sequences are extracted for envelope fusion. The resulting partially safe fused water level difference sequence strictly constitutes the maximum lower bound of the backflow risk in terms of mathematical expectation, thus avoiding the risk underreporting problem caused by average fusion from the algorithmic mechanism.

[0048] Step 106: Based on the partial safety fusion water level difference sequence and the preset backflow judgment threshold, identify the backflow risk period that meets the preset duration.

[0049] In other words, based on this (the partial safety fusion water level difference sequence) and the backflow determination threshold, the backflow risk period that meets the preset duration can be identified.

[0050] As the output of the early warning judgment, the system compares the partially safe fused water level difference sequence obtained in the previous steps with the preset backflow judgment threshold hourly. For example, in the actual scheduling operation of the Pingwang Water Conservancy Project, the backflow judgment threshold can be set to -0.05m. When the value of the fused water level difference sequence drops to -0.05m or below, a single-moment backflow early warning state is triggered.

[0051] Furthermore, the system slides along the forecast timeline to search for time windows that are continuously in the warning state. If the length of the continuous time window reaches or exceeds a preset duration, such as 5 consecutive hours, then the time window is determined to be a high-confidence risk interval. Based on this, the system outputs the start and end timestamps of the risk interval as the backflow risk period and provides it to the flood control dispatchers of the sluice gates to support preventive dispatching decisions such as pre-discharge or gate closure.

[0052] Another example describes a specific implementation method for downstream forecasting and recursive drift correction based on full waveform alignment, particularly a method for decoupling deterministic and empirical components in downstream water level forecasting of sluice gates in plain tidal river networks, and an implementation scheme for tidal recursive drift correction based on full waveform alignment. Based on the overall data foundation, this embodiment separates and corrects the multi-source influences on the downstream section. Specifically:

[0053] Step 201: Based on the long series of observed water levels, extract recent measured data and perform astronomical tide harmonic analysis to separate the periodic fluctuation components and obtain the astronomical tide prediction sequence.

[0054] In plain tidal river networks, water levels at key control sections downstream of sluice gates exhibit periodic fluctuations. To extract deterministic components, the system uses recent measured water level data of a predetermined length prior to the current prediction time to form a training window. The harmonic analysis algorithm is then used to perform least-squares fitting on the water level data within this training window, separating the amplitude and lag parameters of the lunar semi-diurnal tide, solar semi-diurnal tide, lunar diurnal tide, and solar diurnal tide. Based on the extracted tidal parameters, the theoretical astronomical tide levels for each hour within the forecast period are extrapolated, outputting an astronomical tide prediction sequence. This sequence forms the baseline periodic framework for subsequent downstream water level forecasts.

[0055] Step 202: Estimate the runoff response components based on watershed rainfall data and pre-constructed rainfall empirical process relationships to obtain the rainfall-runoff water level variation sequence.

[0056] To address the non-periodic rise or fall of water levels caused by rainfall runoff, this step utilizes previously acquired rainfall data, such as areal rainfall, and substitutes it into an empirical relationship model established in advance through historical long-series statistics. Specifically, univariate or multivariate linear regression is used to fit the relationship between rainfall and daily water level rise, as well as the relationship between flood peak water level and daily receding rate, to calculate the overall water level fluctuation at the target station on a daily scale.

[0057] In the calculation of hourly process curves during the rising and receding periods of rainfall, the system adopts a unified allocation mechanism that evenly distributes the total daily increase or decrease over all 24 hours of the day to meet the real-time dispatch response requirements for hourly water levels within the forecast period. In plain river network areas, hourly water level forecasts based on daily rainfall inherently involve uncertainties. The uniform allocation mechanism simplifies the hourly dynamic calculation of runoff generation and confluence processes, improving the efficiency of engineering calculations. Consequently, hourly process deviations are comprehensively constrained and compensated by subsequent multi-scheme tidal correction and multi-process curve fusion judgment rules. The data sequence generated based on the uniform allocation mechanism is the rainfall-runoff water level variation sequence.

[0058] Step 203: Within the set rolling window, extract the rainfall-runoff water level variation sequence from the long series of observed water levels and extract the tidal residual sequence after removing the runoff influence.

[0059] The system uses a sliding time window, covering one or more complete tidal cycles, to assess the physical distortions caused by recent changes in watershed hydrology on tidal propagation in the river network. Within this window, the system extracts hourly measured water level observation sequences and performs a subtraction stripping operation using estimated rainfall-runoff water level amplitude sequences at the corresponding times. The specific calculation formula is as follows:

[0060] R _t (t)=Z _obs (t)-Z _rain (t);

[0061] Among them, R _t (t) represents the tidal residual sequence value at time t, Z _obs (t) represents the actual extracted water level value from a long series of observations at time t, Z _rain (t) is the estimated value of the rainfall-runoff water level variation sequence at time t.

[0062] This stripping operation suppressed the basement uplift interference caused by the large-scale runoff inflow into the basin. The obtained tidal residual sequence can reflect the actual hydrodynamic tidal wave characteristics of the current river network, providing a data source for subsequent full waveform dynamic alignment.

[0063] Step 204: Compare the tidal residual sequence with the astronomical tide prediction sequence to extract dynamic deformation feature parameters used to characterize phase and amplitude deviations.

[0064] Alternatively, by comparing the tidal residual sequence with the astronomical tide prediction sequence, dynamic deformation characteristic parameters can be extracted to characterize phase deviation, amplitude deviation, and mean deviation.

[0065] Accordingly, this step can be specifically processed by recursive drift correction with full waveform alignment, including: within a preset shift set, calculating the sliding correlation coefficient between the tidal residual sequence and the astronomical tide prediction sequence, and extracting the shift value corresponding to the maximum correlation coefficient as the phase drift parameter.

[0066] Traditional tidal time correction relies solely on comparing time differences between single high or low tide extremes, making it susceptible to interference from local winds, waves, or measurement noise. This step employs a full waveform alignment architecture, constructing a shift set containing multiple integer-hour shifts. The preset shift set can be set to cover a discrete interval from -6 hours to +6 hours. For each shift value in this set, the normalized Pearson correlation coefficient between the astronomical tide prediction sequence after the slip shift and the baseline tidal residual sequence is calculated. The specific calculation formula is as follows:

[0067] Corr(s) = ∑[(R _t (t)-R _mean )×(Z _tide (t+s)-Z _mean )] / sqrt[∑(R _t (t)-R _mean ) 2 ×∑(Z _tide (t+s)-Z _mean ) 2 ];

[0068] Where Corr(s) is the normalized sliding correlation coefficient calculated under the current shift s, t is the discrete sampling time within the rolling window, and R _t (t) represents the tidal residual sequence value at time t, R _mean Z is the arithmetic mean of the tidal residual sequence within the rolling window. _tide (t+s) represents the astronomical tide prediction sequence value after shifting by s time steps, Z _mean ∑ represents the arithmetic mean of the astronomical tide prediction sequence within the calculation window after the translation, and ∑ represents the summation operation performed on all discrete sampling times within the rolling window.

[0069] After traversing the preset shift set and calculating all normalized sliding correlation coefficients, the system compares and extracts the correlation coefficient with the largest value, and determines the specific shift value corresponding to the largest correlation coefficient as the phase drift parameter. This parameter characterizes the overall time lead or lag of the actual tidal wave relative to the theoretically harmonic constant-derived tidal wave.

[0070] Optionally, the astronomical tide prediction sequence is translated and aligned based on the phase drift parameter, and the standard deviation ratio and mean difference between the aligned astronomical tide prediction sequence and the tidal residual sequence are calculated to obtain the amplitude drift parameter and the mean drift parameter.

[0071] After achieving optimal correlation alignment of the waveforms on the time axis, the system further quantifies the attenuation or amplification effect of river network regulation on tidal wave energy and the overall system's baseline deviation. The system calculates the first standard deviation of the tidal residual sequence within the current window, and simultaneously calculates the second standard deviation of the astronomical tide prediction sequence after phase shift alignment within the same window. The ratio of the first standard deviation to the second standard deviation is extracted as the amplitude drift parameter. Further, the mean of the tidal residual sequence is calculated, and the product of the amplitude drift parameter and the mean of the phase-shifted astronomical tide prediction sequence is subtracted from it; the difference is extracted as the mean drift parameter. The specific calculation formula is as follows:

[0072] α _1 =σ _R / σ _Z ;

[0073] β=R _mean -α _1 ×Z _mean ;

[0074] Where, α _1 σ is the amplitude drift parameter. _R σ is the standard deviation of the tidal residual series. _Z R is the standard deviation of the astronomical tide prediction sequence after phase shift alignment, β is the mean shift parameter, and R is the mean shift parameter. _mean This is the arithmetic mean of the tidal residual sequence.

[0075] In other words, β corresponds to the mean shift parameter that characterizes the true average water level and the wind-induced water level rise effect.

[0076] Optionally, the phase drift parameter, amplitude drift parameter, and mean drift parameter are updated recursively and smoothly to output dynamic deformation characteristic parameters.

[0077] Accordingly, the system performs time-series exponential smoothing on each calculated parameter to avoid drastic fluctuations in correction parameters caused by occasional data anomalies within a single rolling window, thus ensuring the robustness of the continuously rolling forecast model. A pre-configured smoothing coefficient is introduced to perform a weighted fusion calculation between the historical parameters stored in the system from the previous update time and the newly extracted parameters for the current time window. Taking the update of the amplitude drift parameter as an example, the specific calculation formula is as follows:

[0078] α _1 (t0)=λ×α _1(t0-1)+(1-λ)×α _1_new ;

[0079] Where, α _1 (t0) represents the amplitude drift parameter calculated at the current update time, λ is the preset smoothing coefficient within the open interval of 0 to 1, and α _1 (t0-1) represents the amplitude drift parameter recorded by the system at the previous update time, α _1_new This is the amplitude drift parameter newly extracted based on the standard deviation ratio for the current calculation window. In other words, t0 corresponds to the current update time, and t0-1 corresponds to the previous update time.

[0080] Assuming the normalization smoothing coefficient is set to 0.8, the historical amplitude drift parameter at the previous rolling update time is 1.05, and the newly extracted amplitude drift parameter in the current window is 1.15, substituting these values ​​into the smoothing formula yields the updated amplitude drift parameter as 0.8 × 1.05 + 0.2 × 1.15 = 1.07. The phase drift parameter and mean drift parameter are updated and calculated separately using a consistent exponential smoothing mechanism. All parameters that have undergone the update operation are then aggregated to form the dynamic deformation characteristic parameters.

[0081] Step 205: Combine and correct the dynamic deformation characteristic parameters, astronomical tide prediction sequence and rainfall-runoff water level amplitude sequence to generate a set of downstream water level forecast curves containing multiple correction modes.

[0082] This step specifically includes:

[0083] The astronomical tide prediction sequence is directly superimposed with the rainfall-runoff water level variation sequence to generate an uncorrected baseline curve.

[0084] By using dynamic deformation characteristic parameters to perform single amplitude scaling and single phase shift on the astronomical tide prediction sequence, and then superimposing the rainfall-runoff water level amplitude sequence, corresponding amplitude adjustment curves and phase adjustment curves are generated.

[0085] The astronomical tide prediction sequence is simultaneously subjected to amplitude scaling and phase shifting using dynamic deformation feature parameters. This is then superimposed with the rainfall-runoff-water level amplitude sequence and integrated with mean drift compensation to generate a dynamic comprehensive correction curve.

[0086] The uncorrected baseline curve, amplitude adjustment curve, phase adjustment curve, and dynamic comprehensive correction curve are combined to form a downstream water level forecast curve set.

[0087] In this step, the decoupled and extracted base sequence and dynamic deformation feature parameters are re-integrated and calculated in multiple dimensions, which can provide physically representative differentiated boundary inputs for subsequent risk fusion based on the error library. Specifically, the first curve is obtained by adding the astronomical tide prediction value and the runoff amplitude value corresponding to the time series scale to obtain the uncorrected baseline curve.

[0088] Furthermore, the second curve introduces tidal range attenuation or amplification constraints. By multiplying the amplitude drift parameter by the mean-based residual term of the astronomical tide prediction sequence, adding back the original mean, and then superimposing the runoff amplitude value, the amplitude adjustment curve is calculated.

[0089] Based on this, the third curve introduces a time axis propagation delay constraint, uses the phase drift parameter to perform translation compensation on the time series index of the astronomical tide prediction sequence, extracts the astronomical tide prediction value at the corresponding translation time, directly superimposes the runoff amplitude value, and calculates the phase adjustment curve.

[0090] Furthermore, the fourth curve integrates all distortion feature constraints, performs amplitude scaling calculations on the astronomical tide prediction values ​​after phase shift compensation, adds back the original mean and mean drift parameter, and superimposes the runoff amplitude value, thereby generating a dynamic comprehensive correction curve that covers the most complete correction information.

[0091] Furthermore, the four forecast sequences generated based on different hydrodynamic assumptions are output in a unified manner and combined to form a set of downstream water level forecast curves.

[0092] In some implementations, for computing devices with limited computing resources or high requirements for real-time flood warning, the preset shift set can be configured with a larger time step size, significantly reducing the number of traversal operations for the sliding correlation coefficient; or a downsampling feature extraction algorithm can be pre-applied to discretize and sparsify the tidal residual sequence and the astronomical tide prediction sequence before performing a dimensionality-reduced full waveform correlation comparison calculation. Introducing such a computational truncation strategy can further reduce the running time of the correction parameter extraction module while sacrificing a small amount of phase alignment accuracy.

[0093] In this embodiment, a multi-station wind field vector parameter decoupling and a full-waveform tidal recursive drift correction mechanism are introduced to address the prediction bias caused by dynamic disturbances from multiple meteorological environmental factors. This mechanism filters out local wind field physical noise and assimilates the tidal wave distortion characteristics of the water network in real time, enabling the tracking of real hydrodynamic evolution under extreme meteorological disturbances and reducing the interference of multi-source environmental noise on the model input quality.

[0094] Another example describes a possible implementation method for downstream forecast correction based on multi-tidal cycle extreme points weighting. This embodiment provides a lightweight data processing method with lower computational resource consumption through feature extraction of discrete extreme points. Specifically, it can be implemented using the following steps:

[0095] Step 301: Identify the high tide extreme events and low tide extreme events of a preset number of consecutive semi-diurnal tide cycles from the tidal residual sequence and the astronomical tide prediction sequence, respectively.

[0096] After acquiring the tidal residual sequence after removing the runoff influence and the corresponding astronomical tide prediction sequence, the system independently performs local extremum retrieval operations on the two sets of time series data. Specifically, for a preset backtracking time window before the current prediction time, the system identifies local maxima and local minima of the water level in the discrete water level time series by performing first-order difference calculations and determining sign zero-crossing points. These extrema correspond to high tide and low tide extreme events at predetermined timestamps. Several consecutive semi-diurnal tidal cycles from the past can be extracted by setting a time threshold. It is recommended to obtain data from more than four semi-diurnal tidal cycles to ensure that the sample size is statistically significant and can fully cover the cyclical characteristics of the phase evolution of spring tides or neap tides.

[0097] In other implementations, considering that there may be local high-frequency water surface disturbances in the measured water level sequence caused by instantaneous gusts or physical measurement errors, the system can use a low-pass filtering algorithm to smooth the original tidal residual sequence before identifying extreme events, remove high-frequency noise components, and prevent transient disturbances from being misjudged as tidal extreme events.

[0098] Step 302: For each semi-diurnal tide cycle, calculate the tidal range scaling ratio and tidal time deviation between the tidal residual sequence and the astronomical tide prediction sequence.

[0099] In other words, for each semi-diurnal tide cycle, the tidal range scaling ratio and tidal time deviation of the tidal residual sequence and the astronomical tide prediction sequence are calculated based on the high tide extreme event and the low tide extreme event.

[0100] After extracting the extreme event sets for each semi-diurnal tidal cycle, the system calculates the geometric morphological deviation parameters between the measured residual tidal wave and the theoretically predicted tidal wave for each discrete semi-diurnal tidal cycle. Specifically, it extracts the high tide extreme event water level values ​​and low tide extreme event water level values ​​corresponding to that cycle from the tidal residual sequence, and calculates the absolute value of the difference between them as the measured tidal range. Similarly, it extracts the high tide extreme event water level values ​​and low tide extreme event water level values ​​for the corresponding cycle from the astronomical tide prediction sequence, and calculates the absolute value of the difference between them as the predicted tidal range. The ratio of the measured tidal range to the predicted tidal range is calculated, and this ratio is extracted as the tidal range scaling factor for the predetermined semi-diurnal tidal cycle.

[0101] Furthermore, the timestamps of the high tide extreme events in the astronomical tide prediction sequence and the corresponding high tide extreme events in the tidal residual sequence are extracted, the absolute time difference between the two is calculated, and this time difference is extracted as the tidal time deviation of the predetermined semi-diurnal tide cycle.

[0102] As an example, if in the first semi-diurnal tidal cycle, the measured high tide level is 1.2 meters and the measured low tide level is 0.4 meters, the corresponding measured tidal range is calculated to be 0.8 meters; while the predicted high tide level for this cycle is 1.0 meter and the predicted low tide level is 0.5 meters, the corresponding predicted tidal range is calculated to be 0.5 meters. The resulting tidal range scaling factor is 0.8 / 0.5 = 1.6. This value is greater than 1, quantifying the physical amplification effect of the current river network hydrodynamic environment on the tidal wave energy of this cycle.

[0103] Step 303: Based on the time span from the current prediction time, assign exponential decay weights to the tidal range scaling ratio and tidal time deviation of each semi-diurnal tidal cycle and perform weighted summation to output dynamic deformation characteristic parameters.

[0104] Among these factors, the relevance of tidal wave distortion characteristics from different historical periods to the current prediction time exhibits a non-linear decreasing trend over time. Based on this, the system introduces an exponential decay weighting mechanism based on temporal distance. Specifically, the system calculates the corresponding exponential decay weighting coefficient based on the number of intervals between each semi-diurnal tidal cycle and the current prediction time. The specific calculation formula is as follows:

[0105] w _j =exp(-λ _1 ×(Mj));

[0106] Among them, w _j The exponentially decaying weights assigned to the j-th semi-diurnal cycle, exp, correspond to the natural exponential function, λ. _1 M is the preset attenuation coefficient, M is the total number of historical semi-diurnal tide cycles extracted by the system, j is the time sequence index of the currently calculated semi-diurnal tide cycle, and (Mj) represents the number of semi-diurnal tide cycles that the extreme event spans from the current prediction time.

[0107] After calculating the exponential decay weight for each period, the system performs a weighted summation calculation on the tidal range scaling ratio for each period, and simultaneously performs a weighted summation calculation on the tidal time deviation for each period. The converged weighted calculation results are packaged and extracted as dynamic deformation feature parameters for output.

[0108] The preset attenuation coefficient can be set to 0.2 to ensure that recent extreme events with shorter time spans have a dominant weight, while also taking into account the benchmark smoothing effect of long-term extreme events.

[0109] In another embodiment, if the amount of historically accumulated extreme tidal wave sample data meets the modeling requirements, the system can also use a linear regression algorithm to construct a dynamic mathematical mapping relationship between tidal time deviation and measured tidal range. The scaling ratio of the tidal range from historical calculation records is extracted as the independent variable, and the corresponding tidal time deviation is used as the dependent variable for multiple linear fitting. The linear regression coefficients are extracted for secondary calibration calculations of various deviations during the prediction period, replacing the exponential decay weighting operation, and extracting a higher-precision physical fitting result as the output of dynamic deformation feature parameters.

[0110] On the one hand, an exemplary scheme for decoupling and dynamic reconstruction of wind fields at multiple stations in upstream large shallow lakes is provided, particularly a specific implementation scheme for the physical decomposition of wind fields in shallow lakes upstream of sluice gates, vector parameterized modeling, joint weighted least squares inversion, and closed-loop generation of upstream forecast curves. Specifically, the following methods can be used:

[0111] Step 401: For each station that constitutes the multi-station observation water level, decompose its observed water level into the true average water level reflecting the overall change in lake storage, the wind-driven water level deviation affected by the local wind field, and the random observation error.

[0112] Random observation errors are statistically filtered out in subsequent joint parameter estimation.

[0113] In plain river network areas, the surface of large, shallow lakes is easily tilted by local wind disturbances, resulting in wind-generated noise superimposed on the measured water levels at individual stations. Therefore, a physical decomposition model can be used to quantify and decompose lake water levels. Specifically, for each discrete station constituting the multi-station water level observations, the observed water level data at a predetermined timestamp is extracted. This observed water level data is mathematically constructed as a linear sum of three independent physical terms, as shown in the following formula:

[0114] Z _i (t)=Z _mean (t)+ΔZ _wind,i (t)+ε _i (t);

[0115] Among them, Z _i Z(t) represents the observed water level data at time t for the i-th station. _mean (t) represents the true average water level at time t, reflecting the overall change in lake storage capacity, ΔZ _wind,i (t) represents the wind-induced water level deviation driven by the local wind field at the i-th station, ε _i (t) represents the random observation error caused by system measurements and environmental random noise. This physical decomposition model extracts a baseline architecture that decouples deterministic water storage elements from wind direction disturbance elements.

[0116] Step 402: Extract the position offset of the pre-configured station relative to the geometric center of the shallow lake, and analyze the wind speed and wind direction angle elements from the wind field data;

[0117] Based on the steady-state wind-induced flow theory of shallow lakes, the square term of wind speed, the directional decomposition component of wind direction angle, and the positional offset are jointly correlated. After introducing the wind-induced water level coefficient, a quantitative expression for wind-induced water level deviation is constructed.

[0118] Alternatively, based on the steady-state wind-induced flow theory of shallow lakes, the square term of wind speed, the directional decomposition component of wind direction angle, and the positional offset are jointly correlated. After setting the wind-induced water level coefficient as the parameter to be estimated, a quantitative expression for the wind-induced water level deviation is constructed.

[0119] Accordingly, the system acquires pre-allocated spatial coordinate parameters and extracts the positional offset of each station relative to the physical geometric center of the shallow lake in two orthogonal two-dimensional directions. Simultaneously, it analyzes and extracts wind speed and wind direction elements at the reference height at the calculation time from the wind field data. Based on the steady-state wind-induced current dynamics principle of shallow lakes, the water surface will tilt along the predetermined wind direction under wind stress. The system constructs the wind-induced water level deviation as a weighted combination operation involving two orthogonal directions, with the specific calculation formula as follows:

[0120] ΔZ _wind,i (t)=α_x×Δx _i ×V(t) 2 ×cos(θ(t))+α _y ×Δy _i ×V(t) 2 ×sin(θ(t));

[0121] Where, ΔZ _wind,i (t) represents the wind-induced water level deviation, α_x is the wind-induced water level increase coefficient in the first orthogonal direction, α _y Let Δx be the wind-induced water increase coefficient in the second orthogonal direction. _i with Δy _i θ(t) represents the positional offset of station i in two orthogonal directions, V(t) represents the wind speed element, and θ(t) represents the wind direction angle element.

[0122] In some scenarios, to maintain the closed-loop characteristics of the physical equations, the unit of measurement for the wind-water increase coefficient is uniformly defined as seconds squared per meter squared.

[0123] Step 403: For high-risk flood control scheduling scenarios with a clear dominant wind direction, the directional decomposition component of the wind direction angle is simplified into the cosine parameter of the angle between the wind direction angle and the preset reference dominant wind direction angle. Combined with the position offset, it is integrated into a single wind sensitivity coefficient that comprehensively represents the station position and the geometric characteristics of the lake. A simplified expression for wind-induced water level deviation is constructed under the scenario of reference dominant wind direction.

[0124] Specifically, for typical engineering flood control scenarios where historical backflow events are concentrated under easterly wind-dominant climate conditions, the system triggers pre-defined dimensionality reduction logic. The system obtains a pre-configured reference prevailing wind direction angle and calculates the relative angle difference between the currently extracted wind direction angle element and this reference prevailing wind direction angle. The cosine parameter of this relative angle difference is calculated and multiplied by a single constant representing fused spatial location offset and other feature attributes to generate a dimensionality-reduced expression. The specific calculation formula is as follows:

[0125] ΔZ _wind,i (t)=k _i ×V(t)×cos(θ(t)-θ _ref );

[0126] Where, ΔZ _wind,i (t) represents the wind-induced water level deviation after dimensionality reduction simplification, k _i To integrate the single wind sensitivity coefficient based on the station location and lake geometry, V(t) is the extracted wind speed element, θ(t) is the extracted wind direction angle element, and θ _ref This is the reference prevailing wind direction angle for easterly wind scenarios. This expression allows for the abandonment of global nonlinear traversal and direct deduction of wind surface fluctuations under high-risk conditions.

[0127] In other words, the simplified expression approximates the squared wind speed term as a first-order wind speed. Specifically, in high-risk flood control scheduling scenarios with a clear dominant wind direction, the wind speed variation range corresponding to relevant historical backflow events is relatively concentrated, and the nonlinear effect of wind speed can be measured by the station-level comprehensive wind sensitivity coefficient k. _i Absorption, measured in seconds, is obtained through calibration using historical data. This simplification reduces the parameter dimensionality of the joint inversion, making it suitable for real-time early warning scenarios with stringent computational efficiency requirements.

[0128] In other words, in scenarios where the range of local wind speed variation is limited and the wind direction is relatively stable, the square term V can be... 2 Approximately linear form k _i ×V(t), improving real-time computing efficiency; where k _i The wind sensitivity coefficient, which corresponds to the integrated monitoring station location and the geometric characteristics of the lake, is obtained through field observation data or numerical simulation calibration, and its dimension is on the order of seconds.

[0129] Step 404: Extract the multi-station observed water levels at each station at the same time to construct the observation vector;

[0130] By utilizing the squared term of wind speed, the directional decomposition component of wind direction angle, and the corresponding position offset, a joint design matrix characterizing spatial features and wind field features is constructed.

[0131] A system of multivariate linear equations is established using the observation vector and the joint design matrix. The weighted least squares method is used for joint parameter estimation, and the true average water level and wind-induced water level coefficient at the target time are simultaneously derived.

[0132] In other embodiments, when a simplified expression is used, the joint design matrix can be adjusted accordingly to a simplified form based on a single wind sensitivity coefficient, and the inversion target is changed to the true average water level and the wind sensitivity coefficient of each station.

[0133] Based on the established quantitative expression for wind-induced water level deviation, the system employs matrix operations to simultaneously remove random observation errors and extract physical parameters. For a specified discrete time, the system extracts water level observations from each station and arranges them to generate an observation vector. Simultaneously, it extracts wind field characteristic parameters and spatial location attributes of each station at the corresponding time, and fills the array according to the array rules to construct a joint design matrix. The observation vector is matched with the joint design matrix to establish a system of multiple linear equations relating the true average water level and the wind-induced water level increase coefficient. A combination of matrix inversion and multiplication is performed; the specific calculation formula is as follows:

[0134] β(t)=(A(t) T ×W×A(t)) -1 ×A(t) T ×W×Z(t);

[0135] Where β(t) is the vector of parameters to be estimated, including the true average water level and the wind-induced water level rise coefficient, and A(t) is the joint design matrix. T Let W be the transpose of the joint design matrix, W be the diagonal weight matrix that assigns weights to each station, and Z(t) be the observation vector. -1 This indicates the attribute of extracting the matrix inversion operation. The operation process can obtain the overall water level after eliminating wind field noise without relying on a single station reference.

[0136] Step 405: Assign physical weights to each station in the observation vector by constructing a diagonal weight matrix. The diagonal elements of the diagonal weight matrix are extracted and assigned values ​​based on the inverse of the variance of the measured observation error of the corresponding station or the spatial representative parameters of the station.

[0137] When performing weighted least squares calculations, to compensate for differences in spatial representativeness and physical measurement accuracy among the stations, the system assigns independent weight constants to each node constituting the matrix equation. The system obtains the measured observation error variance of water levels at each station, which is extracted from historical data beforehand, calculates its mathematical reciprocal, and assigns it to the main diagonal elements of the diagonal weight matrix.

[0138] In another implementation, the system calculates the spatial representativeness parameter of each station based on its proportion of the lake area coverage and assigns this parameter to the diagonal elements. This allocation operation ensures that stations with high measurement accuracy or strong spatial characterization capabilities occupy a set high numerical proportion in the physical parameter inversion.

[0139] Step 406: Extract the rainfall forecast parameters belonging to the watershed rainfall data and the wind field forecast parameters belonging to the wind field data within the forecast period;

[0140] Based on the rainfall forecast parameters and the pre-established empirical relationship, the forecast dynamic increment of the actual average water level is calculated;

[0141] Using wind field forecast parameters, the predetermined spatial location of the upstream control section of the sluice gate, and the wind-induced water level coefficient, the predicted wind-induced water level deviation at the control section is calculated.

[0142] The baseline value of the true average water level, the forecast dynamic increment, and the forecast wind-induced water level deviation are superimposed over time to generate a set of upstream water level forecast curves that reflect the influence of the prevailing wind direction.

[0143] Alternatively, extract rainfall forecast parameters from watershed rainfall data and wind field forecast parameters from wind field data within the forecast period;

[0144] Based on the rainfall forecast parameters and the pre-established empirical relationship, the forecast dynamic increment of the actual average water level is calculated;

[0145] Using wind field forecast parameters, the preset spatial location of the upstream control section of the sluice gate, and the wind-induced water level coefficient, the predicted wind-induced water level deviation at the control section is calculated.

[0146] The true average water level obtained by joint parameter estimation is used as the base value. The forecast dynamic increment and the forecast wind-induced water level deviation are then superimposed over time to generate a set of upstream water level forecast curves that reflect the influence of the prevailing wind direction.

[0147] After completing the parameter inversion of historical observation data, the system initiates the forward extrapolation logic within the forecast period. The system acquires rainfall forecast parameters and wind field forecast parameters such as wind speed and direction within the future forecast time window. Using the initial hydrological conditions of the watershed and the rainfall forecast parameters input into the pre-stored statistical model, the dynamic change of the lake's true average water level is extracted as the forecast dynamic increment. Simultaneously, the fixed coordinates of the actual control section upstream of the sluice gate relative to the center of the lake are extracted. The calculated wind-induced water level coefficient is multiplied and accumulated with the currently acquired wind field forecast parameters to extract the forecast wind-induced water level deviation at the target section.

[0148] Furthermore, the system extracts the true average water level at the target time starting point from the calculated output as the base value, and performs numerical superposition calculation with the forecast dynamic increment and the forecast wind-induced water level deviation. The specific calculation formula is as follows:

[0149] Z _up (t)=Z _mean 0(t)+ΔZ _mean (t)+ΔZ _wind,up (t);

[0150] Among them, Z _up (t) represents the predicted water level at time t at the upstream control section of the sluice gate. _mean 0(t) is the true average water level baseline value starting from 0, and ΔZ _mean (t) represents the calculated and extracted forecast dynamic increment, ΔZ _wind,up (t) represents the calculated wind-induced water level deviation at this cross-section. The hourly accumulated water level time series are combined to generate an upstream water level forecast curve set. Where Z... _up The subscript in (t) corresponds to the upstream control section. For example, in the Taipu River Estuary application scenario, it is the predicted water level of the Taipu River Estuary. The actual station selection can be determined according to the specific lake and water system layout.

[0151] Alternatively, the calculated upstream water level time series constitutes the upstream water level forecast curve set. In other implementations, the corresponding upstream water level time series can be calculated based on multiple different forecast scenarios of rainfall or wind fields, generating an upstream water level forecast curve set containing multiple curves, providing uncertainty boundary samples.

[0152] Furthermore, as an alternative calculation scheme for extracting the true average water level, if the goal is to achieve advanced adaptive tracking capabilities of the system for real-time water flow conditions, a data assimilation operation structure can be used to replace the weighted least squares estimation. The system constructs a Kalman filter model structure based on the state-space update equation, uses the joint design matrix as the input measurement mapping parameter, and performs assimilation update calculations using successively acquired measured sampling data to achieve continuous state correction of predicted feature values.

[0153] On the other hand, this paper describes an optional implementation of a partial safety lower bound fusion early warning system based on conditional quantile calibration, particularly based on long-series error calibration, which transforms the combination of multiple forecast curves into a specific fusion logic for high-confidence risk periods. This embodiment abandons arithmetic mean fusion calculation and constructs a minimum lower bound envelope extraction mechanism. Accordingly, this embodiment includes:

[0154] Step 501: Traverse and extract each downstream forecast curve in the downstream water level forecast curve set and each upstream forecast curve in the upstream water level forecast curve set;

[0155] The extracted downstream forecast curves are combined and paired with the upstream forecast curves.

[0156] For each pairing combination, the difference between the upstream and downstream water levels at each corresponding moment within the forecast period is calculated, forming multiple sequence sets to obtain the predicted water level difference sequence set.

[0157] After obtaining multiple upstream and downstream water level prediction results generated by decoupling correction, the system initiates a cross-combination operation. Specifically, the system extracts a set of downstream forecast curves including uncorrected, amplitude-adjusted, and phase-adjusted modes, and extracts a set of upstream forecast curves generated through deduction. A full permutation and combination operation is performed to form multiple water level forecast pairs. For each water level forecast pair, at each hourly sampling time within the forecast period, the corresponding upstream and downstream water level forecast values ​​are extracted and subjected to a subtraction algebraic operation. The specific calculation formula is as follows:

[0158] D _k (t)=Z _u,j (t)-Z _d,i (t);

[0159] Among them, D _k Z(t) represents the predicted water level difference for the k-th combination at time t. _u,j (t) represents the predicted value of the j-th upstream water level forecast curve at time t, Z _d,i (t) represents the predicted value of the i-th downstream water level forecast curve at time t, and k is the global combination number combining upstream and downstream index items.

[0160] This calculation yields a dynamic time-varying sequence of differences, and the set of these sequences constitutes the predicted water level difference sequence set. Based on this subtraction rule, when the downstream water level is higher than the upstream water level, the predicted water level difference is represented as a negative number, corresponding to the hydrodynamic condition of reverse backflow.

[0161] Step 502: Extract tidal phase elements, precipitation elements and wind field elements at each moment within the forecast period, and integrate them to generate meteorological and hydrological scenario labels that match the current forecast time.

[0162] In other words, tidal phase elements are extracted based on the astronomical tidal cycle characteristics at various times within the forecast period, rainfall elements are extracted based on watershed rainfall data, wind field elements are extracted based on wind field data, and meteorological and hydrological scenario labels that match the current forecast time are integrated to generate them.

[0163] In the error condition distribution database, query the prediction error quantiles corresponding to the meteorological and hydrological scenario tags;

[0164] The prediction error quantiles are superimposed onto the corresponding predicted water level difference sequence set to construct a partially safe lower bound estimation sequence biased towards unfavorable water level conditions for each sequence pairing combination;

[0165] Extract the minimum values ​​of all partial safety lower bound estimation sequences at the same time and perform envelope fusion to generate a single partial safety fused water level difference sequence that is biased towards the safety side.

[0166] Accordingly, the system invokes an error condition distribution library pre-constructed in the offline environment to quantify the physical blind spots of a single deterministic forecast. Specifically, the system analyzes the astronomical tidal phase state, cumulative meteorological rainfall intensity level, and the two-dimensional quadrant interval of wind speed and direction corresponding to the target calculation time within the forecast period. These three types of physical features are then concatenated to generate a pre-defined meteorological and hydrological scenario label for the current forecast time.

[0167] Next, the system performs matching and addressing in the error condition distribution database to extract the prediction error quantiles under the corresponding scenario labels. The parameters of these quantiles are set to meet the boundary constraint condition that the quantile level is greater than 0 and less than 0.5, causing the extracted error statistics to tend towards the negative tail distribution, reflecting the mathematical property of biasing towards more unfavorable backflow states. For a predicted water level difference sequence set containing multiple data sequences, the system directly algebraically adds the extracted prediction error quantiles to the predicted value of each sequence at the current time, thereby generating a partially safe lower bound estimation sequence. The specific calculation formula can be expressed as:

[0168] L _k (t)=D _k (t)+Q _k,g (p);

[0169] Among them, L _k (t) represents the partial safety lower bound estimate constructed for the k-th portfolio at time t, D _k Q(t) represents the initial predicted water level difference for the k-th combination at time t. _k,g (p) represents the prediction error quantile at quantile level p extracted under the meteorological and hydrological scenario label g.

[0170] After obtaining the lower bound estimates of all combinations, the system extracts the minimum value of the entire combination on the same discrete-time section, which serves as the unique output value for fusion convergence. The specific calculation formula is as follows:

[0171] D _safe (t)=min(L _k (t));

[0172] Among them, D _safe (t) represents the value of the generated partial safety fused water level difference sequence at time t, and the min operation operator represents the operation of extracting the minimum value in all combination number dimensions.

[0173] As a numerical calculation example, assuming that after matching the current meteorological and hydrological scenario, the 0.05 quantile prediction error quantile retrieved from the distribution database is -0.02 meters. The system currently contains two predicted water level difference sequences, with initial predicted values ​​of 0.01 meters and -0.01 meters at the target time, respectively. Performing a lower bound bias operation, the estimated partial safety lower bound for the first sequence is calculated to be 0.01 + (-0.02) = -0.01 meters, and the estimated partial safety lower bound for the second sequence is -0.01 + (-0.02) = -0.03 meters.

[0174] Based on this, a minimum value fusion extraction operation is performed, comparing -0.01 meters and -0.03 meters, and extracting the minimum value of -0.03 meters as the output of the partial safety fusion water level difference sequence at that moment. It should be understood that this operation mechanism, through the minimum value extraction operation, realizes a physical envelope of the potential adverse risk baseline.

[0175] Step 503: While generating the partial safety fused water level difference sequence, the corresponding cumulative distribution function can also be extracted using the error cumulative distribution characteristics in the error condition distribution library.

[0176] Based on the cumulative distribution function, the probability of each predicted water level difference sequence falling below the preset backflow judgment threshold is calculated. The maximum probability value is extracted from all sequence combinations at the same time to generate a partially safe fused backflow probability sequence, which is output synchronously as an auxiliary decision support indicator for backflow risk periods.

[0177] Accordingly, the system adds a probability decision branch to the output of absolute numerical boundaries. The system extracts a statistical cumulative distribution function from the error conditional distribution library that matches the current meteorological and hydrological scenario label. Combining the pre-set constant backflow determination threshold with the prediction difference of the current sequence combination, the system calculates the integral probability that the deviation variable falls into the backflow trigger interval. The specific calculation formula is as follows:

[0178] r _k (t)=F _k,g (θ _1 -D_k (t));

[0179] Where, r _k F(t) is the conditional probability of the backflow event occurring in the k-th combination at time t. _k,g Let θ be the cumulative distribution function of the prediction error under the constraint of the predetermined scenario label g. _1 D is the preset backflow detection threshold. _k (t) represents the initial predicted water level difference for the k-th combination at time t.

[0180] After calculating the probability of occurrence for each combination of conditions, the system performs a horizontal comparison across all combinations and extracts the maximum value to generate a partially safe fused backflow probability sequence. This maximum probability value directly quantifies the level of extreme risk exposure under multi-model uncertainty conditions, and is used by the scheduling system backend for risk level classification.

[0181] Step 504: Compare the hourly sequence values ​​in the partial safety fusion water level difference sequence with the backflow determination threshold, and generate a backflow state indication parameter representing possible backflow or non-backflow for each time moment.

[0182] Within the time frame of the foreseeable period, the retrieval backflow status indicator parameter continuously indicates the possible backflow, and the continuous time window is not less than the preset duration.

[0183] Extract all time intervals that are found during the search, along with their corresponding start and end times, and output them as the backflow risk period.

[0184] Optionally, the system performs discrete state determination and temporal consistency verification. Further, the calculated partial safety fused water level difference sequence is extracted and compared hourly with the backflow determination threshold using algebraic comparison operations. The preset backflow determination threshold can be specifically set to -0.05 meters.

[0185] If the fusion difference at the current moment is less than or equal to the threshold, the system assigns a value of 1 to the reverse flow status indicator parameter, indicating that there is a risk; otherwise, it assigns a value of 0, indicating that it is safe.

[0186] Furthermore, the specific calculation process can be described by the following formula:

[0187] When D _safe (t)≤θ _1 S(t) = 1;

[0188] When D _safe (t)>θ _1 S(t) = 0;

[0189] Where S(t) is the reverse flow state indication parameter output at time t, and D _safe(t) represents the value of the partial safety level difference sequence at time t, θ _1 This is the threshold for determining backflow.

[0190] After obtaining the binarized state sequence for the entire forecast period, the system initiates a time-axis sliding search function. The system is configured with a preset duration parameter, which can be set to 5 consecutive hours. The system searches for sequence windows that continuously contain the value 1 and whose cumulative span exceeds the preset duration, thereby eliminating short-term false alarm signals caused by transient disturbances in wind and waves. All closed intervals that meet the span search criteria are used as effective risk windows. The left boundary time point of the interval is extracted as the start time, and the right boundary time point is extracted as the end time. These are combined and packaged for output to obtain the backflow risk period that can be directly invoked by the scheduler.

[0191] In another embodiment, if the emphasis is on further scalability of the fusion model, the probabilistic statistics-based fusion mechanism can be replaced with a weighted fusion strategy based on Bayesian inference. Historical error distribution features are extracted as a prior probability benchmark, and new observation data are gradually introduced during the forecast period to calculate the posterior likelihood function of each sequence combination, thereby generating a full probability density distribution output. This scheme improves the theoretical smoothness of the decision output dimension, but requires more abundant parallel computing resources to meet computational efficiency constraints.

[0192] On the other hand, this embodiment provides a possible implementation process for constructing an offline long series of error condition distribution databases. It involves slicing historical error data according to meteorological and hydrological characteristics using multidimensional conditional classification and extracting offline probability quantiles. Specifically, the following methods can be used:

[0193] Step 601: Obtain the measured upstream and downstream water level sequence within the historical period, and the predicted water level sequence calculated by replaying the historical meteorological and rainfall conditions of the same period. Compare the deviation between the two to extract the historical water level difference prediction error set.

[0194] In other words, the measured upstream and downstream water level sequences within a historical period are obtained, along with the predicted water level sequences calculated by replaying the upstream and downstream water level forecasts using the historical meteorological and rainfall conditions of the same period according to the upstream and downstream water level forecasting method of this method. The deviation between the two is compared to extract the set of historical water level difference prediction errors.

[0195] During the offline preparation phase, when the system is not running online, it extracts observation and playback data from a long historical time window. Specifically, it obtains the measured water level sequence of the upstream and downstream control sections of the sluice gate, calculates the difference between the upstream and downstream measured water levels at the same moment, and uses this difference as the true water level difference.

[0196] Simultaneously, meteorological data such as rainfall, wind speed, and wind direction from corresponding historical periods are used as input conditions for the forecast model to perform playback calculations, generating multiple predicted water level sequences. The predicted water level difference is calculated for each combination of upstream and downstream curve pairings. The actual water level difference at a predetermined time is extracted and compared with the predicted water level difference using the following formula:

[0197] E _k (t)=D _obs (t)-D _k (t);

[0198] Among them, E _k (t) represents the historical water level difference prediction error of combination k at time t, and D _obs (t) represents the actual water level difference at time t, D _k (t) represents the predicted water level difference calculated by combining k at time t using historical meteorological rainfall conditions.

[0199] Based on this, the system performs subtraction comparison operations throughout the entire historical time series, compiling all combinations and error results at all times into a historical water level difference prediction error set.

[0200] Step 602: Based on the tidal phase characteristics, previous cumulative rainfall intensity characteristics, and wind speed and direction quadrant characteristics corresponding to the historical playback calculation time, tidal phase group labels, rainfall scenario group labels, and wind scenario group labels are constructed respectively.

[0201] Accordingly, the system extracts the three-dimensional environmental feature parameters corresponding to the calculation time. The system obtains the astronomical tide basis function output results corresponding to the historical playback time, and classifies the tidal phase characteristics based on the sign of the first derivative of the water level and the extreme value neighborhood characteristics, and maps them into discrete identifiers such as high tide segment, low tide segment, high tide neighborhood or low tide neighborhood, thus forming tidal phase grouping labels.

[0202] The cumulative rainfall within a preset number of days prior to the calculation time is obtained. Based on the watershed hydrological characteristics, the rainfall is divided into intervals. The intensity characteristics of the previous cumulative rainfall are classified and mapped to discrete labels such as drought, moderate humidity, or humidity, forming rainfall scenario grouping labels. Wind speed and direction data at the calculation time are extracted. Based on preset wind speed thresholds and the azimuth quadrant of the wind direction, the wind speed and direction quadrant characteristics are mapped to corresponding discrete labels, forming wind scenario grouping labels.

[0203] Step 603: Construct a combined condition triplet using tidal phase grouping labels, rainfall scenario grouping labels, and wind scenario grouping labels to perform conditional classification slicing on the historical water level difference prediction error set.

[0204] Optionally, the system extracts three independently constructed discrete identifiers and performs a Cartesian product combination operation in mathematical space. The specific definitions are as follows:

[0205] G(t) = (g _tide (t),g _rain (t),g _wind (t));

[0206] Where G(t) is the combination condition triple at time t, g _tide (t) is the tidal phase grouping label at time t, g _rain (t) represents the group label for the rain scenario at time t, g _wind (t) represents the wind scenario grouping label at time t.

[0207] Based on this, the system uses all the combined condition triplets generated by enumeration as index keys to filter and package the historical water level difference prediction error set, and gathers the error samples with the same combined condition triplet attributes into the corresponding physical storage box to complete the condition classification and slicing operation.

[0208] Step 604: For each error sample within a conditional classification slice, perform probability quantile statistical calculations to establish the prediction error quantiles and establish a mapping relationship with the combined conditional triplet to obtain the error conditional distribution library.

[0209] In other embodiments, for each error sample within a conditional classification slice, probability quantile statistics are performed to establish the prediction error quantile and establish a mapping relationship with the combined conditional triplet. At the same time, the cumulative error distribution function features of each conditional classification slice are extracted and stored synchronously to obtain an error conditional distribution library.

[0210] After classifying and aggregating the error samples, the system performs independent statistical analysis on the data sample set within each physical storage box. The system sorts the historical water level difference prediction error samples within the box in ascending order of numerical value and extracts the corresponding values ​​based on the set quantile level parameters. This can be calculated using the following formula:

[0211] Q _k,g (p)=Quantile({E _k (t)|G(t)=g},p);

[0212] In the formula, Q _k,g (p) represents the prediction error quantile of the error sample set of combination k under the combined condition triple g, where Quantile is the probability quantile statistical operator, E _k (t) represents the historical water level difference prediction error sample within the condition slice, G(t)=g represents that the sample satisfies the specified combination condition triplet constraint, and p is the set quantile level parameter.

[0213] In order to support the lower bound estimation with a biased safety, the quantile level parameter p is set to extract a fixed value that is greater than 0 and less than 0.5. The specific value can be set to 0.1.

[0214] After completing the probability quantile statistics, the system uses the calculated and established prediction error quantiles as the retrieval feedback values ​​and the corresponding combined condition triples as the retrieval keys, establishing a dictionary mapping relationship between the two in a relational database or key-value storage system. After traversing all condition classification slices to complete the establishment of all mapping relationships, the encapsulated output data structure is the error condition distribution library.

[0215] Optionally, in addition to extracting a single probability quantile, the system can also perform kernel density estimation or empirical cumulative distribution statistical calculations on the error samples within each conditional classification slice. The system extracts the set of parameters of the fitted cumulative distribution function and simultaneously establishes an additional mapping relationship between it and the combined conditional triplet.

[0216] Based on this, the cumulative distribution function features are stored, and the error condition distribution library can support the output of backflow condition occurrence probability sequences for predetermined judgment thresholds during real-time operation, providing data format support for multi-dimensional scheduling risk assessment.

[0217] The above describes the construction of an error condition distribution library and the adoption of a partial safety lower bound envelope extraction algorithm based on probability quantiles. This technique transforms the multi-curve divergent uncertainty generated by independent decoupling in the early stage into a single safety decision boundary with high confidence. From the underlying logic of the algorithm, it suppresses the risk of missed detection that may be caused by conventional arithmetic mean fusion, and enhances the robustness and reliability of business scheduling decisions under extreme backflow conditions.

[0218] According to one aspect of this application, the specific implementation process of the rapid backflow determination system and its application verification is described, especially the technical solution at the system architecture level, including the module encapsulation mechanism of the backflow time determination system within the plain tidal river network, the physical trend interpretation of empirical statistical relationships, and the coupling gain effect brought about by the decoupling-then-fusion architecture. Specifically, it can be:

[0219] Step 701 provides a system for determining the backflow period of sluice gates in a plain tidal river network, including:

[0220] The first module is used for collecting and managing long-term hydrological and meteorological data;

[0221] The second module is used to construct a water level prediction model for the downstream section of the sluice gate;

[0222] The third module is used to construct a water level prediction model for the upstream section of the sluice gate;

[0223] The fourth module is used for determining and issuing early warnings of sluice gate backflow periods based on the fusion of multiple forecast results;

[0224] The fifth module is used for visualizing forecast results.

[0225] Among them, the system for determining the backflow period of gates in plain tidal river networks encapsulates the method steps of each embodiment into independent computer-executable program code and physical computing nodes.

[0226] Specifically, the first module provides a standardized data access protocol for connecting to the existing integrated platform system for water conservancy basin forecasting and scheduling. It continuously acquires long-series observational water levels, basin rainfall data, and wind field data from upstream and downstream sections of sluice gates, and performs interpolation and logical consistency verification on the time-series data.

[0227] Furthermore, the second module has a built-in harmonic analysis constant solver and correlation coefficient matrix operation unit, which is used to decouple the physical waveforms of astronomical tides and rainfall runoff from the downstream cross-section water level, extract dynamic deformation feature parameters, perform translation and scaling operations, and output a multi-dimensional set of downstream water level forecast curves.

[0228] Furthermore, the third module incorporates a spatial vector decomposition algorithm and a weighted least squares estimation unit. Based on the steady-state wind-driven current theory of shallow lakes, it removes the wind field deviation from the physical observation water level at a single station and extracts the true average water level. It then overlays the wind field forecast parameters to generate a set of upstream water level forecast curves.

[0229] Furthermore, the fourth module calls the error condition distribution library generated offline from historical time series data, performs multi-curve cross-traversal combination operation, extracts the prediction error quantiles under the corresponding meteorological and hydrological scenarios to implement partial safety lower bound calibration, extracts the minimum value to generate a partial safety fused water level difference sequence, and outputs the backflow risk period with clear timestamp boundaries based on the set backflow judgment threshold.

[0230] Furthermore, the fifth module is used to map the output backflow risk period, upstream and downstream water level forecast process lines, and auxiliary decision probability sequence into rendered data for the front-end graphical interface, so that dispatchers can view and extract information interactively.

[0231] Optionally, in the underlying calculation logic of the second module for deriving hourly water level process lines during the flood and receding periods, the system constructs a statistical mapping model between peak flood level and receding rate using historical long-sequence synchronous observation samples. In the regression calculation of actual parameters, the statistical fitting results of this empirical relationship model reflect the negative correlation physical trend that the higher the peak flood level, the faster the receding rate.

[0232] In the real-time flood control dispatch system, this empirical relationship model replaces the long-term numerical solution of partial differential equations, providing a quantitative basis for simplifying the dimensions of the receding water physical process and rapidly estimating the parameters for hourly process line prediction, thus meeting the timeliness constraints of minute-level advanced risk identification.

[0233] Building upon this, the decoupling-then-fusion computational architecture constructed in this embodiment produces a quantitative coupling gain effect. Predictions using a single baseline hydrodynamic model suffer from nonlinear error accumulation and lengthy computation time. This system performs a physical mechanism decoupling operation on the observed water level, generating multiple independent prediction curves with different spatiotemporal physical constraints, providing diverse input samples covering multidimensional boundary attributes for error fusion calculation.

[0234] Next, instead of performing arithmetic averaging for dimensionality reduction on multiple independent curves, the system introduces long-sequence historical error conditional distribution characteristic parameters, extracts the prediction error quantiles biased towards the negative tail water level state to construct a lower bound estimation equation, and extracts the minimum values ​​at the same time sectional area for envelope convergence fusion. This operational mechanism directly transforms the curve divergence uncertainty retained in the previous decoupling extraction process into a single safe decision boundary with high confidence. By combining data statistical models with biased safety fault-tolerant lower bound logic, this system reduces the calculation cycle for determining the sluice gate backflow status from several hours in traditional physical models to the minute level, suppressing the technical problem of the lag in early warning commands on the scheduling time axis.

[0235] According to another aspect of this application, the present invention can also be carried out by the following method:

[0236] Based on multiple modified forecast curves of water level at key sections downstream of the sluice gate, and combined with the forecast results of water level at key sections upstream, a multi-scenario combination analysis is carried out. The safety criterion is used for comprehensive judgment to reduce the uncertainty of the forecast results and improve the reliability and robustness of the identification of backflow periods.

[0237] The safety criterion is defined as follows: if the water level difference sequence is less than -0.05m and lasts for 5 consecutive hours, it is determined that there is a possibility of backflow during this period, and this period is identified as the Taipu Sluice Gate backflow risk period.

[0238] This invention employs a lightweight decision-making architecture that combines physical deterministic decoupling with data statistical driving, eliminating the time-consuming numerical solution process of global partial differential equations in traditional hydrodynamic models. By independently decomposing and empirically extrapolating upstream and downstream water levels, the calculation cycle for backflow risk zones is reduced from hours in traditional physical models to minutes. This mitigates the technical problem of delayed early warning commands on the scheduling timeline and improves the proactive response speed of flood control scheduling.

[0239] 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 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 determining the backflow period of sluice gates in a plain tidal river network, characterized in that, include: Acquire long-term observation water levels at the downstream section of the sluice gate, multi-station observation water levels at shallow lakes upstream, watershed rainfall data, and wind field data; Based on long-term observation of water level and basin rainfall data, the water level at the downstream section is decoupled from astronomical tides and rainfall runoff. Dynamic deformation characteristic parameters are extracted to correct the decoupled waveform, resulting in a set of downstream water level forecast curves. Physical decomposition was performed based on multi-station observed water level and wind field data to extract the true average water level and wind-induced water level deviation, and combined with watershed rainfall data to obtain a set of upstream water level forecast curves; Combine it with the downstream water level forecast curve set to construct a predicted water level difference sequence set; The error condition distribution library is called to extract the prediction error quantiles corresponding to the current meteorological and hydrological scenario. The prediction error quantiles are then used to perform lower bound calibration and minimum value extraction on the predicted water level difference sequence set to obtain a partially safe fused water level difference sequence. Based on this and the backflow determination threshold, backflow risk periods that meet the preset duration are identified.

2. The method according to claim 1, characterized in that, Decoupling of astronomical tides and rainfall runoff at downstream cross-section water levels includes: Based on long-term observed water levels, recent measured data are extracted and astronomical tide harmonic analysis is performed to separate periodic fluctuation components and obtain astronomical tide prediction sequences. Based on watershed rainfall data and empirical rainfall process relationships, the runoff response components are estimated to obtain the rainfall-runoff water level variation sequence.

3. The method according to claim 2, characterized in that, Dynamic deformation feature parameters are extracted to correct the decoupled waveform, resulting in a set of downstream water level forecast curves, including: Within a set rolling window, the rainfall-runoff water level amplitude sequence is extracted from the long series of observed water levels, and the tidal residual sequence after removing the runoff influence is extracted. By comparing the tidal residual sequence with the astronomical tide prediction sequence, dynamic deformation characteristic parameters for characterizing phase and amplitude deviations are extracted. By combining dynamic deformation characteristic parameters, astronomical tide prediction sequences, and rainfall-runoff water level amplitude sequences for correction, a set of downstream water level forecast curves containing multiple correction modes is generated.

4. The method according to claim 3, characterized in that, By comparing the tidal residual sequence with the astronomical tide prediction sequence, dynamic deformation characteristic parameters for characterizing phase and amplitude deviations are extracted. Specifically, this is achieved through recursive drift correction with full waveform alignment, including: Within a preset shift set, the sliding correlation coefficient between the tidal residual sequence and the astronomical tide prediction sequence is calculated, and the shift value corresponding to the maximum correlation coefficient is extracted as the phase drift parameter. The astronomical tide prediction sequence is shifted and aligned based on the phase drift parameter. The standard deviation ratio and mean difference between the aligned astronomical tide prediction sequence and the tidal residual sequence are calculated to obtain the amplitude drift parameter and the mean drift parameter. The phase drift parameter, amplitude drift parameter, and mean drift parameter are updated recursively and smoothly to output dynamic deformation characteristic parameters.

5. The method according to claim 3, characterized in that, By comparing the tidal residual sequence with the astronomical tide prediction sequence, dynamic deformation characteristic parameters for characterizing phase and amplitude deviations are extracted. Specifically, this is achieved through multi-tidal period weighting of extreme points, including: From the tidal residual sequence and the astronomical tide prediction sequence, identify the high tide extreme events and low tide extreme events of a predetermined number of consecutive semi-diurnal tide cycles in the past; For each semi-diurnal tidal cycle, calculate the tidal range scaling ratio and tidal time deviation between the tidal residual sequence and the astronomical tidal prediction sequence; Based on the time span from the current prediction time, the tidal range scaling ratio and tidal time deviation of each semi-diurnal tidal cycle are assigned exponential decay weights and weighted summed to output dynamic deformation characteristic parameters.

6. The method according to claim 3, characterized in that, By combining and correcting dynamic deformation characteristic parameters, astronomical tide prediction sequences, and rainfall-runoff water level amplitude sequences, a set of downstream water level forecast curves containing multiple correction models is generated, including: The astronomical tide prediction sequence is directly superimposed with the rainfall-runoff water level variation sequence to generate an uncorrected baseline curve. By using dynamic deformation characteristic parameters to perform single amplitude scaling and single phase shift on the astronomical tide prediction sequence, and then superimposing the rainfall-runoff water level amplitude sequence, corresponding amplitude adjustment curves and phase adjustment curves are generated. By using dynamic deformation characteristic parameters, amplitude scaling and phase shifting are simultaneously applied to the astronomical tide prediction sequence. The rainfall-runoff-water level amplitude sequence is superimposed and mean drift compensation is integrated to generate a dynamic comprehensive correction curve. The uncorrected baseline curve, amplitude adjustment curve, phase adjustment curve, and dynamic comprehensive correction curve are combined to form a downstream water level forecast curve set.

7. The method according to claim 1, characterized in that, Based on multi-station observed water level and wind field data, physical decomposition is performed to extract the true average water level and wind-induced water level deviation. Specifically, this is done by constructing a physical decomposition model of lake water level, including: For each station that constitutes the multi-station water level observation, its observed water level is decomposed into the true average water level reflecting the overall change in lake storage, the wind-induced water level deviation driven by local wind fields, and random observation errors.

8. The method according to claim 1, characterized in that, By combining this with the downstream water level forecast curve set, a predicted water level difference sequence set is constructed, including: Iterate through and extract each downstream forecast curve in the downstream water level forecast curve set, and each upstream forecast curve in the upstream water level forecast curve set; The extracted downstream forecast curves are combined and paired with the upstream forecast curves. For each pairing combination, the difference between the upstream and downstream water levels at each corresponding moment within the forecast period is calculated, forming multiple sequence sets to obtain the predicted water level difference sequence set.

9. The method according to claim 1, characterized in that, Based on this and the backflow determination threshold, backflow risk periods that meet the preset duration are identified, including: The hourly sequence values ​​in the partial safety fusion water level difference sequence are compared with the backflow determination threshold, and a backflow state indication parameter representing the possible backflow or non-backflow is generated for each moment. Within the time frame of the foreseeable period, the retrieval backflow status indicator parameter continuously indicates the possible backflow, and the continuous time window is not less than the preset duration. Extract all time intervals that are found during the search, along with their corresponding start and end times, and output them as the backflow risk period.

10. The method according to claim 1, characterized in that, The time span for long-term water level observations at the downstream section of the sluice gate shall not be less than thirty days; the time span for rainfall and wind field data in the basin shall not be less than one hydrological year.

Images