A reservoir flood season pre-discharge adaptive decision method

Abstract

The application discloses a reservoir flood season pre-discharge adaptive decision method, comprising: obtaining meteorological ensemble prediction and real-time reservoir water regime data; evaluating flood control margin and extracting pre-discharge trigger index; comparing it with the trigger threshold, if entering the standby state, combining the prediction to deduce the reservoir releasing time constraint to calculate the decision time margin; according to the decision time margin, performing the forward-looking opportunity judgment, when indicating triggering the pre-discharge operation, generating the initial scheduling control parameter based on the multi-objective optimization strategy; during the pre-discharge execution, intercepting the measured data in the real physical time window to calculate the inflow deviation, when satisfying the recalculation condition, performing the multi-dimensional attribution analysis on the deviation source, and according to the attribution level, executing the hierarchical correction response strategy, and outputting the updated scheduling control parameter. The application can optimize the pre-discharge scheduling, reduce the waste of water resources, reduce the correction calculation power consumption of the system in the execution period, and improve the stability of flood control scheduling under variable working conditions.

Description

Technical Field

[0001] This invention relates to the field of reservoir flood control scheduling and water resources management technology, and in particular to an adaptive decision-making method for reservoir pre-release during the flood season. Background Technology

[0002] Pre-release of water from reservoirs during the flood season is a core element in balancing flood control safety and beneficial water storage. Releasing flood control capacity before the arrival of the leading edge of torrential rains and floods can reduce peak flood levels, lower dam operational risks, and alleviate flood pressure on downstream river channels. Pre-release scheduling requires comprehensive processing of multi-source meteorological forecast data and real-time hydrological conditions. While ensuring controllable reservoir capacity adjustment time, it also involves controlling the timing of flood release commands and the allocation of flow rates over specific periods. The rigor and error tolerance of this decision-making process determine the overall stability of the basin's flood control system and the comprehensive utilization efficiency of rainwater resources.

[0003] Current mainstream pre-release decision-making schemes typically rely on preset fixed rainfall thresholds or static water level warning lines. When the rainfall parameters input by the weather forecast exceed the static threshold, the control logic often immediately triggers a full pre-release action. In the optimization generation stage of the scheduling scheme, existing methods mainly construct a static expectation function that includes conventional flood control and beneficial objectives, and call evolutionary algorithms for iterative optimization.

[0004] In actual implementation, due to the inherent spatiotemporal uncertainties of weather forecasts, instantaneous triggering mechanisms based on static fixed thresholds are ill-suited to the differentiated flood control and pressure-bearing capacities of reservoirs at different water storage levels, easily leading to premature water release before rainfall has fully materialized. Furthermore, within the closed-loop execution cycle of pre-release commands, when measured inflow data deviates from the initial forecast baseline, existing systems tend to employ an indiscriminate global recalculation strategy to correct control parameters. This response not only results in significant idle consumption of computational resources but also easily triggers command oscillations when facing systematic parameter errors, lacking the ability to finely address the physical root causes of errors.

[0005] In summary, existing methods still fall short in terms of the flexibility of scheduling triggering mechanisms and the computational efficiency of execution period error response when dealing with complex meteorological uncertainties and variable hydrological conditions, making it difficult to achieve the optimal balance between flood control safety and efficient water resource utilization. Therefore, it is necessary to study a method that can improve the ability to control the timing of reservoir pre-release and the stability of scheduling correction in complex flood scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive decision-making method for pre-release of water during the flood season of a reservoir, in order to solve the aforementioned problems in the prior art.

[0007] Technical solution: An adaptive decision-making method for reservoir flood season pre-release, comprising:

[0008] Acquire real-time meteorological ensemble forecast data and real-time reservoir hydrological data;

[0009] The flood control capacity is assessed based on real-time reservoir hydrological data, and pre-release trigger indicators are extracted by combining real-time meteorological ensemble forecast data.

[0010] The pre-release trigger index is compared with the pre-configured trigger threshold. If it is determined that the pre-release standby state has been entered, the reservoir emptying time constraint is deduced by using real-time meteorological ensemble forecast data and real-time reservoir hydrological data, and the decision-making time margin is calculated.

[0011] Based on the decision-making time margin, the timing of the pre-release is determined, and the result of the pre-release timing determination is output.

[0012] When the pre-discharge timing determination result indicates that the pre-discharge operation is triggered, the initial scheduling control parameters are generated based on the multi-objective optimization strategy.

[0013] During the pre-release execution, the measured data within the real physical time window up to the current moment is extracted to calculate the inflow deviation. When the inflow deviation meets the preset recalculation conditions, a multi-dimensional attribution analysis is performed on the source of the deviation, and the corresponding graded correction response strategy is executed according to the attribution level, and the updated scheduling control parameters are output.

[0014] For example, the steps of assessing flood control capacity based on real-time reservoir hydrological data and extracting pre-release trigger indicators by combining real-time meteorological ensemble forecast data include:

[0015] The actual reservoir capacity is extracted based on real-time reservoir hydrological data, and the flood control margin is calculated by combining the pre-configured upper limit reservoir capacity parameters with the water level. The critical rainfall amount is also estimated.

[0016] The percentage of members whose cumulative areal rainfall reaches the critical rainfall level in the real-time meteorological ensemble forecast data is statistically analyzed to obtain the rainfall exceedance probability, which serves as a pre-release trigger indicator.

[0017] The steps for comparing the pre-leakage trigger metric with the pre-configured trigger threshold include:

[0018] The reservoir capacity utilization rate is calculated by combining the pre-configured lower limit water level reservoir capacity parameters with the current actual reservoir capacity.

[0019] By interpolating the storage capacity utilization rate between the pre-configured lower and upper probability thresholds, an adaptive exceedance probability threshold is generated, which serves as the pre-configured trigger threshold.

[0020] The rainfall exceedance probability is compared with the adaptive exceedance probability threshold to determine whether to enter the pre-discharge standby state.

[0021] For example, the steps of using real-time meteorological ensemble forecast data to extrapolate reservoir emptying time constraints and calculate decision-making time margins include:

[0022] Extract the cumulative areal rainfall process of each forecast member in the real-time meteorological ensemble forecast data one by one;

[0023] The moment when the cumulative areal rainfall of each forecast member first reaches the critical rainfall amount is taken as the flood critical time for the corresponding forecast member;

[0024] The conservative critical time is obtained by extracting the low quantile from the sample set containing each flood critical time.

[0025] The target storage capacity that needs to be freed up is calculated based on the current actual storage capacity and the storage capacity corresponding to the pre-configured target flood level.

[0026] By combining the pre-configured maximum pre-discharge capacity parameter with the current measured inflow extracted from the real-time water level data of the reservoir, the shortest reservoir emptying time required to empty the target reservoir capacity is calculated.

[0027] The decision time margin is obtained by subtracting the shortest emptying time after the pre-configured timing safety factor is amplified from the conservative critical time.

[0028] For example, the process of calculating the decision time margin also includes a strategy for handling extreme operating conditions, specifically including:

[0029] When the cumulative areal rainfall of all forecast members has not reached the critical rainfall amount, the conservative critical time will be set as the end of the effective forecast period.

[0030] When the maximum pre-discharge capacity parameter is not greater than the current measured inflow, it is determined that physical reservoir emptying is not feasible, and an execution command is generated according to the maximum pre-discharge capacity parameter to open the flood discharge facility.

[0031] For example, the steps of performing forward timing determination based on decision-making time margin and outputting the pre-release timing determination result include the following three tiered determination scenarios:

[0032] When the probability of rainfall exceeding the threshold is not less than the pre-configured high confidence probability threshold, a pre-release timing determination result is generated to indicate the triggering of the pre-release operation;

[0033] When the rainfall exceedance probability is between the adaptive exceedance probability threshold and the high confidence probability threshold, and the decision time margin is greater than the pre-configured rolling optimization cycle, a pre-release timing determination result is generated that indicates maintaining the pre-release standby state and postponing the review to the next cycle.

[0034] When the rainfall exceedance probability is between the adaptive exceedance probability threshold and the high confidence probability threshold, and the decision time margin is no greater than the rolling optimization cycle, a pre-release timing determination result is generated, indicating that the latest safe decision time has been reached and triggering the pre-release operation.

[0035] For example, the step of generating initial scheduling control parameters based on a multi-objective optimization strategy includes:

[0036] Construct a multi-objective optimization set, which includes the expected minimization objective of water discharge to reflect the loss of benefits from the reservoir, the expected minimization objective of peak outflow to reflect downstream flood control pressure, and the conditional risk value minimization objective of excess storage capacity to reflect downstream flood control risk.

[0037] Combining the preset reservoir water balance equation constraints and the preset physical discharge capacity constraints, a multi-objective optimization algorithm is used to solve the multi-objective optimization set, generating a set of non-dominated schemes containing multiple non-dominated solutions;

[0038] The spatial distance metric method is used to calculate the comprehensive proximity of each scheme in the non-dominated scheme set. The scheme with the highest comprehensive proximity is selected, and the corresponding target flood level and the upper limit sequence of outflow for each time period are extracted from it as the initial scheduling and control parameters.

[0039] For example, the steps of calculating the inflow deviation by extracting measured data within the actual physical time window up to the current moment, and performing multi-dimensional attribution analysis on the source of the deviation when the inflow deviation meets the preset recalculation conditions, include:

[0040] Determine whether the relative deviation between the measured inflow and the predicted inflow baseline value continues to exceed the allowable ratio within a preset time period. If so, trigger multi-dimensional attribution analysis.

[0041] Extract the measured average rainfall and the forecast rainfall for each forecast member within the actual physical time window;

[0042] Calculate the absolute difference between the measured average rainfall and the corresponding forecast rainfall, and divide the absolute difference by the larger of the measured average rainfall and the pre-configured zero-prevention lower limit to obtain the consistency deviation of each forecast member used for cross-event horizontal comparison.

[0043] Extract the measured inflow and the simulated inflow given by the hydrological model within the same window, record the positive and negative distribution of the difference between the two, and generate an inflow deviation sign sequence.

[0044] For example, before implementing the corresponding graded corrective response strategy based on the attribution level, the specific attribution level classification includes:

[0045] When the consistency deviation of at least one forecast member is no greater than the pre-configured consistency tolerance, and the inflow deviation symbol sequence remains the same sign within the real physical time window, the first-level attribution judgment is output, indicating that it is a hydrological system deviation.

[0046] When the consistency deviation of some forecast members is no greater than the consistency tolerance, while the consistency deviation of the remaining forecast members is greater than the consistency tolerance, the second-level attribution judgment is output, indicating forecast reliability divergence.

[0047] When the consistency deviation of all forecast members is greater than the consistency tolerance, or the risk review at the aforementioned level fails, the third-level attribution judgment is output, indicating that the overall forecast is inaccurate.

[0048] When the meteorological ensemble forecast data stream is detected to be in a state of communication interruption or unavailability, the fourth-level attribution judgment is output, indicating that the data is interrupted.

[0049] For example, before acquiring real-time meteorological ensemble forecast data and real-time reservoir hydrological data, an offline pre-configuration step is also included to support the operation of the method:

[0050] Obtain historical meteorological ensemble forecast reports and measured hydrological data of reservoirs for the target watershed for the same period;

[0051] Calibrate basic hydrological parameters based on historical rainfall-runoff matching relationships;

[0052] The set dispersion distribution of historical rainfall events in the statistical report data is used to extract reference dispersion coefficients;

[0053] Basic hydrological parameters and reference dispersion coefficients are stored as pre-configured parameters in the system for real-time online retrieval.

[0054] For example, after outputting the updated scheduling control parameters, the method further includes:

[0055] Based on the flow-water-level opening relationship curve of the reservoir flood discharge gate, the sequence of upper limit of outflow for each time period in the updated scheduling control parameters is converted into the corresponding physical gate opening instruction sequence.

[0056] According to the set time nodes, the sequence of physical gate opening instructions is sent to the execution control node of the reservoir flood discharge facility to carry out the actual discharge operation.

[0057] Beneficial effects: This invention can optimize pre-discharge scheduling, reduce water waste, reduce the computational power consumption for system correction during execution, and improve the stability of flood control scheduling under variable operating conditions. Attached Figure Description

[0058] The accompanying drawings, which are provided to further illustrate this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application.

[0059] Figure 1 The flowchart of the adaptive decision-making method for reservoir pre-release during the flood season is shown;

[0060] Figure 2 A flowchart for estimating critical rainfall is shown;

[0061] Figure 3 A flowchart for determining the timing of look-ahead operations is shown;

[0062] Figure 4 A flowchart illustrating the generation of initial scheduling control parameters based on a multi-objective optimization strategy is shown.

[0063] Figure 5 A flowchart is shown for the offline pre-configuration steps used to support the operation of the method. Detailed Implementation

[0064] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0065] Furthermore, in the embodiments of this application, the words "exemplary," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the embodiments of this application should not be construed as being better or more advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplary" is intended to present the concept in a concrete manner.

[0066] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0067] Example 1:

[0068] This embodiment provides an adaptive decision-making method for reservoir pre-release during the flood season, such as... Figure 1 As shown, the method includes the following steps:

[0069] Step 101: Obtain real-time meteorological ensemble forecast data and real-time reservoir hydrological data.

[0070] Specifically, the data acquisition system obtains hourly areal rainfall forecast sequences within the effective forecast period from multiple forecast members through a meteorological ensemble forecasting system. Simultaneously, it utilizes the reservoir's on-site telemetry network to receive hydrological dynamic data such as current upstream water level, measured inflow, and actual reservoir capacity. These two dimensions of fundamental data flow to the underlying computational framework, establishing a meteorological forecast and physical state baseline for subsequent flood control timeline simulations and dispatch commands.

[0071] Step 102: Assess flood control capacity based on real-time reservoir hydrological data and extract pre-discharge trigger indicators by combining real-time meteorological ensemble forecast data.

[0072] In this embodiment, the scheduling model extracts the reservoir's current actual capacity and the upper limit of the controlled water level as parameters, and calculates the critical rainfall conditions required for the reservoir to rise to the physical warning line. Furthermore, by combining the cumulative areal rainfall distribution of each meteorological forecast member within the effective forecast period, the consensus probability of the ensemble forecast for the current flood-causing rainfall event is obtained. This data processing process frees the early warning triggering mechanism from a strong dependence on a single, deterministic rainfall value, transforming it into a triggering metric with statistical distribution attributes, thus enhancing the decision-making tolerance for uncertainties in the spatial distribution and intensity of rainfall.

[0073] Step 103: Compare the pre-release trigger index with the pre-configured trigger threshold. If it is determined that the pre-release standby state has been entered, the reservoir emptying time constraint is deduced by using real-time meteorological ensemble forecast data and real-time reservoir water information data, and the decision time margin is calculated.

[0074] Based on this, the indicators are numerically compared with control thresholds dynamically generated based on the current water storage status. Once the conditions are met and the system enters a standby state, the information increment resulting from the convergence of forecast members over time is further explored. The system extracts the defense time boundary marked by early-arriving extreme forecast members and subtracts from this boundary the shortest emptying time required to lower the water level to the target safe level using the maximum capacity of the flood discharge facilities. This difference calculation represents the current time surplus available for waiting for high-confidence forecast information. Under the premise of ensuring the reservoir safely frees up physical storage capacity, it proactively reserves time redundancy to avoid the waste of water resources caused by premature pre-discharge.

[0075] Step 104: Perform forward timing determination based on the decision time margin and output the pre-release timing determination result.

[0076] Accordingly, the forward control module compares the time margin obtained in the previous step with the preset system rolling optimization time span. It calculates a dual quadrant based on the forecast confidence probability interval and the time safety margin, performs branch identification, and ultimately generates a specific instruction to either immediately trigger flood discharge operations or continue to postpone the decision, awaiting the status indicator signal of the next cycle. This diversion decision design changes the static "overshoot equals action" logic mindset, establishing a proactive delay framework that balances forecast quality and flood control safety.

[0077] Step 105: When the pre-release timing determination result indicates that the pre-release operation is triggered, a set of non-dominated schemes containing multiple non-dominated solutions is generated based on real-time meteorological ensemble forecast data and real-time reservoir hydrological data, and initial scheduling control parameters are selected from the set of non-dominated schemes.

[0078] Upon receiving the trigger signal, a mathematical optimization model is constructed using the time series of the target control water level and the pre-released discharge flow as decision-making criteria. The optimization process comprehensively considers three factors: energy loss from water release, the downstream river channel's discharge capacity limit, and the tail probability of over-storage risk due to the dispersion of extreme rainfall forecasts. A multi-objective evolutionary algorithm is used to search for a set of non-dominated solutions within the feasible solution space, and the solution with the optimal comprehensive distance is extracted and issued as a scheduling control command.

[0079] Step 106: Perform pre-leakage operation based on the initial scheduling control parameters; during the pre-leakage execution, extract the measured data within the real physical time window up to the current moment, calculate the inflow deviation of the measured data relative to the forecast inflow benchmark corresponding to the initial scheduling control parameters, and when the inflow deviation meets the preset recalculation conditions, perform multi-dimensional attribution analysis on the source of the deviation, and execute the corresponding graded correction response strategy according to the attribution level, and output the updated scheduling control parameters.

[0080] Furthermore, the pre-release operation enters a continuous closed-loop feedback phase. The monitoring module continuously captures historical measured areal rainfall and inflow sequences up to the current physical moment, excluding forecast periods that have not yet occurred, and then compares the discrete trends of the measured physical quantities with the theoretical forecast baseline. When the deviation exceeds the allowable baseline, the diagnostic logic jointly assesses the spatial deviation of rainfall and the evolution of inflow deviation, diagnosing whether the source of the error is inaccurate meteorological forecasting, errors in hydrological runoff parameters, or communication link interruption. After the diagnostic judgment is completed, the system selectively matches the correction branch strategy with the lowest computational cost, outputting the corrected new round of scheduling and control parameters, reducing the unnecessary computational power consumption and command issuance delays caused by indiscriminately rerunning the optimization model.

[0081] According to one aspect of this application, after outputting the updated scheduling control parameters, the method further includes:

[0082] Optionally, by combining the flow-water-level opening relationship curve of the reservoir's flood discharge gates, the sequence of outflow upper limits for each time period in the initial or updated scheduling control parameters can be converted into the corresponding physical gate opening instruction sequence.

[0083] To facilitate the transfer of logic algorithms to the actual hardware of hydraulic structures, the conversion module addresses pre-set hydrodynamic lookup tables or correlation equations for the flood discharge equipment. This operation maps the upper-level flow rate requirements to specific lifting or opening amplitude parameters of the physical mechanical actuators, transforming data control parameters into engineering quantities that can drive electromechanical components.

[0084] Optionally, according to the set time nodes, the sequence of physical gate opening instructions is sent to the execution control node of the reservoir flood discharge facility for actual discharge operation.

[0085] Furthermore, the industrial communication gateway receives the converted engineering parameters and, based on the pre-configured control timing cycle, transmits the gate opening command to the control cabinet without loss, completing a complete automated engineering closed loop from data detection, timing determination, optimization and correction to physical water flow discharge.

[0086] According to one aspect of this application, in the process of outputting the updated scheduling control parameters, the method further includes:

[0087] Optionally, the status indicator indicating the pre-leakage timing determination result and the deviation source level of the multi-dimensional attribution analysis output can be encapsulated.

[0088] The state aggregation program separates state variables from the decision-making process and assembles and packages the diversion conclusions generated by the look-ahead timing control and the diagnostic level conclusions generated by the execution backtracking phase into a formatted system operation log data message.

[0089] Optionally, a visual scheduling diagnostic early warning signal is generated based on the encapsulated information, and the diagnostic early warning signal is pushed to the reservoir monitoring terminal to assist operators in manual verification.

[0090] Furthermore, the interactive layer acquires the encapsulated data packets, performs graphics rendering and alarm-level color mapping, and transmits the data in real time to the electronic main screen for reservoir command and dispatch or the handheld device of management personnel. This interactive action demonstrates the internal decision-making motivations of the algorithm engine to management personnel, thereby improving the interpretability of intelligent decision-making logic and the availability of manual takeover channels in complex flood response scenarios.

[0091] Example 2:

[0092] Furthermore, this embodiment describes an adaptive pre-release triggering and large-basin spatial analysis mechanism. In one possible implementation, the steps of assessing flood control capacity based on real-time reservoir hydrological data and extracting pre-release triggering indicators in conjunction with real-time meteorological ensemble forecast data include the following steps:

[0093] Step 201: Extract the current actual reservoir capacity based on real-time reservoir hydrological data, calculate the flood control margin by combining the pre-configured upper limit reservoir capacity parameters, and estimate the critical rainfall amount.

[0094] Specifically, the flood control status assessment process involves reading the current actual reservoir capacity data collected by sensors, calculating the difference between this value and the reservoir capacity boundary corresponding to the upper limit of the flood season control water level, to obtain the current flood control margin of the reservoir. This flood control margin represents the incremental water volume that the reservoir can accommodate without exceeding the physical warning line. Based on this flood control margin, the calculation unit calculates the critical rainfall, that is, converts the aforementioned volume capacity into a theoretical rainfall depth index required for the reservoir water level to rise to the upper limit of the flood season control water level.

[0095] According to one aspect of this application, when calculating critical rainfall, such as Figure 2 As shown, it specifically includes:

[0096] Step 2011: Divide the watershed into multiple sub-catchment zones and obtain the area proportion and pre-configured runoff coefficient of each sub-catchment zone.

[0097] Accordingly, when dealing with the uneven spatial distribution of rainfall events in large watersheds or long-duration events, the system performs spatial analysis. The control logic calls geographic information data to divide the total control watershed into several sub-catchment zones and calculates the ratio of the area of ​​a single zone to the total area of ​​the watershed. Simultaneously, it reads the runoff coefficients corresponding to each zone from the historical hydrological feature database. The runoff coefficient is specifically defined as the ratio of the runoff depth of a zone within the forecast period to the average rainfall depth of the same period. Its value is calibrated by backtracking and matching historical rainfall-runoff events, and its value range is defined between 0 and 1.

[0098] Step 2012: Use the area proportion of each sub-sinking zone to perform a weighted summation of the corresponding runoff coefficients to obtain the area-weighted average runoff coefficient.

[0099] Furthermore, the weighted processing module multiplies the corresponding area proportion parameter of each sub-catchment zone with its runoff coefficient, and then linearly sums the product results of all zones. This weighted mapping operation smooths out the spatial heterogeneity of runoff generation and collection capacity within the watershed caused by differences in topography or vegetation, and outputs a global scalar characterizing the comprehensive rainfall-runoff conversion capacity of the entire watershed.

[0100] Step 2013: Divide the flood control margin by the product of the total watershed area and the area-weighted average runoff coefficient to obtain the critical rainfall amount.

[0101] Critical parameters are calculated using the following formula:

[0102] R cr =(V max -V current ) / (A total *φ avg );

[0103] Among them, R cr V represents the calculated critical rainfall amount. max V is the reservoir capacity parameter corresponding to the upper limit of the water level for flood control during the flood season. current A represents the current actual storage capacity. total Represents the total area of ​​the watershed, φ avg This represents the area-weighted average runoff coefficient calculated in the previous step. This calculation expression establishes a quantitative correspondence between flood control physical space and watershed runoff characteristics.

[0104] Step 2014: Calculate the percentage of members in the real-time meteorological ensemble forecast data whose cumulative areal rainfall reaches the critical rainfall level, and obtain the rainfall exceedance probability, which serves as the pre-release trigger indicator.

[0105] Based on this, the feature extraction program traverses the trajectory of each forecast member in the real-time meteorological ensemble forecast data. For a given member, it determines whether its cumulative areal rainfall at the end of the effective forecast period is not less than the calculated critical rainfall value. The counts of members that meet this threshold condition are accumulated, and the total accumulated count is divided by the total number of forecast members. This ratio is the rainfall exceedance probability, reflecting the degree of consensus among the meteorological evolution branches within the ensemble forecast system regarding the occurrence of flood-causing rainfall events, and constitutes a reference system for triggering decision-making.

[0106] According to one aspect of this application, the step of comparing the pre-leakage trigger indicator with a pre-configured trigger threshold includes:

[0107] Optionally, the reservoir capacity utilization rate can be calculated by combining the pre-configured lower limit reservoir capacity parameters, upper limit reservoir capacity parameters, and the current actual reservoir capacity.

[0108] The status assessment module calls the following formula to establish the inventory capacity shortage index:

[0109] r=(V max -V current ) / (V max -V fc );

[0110] In the formula, r is the warehouse capacity utilization rate variable, and V maxV is the reservoir capacity parameter corresponding to the upper limit of the water level for flood control during the flood season. current V represents the current actual storage capacity. fc This is the reservoir capacity benchmark corresponding to the pre-configured lower limit of the flood season control water level. A truncation rule is included during calculation: when the current actual reservoir capacity is lower than the lower limit benchmark, resulting in ample buffer space in the reservoir, the reservoir capacity utilization rate r is truncated to 1 to prevent the threshold from exceeding the preset upper limit.

[0111] Optionally, the storage capacity utilization rate is used to perform interpolation between the pre-configured lower and upper limits of the probability threshold to generate an adaptive exceedance probability threshold, which is then used as the pre-configured trigger threshold.

[0112] The dynamic threshold generator uses the interpolation formula: α adapt =α min +(α max -α min The calculation is performed using )*max(0,min(r,1)).

[0113] Where, α adapt α represents the adaptive exceedance probability threshold. min With α max These represent the pre-configured lower and upper limits of the probability threshold, respectively, and their engineering values ​​can be set as constants within the range of 0.20 to 0.60. The above mechanism establishes a linkage constraint relationship between reservoir water storage status and pre-release trigger sensitivity. When the reservoir water level approaches the upper limit boundary, causing the reservoir capacity utilization rate to tend towards 0, the interpolation result shifts downwards towards the lower threshold, allowing a relatively low consensus probability of rainfall to activate the pre-release command. When the reservoir water level is low, causing the reservoir capacity utilization rate to remain at 1, the threshold shifts upwards towards the upper threshold, raising the trigger threshold to reduce false alarms.

[0114] Optionally, the rainfall exceedance probability can be compared with the adaptive exceedance probability threshold to determine whether to enter the pre-discharge standby state.

[0115] The comparison and judgment module receives the rainfall exceedance probability output from the meteorological terminal and the adaptive exceedance probability threshold output from the hydrological terminal, and performs numerical magnitude judgment. When the measured exceedance probability is lower than the adaptive threshold, the system maintains a normal standby state, suspends the current batch of pre-release judgment flow, and waits for the next data rollout refresh. When the rainfall exceedance probability is greater than or equal to the adaptive threshold, the judgment condition is met, and the control permission flow enters the forward-looking timing pre-release standby analysis state.

[0116] Example 3:

[0117] Furthermore, this embodiment utilizes real-time meteorological ensemble forecast data to extrapolate reservoir emptying time constraints, calculate decision-making time margins, and outline the specific steps for determining the timing of forward-looking assessments, such as... Figure 3 As shown, it includes:

[0118] Step 301: Extract the cumulative areal rainfall process of each forecast member in the real-time meteorological ensemble forecast data one by one.

[0119] As an optional implementation, the time-series data parsing module traverses and decomposes the meteorological data stream, reading the cumulative rainfall sequence corresponding to each forecast member along the time axis. This operation reduces the dimensionality of the multi-dimensional ensemble forecast matrix into multiple independent rainfall evolution trajectories, providing the underlying data sequence foundation for subsequently capturing the time nodes of flood threat.

[0120] Step 302: The moment when the cumulative areal rainfall of each forecast member first reaches the critical rainfall amount is taken as the flood critical time of the corresponding forecast member.

[0121] Furthermore, the operation performs time-domain out-of-bounds detection. For any s-th forecast member, the cumulative rainfall depth from the current time to the future time t is extracted, and the conditional discriminant t is used. s =min(t) solves for the minimum time index, where the search constraint is set to the cumulative areal rainfall of the member being greater than or equal to the critical rainfall estimated from upstream. This calculation process captures the earliest point of occurrence that can trigger the risk of exceeding limits, marking it as the flood-inducing critical time under an independent meteorological evolution branch.

[0122] Step 303: Extract the low quantile from the sample set containing each flood critical time to obtain the conservative critical time.

[0123] The statistical calculation program aggregates and combines the flood-inducing critical times output by all valid forecast members into a one-dimensional array distribution, and uses the quantile calculation expression t. arr =quantile(t s ,p a Extract the lower boundary of the defense time. Parameter p a The pre-configured quantile limits are typically set between 0.05 and 0.10 in engineering terms. This calculation assigns higher decision weight to a small number of extreme forecast members that arrive earlier, ensuring that there is sufficient time leeway to counter severe weather conditions.

[0124] In extreme cases where the sample set is empty, the control flow triggers alternative handling strategies. The calculation of the decision time margin also includes strategies for handling extreme cases, specifically: when the cumulative areal rainfall of all forecast members has not reached the critical rainfall amount, a conservative critical time is set to the end of the effective forecast period. This strategy avoids computational interruptions caused by empty arrays and guides the decision network to continue evolving downstream by assigning a conservative constant to the time axis.

[0125] Step 304: Calculate the target storage capacity that needs to be freed up based on the current actual storage capacity and the storage capacity corresponding to the pre-configured target flood level.

[0126] Accordingly, the volume difference calculation unit reads the real-time capacity of the reservoir and introduces a predefined water level benchmark representing the lower tolerance limit for subtraction. To ensure that this calculation process is decoupled from the scheduling optimization variables in subsequent algorithm modules, the reservoir capacity corresponding to the target flood level called in this stage is taken from the pre-calibrated static safety lower limit control line.

[0127] Step 305: Combine the pre-configured maximum pre-discharge capacity parameter with the current measured inflow rate extracted from the real-time water level data of the reservoir to calculate the shortest reservoir emptying time required to empty the target reservoir capacity.

[0128] The emptying time is calculated using the following relationship τ ev =(V current -V h1 ) / (Q max_pre -Q in_0 Obtain the time parameter.

[0129] In the formula, τ ev V represents the shortest duration of the Tengku (or Tengku) process. current V represents the current actual storage capacity. h1 Q refers to the reservoir capacity corresponding to the aforementioned target flood level. max_pre Q is the maximum pre-discharge capacity parameter of the system. in_0 This represents the current measured inflow rate transmitted back by the sensor. Based on the local assumption of a stable inflow trend within a short-term window, this calculation derives the theoretical minimum duration required to achieve storage capacity clearance under extreme emission conditions.

[0130] Under extreme operating conditions with severely limited hydraulic capacity, the dispatching system executes a pre-set intervention alternative branch. When the maximum pre-discharge capacity parameter is not greater than the current measured inflow, it is determined that physical reservoir emptying is not feasible, and an execution command is generated according to the maximum pre-discharge capacity parameter to activate the flood discharge facilities. This response avoids anomalies caused by divisors approaching zero or being negative, activating the maximum defensive posture under extreme water balance conditions.

[0131] Step 306: Subtract the shortest emptying time after the pre-configured timing safety factor is amplified from the conservative critical time to obtain the decision time margin.

[0132] Furthermore, the look-ahead discrimination module, based on formula T... mar =t arr -κ*τ ev Complete the synthesis of the margin index. Variable T marThis refers to the decision-making time margin, with coefficient κ used to compensate for the additional time overhead caused by the delay in opening and closing mechanical gates and the lag in reservoir water flow evolution. In one example, assuming the extracted conservative critical time is 6.0 units, the minimum emptying time required to drain the target reservoir capacity is calculated to be 2.0 units, and the pre-configured timing safety factor is set to 1.5, the calculated decision-making time margin is 6.0 - 2.0 * 1.5 = 3.0 units. This positive value index represents the redundancy tolerance that the system can use to delay decisions and wait for high-confidence weather forecasts while meeting the emptying safety constraints.

[0133] Step 307: Perform forward timing determination based on the decision time margin and output the pre-release timing determination result.

[0134] Upon entering the timing determination level, the system terminates the serial progression and, based on the multi-condition concurrent discrimination network, introduces three mutually exclusive parallel state output branches.

[0135] In the first scenario, when the probability of rainfall exceeding the threshold is not less than a pre-configured high-confidence probability threshold, a pre-release timing determination result is generated, indicating that a pre-release operation will be triggered. This scenario indicates that the meteorological hazard has reached a high degree of certainty, and the system issues a flood control release command after exceeding the time margin determination procedure.

[0136] In the second scenario, when the rainfall exceedance probability falls between the adaptive exceedance probability threshold and the high confidence probability threshold, and the decision time margin is greater than the pre-configured rolling optimization cycle, a pre-release timing determination result is generated, indicating that the pre-release standby state should be maintained and the review should be postponed to the next cycle. Within this state space, the time redundancy supports the waiting strategy, thereby allowing the system to utilize the convergent characteristics of ensemble forecasts over time to obtain better decision information sources and avoid unnecessary water waste.

[0137] In the third scenario, when the rainfall exceedance probability is between the adaptive exceedance probability threshold and the high confidence probability threshold, and the decision time margin is no greater than the rolling optimization cycle, a pre-release timing determination result is generated, indicating that the latest safe decision time has been reached and triggering the pre-release operation. This scenario indicates that the system has approached the lower limit of the safe storage time, the control flow terminates the waiting action, and enters the execution scheduling intervention.

[0138] Example 4:

[0139] Furthermore, this embodiment describes the specific construction steps for generating initial scheduling control parameters based on a multi-objective optimization strategy, and minimizing the conditional risk value of excess storage capacity.

[0140] In one possible implementation, the step of generating initial scheduling control parameters based on a multi-objective optimization strategy is as follows: Figure 4 As shown, it includes the following steps:

[0141] Step 401: Construct a multi-objective optimization set, which includes the expected minimization objective of water discharge to reflect the loss of benefits from the reservoir, the expected minimization objective of peak outflow to reflect downstream flood control pressure, and the conditional risk value minimization objective of excess storage capacity to reflect downstream flood control risk.

[0142] Specifically, the computation module establishes three parallel optimization indices within the feasible solution space. The expected minimization objective for water release represents the total net discharge volume when the reservoir outflow exceeds the inflow, used to assess the loss of stored water resources. The expected minimization objective for peak outflow extracts extreme value variables from the outflow sequence for each time period, used to constrain downstream flow and ensure the safety of downstream river embankments and protected areas. The conditional risk value minimization objective for excess storage capacity focuses on low-probability, high-risk events at the tail end of the meteorological distribution, calculating the expected cumulative excess when the reservoir capacity exceeds the upper limit of the flood season control water level for each time period.

[0143] The discretized forms of the three objective functions mentioned above are as follows: Let the decision variable be the reservoir capacity V corresponding to the target flood level. h and the upper limit sequence of outbound flow for each time period Q out_k The forecast member index is s, the time period index is k, and the total number of set members is N. s The total number of time periods is K.

[0144] The objective of minimizing the expected amount of water to be discarded is expressed as: min f1 =(1 / N s )*Σ s *Σ k max(Q out_k_s -Q in_k_s ,0)*Δt, where Q out_k_s Let Q be the actual outbound flow rate in the k-th time period under the scenario of the s-th member. in_k_s This corresponds to the inbound flow.

[0145] The objective of minimizing the peak outbound flow rate is expressed as: min f2 =(1 / N s )*Σ s max k (Q out_k_s ).

[0146] The objective of minimizing the conditional value at risk of excess storage capacity is expressed as: min f3 =CVaR β [L], where L=Σ k max(V k_s -V max ,0) represents the cumulative excess storage capacity of a single member, CVaR β This represents the conditional risk value of the total excess reserves of all members at a confidence level β.

[0147] Where V k_s V is obtained recursively from the water balance equation. k_s =V (k-1)_s +(Q in_k_s -Q out_k_s )*Δt.

[0148] According to one aspect of this application, when constructing the conditional value-at-risk minimization objective for excess storage capacity, its confidence level parameter is determined through the following dynamic adaptive method:

[0149] Step 4011: Extract the cumulative inflow of water into the reservoir for each forecast member within the effective forecast period from the real-time meteorological ensemble forecast data, and calculate the mean and standard deviation of the cumulative inflow of water into the reservoir for each forecast member.

[0150] Accordingly, the feature sampling unit traverses all ensemble forecast branches in the meteorological data stream, accumulating the input water volume for each forecast member throughout the entire forecast window. After collecting multiple sets of volume variables, the statistical analysis program calculates the mean of the cross-member mathematical expectation of the set of values, and simultaneously solves the standard deviation of the set of values ​​from the mean, obtaining the basic statistical measures reflecting the distribution characteristics of the forecast sample.

[0151] Step 4012: Divide the standard deviation by the mean to obtain the coefficient of variation of the current weather ensemble forecast.

[0152] To reduce the impact of absolute values ​​at different rainfall levels, the variability assessment module performs a division operation. Specifically, this is done through the expression CV=σ. W / μ W Complete the conversion, where CV represents the discrete coefficients, and σ W μ represents the standard deviation of the output calculated in the previous step. W This represents the mean. This dimensionless coefficient is used to assess the degree of differentiation and uncertainty among members within a given weather ensemble forecast.

[0153] Step 4013: Using the ratio of the discrete coefficients to the pre-configured reference discrete coefficients, perform dynamic interpolation between the pre-configured lower and upper confidence levels to generate the current confidence level parameters.

[0154] The parameter regulator performs dynamic adaptive reconstruction based on pre-stored historical benchmarks, and the calculation expression is β. t =β min +(β max -β min )*min(CV / CV ref,1 );

[0155] In the formula, β t For the generated current confidence level parameter, β min With βmax These are the pre-configured lower and upper confidence levels, respectively, with an engineering setting range typically between 0.90 and 0.99. (CV) ref The reference dispersion coefficients are calibrated from historical return data. This truncated interpolation operation allows the output parameters to fluctuate synchronously with meteorological uncertainties, avoiding the rigid behavior of static constant parameters under complex meteorological conditions.

[0156] Step 4014: Input the current confidence level parameter as the risk tail truncation boundary into the conditional risk value minimization objective of the excess storage capacity, so that the flood control risk control intensity is automatically increased when the uncertainty of weather forecast increases.

[0157] Furthermore, this data injection action modifies the tail boundary constraints in the multi-objective optimization ensemble. When the ensemble dispersion is high, the confidence level parameter approaches its upper limit, forcing the conditional value of risk calculation to focus more deeply on the extreme and severe scenarios of the distribution curve, increasing the penalty weight for over-storage risk. As the forecast tends to converge, this parameter falls back, guiding the optimization model to release excess flood control redundancy to also consider beneficial water storage.

[0158] Step 402: Based on the inflow scenarios of each forecast member driven by real-time meteorological ensemble forecast data, combined with the preset reservoir water balance equation constraints and preset physical discharge capacity constraints, a multi-objective optimization algorithm is used to solve the multi-objective optimization set to generate a set of non-dominated schemes containing multiple non-dominated solutions.

[0159] Before performing the optimization calculation, physical boundary conditions are applied to the constraint construction layer. The reservoir water balance equation constraint is expressed as V. k =V (k-1) +(Q in -Q out )*△ t This constrains the material conservation relationships between reservoir capacity, input flow, output flow, and time step size in adjacent time periods. Physical discharge capacity constraints set that the outflow in each time period does not exceed the facility hardware limits and downstream safety discharge limits. Furthermore, the solver employs a multi-objective evolutionary algorithm framework for iterative processing. The engineering guidance value for the population size is set between 100 and 200, the number of iterations is limited to 200 to 500 generations, and the crossover probability is set to 0.9.

[0160] In some alternative implementations, to address the time-sensitive needs of emergency scenarios, the system provides a branch option to activate a fast computation mode. When this mode is enabled, the computing unit automatically reduces the population size and maximum number of iterations to 50% of the conventional baseline values. This alternative reduces the computational convergence time by decreasing the global search breadth and depth of the algorithm, thus meeting the high-frequency response requirements of extreme proximity scheduling commands.

[0161] Step 403: Calculate the comprehensive proximity of each scheme in the non-dominated scheme set using the spatial distance metric method, select the scheme with the highest comprehensive proximity, and extract the corresponding target flood level and the upper limit sequence of outflow for each time period as the initial scheduling and control parameters.

[0162] The decision-making and screening module evaluates the distances to multiple feasible scheduling boundaries within the set of non-dominated solutions. Considering the inconsistencies in the physical units of the three objectives (cubic meters, cubic meters per second, and expected risk), the program processes the objective matrix using range normalization or vector normalization operators before calculating the distances, mapping each objective value to a dimensionless space of the same scale. After dimension elimination, the specific calculation process is as follows: for the normalized objective matrix, the positive ideal solution f is determined. pos (A vector consisting of the minimum values ​​of each objective across all possible solutions) and the negative ideal solution f neg (A vector consisting of the maximum values ​​of each objective across all solutions). For the i-th solution, calculate its Euclidean distance D to the positive ideal solution. i_pos =sqrt(Σ j (f i_j -f pos_j ) 2 The Euclidean distance D to the negative ideal solution i_neg =sqrt(Σ j (f i_j -f neg_j ) 2 ), where j is the target index. Using formula C i =D i_neg / (D i_pos +D i_neg ) Calculate the overall similarity, C i The value ranges from 0 to 1, with a larger value indicating that the solution is closer to the ideal solution. The sorter selects the data sequence with the highest degree of closeness as the basis for the scheduling decision parameters to be physically executed, thus completing the entity transformation from algorithm generation to engineering parameters.

[0163] Example 5:

[0164] In one possible implementation, this embodiment further details the specific steps for calculating the inflow deviation from measured data within the actual physical time window up to the current moment, as well as the multi-dimensional attribution analysis and graded correction response strategy, including:

[0165] Optionally, it can determine the relative deviation between the measured inflow and the predicted inflow baseline value corresponding to the initial scheduling control parameters, and whether it continuously exceeds the allowable ratio within a preset time period. If so, it can trigger a multi-dimensional attribution analysis.

[0166] As an optional implementation, the deviation monitoring module maintains high-frequency monitoring throughout the pre-release execution lifecycle. This module extracts the measured inflow sequence within the rolling cycle and compares it with the forecast inflow baseline value upon which the initial scheduling control parameters were generated. The forecast inflow baseline value is the weighted arithmetic mean of the inflow sequences of each forecast member, representing the central trend estimate of the ensemble forecast.

[0167] To filter high-frequency noise caused by instantaneous fluctuations in water body evolution, a sliding time observation benchmark is set in the computational logic. The state machine only outputs a trigger command to enter downstream deep-source analysis when the absolute value of the relative error exceeds a pre-configured tolerance ratio for more than half of the statistical time periods, thus avoiding frequent computational interruptions due to short-term pulsations. The tolerance ratio is determined as follows: based on historical return data obtained during the offline pre-configuration phase, the temporal distribution of the relative deviation between the measured inflow and the corresponding forecast inflow benchmark value in each historical period is statistically analyzed. The upper quantile of its normal random fluctuation range is taken as the tolerance ratio to distinguish between systematic deviations and random fluctuations. Those skilled in the art can determine an appropriate value based on the forecast accuracy characteristics of the target watershed using the above statistical method.

[0168] According to one aspect of this application, when extracting measured average rainfall from a real physical time window, a safeguard mechanism for data quality control is also included:

[0169] Optionally, when missing or bad data is detected in the reservoir's on-site telemetry system, spatial interpolation is performed using the historical rainfall distribution weights of spatially adjacent observation stations to fill the data gaps within the actual physical time window, ensuring the continuity of the data sequence input for multi-dimensional attribution analysis.

[0170] The quality control program performs integrity checks on the raw observation arrays returned by the physical sensor links. Since extreme rainfall conditions can easily induce communication disruptions at individual stations, the extraction logic, upon detecting a sequence breakpoint, uses a weighted inversion based on the pre-set topological distances between adjacent stations and the spatial correlation coefficients of historical rainfall in the static database. This interpolation process generates surrogate observations to fill gaps, maintaining the temporal continuity of the matrix and providing a data foundation for subsequent consistency comparisons.

[0171] Optionally, the measured average rainfall and the forecast rainfall for each forecast member can be extracted from the actual physical time window.

[0172] Based on this, the data sampling unit performs a time-domain truncation operation. The right boundary of the real physical time window is locked at the current physical timestamp, and the algorithm does not extract measured climate data from future prediction windows. By limiting the observation interval, the system calculates the mean vector of actual areal rainfall within the historical closed interval and the rainfall forecast vectors of each meteorological branch for the corresponding time period, establishing the lateral comparison caliber between the two.

[0173] Optionally, the absolute difference between the measured average rainfall and the corresponding forecast rainfall is calculated, and the absolute difference is divided by the larger of the measured average rainfall and the pre-configured zero-prevention lower limit to obtain the consistency deviation of each forecast member used for cross-event horizontal comparison.

[0174] The computational kernel executes the relation e s =abs(P obs -P s ) / max(P obs Extract consistency parameters (P0).

[0175] Among them, e s P represents the calculated consistency deviation. obs P represents the measured average rainfall. s P represents the forecast isal rainfall, and P0 represents the pre-configured lower limit parameter for zero-rainfall protection. Under conditions of very weak or no actual rainfall, P... obs When the denominator approaches zero, the max operator can be used to perform division operations using P0. This zero-prevention design prevents mathematical overflow of deviation values ​​caused by an excessively small denominator, and the dimensionless ratio output supports objective evaluation of rainfall events across different magnitudes.

[0176] Optionally, the measured average rainfall is input into a pre-configured rainfall-runoff model for simulation to obtain the simulated inflow; the measured inflow and simulated inflow within the same window are extracted, the positive and negative distribution of their difference is recorded, and an inflow deviation sign sequence is generated.

[0177] Specifically, the feature generation module calls a pre-configured watershed rainfall-runoff model for synchronous simulation. The process is as follows: the measured average isothermal rainfall within the actual physical time window is used as the rainfall-driving input; the runoff coefficient, a basic hydrological parameter calibrated offline in stages, is used as the basis for runoff generation and conversion; and the simulated inflow process line is output after runoff generation and confluence calculation. The rainfall-runoff model can be implemented using a conventional conceptual hydrological model in this field; those skilled in the art can select the specific model type based on the watershed characteristics. The conceptual model receives the measured isothermal rainfall from the same period as the driving input and outputs the simulated flow process line after runoff generation and confluence evolution. The comparison module subtracts the simulated flow from the sensor-returned measured flow, removes specific amplitude variables, and extracts only the mathematical sign features of the difference column, forming a one-dimensional array with alternating positive and negative signs, used to determine whether there is a systematic unidirectional bias tendency in the hydrological response process.

[0178] Before performing attribution level determination, two pre-configured parameters need to be determined. The first is the consistency tolerance, which is determined as follows: based on historical return data obtained during the offline pre-configuration phase, calculate the consistency deviation of each forecast member for each historical rainfall event, statistically analyze the empirical distribution of the consistency deviation of all members in events where the forecast successfully tracks the actual situation, and take the specified upper quantile of this distribution as the judgment boundary distinguishing between weather forecast tracking of the actual situation and weather forecast deviating from the actual situation. The second is the scaling parameter e0 in the exponential kernel function, which physically controls the rate at which the confidence weight decays with the consistency deviation. It can be set to the same order of magnitude as the consistency tolerance, and those skilled in the art can determine it through conventional adjustments based on the concentration requirements of the weight allocation.

[0179] Optionally, the specific attribution level classification logic and graded correction response strategy include the following parallel discriminant branches.

[0180] In the first attribution scenario, when the consistency deviation of at least one forecast member is no greater than the pre-configured consistency tolerance, and the inflow deviation sign sequence remains the same within the actual physical time window, a first-level attribution decision is output, indicating a hydrological system deviation. For the first-level attribution decision, the tiered correction response strategy is as follows: when the attribution is a first-level attribution decision, the ratio of measured inflow to simulated inflow within the actual physical time window is calculated, and the hydrological correction factor is updated using the following exponentially weighted filtering recursive formula: λ new =α f *(Q obs_k / Q sim_k )+(1-α f )*λ old ;

[0181] Where, λ new For the updated hydrological correction factor, λ old Q is the hydrological correction factor for the previous time period, with an initial value set to 1. obs_k With Q sim_k These represent the measured inflow and simulated inflow during the k-th time period within the real physical time window, respectively, α. f The filtering smoothing coefficient can be determined by those skilled in the art using conventional methods based on the number of effective sample periods within the time window. The updated hydrological correction factor λ is then used. new Multiply the inflow sequence corresponding to each forecast member by λ. new This completes the synchronous correction of the inflow scenario; simultaneously multiplying the area-weighted average runoff coefficient by λ. new Runoff parameters are corrected to align subsequent water balance projections with measured responses. Before retaining the initial scheduling and control parameters, the system performs a tail-end risk verification operation.

[0182] Specifically, this includes: utilizing the updated hydrological correction factor λ newThe inflow sequence of each forecast member is corrected, and the corrected inflow scenarios of each member are substituted into the reservoir water balance equation V for each time period. k =V (k-1) +(Q in，修正 -Q out )*△ t The evolution of reservoir capacity for each forecast member is deduced; based on the current confidence level parameter β... t Calculate the reservoir capacity exceeding the upper limit of the water level V for each time period. max The conditional risk value is determined. If the conditional risk value is not greater than the preset safety risk tolerance, the tail risk review is deemed passed, and the initial scheduling control parameters are retained as the updated scheduling control parameters. If the conditional risk value is greater than the preset safety risk tolerance, the tail risk review is deemed failed, and the control flow is transferred to the full recalculation branch corresponding to the third-level attribution judgment. The safety risk tolerance can be determined by those skilled in the art based on the reservoir's safety level and the importance of downstream protected objects through engineering experience.

[0183] When it is determined that the weather forecast has not deviated from its overall course and the model exhibits unidirectional systematic errors, the system initiates lightweight parameter iteration. It should be noted that the simplified operation of modifying the runoff coefficient with the inflow correction factor has applicable boundaries, primarily constrained to implementation in engineering scenarios where the pre-release time window is short and total quantity control takes precedence over process allocation. This branch completes state correction with relatively low computational overhead.

[0184] In the second attribution scenario, when the consistency deviation of some forecast members is no greater than the consistency tolerance, while the consistency deviation of the remaining forecast members is greater than the consistency tolerance, a second-level attribution judgment is output, indicating forecast reliability divergence. For the second-level attribution judgment, the hierarchical correction response strategy is as follows: A pre-configured exponential kernel function is used to convert the consistency deviation of each forecast member into a corresponding reliability weight, ensuring that members with smaller consistency deviations receive higher weight allocations; the reliability weights of each forecast member are used to replace the equal-weight coefficients to reconstruct the expected value and conditional risk value of the scheduling optimization objective function; within the non-dominated scheme set, the comprehensive proximity of each scheme is recalculated based on the reconstructed objective function, and the scheme with the highest proximity is selected. The control variables corresponding to this scheme are extracted as updated scheduling control parameters, avoiding the need to rerun the optimization iteration algorithm.

[0185] The calculation uses the formula w s =exp(-e s / e0) / Σ(exp(-e j / e0)) performs the mapping transformation.

[0186] In the formula, the variable w s Represents the reliability weight, e s Represents the consistency deviation, and e0 is the scale parameter that controls the decay rate.

[0187] In the third attribution scenario, when the consistency deviation of all forecast members exceeds the consistency tolerance, or after the corrective response corresponding to the first or second level attribution judgment is executed, the corrected inflow scenario is substituted into the water balance equation to calculate the predicted reservoir capacity for each time period. When the conditional risk value of the reservoir capacity exceeding the pre-configured upper limit water level reservoir capacity parameter exceeds the preset safety risk tolerance, and the risk review is deemed to have failed, a third level attribution judgment is output, indicating that the overall forecast is inaccurate. When the attribution is a third level attribution judgment, the latest acquired real-time meteorological ensemble forecast data and real-time reservoir hydrological data are used to cover the initial optimized input conditions, and the multi-objective optimization algorithm is re-run for full optimization iteration to generate updated scheduling and control parameters.

[0188] Faced with extreme conditions where the meteorological input field completely deviates from the actual evolution trajectory, the control engine triggers a full-load operation and calls up the population and generation reduction modes in conjunction with it to find a safe and feasible solution through a new round of evolutionary calculations.

[0189] In the fourth attribution scenario, when the meteorological ensemble forecast data stream is detected to be in a communication interruption or unavailable state, a fourth-level attribution judgment is output, indicating a data interruption. For the fourth-level attribution judgment, the tiered corrective response strategy is as follows: Deterministic single-value meteorological forecast data is obtained through a backup data channel; the triggering logic of the dynamic adaptive rainfall threshold is masked; and the pre-release trigger judgment is taken over using a fixed rainfall forecast threshold; based on the single-value meteorological forecast data and the pre-configured empirical data saturation time parameter, updated scheduling control parameters are generated according to the maximum pre-release capacity parameter.

[0190] As a fallback alternative, the system automatically reverts to single-value response mode when high-dimensional data sources are lost. Under this control chain, the computation process skips the complex deduction steps of look-ahead time margin and instead calls the pre-configured experience-based time allocation parameter to statically take over the allocation of the total pre-leakage period, ensuring that the underlying defense mechanism still maintains its basic operational capability in a harsh information environment.

[0191] Taking a flood season scheduling process as an example, assuming a reservoir's current capacity is close to its flood limit and the capacity utilization rate (r) is calculated to be 0.15, the system generates a low adaptive exceedance probability threshold. If a member of the current meteorological ensemble forecast predicts that the cumulative areal rainfall will exceed the critical rainfall amount, the system enters a pre-release standby state. Time margin calculation reveals that the current margin is greater than the rolling optimization period, so the system chooses to postpone the decision. In the next rolling period, the margin has shrunk to no greater than the rolling period, the system determines that the latest safe decision time has been reached and triggers the pre-release operation, generating initial scheduling control parameters and executing the release. After two periods, the measured inflow remains consistently high. Consistency deviation analysis shows that some members have smaller deviations while others have larger ones, attributed to the differentiation in the reliability of the second-level forecast. The system re-selects a scheme from the existing non-dominated scheme set through reliability weight reconstruction, completing the error correction without rerunning.

[0192] Example 6:

[0193] Based on the above embodiments, this embodiment provides offline pre-configuration steps to support the operation of the method. In one possible implementation, before acquiring real-time meteorological ensemble forecast data and real-time reservoir hydrological data, such as... Figure 5 As shown, it also includes offline pre-configuration steps to support the operation of the method:

[0194] Step 601: Obtain historical meteorological ensemble forecast reports for the target watershed and measured hydrological data of reservoirs during the same period.

[0195] The data preprocessing module connects to meteorological and hydrological databases, batch loading ensemble forecast return sequences of the target watershed over the past few years, along with actual reservoir inflow and water level logs for the corresponding time periods. The return data mentioned in this paper specifically refers to ensemble forecast data generated by posterior simulations of historical rainfall events using existing meteorological forecasting models. This offline data alignment operation provides the underlying sample foundation for the system to extract the steady-state mapping relationship between climate and hydrology.

[0196] Step 602: Calibrate basic hydrological parameters based on historical rainfall-runoff matching relationships.

[0197] The offline calibration engine extracts multiple typical rainstorm and flood events recorded in historical logs. For a single rainfall event, the computing nodes calculate the measured total net inflow runoff depth and the measured average isothermal rainfall depth during the same period. By performing nonlinear fitting or parameter optimization on the input-output relationship of multiple historical sample events, the runoff-generation conversion ratio with the minimum fitting error is obtained. This extracted ratio is defined as a basic hydrological parameter, namely the runoff coefficient, which reflects the background runoff mapping characteristics of the target watershed under normal climatic conditions.

[0198] Step 603: Analyze the set dispersion distribution of each historical rainfall event in the statistical report data and extract the reference dispersion coefficient.

[0199] When constructing a baseline for forecast uncertainty, a statistical algorithm iterates through each independent rainfall event in the historical return data. For a single event, the system reads the predicted inflow of water for each forecast member, calculates the standard deviation and mean of the predicted sample, and takes the quotient of the two to output the degree of discrete variability of the single event.

[0200] Furthermore, a statistical distribution map is constructed by aggregating all single-iteration variation data from the historical sample database. The data extraction layer extracts quantiles or medians from this distribution sequence and labels them as reference coefficients of dispersion. This feature parameter characterizes the level of normalized uncertainty in the ensemble weather forecast for the target watershed.

[0201] Step 604: Store the basic hydrological parameters and reference dispersion coefficients as pre-configured parameters in the system for real-time online retrieval.

[0202] The data persistence module writes the basic hydrological parameters and reference dispersion coefficients generated from the aforementioned offline calibration, as well as the upper and lower limits of probability thresholds and physical reservoir capacity boundaries pre-specified during the system design phase, into the static configuration storage area of ​​the scheduling platform. After the system switches to online business mode, the real-time computing kernel reads this memory area by calling the corresponding port to complete parameter addressing, using these normalized benchmark constants as anchor points for online dynamic interpolation and risk assessment. This process clearly defines the technical boundaries between time-consuming offline calibration and high-frequency online adaptive computation in the engineering architecture.

[0203] This solution addresses the problem of premature water release before the arrival of the rainfall front in traditional scheduling methods, which rely on fixed thresholds and immediate triggering mechanisms. Instead, it employs an adaptive triggering threshold generated by interpolation based on actual reservoir capacity utilization, and extrapolates the reservoir release time constraint to calculate the decision-making time margin. This method enables the system to proactively reserve time redundancy to confirm high-confidence weather forecasts while ensuring flood control safety, filtering out false alarms about invalid rainfall, and minimizing unnecessary water resource losses.

[0204] Meanwhile, a method for extracting consistency deviations and symbol sequences from real physical time windows is introduced to construct a multi-dimensional attribution analysis logic, improving the computational waste and instruction oscillation problems caused by indiscriminate recalculation when facing forecast deviations during the execution period. This method distinguishes between different physical error sources such as meteorological inaccuracies and hydrological parameter errors, and compresses the computational overhead of the global optimization algorithm and shortens the issuance delay of emergency correction instructions by matching lightweight parameter updates or selecting optimization without reruns in a hierarchical response strategy. For the problem that static scheduling optimization is difficult to adapt to the uncertainty of variable weather, a method is adopted to dynamically adjust the conditional risk value confidence level parameter based on the discrete coefficient of meteorological ensemble forecast. This method enables the optimization model to automatically tighten the tail flood control sensitivity when the divergence of extreme weather increases, and release the water level safety space when the water situation is stable, realizing a dynamic and elastic balance between flood control safety and beneficial water storage.

[0205] In this application, the relevant selections, parameter values, and process steps given are merely exemplary embodiments and are not intended to limit the scope of protection of this invention. Based on the core technical logic of this invention, the devices, parameters, and implementation methods can be equivalently replaced and adjusted according to the actual application scenario.

[0206] The order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments.

[0207] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0208] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An adaptive decision-making method for pre-release of water during the flood season of a reservoir, characterized in that, include: Acquire real-time meteorological ensemble forecast data and real-time reservoir hydrological data; The flood control capacity is assessed based on real-time reservoir hydrological data, and pre-release trigger indicators are extracted by combining real-time meteorological ensemble forecast data. The pre-release trigger index is compared with the pre-configured trigger threshold. If it is determined that the pre-release standby state has been entered, the reservoir emptying time constraint is deduced by using real-time meteorological ensemble forecast data and real-time reservoir hydrological data, and the decision-making time margin is calculated. Based on the decision-making time margin, the timing of the pre-release is determined, and the result of the pre-release timing determination is output. When the pre-discharge timing determination result indicates that the pre-discharge operation is triggered, the initial scheduling control parameters are generated based on the multi-objective optimization strategy. During the pre-release execution, the measured data within the real physical time window up to the current moment is extracted to calculate the inflow deviation. When the inflow deviation meets the preset recalculation conditions, a multi-dimensional attribution analysis is performed on the source of the deviation, and the corresponding graded correction response strategy is executed according to the attribution level, and the updated scheduling control parameters are output.

2. The method according to claim 1, characterized in that, The steps for assessing flood control capacity based on real-time reservoir hydrological data and extracting pre-release trigger indicators by combining real-time meteorological ensemble forecast data include: The actual reservoir capacity is extracted based on real-time reservoir hydrological data, and the flood control margin is calculated by combining the pre-configured upper limit reservoir capacity parameters with the water level. The critical rainfall amount is also estimated. The percentage of members whose cumulative areal rainfall reaches the critical rainfall level in the real-time meteorological ensemble forecast data is statistically analyzed to obtain the rainfall exceedance probability, which serves as a pre-release trigger indicator. The steps for comparing the pre-leakage trigger metric with the pre-configured trigger threshold include: The reservoir capacity utilization rate is calculated by combining the pre-configured lower limit water level reservoir capacity parameters with the current actual reservoir capacity. By interpolating the storage capacity utilization rate between the pre-configured lower and upper probability thresholds, an adaptive exceedance probability threshold is generated, which serves as the pre-configured trigger threshold. The rainfall exceedance probability is compared with the adaptive exceedance probability threshold to determine whether to enter the pre-discharge standby state.

3. The method according to claim 2, characterized in that, The steps for extrapolating reservoir emptying time constraints and calculating decision-making time margins using real-time meteorological ensemble forecast data include: Extract the cumulative areal rainfall process of each forecast member in the real-time meteorological ensemble forecast data one by one; The moment when the cumulative areal rainfall of each forecast member first reaches the critical rainfall amount is taken as the flood critical time for the corresponding forecast member; The conservative critical time is obtained by extracting the low quantile from the sample set containing each flood critical time. The target storage capacity that needs to be freed up is calculated based on the current actual storage capacity and the storage capacity corresponding to the pre-configured target flood level. By combining the pre-configured maximum pre-discharge capacity parameter with the current measured inflow extracted from the real-time water level data of the reservoir, the shortest reservoir emptying time required to empty the target reservoir capacity is calculated. The decision time margin is obtained by subtracting the shortest emptying time after the pre-configured timing safety factor is amplified from the conservative critical time.

4. The method according to claim 3, characterized in that, The steps for determining the timing of pre-emptive release based on the decision-making time margin include the following three tiered determination scenarios: When the probability of rainfall exceeding the threshold is not less than the pre-configured high confidence probability threshold, a pre-release timing determination result is generated to indicate the triggering of the pre-release operation; When the rainfall exceedance probability is between the adaptive exceedance probability threshold and the high confidence probability threshold, and the decision time margin is greater than the pre-configured rolling optimization cycle, a pre-release timing determination result is generated that indicates maintaining the pre-release standby state and postponing the review to the next cycle. When the rainfall exceedance probability is between the adaptive exceedance probability threshold and the high confidence probability threshold, and the decision time margin is no greater than the rolling optimization cycle, a pre-release timing determination result is generated, indicating that the latest safe decision time has been reached and triggering the pre-release operation.

5. The method according to claim 1, characterized in that, The steps for generating initial scheduling control parameters based on a multi-objective optimization strategy include: Construct a multi-objective optimization set, which includes the expected minimization objective of water discharge to reflect the loss of benefits from the reservoir, the expected minimization objective of peak outflow to reflect downstream flood control pressure, and the conditional risk value minimization objective of excess storage capacity to reflect downstream flood control risk. Combining the preset reservoir water balance equation constraints and the preset physical discharge capacity constraints, a multi-objective optimization algorithm is used to solve the multi-objective optimization set, generating a set of non-dominated schemes containing multiple non-dominated solutions; The spatial distance metric method is used to calculate the comprehensive proximity of each scheme in the non-dominated scheme set. The scheme with the highest comprehensive proximity is selected, and the corresponding target flood level and the upper limit sequence of outflow for each time period are extracted from it as the initial scheduling and control parameters.

6. The method according to claim 5, characterized in that, When constructing the conditional value-at-risk (VAT) minimization objective for excess storage capacity, its confidence level parameter is determined through the following dynamic adaptive method: Extract the cumulative inflow of water into the reservoir for each forecast member within the effective forecast period from the real-time meteorological ensemble forecast data, and calculate the mean and standard deviation of the cumulative inflow of water into the reservoir for each forecast member; Divide the standard deviation by the mean to obtain the coefficient of variation of the current weather ensemble forecast; By using the ratio of the discrete coefficients to the pre-configured reference discrete coefficients, dynamic interpolation is performed between the pre-configured lower and upper confidence levels to generate the current confidence level parameters. By using the current confidence level parameter as the risk tail cutoff boundary input into the conditional risk value minimization objective of excess storage capacity, the flood control risk control intensity is automatically increased when the uncertainty of weather forecasts increases.

7. The method according to claim 1, characterized in that, The steps for calculating the inflow deviation by extracting measured data within the actual physical time window up to the current moment, and performing multi-dimensional attribution analysis on the source of the deviation when the inflow deviation meets the preset recalculation conditions, include: Determine whether the relative deviation between the measured inflow and the predicted inflow baseline value continues to exceed the allowable ratio within a preset time period. If so, trigger multi-dimensional attribution analysis. Extract the measured average rainfall and the forecast rainfall for each forecast member within the actual physical time window; Calculate the absolute difference between the measured average rainfall and the corresponding forecast rainfall, and divide the absolute difference by the larger of the measured average rainfall and the pre-configured zero-prevention lower limit to obtain the consistency deviation of each forecast member used for cross-event horizontal comparison. Extract the measured inflow and the simulated inflow given by the hydrological model within the same window, record the positive and negative distribution of the difference between the two, and generate an inflow deviation sign sequence.

8. The method according to claim 7, characterized in that, Before implementing the corresponding graded corrective response strategy based on the attribution level, the specific attribution level classification includes: When the consistency deviation of at least one forecast member is no greater than the pre-configured consistency tolerance, and the inflow deviation symbol sequence remains the same sign within the real physical time window, the first-level attribution judgment is output, indicating that it is a hydrological system deviation. When the consistency deviation of some forecast members is no greater than the consistency tolerance, while the consistency deviation of the remaining forecast members is greater than the consistency tolerance, the second-level attribution judgment is output, indicating forecast reliability divergence. When the consistency deviation of all forecast members is greater than the consistency tolerance, or the risk review at the aforementioned level fails, the third-level attribution judgment is output, indicating that the overall forecast is inaccurate. When the meteorological ensemble forecast data stream is detected to be in a state of communication interruption or unavailability, the fourth-level attribution judgment is output, indicating that the data is interrupted.

Images