Short, medium and long term coupled digital twin reservoir group collaborative scheduling method and system

Abstract

The application discloses a short-medium-long-term coupled digital twin reservoir group collaborative scheduling method and system, and particularly relates to the technical field of reservoir group scheduling. The method comprises the following steps: acquiring multi-source water and rainfall conditions and working condition data related to the reservoir group, performing unified time scale alignment and quality control, and generating a twin initial state package; generating a candidate rainfall sequence based on the twin initial state package for rainfall input disturbance, driving a digital twin reservoir group model to form a time scenario set, and generating a time scenario credibility weight set from rainband drop zone drift, rising edge alignment residual error, propagation delay residual error and key data freshness; under the constraint of executability, a candidate collaborative peak-shaving discharge scheme set is generated and multiple scenarios are deduced, a weighted peak-shaving super-alarm risk index is obtained by weighting according to the credibility weight, a robust peak-shaving scheme is selected, a collaborative scheduling instruction set is generated, and the instruction set is issued to an on-site control system; and the system is used for implementing the method.

Description

Technical Field

[0001] This invention relates to the field of reservoir group scheduling technology, and more specifically, to a short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method and system. Background Technology

[0002] Reservoir group coordinated scheduling technology is mainly used to automate and refine the timing and volume of flood discharge from cascade reservoirs during floods. This allows for peak shaving and staggering, reducing downstream risks, and simultaneously achieving power generation and water supply goals, all while meeting reservoir safety, downstream control section water level constraints, and engineering operational constraints. This technology is widely applied in areas such as river basin flood control, water resource allocation, and water conservancy informatization. However, under short-duration severe convective weather conditions, the upstream tributary confluence time is short, rainfall is sudden and intense, and the flood peak exhibits a sharp, narrow shape. Furthermore, weather radar and short-term forecasts have errors in estimating the location and intensity of rainfall, making it easy for the flood peak to shift upon arrival. The sensitivity of cascade reservoir group coordinated peak shaving to timing is significantly increased, urgently requiring more robust short-term scheduling and control methods.

[0003] The existing technology has the following shortcomings:

[0004] In the past, when cascade reservoir groups faced short-duration, strong convective scenarios, scheduling and staggered release arrangements were often based on a single inflow forecast process line. Candidate schemes were primarily compared using indicators such as peak size and highest water level, without constraining the uncertainty of the flood peak arrival time or considering its distribution and the resulting downstream superimposed risks as core optimization objectives or constraints. Consequently, when the actual flood peak arrival time deviates from the forecast by approximately 30 to 60 minutes, upstream reservoirs may release water ahead of schedule, while downstream reservoirs release water simultaneously. This results in the superposition of multiple reservoir releases at the downstream control section, causing the water level at the control section to rise rapidly within a short period, approaching or exceeding the warning level. Emergency responses may necessitate temporary gate closures, reduced power output, or the activation of unconventional flood diversion measures, further exacerbating backwater backflow, increased water wastage, and widening grid planning deviations, impacting scheduling safety and project feasibility.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method and system to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method includes the following steps:

[0009] Acquire multi-source water and rainfall data and engineering data related to the reservoir group; perform unified time-scale alignment and quality control on the multi-source water and rainfall data and engineering data to generate a twin initial state package to drive the digital twin reservoir group model;

[0010] Based on the twin initial state package, multiple candidate rainfall sequences are generated by perturbation rainfall input, and each candidate rainfall sequence is input into the digital twin reservoir group model to obtain the corresponding inflow prediction sequence and propagation arrival parameters, thereby constructing the arrival scenario set; based on multiple pieces of evidence related to arrival deviation, the component credibility of each arrival scenario is generated and fused and normalized to obtain the arrival scenario credibility weight set corresponding to the arrival scenario set;

[0011] A set of candidate coordinated peak-shaving release schemes that meet the executability constraints is generated. For each candidate coordinated peak-shaving release scheme, a digital twin reservoir group model is used to perform multi-scenario simulations under various arrival scenarios in the arrival scenario set to obtain the downstream control section water level process and the reservoir group release process. Based on the simulation results, scenario losses under each arrival scenario are constructed, and the scenario losses are weighted and summarized according to the arrival scenario credibility weight set to obtain the weighted peak-shaving exceedance risk index corresponding to each candidate coordinated peak-shaving release scheme. Candidate coordinated peak-shaving release schemes whose weighted peak-shaving exceedance risk index meets the preset optimization criteria are selected as robust peak-shaving schemes, and a coordinated scheduling instruction set is generated.

[0012] The coordinated scheduling instruction set is sent to the local control system and execution receipt information is received. Execution consistency verification is performed based on the execution receipt information. The verification results and newly added observation data are fed back into the digital twin reservoir group model to update the state. The arrival scenario set and the arrival scenario trusted weight set are updated within a preset rolling cycle. When the triggering condition is met, a rolling re-optimization process is initiated based on the updated twin initial state package, the arrival scenario set, and the arrival scenario trusted weight set to update the robust peak-shaving scheme and the coordinated scheduling instruction set.

[0013] In a preferred embodiment, the multi-source water and rainfall data and engineering data include: rainfall information corresponding to radar extrapolation rainfall products and gridded data of short-term rainfall forecasts from numerical weather prediction, hydrological information corresponding to upstream and downstream control section observations, and engineering information corresponding to reservoir water levels, gate status, and unit status.

[0014] In a preferred embodiment, quality control includes one or more of the following: marking and processing outliers, imputing missing data, and statistical analysis of critical data freshness; critical data freshness reflects at least one of data arrival delay and missing data rate.

[0015] In a preferred embodiment, when constructing the time-bound scenario set, the perturbations performed on the rainfall input include the area offset perturbation and the rainfall intensity perturbation.

[0016] In a preferred embodiment, the method further includes online propagation time correction, which is achieved by calculating the cross-correlation or characteristic time difference of the observation sequences of upstream and downstream control sections.

[0017] In a preferred embodiment, multiple pieces of evidence related to arrival deviation include rainband drift, rising edge alignment residual, propagation delay residual, critical data arrival delay, and critical data missing rate.

[0018] In a preferred embodiment, the enforceability constraints include the downstream control section warning water level, the reservoir flood control limit water level, the minimum ecological discharge flow, the gate opening and closing rate limit, and the unit start-up and shutdown time constraint.

[0019] In a preferred embodiment, the scenario loss is composed of the over-alarm penalty amount reflecting the risk of the downstream control section exceeding the alarm level, the superimposed over-limit penalty amount reflecting the risk of the reservoir group releasing water, and the insufficient peak-shifting penalty amount reflecting the risk of insufficient peak-shifting interval.

[0020] In a preferred embodiment, the consistency verification includes: aligning the target control sequence and execution receipt in the collaborative scheduling instruction set on the time axis to obtain the control deviation; generating an execution deviation flag when the control deviation exceeds a preset threshold; and using the execution deviation flag as one of the triggering conditions for triggering the rolling re-optimization process.

[0021] In a preferred embodiment, the following modules are included:

[0022] The access verification module is used to acquire multi-source water and rainfall data and engineering data related to the reservoir group; it performs unified time-scale alignment and quality control on the multi-source water and rainfall data and engineering data to generate a twin initial state package for driving the digital twin reservoir group model;

[0023] The scenario weight module is used to generate multiple candidate rainfall sequences based on the twin initial state package and the perturbation rainfall input. Each candidate rainfall sequence is then input into the digital twin reservoir group model to obtain the corresponding inflow prediction sequence and propagation arrival parameters, thereby constructing the arrival scenario set. Based on multiple pieces of evidence related to the arrival deviation, the component credibility of each arrival scenario is generated and fused and normalized to obtain the arrival scenario credibility weight set corresponding to the arrival scenario set.

[0024] The simulation and optimization module generates a set of candidate coordinated peak-shaving and discharge schemes that meet the executability constraints. It then uses a digital twin reservoir group model to perform multi-scenario simulations of each candidate scheme under various arrival scenarios in the arrival scenario set, obtaining the downstream control section water level process and the reservoir group discharge process. Based on the simulation results, it constructs the scenario loss for each arrival scenario and weights and summarizes the scenario losses according to the arrival scenario credibility weight set to obtain the weighted peak-shaving and discharge risk index corresponding to each candidate scheme. Candidate coordinated peak-shaving and discharge schemes whose weighted peak-shaving and discharge risk index meets the preset optimization criteria are selected as robust peak-shaving schemes, and a coordinated scheduling instruction set is generated.

[0025] The closed-loop control module is used to send the collaborative scheduling instruction set to the local control system and receive execution feedback information, and to perform execution consistency verification based on the execution feedback information; to feed the verification results and newly added observation data back into the digital twin reservoir group model to update the state; to update the arrival scenario set and the arrival scenario trusted weight set within a preset rolling cycle, and to start the rolling re-optimization process based on the updated twin initial state package, arrival scenario set and arrival scenario trusted weight set when the trigger condition is met, so as to update the robust peak-shaving scheme and collaborative scheduling instruction set.

[0026] The technical effects and advantages of this invention are as follows:

[0027] This invention addresses the scheduling challenges posed by short-duration, strong convection conditions, such as the tendency for flood peak arrival times to deviate, the ease with which field data lags, and the high sensitivity of reservoir group peak shifting to time sequence. By performing unified time-scale alignment and quality control on multi-source water and rainfall data and engineering data, and statistically analyzing the freshness of key data, a traceable twin initial state package is formed. This ensures that the starting input and state caliber of the digital twin reservoir group model are consistent, reducing the accumulation of starting deviations caused by supplementary reporting, missing measurements, and delayed feedback, thereby providing a stable and reliable state foundation for subsequent collaborative peak shifting simulations.

[0028] This invention further constructs a set of arrival scenarios covering a reasonable deviation range by performing area offset perturbation and rainfall intensity perturbation on the rainfall input, and introduces an online propagation time correction mechanism to dynamically correct the propagation delay, so that the scenario extrapolation process can closely reflect the real-time river propagation conditions. At the same time, based on evidence such as rainband area drift, rising edge alignment residual, propagation delay residual, and key data arrival delay and missing rate, a set of reliable weights for arrival scenarios is generated, so that the arrival uncertainty is explicitly modeled and participates in decision-making in the form of interpretable weights, avoiding excessive reliance on a single inflow forecast process line for coordinated scheduling, thereby improving the robustness to peak arrival deviation.

[0029] In terms of scheme generation and evaluation, this invention forms a set of candidate coordinated staggered peak release schemes under executability constraints, and conducts multi-scenario simulations under various arrival scenarios to obtain the downstream control section water level process and reservoir discharge process. By constructing scenario losses and weighting them according to the reliability weights of the arrival scenarios, a weighted staggered peak exceeding warning risk index is formed for the candidate schemes. Robust staggered peak release schemes are selected based on this index, and a coordinated scheduling instruction set is generated. This mechanism can compare and select the best schemes for key risks such as downstream control section exceeding warning risk, reservoir discharge superposition risk, and insufficient staggered peak interval risk using unified indicators. This ensures that the selected scheduling scheme can still maintain the downstream control section warning water level and suppress the superposition of discharge peaks even under more likely arrival scenarios, thereby improving the safety and stability of flood control scheduling. Attached Figure Description

[0030] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0031] Figure 1 This is a flowchart illustrating the short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method of the present invention;

[0032] Figure 2 This is a schematic diagram of the structure of the short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling system of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Example 1: The present invention provides a short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method, such as... Figure 1 As shown, it includes the following steps:

[0035] Step 1: Integrate multi-source water and rainfall data with engineering data;

[0036] This step focuses on the cascade reservoir group and its upstream catchment area at the current scheduling time, establishing the starting base for the digital twin reservoir group model. The cascade reservoir group includes Reservoirs A, B, and C. Under short-duration, strong convective conditions, the rapid movement of rainbands and the shift in their landing areas significantly affect the arrival time of reservoir inflows, and on-site communication may experience issues such as re-reporting, disconnections, or delayed acknowledgments. If the inputs and states at the starting time are inconsistent, subsequent scenario sets and peak-shifting simulations will accumulate deviations at the starting point, leading to distorted scheme evaluations. Therefore, this step organizes rainfall, multi-section hydrological observations, and reservoir group operational conditions into a consistent input under the same time reference and synchronously writes them into the twin, outputting a traceable twin initial state package.

[0037] In terms of data access, the system accesses minute-level or 5-minute-level rainfall sequences from the upstream confluence area rain gauge network, weather radar mosaics and radar extrapolation rainfall products, and gridded data from numerical weather prediction short-term rainfall forecasts. Rain gauges provide constraints on measured intensity and cumulative amount at specific locations, radar products provide spatial distribution of areal rainfall and rainband morphology, radar extrapolation is used to describe the short-term movement trend of the rainband, and short-term forecasts are used to constrain the persistence and intensification / attenuation tendencies of rainfall. The system uniformly maps the above rainfall inputs to a preset watershed grid or sub-watershed unit, making it consistent with the spatial input caliber of the digital twin reservoir group model. Considering that radar quantitative rainfall may have systematic deviations under strong convective conditions, the system performs consistency calibration on radar rainfall and surface rainfall: keeping the spatial distribution pattern of radar rainfall unchanged, using the aggregated areal rainfall from rain gauges as a scale reference, and scaling or segmenting the radar estimated areal rainfall, so that the rainfall input snapshot retains the rainband structure while being constrained by measured magnitude, thereby reducing the amplification effect of rainfall input deviations on reservoir extrapolation.

[0038] In terms of hydrological and engineering data integration, the system synchronously accesses upstream control section water level or flow sequences and downstream control section water level sequences. Upstream control sections are used to characterize the actual arrival and rate of change of the flood rise, while downstream control sections are used for subsequent verification of warning water level constraints throughout the entire process. The system also accesses reservoir water levels, gate opening receipts, gate operation status, generator output and start / stop status, and head information (measured discharge flow or required for discharge flow conversion) from the on-site monitoring systems of Reservoirs A, B, and C. To maintain consistency in subsequent model simulations and execution verification, this step preferentially uses the total discharge flow as the unified outflow: the gate discharge portion is calculated based on the gate opening and head conditions combined with the gate discharge characteristics; the generator flow portion is calculated based on the flow curve corresponding to the generator output and head. These two are combined to form the total discharge flow for each reservoir, and are saved along with the gate opening and generator output data so that the control status corresponding to the same flow can be traced back when needed.

[0039] To ensure consistent input from multiple data sources at the current scheduling moment, the system performs unified time-scale alignment and quality control on the incoming data. Each data source is aligned to a preset short-term scheduling time step, such as 1 minute or 5 minutes, and outliers are processed according to physical boundaries and reasonable rates of change: for observations that clearly exceed limits or jump, the system retains the original record while marking them as outliers and using the removed or downweighted values ​​for twin synchronization; for short-term missing data segments, the system performs interpolation under amplitude limiting and smoothing constraints to ensure continuous twin computation; for supplementary data, the system distinguishes between the observation time and the arrival time to avoid treating supplementary data as real-time input directly driving scheduling. Simultaneously, the system statistically analyzes the freshness of key data to describe the support strength of the input at the current moment. The statistical results of key data freshness include at least the arrival delays of key rain gauges, upstream control sections, downstream control sections, and reservoir operation reports, as well as the missing data rate within the most recent statistical window, and are bound to the data source identifier and the start and end times of the statistical window, serving as one of the direct input sources for subsequent calculations of the reliable weights of the arriving scenarios.

[0040] After data alignment is completed, the system synchronizes the state of the digital twin reservoir group model. The system writes the current water levels of reservoirs A, B, and C into the twin, and uses the pre-stored reservoir capacity and water level curves of each reservoir to convert the reservoir water levels into the current reservoir capacity. The conversion relationship is expressed as follows: ;

[0041] Where i represents the reservoir number (Reservoir A, Reservoir B, Reservoir C). Let i be the current water level of the i-th reservoir. The reservoir capacity corresponding to the water level. The reservoir's capacity-to-water-level mapping function or piecewise interpolation curve is defined as follows: Preferably, the data is constructed from a discrete table of the reservoir's capacity and water level, and the discrete table contains at least several sets of data. , The calibration point is determined and satisfies the monotonicity constraint. Piecewise linear interpolation or spline interpolation is preferred when... When it falls within the interval of adjacent calibration points, it is obtained by interpolation. When Hi exceeds the endpoint range of the discrete table, endpoint truncation or calculation according to a preset extrapolation rule is preferred, and this out-of-bounds situation is written into the anomaly flag field of the twin's initial state package. The twin obtains the current water storage status of each reservoir based on this, and can further link it with constraints such as flood control limit water levels to form an initial regulation boundary that can be used for short-term peak-shaving simulations. Simultaneously, the system writes the total outflow from each reservoir and the executable capacity boundaries of gates and generating units into the control constraint layer of the twin model, and incorporates the latest observations of the upstream and downstream control sections as boundary conditions for the river propagation sub-model, enabling the twin model to reflect the impact of current river baseflow and backwater conditions on flood peak propagation in subsequent simulations.

[0042] Through the aforementioned access, alignment, calibration, and state synchronization, this step outputs a twin initial state package. This twin initial state package includes at least: the current water levels and converted storage capacities of reservoirs A, B, and C; the total outflow from each reservoir, gate opening, and generator output and start / stop status; snapshots of observation sequences from upstream and downstream control sections; snapshots of rainfall inputs that have undergone consistency calibration and spatial mapping; and key data freshness statistics, including arrival delay and missing measurement rate. This twin initial state package serves as the unified starting point for constructing the arrival scenario set and generating reliable weights for the arrival scenarios in step two. This enables subsequent twin simulations and coordinated peak-shifting scheduling to start based on the same state at the same time and with the same caliber, thereby providing reliable support for suppressing the superimposed peak discharge and reducing the risk of downstream control sections exceeding warning levels.

[0043] Step 2: Construct the time-to-date scenario set and generate the time-to-date scenario credibility weights;

[0044] This step addresses the core contradiction of peak arrival deviation under short-duration strong convection conditions: radar can see the rainband, but if the rainband's landing area shifts slightly or the rainfall intensity is estimated slightly, the arrival time of the flood peak entering Reservoir A will be delayed or advanced, thus disrupting the staggered peak release schedule originally arranged by Reservoirs A, B, and C according to the propagation time.

[0045] The goal of this step is not to provide a single inbound forecast process line, but to break down and present the uncertainties, forming a set of arrival scenarios that cover a reasonable range of deviations, and to assign interpretable and credible weights to each arrival scenario. In this way, the quality of subsequent coordinated scheduling schemes no longer depends on whether a particular forecast is accurate, but rather on whether the downstream control section warning water level can still be maintained and the superposition of discharges can be avoided under more likely scenarios.

[0046] The input for this step comes from the twin initial state package generated in step one, mainly including: a spatially mapped and calibrated snapshot of rainfall input, the latest water level or flow sequence of the upstream and downstream control sections, the current water level and discharge status of reservoirs A, B, and C, and key data freshness statistics. In addition, this step integrates radar extrapolated rainfall products and short-term forecast gridded data, and optionally calls upon a historical strong convective event sample library to provide the prior disturbance range and parameter scale for similar morphologies.

[0047] When constructing the time-bound scenario set, the system focuses on two of the most common and critical sources of uncertainty: rainfall area shift and rainfall intensity disturbance. First, the system primarily uses radar extrapolation rainfall products to extract the location of the main rain area, the direction of the rain belt's main axis, and its movement speed. Using key confluence areas as anchor points, it sets several spatial shift disturbances along the rain belt's movement direction and lateral directions, causing the main rain area to shift within a reasonable range, thus generating multiple rainfall sequences with perturbed rainfall areas. This shift range is not arbitrarily set but is determined by combining watershed scale, radar positioning error statistics, and the instability characteristics of strong convective rainfall areas, ensuring that the shift covers common error ranges without generating obviously impossible scenarios. Second, while maintaining the spatial structure of radar rainfall, the system performs tiered perturbations on rainfall intensity: using the duration and intensity trends of heavy rainfall given in short-term forecasts as constraints, it appropriately amplifies or reduces the rainfall intensity in the main rain area and slightly stretches or shrinks the peak duration, thereby generating rainfall sequences with perturbed rainfall intensity. For short-duration strong convection, this step is significant because it includes both sharper and narrower or wider peak morphological differences within the coverage, so that the coupling relationship between the arrival time of the flood peak and the morphology of the flood peak can be reflected in the scenario set.

[0048] In addition to the two types of disturbances mentioned above, to avoid insufficient scenario coverage due to relying solely on a single radar extrapolation, the system can optionally retrieve similar events from a historical strong convective event sample database as a supplementary scenario source. The similar event retrieval preferably considers three types of features simultaneously: radar echo morphology, rainband movement direction and velocity range, and the rising edge morphology of the upstream control section in the early stages of the event. After retrieving similar events, the system maps their rainfall spatial structure, inflow response morphology, propagation delay interval, and other features to the current watershed state, and writes them into the scenario set as additional candidate rainfall sequences combined with propagation parameters. This ensures that the scenario set includes both disturbance branches based on the current radar and comparable branches that have actually occurred historically.

[0049] Once the candidate rainfall sequences are determined, the system inputs each candidate rainfall sequence into the runoff generation and channel propagation module of the digital twin reservoir group model to obtain the corresponding short-term inflow prediction sequence for reservoir A. It then extrapolates the propagation process from reservoir A to reservoir B, and from reservoir B to reservoir C, thus forming a complete time-delay scenario for each candidate branch. For ease of consistent representation, the system denotes the scenario numbers in the time-delay scenario set as follows: For each arrival scenario k, the system saves at least the following: the rainfall input sequence for that scenario, the predicted sequence of the A reservoir input, the propagation delay from A reservoir to B reservoir and the propagation delay from B reservoir to C reservoir, and the propagation arrival parameters for subsequent extrapolation.

[0050] To prevent the scenario set from deviating further as the event progresses, the system introduces an online propagation time correction mechanism to dynamically adjust the propagation delay in the scenario. The system continuously receives measured water level or flow sequences from upstream and downstream control sections and extracts standardized waveform segments from these sections within the flood rise edge window. It preferentially uses cross-correlation matching to calculate the online propagation delay estimate: the cross-correlation function is calculated for the waveform sequences of the two sections within a preset time shift search range, and the time shift corresponding to the correlation peak is taken as the online propagation delay estimate. When the correlation peak is lower than a preset confidence threshold or there are insufficient rise edge segments, the propagation delay is preferentially estimated by degenerating to the rise edge arrival time difference or peak time difference, and the degradation marker is written into the online propagation delay estimation result of this round. This online propagation delay reflects the actual timescale of flood wave propagation from upstream to downstream under the current river channel conditions. The system compares the online propagation delay with the propagation delay used in each time scenario k to obtain the propagation delay residual, and corrects the propagation time parameters of the scenario accordingly, so that the scenario is closer to the real-time river propagation conditions in subsequent rolling simulations, thereby reducing the dependence of peak-shifting scheduling on fixed propagation time experience values.

[0051] After completing scenario construction and propagation correction, the system generates a maturity scenario credibility weight for each maturity scenario. This serves as a weighting factor for subsequent scheme evaluation. The credibility weight is not based on subjective scoring, but rather on a comprehensive analysis of multiple observables, with the rainband drift being the preferred choice. Rising edge alignment residual Propagation delay residual Delay in arrival of critical data and the rate of missing key data Five types of input.

[0052] To ensure the reproducibility of the above five types of input quantities, it is preferable to calculate them according to the following criteria: This represents the equivalent drift distance of scenario k relative to the currently observed main rain area in the key confluence area coordinate system; the main rain area can be determined by the connected region formed by the radar quantitative rainfall field above a preset heavy rainfall threshold, and its feature points are taken as the area-weighted center coordinates of this connected region. The Euclidean distance between the feature points of the main rain area in scenario k and the feature points of the observed main rain area can be used. When using sub-basin units, the distance can also be converted into the equivalent distance corresponding to the number of sub-basin grids. This represents the fitting residual of scenario k to the arrival time of the rising edge of the flood at the upstream control section. The arrival time of the rising edge is preferably defined as the moment when the flow sequence first exceeds the baseflow level threshold and the first difference exceeds the rate of change threshold within the rolling window. Take the difference between the predicted rising edge time and the actual rising edge time. The propagation delay residual is represented by the difference between the propagation delay used in scenario k and the online propagation delay estimate. This indicates the arrival delay of key data, and the preferred method is to take the weighted average delay of the arrival time and observation time of each key data source within the most recent statistical window; This indicates the missing rate of key data. The preferred method is to take the proportion of missing samples from key data sources within the most recent statistical window to the total number of samples that should be received. The start and end times of the statistical window are then bound together with the data source identifier and written into the twin initial state package.

[0053] here, Used to depict the degree of deviation of the rainband's landing area relative to the key confluence area; Used to characterize the fit of the scenario to the arrival time of the rising edge of the upstream control section. It can be obtained from the difference between the predicted rising edge time and the measured rising edge time. Used to characterize the consistency between context propagation delay and online propagation delay estimation; and The key data freshness statistics from step one are used to reflect the real-time nature and completeness of the input data upon which this scenario depends. To map the above quantities into fusionable confidence components, the system constructs component confidence using an exponential decay method: , , , ;

[0054] in, This is a scale parameter for the amount of rainband drift. The scaling parameter for the rising edge aligned residual. The scale parameter for the propagation delay residual. The scaling parameter for data arrival delay can be obtained from historical strong convective events or set empirically according to scheduling procedures. It is used to uniformly map offsets and residuals of different dimensions to a confidence level within the range of 0 to 1. For example... It can be taken from 5 to 20 km; It can be taken for 10 to 30 minutes; It can be taken for 5 to 15 minutes; The timeframe can be 2 to 10 minutes. The closer the confidence level of a component is to 1, the more credible the scenario is in the corresponding dimension; the closer it is to 0, the less credible the scenario is in the corresponding dimension.

[0055] After obtaining the component credibility, the system uses a multiplicative approach for fusion to reflect the engineering principle that when any key dimension is obviously unreliable, the overall scenario credibility should decrease significantly, thus constructing unnormalized credibility weights: ;

[0056] in, , , , This is a weighting index used to adjust the degree of influence of different components in the overall credibility; for example... , , , Considering that this embodiment focuses on the peak arrival time deviation, the weighting index can be appropriately increased for time-related components under short-duration strong convection conditions, making the rising edge alignment residual and propagation delay residual more sensitive to the reliable weights. Finally, the system... After normalization, the confidence weights of the scenarios at that time are obtained: ;

[0057] Thus guarantee This allows it to be directly used as a weighting factor in the calculation of the weighted peak-shifting and over-alert risk index in step three.

[0058] Through the above processing, this step outputs the set of time-delayed scenarios and its set of reliable weights for those scenarios. The output enables subsequent coordinated scheduling to no longer rely on a single inflow forecast process line, but to assess the risk of downstream control section exceeding the warning level and the risk of leakage superposition under multiple scenario coverage, and to place more likely scenarios in a more important position through reliable weights, thereby laying the foundation for robust peak-shifting scheduling under short-duration strong convection conditions.

[0059] Step 3: Multi-scene simulation and optimization;

[0060] This step addresses the coordinated staggered peak release decision-making for reservoirs A, B, and C within a short-term scheduling window. When there are deviations in the arrival time of flood peaks under strong convection, simply looking at the predicted peak size is often insufficient: if the arrival time is slightly later or earlier, the seemingly reasonable release arrangements of reservoirs A, B, and C may create superimposed peaks at downstream control sections, causing a rapid rise in water levels within a short period. This step addresses the effectiveness of the proposed solutions by quantifying and comparing them: digital twin simulations are performed on the candidate coordinated staggered peak release schemes under various arrival scenarios output in step two, obtaining the downstream control section water level process and the reservoir group release process; then, the risks of exceeding warning levels, superimposed exceeding limits, and insufficient staggered peak intervals are combined into a scenario loss, and the scenario's reliability weights are applied. Perform weighted aggregation to form a scheme-level weighted peak-shifting and peak-exceeding risk index. Based on this, the system selects a robust peak-shifting scheme and generates a collaborative scheduling instruction set for direct execution.

[0061] The inputs for this step include: the twin initial state package generated in step one, which includes reservoir water level, reservoir capacity, discharge capacity boundary, river boundary conditions, etc.; and the time-delay scenario set and the time-delay scenario confidence weight set generated in step two. And the warning water level at the downstream control section given in the dispatching procedure. Hard constraints include the flood control limit water level of each reservoir, the minimum ecological discharge requirement, the upper limit of the gate opening and closing rate, and the unit start-up and shutdown constraints. The system first determines the short-term scheduling window, such as 0 to 6 hours or 0 to 12 hours in the future, and organizes the control sequence according to the time step of the window to ensure that the subsequent simulation and command issuance are consistent in time.

[0062] During the generation phase of candidate collaborative peak-shifting and load-dissipation schemes, the system constructs a set of candidate schemes. Each candidate solution Provide time-segmented discharge flow sequences for reservoirs A, B, and C, or provide gate opening and unit output control sequences consistent with the local control system. The candidate scheme generation process does not involve pre-generating and then screening; instead, it embeds executability constraints during the generation process: the gate opening change rate does not exceed the preset upper limit of the opening and closing rate; unit start-up and shutdown meet minimum start-up and shutdown times and minimum output limits; and during the simulation period, the water level of each reservoir must not exceed the flood control limit level, while the discharge flow of each reservoir must not be lower than the minimum ecological discharge requirement. By placing these boundary conditions forward in the candidate scheme generation stage, the candidate set naturally possesses feasibility, avoiding the situation where a scheme is optimal in the simulation but cannot be implemented on-site.

[0063] After the candidate set is established, the system performs simulations of each candidate scheme s under each arrival time scenario k using a digital twin reservoir group model. During the simulation, the system writes the short-term inflow prediction sequence and propagation arrival time parameters corresponding to scenario k into the twin model's boundary conditions, and writes the control sequence of candidate scheme s into the twin model's control terminal. The system jointly simulates reservoir group water level changes, discharge response, and river propagation to obtain the key outputs under the scenario and scheme combination. The system must at least output the downstream control section water level process and extract the maximum water level within the simulation window. ,in This indicates the maximum water level at the downstream control section within the simulation window when candidate scheme s is executed under scenario k. The system also outputs the discharge flow process of each reservoir within the simulation window. ,in This indicates the reservoir number, and t represents the time point in the simulation. Let represent the discharge flow of the i-th reservoir at time t under scenario k and when implementing scheme s. Based on the discharge flow process of each reservoir, the system calculates the superimposed discharge process of the reservoir group. And identify the peak discharge time of each reservoir during the discharge process. This is used to subsequently measure whether peak shifting is sufficient.

[0064] After completing the simulation, the system converted the results into three risk parameters: whether the downstream control section shows a tendency to exceed the warning level, whether the reservoir discharge exceeds the limit cumulatively, and whether the staggered peak interval is insufficient. First, the system uses the warning water level... Calculate the over-warning margin using hard constraint anchor points. ;

[0065] in This represents the excess of the maximum water level at the downstream control section relative to the warning water level; a positive value indicates a risk of exceeding the warning level, while a negative value indicates that there is still a safety margin.

[0066] To avoid unnecessary rewards for safety margins and to emphasize the flood control baseline, the system uses a positive function to construct the over-alert penalty. This ensures that the penalty is 0 when the warning level is not exceeded, and the penalty is equal to the extent of exceeding the warning level when the warning level is exceeded.

[0067] Secondly, the system uses the superimposed flow peak to characterize the intensity of the leakage superposition, and calculates... ;in To extrapolate the maximum combined discharge value of the reservoirs within the window, a larger peak value indicates a more severe overlap between reservoirs A, B, and C on the timeline. If there is an upper limit to the allowable combined discharge volume in the engineering design... This upper limit can be derived from the downstream river channel's flood control capacity or the relationship between water level and flow rate at the downstream control section. The system further monitors whether the limit is exceeded and constructs a superimposed penalty for exceeding the limit. Therefore, the penalty is 0 when the superimposed peak value does not exceed the allowed upper limit, and the penalty is equal to the excess portion when it exceeds the upper limit.

[0068] Secondly, to ensure that peak shifting is truly implemented on a time scale, the system calculates the minimum peak shifting interval. ;in It represents the minimum time interval between the peak discharge times of any two reservoirs. The smaller the interval, the easier it is for the peaks to overlap.

[0069] The system is set to the minimum expected peak-shaving interval. This value can be determined by combining the river channel propagation time scale with the downstream cross-sectional storage capacity, and a penalty for insufficient peak shifting can be constructed. This ensures that the penalty is 0 when the actual peak-shifting interval meets or exceeds the expected interval, and the penalty is equal to the insufficient amount when the interval is insufficient.

[0070] Based on the three penalty quantities mentioned above, the system integrates them into a scenario loss function, using a single quantity to describe the overall risk level of solution s under scenario k: ;

[0071] in , , These are weighting coefficients used to reflect the priority of different risk items. Following the principle of prioritizing flood control safety, The optimal choice is to select a larger value, which would significantly increase the scenario loss if any trend exceeds the warning level; under the premise of not exceeding the warning level, through... and Suppress superposition exceeding limits and maximize peak spacing to avoid strategies that appear safe but have very high superposition spikes becoming unstable under time deviations; for example: Version 1.0 is acceptable. and 0.5 and 0.3 are acceptable values.

[0072] Since the credibility of each scenario at the time of arrival is different, the system does not use a simple average, but instead uses the credibility weights of the scenarios at the time of arrival generated in step two. The scenario losses are weighted and aggregated to obtain the weighted peak-shifting and over-alert risk index of candidate solution s. ;

[0073] Where K is the number of scenarios at that time, and This weighting method allows more credible arrival scenarios to contribute more to the evaluation of the schemes, thereby enabling the selected schemes to maintain the warning water level at the downstream control section and reduce the superposition of discharge even under more likely arrival deviations.

[0074] The system in the candidate set Select to The minimum solution is considered the robust peak-shifting solution: Determine a robust peak-shifting scheme. Subsequently, the system encapsulates this into a collaborative scheduling instruction set and binds it to the current round's arrival scenario set version identifier, the arrival scenario credibility weight set, and risk assessment tags. The collaborative scheduling instruction set preferably includes: time-segmented discharge flow instructions or gate opening and unit output control instructions for Repositories A, B, and C within the short-term scheduling window; the start and end times of instruction effectiveness and update cycle; and safety margin identifiers related to downstream control section warning water level constraints and reservoir group superimposed control identifiers, so that consistency verification and rolling updates can be performed using the same caliber for recalculation and comparison. Through this step, the system transforms the arrival uncertainty into an executable scheduling decision with the lowest weighted risk, enabling collaborative peak shifting to no longer rely on the accuracy of a single forecast process line, but rather possessing robustness to peak arrival deviations.

[0075] Step 4: Issue the reinjection closed-loop control;

[0076] This step focuses on the continuous effectiveness of the coordinated scheduling instruction set in the field execution process of reservoirs A, B, and C, as well as the evolution of strong convective processes. The reality of short-duration strong convection is that adjustments are made while rainfall is occurring: radar and rain gauge observations are constantly updated, and the judgment of the inflow process line and propagation arrival time converges over time; simultaneously, there are response delays in gate opening and closing, and unit start-up and shutdown, and local feedback may also arrive late. If the robust peak-shaving scheme obtained in step three is still used as a one-time fixed scheme, two types of problems are likely to occur: first, the scenario credibility weights have changed significantly but the scheme has not been updated, leading to a decrease in robustness to arrival time deviations; second, the instructions deviate from the plan during field execution and are not identified in a timely manner, causing the twin simulation to become disconnected from the actual hydraulic response, thus misjudging the peak-shaving effect. Therefore, this step establishes a closed loop of issuance, feedback, verification, reinjection, and rolling updates, ensuring that the digital twin reservoir group model always operates closely with real operating conditions, and that the coordinated scheduling instruction set is rolled out as necessary.

[0077] The system first sends the collaborative scheduling instruction set generated in step three to the local control systems of reservoirs A, B, and C via the scheduling communication link. The collaborative scheduling instruction set preferably provides a time-segmented control sequence based on a unified time base, including the effective start time, scheduling step size, and update cycle, enabling each reservoir to switch to the corresponding control state at the same time. After the instructions are issued, the system continuously receives execution feedback information returned by the local control systems. This execution feedback includes at least the gate opening position, gate action status, unit output feedback, unit start / stop status, and necessary protection alarm status. Simultaneously, the system synchronously accesses the real-time reservoir water level and measured discharge flow of each reservoir, and continuously accesses water level or flow observations at upstream and downstream control sections, as well as continuously updated rainfall information such as radar rainfall, rain gauge rainfall, and short-term forecasts. Before new observations enter the closed loop, the unified time-scale alignment and quality control mechanism described in step one is used to ensure that the data accuracy entering the twin is not corrupted by the rolling process.

[0078] To avoid the common misjudgment of assuming execution is complete upon issuance of instructions, this step introduces an execution consistency verification mechanism. The system aligns the target control sequence and execution receipts in the coordinated scheduling instruction set on the time axis to obtain gate opening deviations and unit output deviations, and further maps them to discharge response deviations. Simultaneously, it performs cross-verification by combining reservoir water level changes and measured discharge flow: when situations arise such as gate opening receipts being in place but measured discharge not changing as expected, or significant inconsistencies between unit output feedback and overcurrent conversion, the system marks these situations as execution deviations. This execution deviation marker does not directly negate the scheduling strategy, but rather serves as one of the triggering information for rolling updates, indicating that if the current twin continues to extrapolate according to the planned control sequence, it will deviate from the actual response.

[0079] While receiving confirmations and real-world observations, the system performs state recharge on the digital twin reservoir group model to keep the twin synchronized with the physical reservoir group. During the recharge process, the system updates the reservoir capacity status of the twin model with the latest reservoir water level, updates the outflow boundary of the twin model with the measured or consistency-verified discharge flow, and updates the river propagation boundary conditions with the latest upstream and downstream control section observations, while also updating the rainfall input snapshot. The initial state package of the twin is refreshed within the rolling cycle, and its field meanings and conversion relationships follow the standards described in step one, ensuring that subsequent simulations always start from the latest real-world conditions, rather than using the state from an older time period.

[0080] As the strong convection process progresses, the confidence weights of the arrival scenarios are the part that most needs to be changed accordingly. This step performs incremental updates to the arrival scenario set within each rolling cycle: the online propagation time correction follows the mechanism described in step two, using the latest upstream and downstream control section observations to update the online propagation delay estimates, and accordingly updating the propagation delay residuals of each arrival scenario. Simultaneously, the rising edge alignment residual is updated based on the latest rising edge information from the upstream control section. The drift of the rainband's landing area is updated by combining the changes in the location of the main rain area on the radar. The key data freshness statistics are also updated as observations are refreshed. Based on the above updates, the system generates the confidence weights for the current scenario according to the confidence weight generation model described in step two. By performing rolling recalculations, the credibility weights gradually converge to a more realistic distribution as events evolve, allowing subsequent evaluations to place greater emphasis on the more likely scenarios to occur.

[0081] Once the scenario confidence weights are updated, the system performs a necessary reassessment of the currently executing collaborative scheduling instruction set. This reassessment does not involve a complete recalculation of all candidate schemes each time; instead, it is triggered by risk signals: when an execution deviation is flagged, the downstream control section water level rises abnormally, or the twin simulation shows that the safety margin of the downstream control section relative to the warning water level is rapidly converging, the system initiates a rolling re-optimization process. During rolling re-optimization, the generation and executability constraints of candidate collaborative peak-shaving release schemes follow the mechanism described in step three, preferably performing incremental updates or local disturbance expansions near the original robust peak-shaving scheme to reduce instruction jumps and shorten computation convergence time. Subsequently, under the updated twin initial state and scenario set, the system reassesses the candidate schemes according to the multi-scenario simulation and weighted risk assessment model described in step three, and updates the robust peak-shaving scheme and collaborative scheduling instruction set accordingly. The emphasis here is on updates under the same caliber: the risk assessment still uses the weighted staggered peak over-alert risk index as the basis for comparison, but its input has been changed to the latest observation-driven twin initial state and the updated time-to-date scenario credibility weight, so as to ensure that the evaluation scale is consistent before and after the rolling update, and will not have the unexplainable problem of using this standard in the previous time and changing the standard in the next time.

[0082] To mitigate the field disturbances caused by frequent rolling corrections, this step sets control smoothing constraints on the collaborative scheduling instruction set after rolling updates. This limits the range of gate opening and unit output changes between adjacent instruction sets. Under the premise of meeting downstream control section warning water level constraints and peak-shaving targets, it prioritizes updating schemes with more continuous control actions to reduce the adverse effects of high-frequency gate actions and frequent unit start-ups and shutdowns on equipment and operational safety. After each rolling cycle, the system archives the current round's observation snapshot, execution receipt, execution deviation marker, the set of credible weights for the current scenario, and the key evaluation results of the schemes before and after the update into a historical strong convective event sample library. This data is used for subsequent statistical calibration or adaptive adjustment of the scale parameters and weight indices in step two, and the weight coefficients in step three, enabling the system to gradually form an error scale and scheduling preference that is closer to the local watershed and reservoir group during multiple strong convective events.

[0083] Through the issuance and execution of the aforementioned instructions, consistency verification, twin state reinjection, rolling updates of the reliability weights for the time-bound scenarios, and necessary reassessment of the scheme and rolling correction of the instruction set, this step transforms the collaborative scheduling of the digital twin reservoir group from a one-time calculation to a process closely monitored on-site. Under conditions where short-duration strong convection leads to peak arrival deviations, this closed-loop mechanism can promptly identify signs of peak-shaving failure, suppress the superposition of peak discharges from reservoirs A, B, and C, and continuously reduce the risk of downstream control sections exceeding warning levels while ensuring executability and control continuity.

[0084] Example 2: The design of the short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling system of the present invention is based on the method in Example 1, specifically as follows... Figure 2 The following modules are shown:

[0085] The access verification module is used to perform the multi-source water and rainfall information and engineering information access function described in step one of Embodiment 1. This module is configured to access minute-level or 5-minute-level rainfall sequences, weather radar mosaics and radar extrapolation rainfall products, and short-term rainfall forecast grid data from the upstream confluence area rain gauge network, and to perform consistency calibration on radar rainfall and surface rainfall, uniformly mapping the rainfall input to a preset watershed grid or sub-watershed unit. The module also synchronously accesses the water level or flow sequence of the upstream control section and the downstream control section. The cross-sectional water level sequence is obtained, and the reservoir water level, gate opening receipts, gate action status, unit output and start-up / shutdown status are accessed from the on-site monitoring systems of reservoirs A, B, and C. The total discharge flow of each reservoir is calculated by optimization. This module further performs unified time scale alignment and quality control, statistically analyzes the freshness of key data, and writes the processed data into a digital twin reservoir group model to generate a twin initial state package containing rainfall input snapshots, cross-sectional observation sequence snapshots, reservoir water levels and converted reservoir capacity, total discharge flow, and statistical results of key data freshness.

[0086] The scenario weighting module is used to perform the function of constructing the arrival scenario set and generating the arrival scenario credibility weights as described in step two of embodiment 1. This module is configured to generate multiple candidate rainfall sequences based on the rainfall input snapshot, cross-sectional observation sequence, and key data freshness statistics in the twin initial state package, and combined with radar extrapolated rainfall products and short-term forecast grid data, through area offset perturbation and rainfall intensity perturbation, and can optionally call the historical strong convective event sample library to supplement similar scenarios. This module inputs each candidate rainfall sequence into the runoff generation and river propagation process of the digital twin reservoir group model to obtain the corresponding inflow prediction sequence and propagation arrival parameters, forming an arrival scenario set containing multiple scenarios. This module also introduces an online propagation time correction mechanism to dynamically correct the propagation delay based on observations of the upstream and downstream control sections. This module further generates component credibility based on the rainband area drift, rising edge alignment residual, propagation delay residual, key data arrival delay, and missing measurement rate, and performs multiplicative fusion and normalization processing to output the arrival scenario credibility weight set.

[0087] The simulation and optimization module is used to execute the multi-scenario simulation and optimization function described in step three of embodiment 1. This module is configured to generate a set of candidate coordinated peak-shaving release schemes that meet the executability constraints based on the twin initial state package, the arrival scenario set, and the arrival scenario credible weight set, and in combination with hard constraints such as the downstream control section warning water level, the flood control limit water level of each reservoir, the minimum ecological discharge requirement, the upper limit of the gate opening and closing rate, and the unit start-up and shutdown constraints. This module calls the digital twin reservoir group model to simulate each candidate scheme under each arrival scenario to obtain the downstream control section water level process and the reservoir group discharge process. Based on this, the module constructs the scenario loss and weights it according to the arrival scenario credible weight set to obtain the weighted peak-shaving over-warning risk index of the candidate schemes, and selects the scheme with the smallest weighted peak-shaving over-warning risk index as the robust peak-shaving scheme, and then encapsulates and generates a set of coordinated scheduling instructions bound to the arrival scenario set version identifier and the arrival scenario credible weight set.

[0088] The closed-loop control module is used to execute the recharge closed-loop function described in step four of Embodiment 1. This module is configured to send the collaborative scheduling instruction set to the local control systems of reservoirs A, B, and C through the scheduling communication link, and continuously receive execution feedback information. This module synchronously accesses the real-time reservoir water level, measured outflow, and water level or flow observations at the upstream and downstream control sections of each reservoir, and accesses continuously updated rainfall information such as radar rainfall, rain gauge rainfall, and short-term forecasts. It uses a unified timescale alignment and quality control mechanism to maintain data consistency. This module introduces an execution consistency verification mechanism to align and compare the instruction target and the control feedback and identify execution deviations. This module performs state recharge on the digital twin reservoir group model to refresh the initial state package of the twin, and performs incremental updates on the time-bound scenario set and rolling recalculation of the time-bound scenario confidence weight set within the rolling cycle. When risk signals such as execution deviation, abnormal rise in water level at downstream control section, or rapid convergence of safety margin are detected, the rolling re-optimization process is triggered to update the robust peak-shifting scheme and the collaborative scheduling instruction set, and control smoothing constraints are set on the updated instruction set to reduce instruction jumps. After each round of rolling, the observation snapshot, execution receipt, time-bound scenario confidence weight set, and evaluation results before and after the update are archived to the historical strong convective event sample library for subsequent parameter statistical calibration or adaptive adjustment.

[0089] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0091] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0093] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method, characterized in that, Includes the following steps: Acquire multi-source water and rainfall data and engineering data related to the reservoir group; Unified time-scale alignment and quality control are performed on multi-source water and rainfall data and engineering data to generate a twin initial state package for driving the digital twin reservoir group model; Based on the twin initial state package, multiple candidate rainfall sequences are generated by perturbation rainfall input, and each candidate rainfall sequence is input into the digital twin reservoir group model to obtain the corresponding inflow prediction sequence and propagation time parameters, thereby constructing the time scenario set; Based on multiple pieces of evidence related to the arrival deviation, the component credibility of each arrival scenario is generated and then fused and normalized to obtain the set of arrival scenario credibility weights corresponding to the set of arrival scenarios. A set of candidate coordinated peak-shaving release schemes that meet the executability constraints is generated. For each candidate coordinated peak-shaving release scheme, a digital twin reservoir group model is used to perform multi-scenario simulations under various arrival scenarios in the arrival scenario set to obtain the downstream control section water level process and the reservoir group release process. Based on the simulation results, scenario losses under each arrival scenario are constructed, and the scenario losses are weighted and summarized according to the arrival scenario credibility weight set to obtain the weighted peak-shaving exceedance risk index corresponding to each candidate coordinated peak-shaving release scheme. Candidate coordinated peak-shaving release schemes whose weighted peak-shaving exceedance risk index meets the preset optimization criteria are selected as robust peak-shaving schemes, and a coordinated scheduling instruction set is generated. The coordinated scheduling instruction set is sent to the local control system and execution receipt information is received. Execution consistency verification is performed based on the execution receipt information. The verification results and newly added observation data are fed back into the digital twin reservoir group model to update the state. The arrival scenario set and the arrival scenario trusted weight set are updated within a preset rolling cycle. When the triggering condition is met, a rolling re-optimization process is initiated based on the updated twin initial state package, the arrival scenario set, and the arrival scenario trusted weight set to update the robust peak-shaving scheme and the coordinated scheduling instruction set.

2. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: Multi-source water and rainfall data and engineering data include: rainfall information corresponding to radar extrapolated rainfall products and gridded data of short-term rainfall forecasts in numerical weather prediction; hydrological information corresponding to upstream and downstream control section observations; and engineering information corresponding to reservoir water levels, gate status, and unit status.

3. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: Quality control includes one or more of the following: marking and handling outliers, imputation of missing data, and statistics on the freshness of key data; the freshness of key data reflects at least one of the following: data arrival delay and missing rate.

4. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: When constructing the time-based scenario set, the perturbations applied to the rainfall input include the area offset perturbation and the rainfall intensity perturbation.

5. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: The method also includes online propagation time correction, which is achieved by calculating the cross-correlation or characteristic time difference of the observation sequences of upstream and downstream control sections.

6. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: Multiple pieces of evidence related to arrival bias include rainband drift, rising edge alignment residual, propagation delay residual, critical data arrival delay, and critical data missing rate.

7. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: The enforceability constraints include the downstream control section warning water level, the reservoir flood control limit water level, the minimum ecological discharge flow, the gate opening and closing rate limit, and the unit start-up and shutdown time constraints.

8. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: The scenario loss is composed of the penalty amount for exceeding the warning level, which reflects the risk of the downstream control section exceeding the warning level; the penalty amount for superimposed exceeding the limit, which reflects the risk of the reservoir group releasing water; and the penalty amount for insufficient peak shifting, which reflects the risk of insufficient peak shifting interval.

9. The short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling method according to claim 1, characterized in that: The consistency verification process includes: aligning the target control sequence and execution receipt in the collaborative scheduling instruction set on the time axis to obtain the control deviation; generating an execution deviation flag when the control deviation exceeds a preset threshold; and using the execution deviation flag as one of the triggering conditions for triggering the rolling re-optimization process.

10. A short-, medium-, and long-term coupled digital twin reservoir group collaborative scheduling system, characterized in that: The scheduling system is used to implement the method according to any one of claims 1-9, and includes the following modules: The access verification module is used to acquire multi-source water and rainfall data and engineering data related to the reservoir group; it performs unified time-scale alignment and quality control on the multi-source water and rainfall data and engineering data to generate a twin initial state package for driving the digital twin reservoir group model; The scenario weighting module is used to generate multiple candidate rainfall sequences based on the twin initial state package and the perturbation rainfall input. Each candidate rainfall sequence is then input into the digital twin reservoir group model to obtain the corresponding inflow prediction sequence and propagation time parameters, thereby constructing the time scenario set. Based on multiple pieces of evidence related to the arrival deviation, the component credibility of each arrival scenario is generated and then fused and normalized to obtain the set of arrival scenario credibility weights corresponding to the set of arrival scenarios. The simulation and optimization module generates a set of candidate coordinated peak-shaving and discharge schemes that meet the executability constraints. It then uses a digital twin reservoir group model to perform multi-scenario simulations of each candidate scheme under various arrival scenarios in the arrival scenario set, obtaining the downstream control section water level process and the reservoir group discharge process. Based on the simulation results, it constructs the scenario loss for each arrival scenario and weights and summarizes the scenario losses according to the arrival scenario credibility weight set to obtain the weighted peak-shaving and discharge risk index corresponding to each candidate scheme. Candidate coordinated peak-shaving and discharge schemes whose weighted peak-shaving and discharge risk index meets the preset optimization criteria are selected as robust peak-shaving schemes, and a coordinated scheduling instruction set is generated. The closed-loop control module is used to send the collaborative scheduling instruction set to the local control system and receive execution feedback information, and to perform execution consistency verification based on the execution feedback information; to feed the verification results and newly added observation data back into the digital twin reservoir group model to update the state; to update the arrival scenario set and the arrival scenario trusted weight set within a preset rolling cycle, and to start the rolling re-optimization process based on the updated twin initial state package, arrival scenario set and arrival scenario trusted weight set when the trigger condition is met, so as to update the robust peak-shaving scheme and collaborative scheduling instruction set.

Images