Reservoir group scheduling strategy construction method based on medium and long term forecast

Abstract

The invention discloses a reservoir group scheduling strategy construction method based on medium and long term forecasting, and relates to the technical field of hydraulic engineering scheduling. The method comprises the following steps: mapping medium and long term forecast data to a typical scheduling scene, and constructing a forecast evaluation data set; calculating forecast precision evaluation results of the sub-scenes; constructing a dispatching application-oriented applicability evaluation model, and generating a disturbance scene by utilizing forecast error distribution under a flood risk budget framework; a disturbance scene is converted into an equivalent warehousing process including upstream adjustment influence through an upstream library group equivalent warehousing influence module, and a scheduling response model is driven to perform multi-scene simulation; and carrying out statistics on the scene proportion meeting the risk budget upper limit, calculating the applicability probability and generating a scheduling strategy. According to the method, through risk budget simulation and equivalent warehousing correction, the problems of separation of forecasting precision and scheduling effect and upstream adjustment interference assessment are solved, and scientific scheduling based on risk quantification is realized.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy project scheduling technology, and in particular, it is a method for constructing a reservoir group scheduling strategy based on medium and long-term forecasts. Background Technology

[0002] Medium- and long-term hydrological and meteorological forecasts are a key support for large-scale cascade reservoir groups to achieve efficient water resource utilization and drought and flood disaster prevention. They help to extend the forecast period for scheduling, optimize the timing of water storage at the end of the flood season, and improve power generation efficiency.

[0003] Currently, reservoir group scheduling relies heavily on conventional scheduling charts or deterministic rules based on historical statistical patterns. Medium- and long-term forecast products are typically used only as qualitative consultation references and are not truly incorporated into decision-making models. Existing forecast evaluation methods mainly follow traditional statistical indicators from the meteorological field, such as mean square error, anomaly percentage, or pass rate, focusing on evaluating the degree of dispersion between forecast and measured values ​​from a meteorological or hydrological statistical perspective.

[0004] In practical applications, existing technologies mainly suffer from a nonlinear disconnect between forecast accuracy and scheduling risk, as well as issues with the assessment benchmark for upstream cascade regulation interference. Therefore, further research and innovation are needed to address these problems in existing technologies. Summary of the Invention

[0005] Purpose of the invention: In view of the above-mentioned problems in the prior art, this application provides a method for constructing a reservoir group scheduling strategy based on medium and long-term forecasts.

[0006] Technical solution: According to one aspect of this application, a method for constructing a reservoir group scheduling strategy based on medium- and long-term forecasts includes:

[0007] Medium- and long-term water and rainfall forecast data are mapped to predefined typical scheduling scenarios and watershed control sections to generate a forecast evaluation dataset containing scenario labels and spatial attributes.

[0008] Based on the forecast assessment dataset, the forecast accuracy evaluation results for different scenarios are calculated, and the statistical deviation and process characteristic deviation between the forecast values ​​and the measured hydrological and meteorological data are quantified.

[0009] The forecast accuracy evaluation results are used to drive the applicability assessment model for scheduling applications. Multi-scenario scheduling simulation is performed under the flood risk budget framework to calculate the risk budget consumption caused by forecast errors in the reservoir group scheduling response.

[0010] The proportion of simulation scenarios that meet the risk budget ceiling is statistically analyzed, scenario-based forecast applicability indicators are generated, and reservoir group scheduling strategies are constructed based on the forecast applicability indicators.

[0011] Beneficial effects: This invention solves the problems of disconnect between forecast accuracy and scheduling effectiveness, as well as upstream regulation interference assessment, through risk budget simulation and equivalent data entry correction, thus realizing scientific scheduling based on risk quantification. The related technical effects will be described in detail below with reference to specific embodiments. Attached Figure Description

[0012] Figure 1 A flowchart illustrating a method for constructing a reservoir group scheduling strategy based on medium- and long-term forecasts, provided in this application embodiment.

[0013] Figure 2 A flowchart illustrating the risk budget consumption of the calculation forecast error in the reservoir group scheduling response, provided in the embodiments of this application.

[0014] Figure 3 A flowchart for providing the statistical proportion of simulation scenarios that meet the risk budget ceiling and generating scenario-based forecast applicability indicators for embodiments of this application.

[0015] Figure 4 This is a flowchart illustrating the scenario-by-scenario evolution calculation of inputting a set of disturbance forecast scenarios into a scheduling response model, as provided in an embodiment of this application.

[0016] Figure 5 A flowchart for identifying the set of upstream key reservoirs involved in the calculation, provided for an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] To address the aforementioned issues, the applicant conducted in-depth searches and analyses, and discovered:

[0020] Correspondingly, traditional statistical errors cannot quantify the destructive power of forecast deviations on scheduling objectives. Small meteorological errors, if they occur during critical flood control periods, may lead to significant scheduling risks, such as exceeding warning levels or failure to store water. There is a lack of a quantitative mechanism to directly map forecast uncertainties into the consumption of flood control risk budget.

[0021] Next, in a large cascade reservoir system, the inflow boundary of downstream reservoirs is greatly affected by the human regulation of upstream reservoirs. Existing assessments are often based on natural runoff and cannot separate the interference of upstream regulation behavior on inflow forecasts. This makes it impossible to accurately assess the true applicability of forecasts in complex joint regulation environments and to generate robust joint regulation strategies.

[0022] In some embodiments,

[0023] To solve these problems, combined with Figures 1 to 5 The present invention will be specifically described through the following embodiments.

[0024] Example 1 provides an exemplary scheme for constructing a reservoir group scheduling strategy based on medium- and long-term forecasts. While existing medium- and long-term hydrological and rainfall forecasts provide some predictive information from a meteorological perspective, the lack of a direct mapping mechanism to specific reservoir scheduling risks makes it difficult for scheduling decision-makers to quantify the impact of forecast errors on specific objectives such as flood control and water storage, leading to the adoption of conservative strategies. This example introduces simulation evaluation under a risk budget framework, transforming the uncertainty of meteorological forecasts into the risk cost of scheduling decisions, and generating a scientific scheduling strategy.

[0025] Step 101: Map medium- and long-term water and rainfall forecast data to predefined typical scheduling scenarios and watershed control sections to generate a forecast evaluation dataset containing scenario labels and spatial attributes.

[0026] Specifically, this step constructs a multi-dimensional forecast data storage structure to accommodate forecast information from different sources and at different scales. This data structure can be represented as a five-dimensional tensor D(S, T). _pub L, R, δ _t ), where S represents the forecast scheme identifier, T _pub L represents the forecast release time, R represents the forecast lead time, and δ represents the spatial unit. _tThis represents the time step. In medium- and long-term forecasts, spatial units can be grid points, sub-basins, or administrative divisions. To interface with reservoir scheduling models, the original grid point data needs to be mapped to watershed control sections. For example, using Kriging interpolation or Thiessen polygon methods, the rainfall forecast data from multiple grid points covering a reservoir's catchment area can be weighted and averaged to obtain the areal average rainfall above the reservoir dam site. Alternatively, a confluence model can be used to directly convert rainfall into an inflow sequence.

[0027] Furthermore, the forecast data is assigned scenario labels for different scheduling objectives. For example, if the forecast period falls within the main flood season, the system automatically labels it as a flood control-dominant scenario based on the timestamp; if it falls within the end-of-flood transition period, it is labeled as an end-of-flood water storage scenario. The generated forecast evaluation dataset not only contains numerical sequences but also carries the aforementioned data, providing a structured foundation for subsequent scenario-specific evaluations.

[0028] Step 102: Calculate the forecast accuracy evaluation results for each scenario based on the forecast assessment dataset, quantifying the statistical deviation and process characteristic deviation between the forecast values ​​and the measured hydrological and meteorological data. This is used to measure forecast quality.

[0029] In this embodiment, a two-tiered evaluation system combining basic statistical indicators and process characteristic indicators is adopted. Basic statistical indicators, used to reflect the overall consistency of the sequences, include mean squared error (MSE) and correlation coefficient. Specifically, MSE = (1 / n) × ∑((P...) _i -O _i ) 2 ), P _i For the predicted value at the i-th time step, O _i is the measured value, and n is the sample size.

[0030] Furthermore, process characteristic deviations focus more on morphological features that have a critical impact on dispatch safety. For example, in flood control scenarios, attention is paid to the relative error of peak flow and the deviation of peak occurrence time; in water storage scenarios, attention is paid to the deviation of the cumulative inflow within a time period. Through hierarchical calculation, the forecast accuracy evaluation result is no longer a single score, but a vector containing multi-dimensional error components, which can characterize the performance of the forecast from different aspects. The calculation of the indicators is based on historical back-forecast data and measured data from the same period, making the evaluation results statistically significant.

[0031] Step 103: Utilize the forecast accuracy evaluation results to drive an applicability assessment model for scheduling applications. Perform multi-scenario scheduling simulations within a flood risk budget framework to calculate the risk budget consumption caused by forecast errors in the reservoir group scheduling response. This step constructs a virtual scheduling simulation environment for trial and error. The applicability assessment model is a simulator that couples an error generator and scheduling rules.

[0032] Accordingly, based on the accuracy evaluation results, such as MSE and process error distribution, a large number of forecast scenarios with random perturbations are generated using the Monte Carlo method. Next, these scenarios are input into the reservoir group scheduling response model. This response model incorporates the reservoir's water balance equations:

[0033] V(t+1)=V(t)+(Q _in (t)-Q _out (t))×δ _t ,

[0034] In the formula, V(t) is the storage capacity at time t, V(t+1) is the storage capacity at the next time step, and Q... _in For inbound traffic, Q _out This refers to the outflow from the reservoir. Simulations can yield a set of possible water level events at the downstream control section when a certain forecast is used for scheduling. The risk budget consumption is a quantitative measure of the simulation results; for example, it's the probability that the downstream water level will exceed the warning level in all simulation scenarios, or the probability that the reservoir will not be fully filled by the end of the flood season. The probability value represents the safety margin or risk budget consumed by using this forecasting scheme.

[0035] Step 104: Calculate the proportion of simulation scenarios that meet the risk budget ceiling, generate scenario-based forecast applicability indicators, and construct a reservoir group scheduling strategy based on the forecast applicability indicators.

[0036] In this step, the system pre-sets budget limits for various risk indicators, such as the downstream exceedance probability not exceeding 5%. By comparing the risk consumption calculated from the simulation with the budget limits, the proportion of scenarios satisfying the constraints to the total number of scenarios is statistically determined, denoted as P. _ok This indicator reflects the confidence level in executing predetermined scheduling rules given the current forecast accuracy. If P _ok If the level is high, such as greater than 90%, the generated scheduling strategy can be proactive, such as dynamically storing water above the flood control limit; if P _ok If the forecast is low, the strategy automatically adjusts to a conservative and defensive approach, such as pre-emptively releasing resources. This mechanism ensures that the scheduling strategy always matches the actual availability forecast, avoiding both blind risk-taking and resource waste caused by excessive conservatism.

[0037] Example 2 provides an optional implementation method for scenario-based data preprocessing and process-oriented indicator construction. Specifically, it addresses the scenario mapping logic in data preprocessing and the method for extracting process-oriented indicators for flood control and water storage scenarios. This example solves the problem of neglecting the differences in scheduling scenarios in traditional evaluation methods through data labeling and feature extraction.

[0038] Step 201: Obtain the applicable time window and associated watershed spatial range of predefined typical scheduling scenarios; read the forecast release time, forecast period, and forecast grid coordinate information of medium- and long-term hydrological and rainfall forecast data; determine whether the forecast period falls within the applicable time window and whether the forecast grid coordinates are within the watershed spatial range, and establish a mapping relationship between forecast records and typical scheduling scenarios; attach corresponding scenario labels to each forecast record that conforms to the mapping relationship, and generate a forecast evaluation dataset. Scenario labels are used to trigger the calculation of corresponding evaluation indicators and the definition of risk budgets in subsequent steps. This is used to achieve intelligent data classification.

[0039] The system backend maintains a scheduling scenario definition table, which defines scenarios such as the primary flood control scenario in the lower reaches of the Jinsha River. Its applicable time window is from June 1st to August 31st each year, and its associated spatial range is the catchment area of ​​the cascade reservoirs in the lower reaches of the Jinsha River. When a forecast record published on July 1st, predicting rainfall for the next 30 days, is read, the system compares the time attributes to confirm that its coverage period falls within the June-August flood control window. Then, using GIS spatial calculations, it determines that the grid coordinates of the forecast data fall within the lower reaches of the Jinsha River basin.

[0040] Furthermore, if both conditions are met simultaneously, the system will assign a scenario label to the data. This label not only identifies the purpose of the data but also links it to the specific evaluation algorithm library to be called in subsequent steps, such as calculating only the peak flow error and not the low-water flow error.

[0041] Step 202: Select forecast records and corresponding measured records with the scene label of flood control as the dominant scene from the forecast assessment dataset; extract key features of the flood process from the forecast records and measured records, including peak flow, peak occurrence time and duration of water level exceeding the warning level; calculate the relative error of peak flow, peak time error and duration error of exceeding the standard between the forecast and the measured data respectively, and use the set of errors as the flood control process error index.

[0042] In this step, the system automatically executes a flood feature extraction algorithm for data tagged with flood control information. Taking a specific flood event as an example, the predicted flow rate is denoted as Q. _pre (t), the measured flow rate process is denoted as Q. _obs (t). The algorithm searches for the maximum value and its corresponding time index in the two sequences to obtain the predicted flood peak Q. _p_pre Forecast peak time T _p_pre And the measured flood peak Q _p_obs Measured peak time T _p_obs Relative error of peak flow rate E _Qp The calculation formula is:

[0043] E _Qp =(Q _p_pre-Q _p_obs ) / Q _p_obs .

[0044] For example, if the predicted flood peak is 15,000 cubic meters per second, and the actual measured value is 16,000 cubic meters per second, then E _Qp =(15000-16000) / 16000=-6.25%, indicating that the forecast was underestimated. Correspondingly, the error in the peak flood time is E. _Tp The calculation formula is: E _Tp =T _p_pre -T _p_obs The unit is usually hour or day. If the forecasted flood peak occurs on the 10th, but the actual occurrence is on the 11th, then E _Tp =-1 day indicates that the flood peak is predicted to arrive earlier than expected.

[0045] Furthermore, if a warning flow Q is set... _warning The algorithm also counts the number of sequences greater than Q. _warning The number of time steps is used to calculate the error E of the duration of exceeding the standard. _Dur The above indicators together constitute a precise profile of the flood control scenario.

[0046] Step 203: Select forecast records and corresponding measured records with the scenario label of flood season water storage scenario from the forecast assessment dataset; extract resource replenishment process features from the forecast records and measured records, including the cumulative inflow at the end of the flood season and the reservoir water level on the key control date of the water storage period; calculate the relative error of the cumulative inflow and the water level deviation on the key date between the forecast and the measured data respectively, and use the set of errors as the water storage process error index.

[0047] Specifically, for data labeled as end-of-flood storage scenarios, typically in September and October, the assessment focus shifts to water resource volume. The algorithm integrates or sums the flow series within the forecast period to obtain the forecast cumulative inflow volume Vol. _pre =∑(Q _pre (t)×δ _t Similarly, the measured cumulative water volume Vol can be obtained. _obs .

[0048] Furthermore, the cumulative relative error E of the inflow volume _Vol Calculated as: E _Vol =(Vol _pre -Vol _obs ) / Vol _obs This indicator reflects whether there is sufficient water storage. Simultaneously, the system monitors water level deviations on key control dates, such as September 30th or October 31st. Using the reservoir's capacity curve, the theoretical final water level W can be calculated based on the initial water level and cumulative inflow. _pre_end Compare it with the measured end-of-period water level W _obs_endBy comparison, the water level deviation E is obtained. _W_end =W _pre_end -W _obs_end For example, if the calculated water level deviation is -2 meters, it means that according to the forecasted water inflow, the final water level will be 2 meters lower than the actual level, which has a direct warning effect on the completion of the water storage task.

[0049] Example 3 describes an optional implementation process for an applicability assessment simulation based on a risk budget framework, particularly the risk budget simulation mechanism. It describes how to transform abstract probabilistic forecast errors into concrete, operable scheduling risk indicators. This example constructs an evaluation system that includes a flood risk budget framework, a scheduling response model, and a multi-scenario simulation module. It can simulate the operational status of a reservoir group under different forecast accuracies and calculate the safety budget consumed by the forecast error. Exemplarily, this example includes:

[0050] Step 301: The applicability assessment model for scheduling applications includes: a flood risk budget framework, used to define the calculation logic of risk indicators and the corresponding risk budget upper limit under the watershed scheduling scenario; a scheduling response model, used to parametrically describe the reservoir group's reservoir water level and outflow response relationship under the constraints of input flow sequence and scheduling rules; and a multi-scenario simulation module, configured to generate random disturbance inputs by combining forecast accuracy evaluation results, driving the scheduling response model to output risk assessment results.

[0051] Specifically, for the flood control-dominated scenario, the risk indicator is defined as the probability that the water level at the downstream control section exceeds the warning level, and the maximum allowed probability of exceeding the warning level is set as the upper limit of the risk budget. For the flood season water storage scenario, the risk indicator is defined as the probability that the reservoir water level at the end of the flood season is lower than the target water storage level, and the maximum allowed probability of insufficient water storage is set as the upper limit of the risk budget. The risk indicator calculation logic is used to guide the multi-scenario simulation module to count the frequency of events that trigger the above-defined events when calculating the risk budget consumption.

[0052] In this step, a digital twin environment for simulation evaluation was constructed. Correspondingly, the flood risk budget framework, a data structure instantiated within the system, includes predefined risk event definitions E. _i and its corresponding allowed occurrence probability threshold P _limit_i For example, regarding flood control in the Yangtze River basin, E is defined as... _flood With the water level at Shashi Station exceeding 43.0 meters, a budget ceiling P is set for it. _limit_flood The probability is 0.05, or 5%. A scheduling response model is used, which is part of the simulation's physics engine. In this embodiment, the model can be constructed based on a discretized water balance equation, as follows:

[0053] V _(t+1) =V _t +(Q _in_t-Q _out_t )×Δt-Loss _t .

[0054] Among them, V _t Let V be the reservoir's water storage at time t. _(t+1) Q represents the reservoir's water storage capacity at the next moment. _in_t This refers to the inflow or the flow after upstream correction; Q _out_t The outflow rate is constrained by the reservoir's discharge capacity curve and downstream safe discharge limits; Loss _t The value represents evaporation and seepage losses; Δt is the simulation step size (e.g., 6 hours or 24 hours). The reservoir water level is calculated in real time using the reservoir capacity-water level relationship curve Z=f(V), where Z is the water level and V is the reservoir capacity.

[0055] Based on this, the multi-scenario simulation module connects the input and the model, and manages the iterative process of the Monte Carlo simulation.

[0056] Optionally, the scheduling response model can further include a hydraulic evolution module (such as the Muskingan method) to calculate the flow and water level at the downstream control section after the outflow has evolved through the river channel, thus more accurately assessing downstream risks. Furthermore, the risk budget framework can be designed to be dynamically adjusted, automatically tightening or loosening the risk budget ceiling based on the current soil moisture content or previous rainfall conditions in the watershed.

[0057] Step 302: Analyze the statistical bias and process characteristic bias in the forecast accuracy evaluation results, and construct error probability distribution data to describe the probability characteristics of forecast errors; based on the error probability distribution data, perform multiple random sampling perturbations on the forecast sequences in the forecast evaluation dataset to generate a perturbation forecast scenario set containing multiple sample sequences.

[0058] In this step, the system transforms deterministic forecasts into a set of probabilistic scenarios. Specifically, the system reads error indicators, such as the mean μ of the relative error of peak flow. _Q and standard deviation σ _Q Assuming the error follows a certain probability distribution, such as a normal distribution or a Pearson type III distribution, construct the error model N(μ). _Q , σ _Q When generating disturbance scenarios, the system performs random sampling. For example, for the k-th sample, a random error factor ε is generated. _k This is then superimposed onto the forecast flow process.

[0059] Optionally, employing a process correction method instead of simply adding noise point by point can maintain the continuity and physical rationality of the hydrological process. Specifically, based on the peak flood error ε obtained from sampling... _Qp Peak occurrence time error ε _TpThe original forecast process line is scaled and translated to generate a new perturbation process Q. _sim_k (t). By repeating this process N times (e.g., N=1000), a set of 1000 possible inflow processes is obtained, which is the perturbation forecast scenario set. This set statistically covers all hydrological conditions that may occur at the current forecast accuracy.

[0060] Step 303: Input the disturbance forecast scenario set into the scheduling response model for scenario-by-scenario evolution calculation, and output the control section state variables corresponding to each sample sequence; according to the risk index calculation logic defined in the flood risk budget framework, transform the control section state variables into scenario risk index values ​​for each disturbance scenario, and use the scenario risk index values ​​as the risk budget consumption.

[0061] Specifically, the system traverses each sequence Q in the set of disturbance prediction scenarios. _sim_k (t) is used as a boundary condition input to the scheduling response model. The model evolves step-by-step according to predetermined scheduling rules, such as controlling water levels according to flood control limits or generating power at maximum output, to calculate the reservoir's operational trajectory under this scenario, with water level Z. _k (t), Outbound Q _out_k (t) and the status of downstream control sections, such as the water level Z in Shashi. _control_k (t). After the evolution is complete, the system calls the logic in the risk budget framework to make a judgment. For example, for the k-th scenario, it checks the maximum downstream water level max(Z). _control_k Has the water level exceeded the warning level? _warn If the value exceeds the limit, record the risk indicator value I for that scenario. _risk_k A value of 1 indicates a risk event has occurred; if the risk is not exceeded, a value of 0 indicates risk consumption. Alternatively, the specific extent of the exceedance can be calculated as a measure of risk consumption.

[0062] Step 304: Read the risk budget ceiling corresponding to the current dispatch scenario from the flood risk budget framework; compare the scenario risk index value of each disturbance scenario with the risk budget ceiling one by one, and filter out the disturbance scenarios in which all risk index values ​​do not exceed the risk budget ceiling, and count the number of scenarios that meet the budget; calculate the proportion of the number of scenarios that meet the budget to the total number of samples in the disturbance forecast scenario set, and use the proportion as the scenario-based forecast applicability index.

[0063] In this step, the convergence from microscopic physical quantities to macroscopic decision indicators is achieved. Assume a total of N simulation scenarios. _total The number is 1000. The system counts the total number N scenarios where no risk event occurred or the risk loss was within the acceptable range. _ok For example, if the downstream water level does not exceed the warning level in 950 scenarios, then N _ok =950. Applicability probability P _okCalculated as: P _ok =N _ok / N _total =0.95.

[0064] Next, the system will P _ok Compare with a preset confidence threshold, such as 0.90. If P _ok A value ≥0.90 indicates that, given the current forecast accuracy, the probability of the predetermined scheduling plan triggering a flood risk is only 5%, lower than the system's risk budget limit. Assuming a 10% risk is allowed, the forecast plan is deemed applicable in the current scenario. Conversely, if P... _ok With a value of only 0.70, there is a 30% chance that the value will exceed the limit. In this case, the system will determine that the forecast is not applicable and trigger a more conservative scheduling strategy.

[0065] Example 4 provides an exemplary scheme for equivalent inflow correction and parameter calibration of upstream reservoir groups. It describes a logical correction module for cascade reservoir groups. In giant cascade systems such as the Jinsha River downstream-Three Gorges Dam, the inflow to downstream reservoirs is significantly affected by upstream reservoir regulation, and directly using natural inflow forecasts will lead to errors. This example eliminates upstream human interference and restores the inflow boundary by constructing a parameterized equivalent inflow model. As an example, this example can be implemented using the following steps:

[0066] Step 401: Call the upstream reservoir group equivalent inflow impact module. The module is configured with the equivalent mapping relationship between the outflow of the upstream key reservoir and the natural inflow of the downstream control section. Using this impact module (upstream reservoir group equivalent inflow impact module), each sample sequence in the disturbance forecast scenario set is converted into a disturbance equivalent inflow process that includes the upstream regulation impact.

[0067] In this step, a preprocessing stage is embedded in the simulation loop. This involves applying the perturbation scenario Q... _sim_k (t) Before being input into the scheduling model, the model is shaped by this module. The module's input includes not only the interval natural inflow generated based on rainfall forecasts, but also the operational plan data of key upstream reservoirs, i.e., the expected outflow sequence from the upstream reservoirs. The module performs hydraulic superposition and calculation on the two flows, outputting the actual inflow to the target reservoir. This process allows the simulation model to perceive whether the upstream reservoir is storing water or releasing floodwater, correctly assessing the downstream risks. For example, when an upstream reservoir plans to intercept a large-scale flood peak, even if the rainfall forecast is high, the corrected equivalent inflow will be reduced, lowering the calculated risk index and avoiding misjudgment.

[0068] Step 402: The upstream reservoir group equivalent inflow impact module uses a linear superposition function with time delay parameters to calculate the disturbance equivalent inflow process. Its calculation logic is as follows: For each time step, the outflow of each reservoir in the upstream key reservoir set is time-delayed shifted and weighted summed, and superimposed with the weighted natural inflow forecast data to obtain the equivalent inflow of the control section.

[0069] In this step, the structure of the linear superposition function satisfies:

[0070] Q _equ (t)=∑[a _k ×Q _k_out (t-τ _k )]+b×Q _nat (t);

[0071] Among them, Q _equ Q(t) represents the equivalent inflow rate at time t. _k_out Let τ be the outflow from the k-th upstream key reservoir. _k Let a be the propagation time delay corresponding to the k-th reservoir. _k Let Q be the influence weight of the k-th reservoir. _nat (t) represents the predicted natural water inflow at time t, and b is the natural water inflow correction factor.

[0072] In other embodiments, it may also be described as:

[0073] Q _equ (t)=∑(a _j ×Q _j_out (t-τ _j ))+b×Q _nat (t);

[0074] Where j represents the j-th key reservoir upstream; Q _j_out τ is the outflow rate of the reservoir; _j The average lag time (usually in hours or days) from the reservoir to the target section; a _j is the weight of the reservoir's influence on the target section (usually reflecting water loss along the course or the proportion of water flowing into the section); b is the correction coefficient for natural inflow forecast (used to correct systematic biases in rainfall-runoff models).

[0075] For example, the inflow to the Three Gorges Reservoir is directly controlled by the outflow from the upstream Xiangjiaba Reservoir. Assuming the propagation time from Xiangjiaba to the Three Gorges is 24 hours, the formula becomes:

[0076] Q _TG_in (t) = 1.05 × Q _XJB_out (t-24h)+1.1×Q _interval (t).

[0077] In the formula, 1.05 may reflect the contribution of tributaries flowing into the area between Xiangjiaba and the Three Gorges. That is, 1.1 corresponds to b, Q _interval (t) represents the natural water inflow forecast data for this interval corresponding to time t, Q _XJB_out Corresponding to the outflow from the upstream reservoir, Q _TG_in (t) represents the equivalent inflow to the reservoir at time t.

[0078] Step 403: The construction of the upstream reservoir group equivalent inflow impact module includes identifying the set of upstream key reservoirs participating in the calculation. This identification specifically involves: obtaining the historical operation records of each reservoir in the upstream basin of the target control section and the historical measured flow sequence of the control section; calculating the cross-correlation coefficient and lag response time between the outflow of each reservoir and the historical measured flow sequence of the control section; and, based on the significance of the cross-correlation coefficient and the reservoir regulation capacity, selecting reservoirs that have a significant impact on the flow of the control section from among the reservoirs in the upstream basin to form the set of upstream key reservoirs.

[0079] In other words, it is possible to determine the maximum cross-correlation coefficient R. _max Perform a significance test, such as a t-test, where p < 0.05 is considered significant. If the maximum cross-correlation coefficient R0.05 of a certain reservoir... _max If the value is greater than a preset threshold, such as 0.6, the significance test is passed.

[0080] Furthermore, given the numerous upstream reservoirs, a selection process is necessary to reduce model complexity. This step employs a data-driven approach. Historical outflow sequences from all upstream reservoirs and measured flow sequences at the target cross-section are collected. The cross-correlation function R(τ) is calculated, and the lag time τ that maximizes the correlation is found. _max If the maximum cross-correlation coefficient R of a certain reservoir is... _max If a reservoir's regulation capacity exceeds a preset threshold (e.g., 0.6) and its proportion of the annual runoff at the target section exceeds a certain threshold (e.g., 5%), it is included in the set of key reservoirs. For example, in the Jinsha River basin, although reservoirs such as Liyuan and Ahai are located further upstream, analysis revealed that their regulation signals had attenuated after long-distance propagation, resulting in low correlation. In contrast, reservoirs such as Xiluodu and Xiangjiaba showed extremely high correlation with no significant attenuation. Therefore, only Xiluodu and Xiangjiaba were selected as key reservoirs, and equivalent models were constructed.

[0081] Step 404, Propagation delay τ in linear superposition functions _k Influence weight a _kThe natural inflow correction factor b is determined using a parameter calibration method: a training sample set containing historical data on outflows from key upstream reservoirs, natural inflows, and measured flows at control sections is used; with the training sample set as input, and aiming to minimize the sum of squared errors between the calculated values ​​of the linear superposition function and the measured flows at control sections, a parameter estimation algorithm is employed to adjust the propagation time delay τ. _k Influence weight a _k The model parameters were obtained by jointly optimizing and calibrating the model with the natural water correction factor b.

[0082] In this step, in order to ensure parameter a _k τ _k To ensure the accuracy of b, this embodiment uses historical data inversion calibration. A training set containing historical flood season data from the past 5-10 years is constructed, including known upstream outflow Q. _up_hist Measured water inflow Q _nat_hist and downstream actual warehousing Q _in_obs .

[0083] Furthermore, we define the objective function J = ∑((Q) _equ_calc (t)-Q _in_obs (t)) 2 ), where Q _equ_calc (t) is the calculated equivalent inflow rate at time t. The least squares method or genetic algorithm is used to search for the parameter combination that minimizes J. For example, after training, the following parameter set might be obtained: a _1 =0.98, τ _1 =30 hours, b=1.2. This set of parameters is fixed in the module, enabling the model to memorize the hydraulic characteristics of this specific watershed.

[0084] Example 5 provides an exemplary scheme for comprehensive evaluation and strategy generation. Specifically, it generates scheduling strategies based on the evaluation results. This includes:

[0085] Step 501: Construct a comprehensive evaluation rule that includes accuracy grading thresholds and applicability probability thresholds; combine the statistical bias and process characteristic bias in the forecast accuracy evaluation results, as well as the scenario-based forecast applicability index, and use the comprehensive evaluation rule to conduct a multi-dimensional comprehensive score for the forecast scheme; classify the forecast scheme into a preset comprehensive evaluation level based on the comprehensive score results.

[0086] Specifically, this step establishes a two-dimensional evaluation matrix. The horizontal axis represents the accuracy score, based on MSE and E. _Qp The indicators are as follows, with the vertical axis representing the applicability probability, based on P. _ok The system can set the threshold using the following methods:

[0087] If P _okIf the accuracy score is ≥0.90 and the accuracy score is ≥80, it is rated as Level I (Highly Applicable).

[0088] If 0.70 ≤ P _ok <0.90, rated as Level II (generally applicable);

[0089] If P _ok <0.70, rated as Level III (not suitable for use).

[0090] Accordingly, this classification not only considers objective error, but also uses P... _ok A sensitivity assessment of the current scheduling conditions was introduced. For example, when the reservoir water level is low, even if the forecast error is large, P is still far from the risk boundary. _ok The level may still be very high, and the forecast is still considered applicable; however, when the reservoir level approaches the warning line, the same error will lead to P _ok The decline was sharp, and the forecast level was subsequently lowered.

[0091] Step 502: Based on the comprehensive assessment level, generate forecast application recommendations for the forecast scheme in the current scheduling scenario; if the comprehensive assessment level is highly applicable, mark the forecast scheme as the main basis for scheduling decisions in the reservoir group scheduling strategy; if the comprehensive assessment level is generally applicable, recommend using multi-source forecast joint analysis or reducing the weight of the forecast scheme in decision-making in the reservoir group scheduling strategy; if the comprehensive assessment level is not suitable for use, recommend activating emergency plans or relying on short-term forecasts for rolling correction in the reservoir group scheduling strategy.

[0092] Specifically, the system outputs specific strategy instructions based on the rating. For Level I forecasts, the forecasts are highly reliable and can be used as the primary basis for decision-making. It is recommended to adopt a risk-neutral strategy, and under the premise of ensuring flood control safety, the operating water level can be appropriately raised to increase power generation efficiency.

[0093] Furthermore, for Level II forecasts, the forecasts are usable but there are risks. It is recommended to use multi-source forecast fusion and reserve an additional 10% flood control capacity as a safety margin.

[0094] Furthermore, Level III forecasts are unreliable and should be downgraded for use, serving only as a trend reference. Dispatch decisions should revert to conventional rules based on actual hydrological conditions, initiating short-term rolling correction plans and increasing monitoring frequency.

[0095] Optionally, when generating scheduling strategies, not only can qualitative suggestions be provided, but a joint optimization model can also be directly solved. This model aims to maximize overall benefits.

[0096] Max Object = w1 × E _gen +w2×V _storage ;

[0097] In the formula, Max Object corresponds to the target value to be maximized, and E _gen The total power generation benefit during the corresponding dispatch period, V _storage The reservoir water storage at the end of the corresponding scheduling period (or key node) represents the water resource reserve benefit, and w1 and w2 represent the corresponding weight coefficients.

[0098] Meanwhile, embedding risk budgeting as a hard constraint into the model can be described as follows:

[0099] Risk _total ≤Budget _limit ;

[0100] Among them, Risk _total The total risk (dimensionless, or in units corresponding to the risk type) during the corresponding scheduling period is a comprehensive quantitative value of various scheduling risks. _limit Corresponding to the preset risk budget threshold (and Risk) _total (Unit consistency) is the upper limit of hard constraints that the model cannot break, which can be determined by engineering design standards, watershed management requirements or historical risk assessment results.

[0101] In particular, this model allows for the allocation of risk budgets among cascade reservoirs. For example, if an upstream reservoir has significant regulation potential, the optimization algorithm can automatically allocate more risk quotas to the upstream, allowing for higher upstream water levels and reducing pressure on the downstream. By solving this optimization problem, the optimal water level control trajectory that satisfies the risk budget constraints is directly output as the scheduling strategy execution plan. This achieves a closed loop from assessment to automatic decision-making.

[0102] Example 6 describes the specific implementation process of data cleaning, spatiotemporal matching, and multi-scale statistical evaluation. In particular, it details the basic data processing flow for constructing the forecast assessment dataset and calculating statistical biases. This ensures a robust data foundation and accurate statistical calculations.

[0103] Step 601: Collect and organize the raw data of medium- and long-term water and rainfall forecasts and measured hydrological and meteorological data, and carry out quality control and outlier processing.

[0104] Specifically, the data sources processed in this step include precipitation forecast gridded products, flow forecast sequences, and corresponding measured data from rain gauges and meteorological stations. Before the data is entered into the database, the system performs consistency decoding to uniformly map data formats from different sources, such as GRIB, NetCDF, and ASCII, to the internal standard format.

[0105] Next, quality control procedures are implemented. For measured data, outlier detection is performed using the 3-sigma principle or comparison with neighboring stations. For example, if a rain gauge station's daily rainfall exceeds three times the average of neighboring stations and surpasses historical extremes, it is marked as suspicious and corrected using inverse distance weighted interpolation data from surrounding stations. For forecast data, the integrity of its time labels is carefully checked, ensuring that each forecast file includes a clear forecast release time and forecast validity period. Records with missing metadata are directly discarded.

[0106] Furthermore, the discrete station-measured data are converted into areal rainfall data with the same spatial resolution as the forecast grid points by using the Kriging interpolation method, thus completing the initial alignment of spatial dimensions.

[0107] Step 602: Establish the spatiotemporal matching relationship between forecasts and actual measurements to form a forecast-actual measurement matching dataset for error analysis.

[0108] This step requires addressing the misalignment between forecast and observed data on the timeline. The system follows the logical formula: Target Verification Time = Forecast Release Time + Forecast Lead Time for matching. Specifically, for each forecast record, the system reads its forecast release time and lead time (e.g., day 1 or day 10 in the future) and calculates the corresponding actual target date T. _target Then, the system retrieves a timestamp equal to T from the measured database. _target The forecast and observed records are only paired and stored in the matching dataset if they match in spatial cell (e.g., the same sub-basin ID) and time step (e.g., both are daily averages). If a forecast lacks corresponding observed data (e.g., observed data is missing), the forecast record is marked as invalid or deleted from the evaluation set, and the deletion percentage is recorded for data quality analysis. This process ensures that each subsequent pair (P) is calculated... _i O _i Both correspond to each other in a physical sense.

[0109] Step 603: Construct a hierarchical structure of time and spatial scales, and calculate traditional statistical precision indicators.

[0110] Specifically, in order to meet the needs of different scheduling levels, this embodiment performs multi-scale aggregation on the matching dataset. In the time dimension, the system not only retains the daily scale, but also aggregates the data into a ten-day scale (sum of 10 days), a monthly scale (sum of 30 days), and a scale for the entire forecast period.

[0111] Spatially, the system aggregates from single-grid scales to sub-basin scales, and then to the scale of cascade reservoir-controlled watersheds. At each scale level, the system calculates statistical indicators. In addition to the aforementioned mean squared error, this embodiment calculates correlation coefficients in detail to measure the consistency of forecast trends.

[0112] Furthermore, the correlation coefficient can be calculated as follows: the numerator is the sum of the products of forecast bias and observed bias, and the denominator is the product of the square root of the sum of squares of forecast bias and the square root of the sum of squares of observed bias. Here, bias refers to the value at a certain moment minus the average value of the sequence. If the correlation coefficient is close to 1, it indicates that the predicted fluctuation trend is highly consistent with the observed trend; if the correlation coefficient is close to 0, it indicates that the two are unrelated. Furthermore, the mean error ME is also calculated, which is the average of the forecast value minus the observed value, used to reflect whether the forecast has a systematic tendency to be overestimated or underestimated.

[0113] Based on this, a hierarchical indicator system can be constructed, allowing dispatchers to see which time scale (e.g., ten-day forecasts are more accurate than daily forecasts) or which spatial region (e.g., upstream is more accurate than downstream) the forecasts perform better, enabling them to adopt the forecasts more selectively.

[0114] According to one aspect of this application, some methods of the present invention may also be:

[0115] In flood control-dominated scenarios, for key reservoirs, a water balance equation is used to link inflow, outflow, and water level changes. The formula can be described as follows:

[0116] V(t+1)=V(t)+(Q _in (t)-Q _out (t))×Δt;

[0117] Where V(t) is the storage capacity at time t, V(t+1) is the storage capacity at the next time step, and Q... _in (t) represents the inflow rate, Q _out (t) represents the outbound flow rate, and Δt represents the time step.

[0118] Based on this, the water level is obtained according to the reservoir capacity-water level curve. Using simple hydraulic or empirical relationships, the outflow and upstream inflow are transmitted through the river channel to the downstream control section water level.

[0119] In the scenario of water storage and supply at the end of the flood season, the focus is on describing how the cumulative inflow and outflow affect the water level at the end of the period, and how the flow rate during the water supply period affects the downstream water supply security indicators; the basic reservoir capacity-water level-outflow relationship is retained, but the river transmission part can be simplified.

[0120] In the context of power generation and ecology, a simplified relationship between power generation output and water head and flow can be introduced; the impact of ecological discharge flow constraints on the degree to which downstream ecological water demand is met can also be introduced.

[0121] Furthermore, key relationships in each scenario are stored in a parameterized form, including equation form, coefficients, and constraint boundaries. The output is callable scheduling response model parameter data. It can be integrated with a risk budget framework for multi-scenario simulation.

[0122] Optionally, risk indicators can be defined for the main flood control scenarios, such as the probability that the water level at the control section exceeds the warning level, the probability that the water level exceeds the guaranteed level, and the probability that the duration of the warning exceeds the threshold. Such indicators can be formalized as the probability of an event occurring.

[0123] Optionally, risk indicators can be defined for the flood season water storage scenario, such as the probability that the reservoir water level will be lower than the target water level at the end of the flood season; events where the reservoir water level does not meet the target can be used as risk events.

[0124] Optionally, risk indicators can be defined for water supply security scenarios, such as the probability that the water shortage exceeds a specified threshold, or the probability that the downstream flow is lower than the ecological flow for several consecutive days.

[0125] Based on this, a calculation method is defined for each risk indicator, such as calculating event frequency from multi-scenario simulation results. The output includes risk indicator definition data containing the names and calculation definitions of various scenario risk indicators. This risk indicator definition data can be used to set budget limits and calculate risk results.

[0126] Furthermore, for each risk indicator, a benchmark level is determined with reference to national or industry standards, such as the probability of flood exceeding the warning level not exceeding a certain threshold.

[0127] Furthermore, based on the risk tolerance allowed by the management, the benchmark level is adjusted to form an acceptable upper limit for the risk budget; the risk indicator names are paired with their corresponding upper limits to form a structured record. The output contains the upper limit data for the risk budget for each risk indicator. This data can be incorporated into the assessment model or used as a comparison basis when calculating applicability.

[0128] Based on this, the scheduling response equation for each typical scheduling scenario is bound to the corresponding risk indicator, so that during simulation, each simulated path can simultaneously output state variables such as water level, flow rate, and water supply status, as well as the corresponding risk indicator values. A budget upper limit is attached to each risk indicator, forming a comparison structure between the current simulation result risk value and the budget upper limit. The above information is summarized into a unified data structure, namely the flood risk budget framework data, which clearly defines the calculable variables, risk indicator formulas, and budget thresholds for each scenario, and outputs the flood risk budget framework data.

[0129] According to another aspect of this application, some methods of the present invention may also be:

[0130] Accordingly, for each control section, a set of reservoirs that may be affected upstream is listed. The correlation and time lag characteristics between the outflow from each reservoir and the flow at the control section are analyzed; the correlation coefficients for typical flood processes and non-flood processes are calculated; and the main influencing time lags are determined.

[0131] Furthermore, based on correlation strength and reservoir capacity, reservoirs are screened to select key reservoirs that have a significant regulatory effect on a specific control section. The correspondence between control sections and the set of key reservoirs is recorded. The key reservoir set data organized by control section is output. This data can be used to construct equivalent reservoir inflow relationships.

[0132] Based on this, select a suitable function structure, for example:

[0133] Linear superposition form, such as the equivalent inflow being the weighted sum of the outflows from each key reservoir; or simplified response function form, such as mapping future water error to outflow changes and then to cross-sectional flow changes.

[0134] Furthermore, we preliminarily define the function form, for example:

[0135] Qequ(t)=∑ k∈K a _k ×Q _k_out (t-τ _k )]+b×Q _nat (t)

[0136] Among them, Q _equ (t) represents the equivalent inflow at the key cross-section, K represents the set of key reservoirs, and Q _k_out Let a be the outflow from the k-th reservoir. _k τ is the weighting coefficient. _k For time delay, Q _nat denoted as , where is the natural inflow rate and 'b' is a coefficient.

[0137] Based on this, the selected function structure and initial parameter form are recorded as equivalent database function form data. The output describes the equivalent database function form data describing the equivalent relational structure.

[0138] Furthermore, a sample set is constructed over several representative historical periods. The input samples are the outflow from key reservoirs and the natural inflow; the output samples are the measured flow at the corresponding time-lag control sections. The parameter 'a' is optimized using least squares or other parameter estimation methods. _k τ _k Step 1, b, to optimize the equivalent inflow function by fitting the control section flow rate. Verify that the fitting error is within an acceptable range, adjusting the function form or introducing nonlinear terms if necessary. Bind and store the optimized parameter values ​​with the corresponding control sections and reservoir sets. Output the calibrated equivalent inflow model parameter data, which can be used to convert forecasted water and rainfall conditions into an equivalent inflow process.

[0139] Furthermore, the equivalent inflow function and corresponding parameters for each control section are combined into a module interface. The input is the outflow from each key reservoir and the natural inflow; the output is the equivalent inflow at that control section. The module interface definition, parameters, and calling method are packaged into a callable data structure. This outputs the equivalent inflow impact module data.

[0140] According to another aspect of this application, a specific implementation method for calculating forecast applicability indicators based on multi-scenario simulation is provided.

[0141] Accordingly, for each forecast scheme and each forecast lead time, statistical indicators such as mean square error, bias, and correlation coefficient are extracted over a certain period. These statistical indicators are then transformed into error distribution parameters: for example, it is assumed that the location error follows a zero-mean normal distribution, with the variance given by the mean square error; for process-related indicators, such as peak deviation and time deviation, corresponding one-dimensional distributions can be constructed respectively. If sufficient samples are available, the error distribution can be directly constructed using empirical distributions or kernel density estimation methods, thus reducing the constraints of the normality assumption. The error distribution parameters corresponding to each forecast scheme under different lead times and scenarios are recorded. Error probability distribution data characterizing the error features are output, which can be used to generate perturbation forecast scenarios.

[0142] Based on this, using error probability distribution data and forecast evaluation scenario datasets, a large number of perturbation forecast sequences are generated for each forecast scheme, forming a perturbation forecast scenario dataset.

[0143] Specifically, a specific forecast scheme and forecast lead time are selected, and the corresponding original forecast sequence and error distribution parameters are read. The original forecast sequence represents the forecast flow process. A certain number of error sequences (e.g., hundreds or thousands) are sampled; for example, for each time step, an error value is randomly selected from the error distribution. For process-related features (such as peak flow error), the overall offset can be extracted, perturbing the original forecast in a process-based manner. The extracted errors are added to the original forecast sequence to obtain the perturbed forecast process.

[0144] Furthermore, precipitation and flow forecasts are processed separately to ensure that the post-perturbation forecasts do not violate physical principles (e.g., flow rates are not negative). The above perturbation process is repeated for forecast sequences of different scenarios and control sections to generate a set of perturbation scenarios containing forecast schemes, lead times, scenarios, and control section labels. This yields a dataset of perturbation forecast scenarios that can be used as simulation input.

[0145] Furthermore, for each disturbance scenario in the disturbance forecast scenario dataset, the equivalent inflow impact module data is used to convert it into the equivalent inflow flow process of the key section, thus obtaining the disturbance equivalent inflow process data.

[0146] Accordingly, for each disturbance forecast scenario, the corresponding precipitation and natural inflow forecast sequences are extracted. Based on the input requirements of the equivalent inflow module, a simplified description of the outflow behavior of key reservoirs is obtained. Historical average scheduling behavior or typical procedure simulations can be used to generate the expected outflow process. When errors affect upstream inflow, the outflow process will be reflected in the model. The outflow process of key reservoirs and the natural inflow process are input into the equivalent inflow function, and the equivalent inflow flow process of the control section is calculated using the equivalent inflow model parameter data. The above process is repeated for all disturbance scenarios and all relevant control sections. Disturbance equivalent inflow process data containing the equivalent inflow flow process of key control sections under each disturbance scenario is obtained. This data can be used to drive scheduling response and risk simulation.

[0147] For example, using data on the equivalent inflow process during disturbances and flood risk budget framework data, simplified scheduling simulations and risk calculations are performed for each disturbance scenario to generate scenario risk result data. For each disturbance scenario, its corresponding equivalent inflow process is read, and the corresponding scheduling response model is selected (based on scenario coding and control section coding).

[0148] Next, the reservoir water level, outflow, and downstream control section water level are calculated step-by-step using the scheduling response model. Based on these state variables, the values ​​of each risk indicator are calculated according to the formulas defined in the risk indicator data. For example:

[0149] Has the downstream water level exceeded the warning level?

[0150] Whether there have been any incidents where the water level at the end of the flood season did not meet the standard;

[0151] Has a water shortage occurred?

[0152] Based on this, the risk indicator values ​​for each scenario are compared with the risk budget ceiling data to indicate whether the scenario meets the various risk budget constraints. Scenario risk outcome data, recording the risk indicator values ​​and budget fulfillment status for each disturbance scenario, is obtained. This data can be used to statistically determine applicability probabilities.

[0153] Based on this, and according to the scenario risk results data, the proportion of disturbance scenarios that meet the risk budget requirements under each forecast scheme, each lead time, and each scenario is statistically analyzed to obtain forecast applicability index data. Statistical analysis is performed on all disturbance scenarios under the same forecast scheme, the same lead time, and the same scheduling scenario, as follows:

[0154] Count the number of scenarios in which all risk indicators do not exceed the budget limit;

[0155] Count the total number of scenarios.

[0156] Furthermore, calculate the proportion that meets the risk budget, for example, the formula for the proportion is:

[0157] P _ok =N _ok / N _total ;

[0158] Among them, P _ok To meet the proportion of the risk budget, N _ok N represents the number of scenarios where all risk indicators are within the budget. _total This represents the total number of scenarios.

[0159] Based on this, P _ok As the applicability probability of a forecast scheme under this scenario and lead time, multiple applicability indicators (such as flood control applicability probability, water storage applicability probability, etc.) can be calculated separately based on different risk indicators. All applicability probabilities and related statistical information are organized and stored according to forecast scheme, lead time, and scenario code. The output includes forecast applicability indicator data for different forecast schemes, different lead times, and different scheduling scenarios. This data can be used for comprehensive evaluation and application recommendation generation.

[0160] This application employs a multi-scenario simulation technique based on a flood risk budget framework to address the nonlinear disconnect between forecast accuracy and scheduling risk. Accordingly, a scheduling response model incorporating water balance equations is constructed, transforming the abstract forecast error probability distribution into specific downstream water level and flow processes, which are then compared with a preset risk budget upper limit (such as the allowable probability of exceeding warning levels). Meteorological statistical errors are nonlinearly mapped to scheduling risk indicators (applicability probability), achieving a leap from assessing forecast accuracy to assessing forecast usability, thus resolving the problem that purely statistical indicators cannot reflect scheduling safety risks.

[0161] To address the issue of assessment benchmarks for upstream cascade regulation interference, this implementation plan introduces an equivalent inflow impact module for the upstream reservoir group. Through parameterized linear superposition and time-delay functions, combined with the operational plans of key upstream reservoirs, the natural inflow forecast is dynamically corrected to an equivalent inflow that incorporates upstream regulation effects. This eliminates the interference of upstream human scheduling on downstream inflow forecast assessment, restores the true boundary conditions under cascade joint operation, and ensures the accuracy of assessment results in complex cascade environments.

[0162] The optional embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solution of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for constructing a reservoir group scheduling strategy based on medium- and long-term forecasts, characterized in that, include: Medium- and long-term water and rainfall forecast data are mapped to predefined typical scheduling scenarios and watershed control sections to generate a forecast evaluation dataset containing scenario labels and spatial attributes. Based on the forecast assessment dataset, the forecast accuracy evaluation results for different scenarios are calculated, and the statistical deviation and process characteristic deviation between the forecast values ​​and the measured hydrological and meteorological data are quantified. The forecast accuracy evaluation results are used to drive the applicability assessment model for scheduling applications. Multi-scenario scheduling simulation is performed under the flood risk budget framework to calculate the risk budget consumption caused by forecast errors in the reservoir group scheduling response. The proportion of simulation scenarios that meet the risk budget ceiling is statistically analyzed, scenario-based forecast applicability indicators are generated, and reservoir group scheduling strategies are constructed based on the forecast applicability indicators.

2. The method according to claim 1, characterized in that, A suitability evaluation model for scheduling applications includes: A flood risk budget framework is used to define the calculation logic of risk indicators and the corresponding risk budget ceiling in watershed scheduling scenarios. The scheduling response model is used to parametrically describe the relationship between reservoir water level and outflow response of a reservoir group under the constraints of input flow sequence and scheduling rules; The multi-scenario simulation module is configured to generate random disturbance inputs by combining forecast accuracy evaluation results, and drive the scheduling response model to output risk assessment results.

3. The method according to claim 2, characterized in that, Using forecast accuracy evaluation results to drive an applicability assessment model for scheduling applications, multi-scenario scheduling simulations are performed within a flood risk budget framework. The risk budget consumption resulting from forecast errors in reservoir group scheduling responses is calculated, including: We analyze the statistical bias and process characteristic bias in the forecast accuracy evaluation results and construct error probability distribution data that describes the probability characteristics of forecast errors. Based on the error probability distribution data, the forecast sequences in the forecast evaluation dataset are subjected to multiple random sampling perturbations to generate a perturbation forecast scenario set containing multiple sample sequences; The disturbance forecast scenario set is input into the scheduling response model for scenario-by-scenario evolution calculation, and the control section state variables corresponding to each sample sequence are output. Based on the risk indicator calculation logic defined in the flood risk budget framework, the state variables of the control section are transformed into scenario risk indicator values ​​for each disturbance scenario, and the scenario risk indicator values ​​are used as the risk budget consumption.

4. The method according to claim 3, characterized in that, The proportion of simulation scenarios that meet the upper limit of the risk budget is statistically analyzed, and scenario-based forecast applicability indicators are generated, including: Read the upper limit of the risk budget corresponding to the current scheduling scenario from the flood risk budget framework; The scenario risk index value of each disturbance scenario is compared with the risk budget limit one by one. Disturbance scenarios in which all risk index values ​​do not exceed the risk budget limit are selected, and the number of scenarios that meet the budget is counted. The proportion of scenarios that meet the budget to the total number of samples in the perturbation forecast scenario set is calculated, and this proportion is used as an indicator of the applicability of scenario-based forecasts.

5. The method according to claim 2, characterized in that, The flood risk budget framework defines the calculation logic for risk indicators and the upper limit of the risk budget, including: For flood control-dominated scenarios, the risk indicator is defined as the probability that the water level at the downstream control section exceeds the warning level, and the maximum allowable probability of exceeding the warning level is set as the upper limit of the risk budget. For the scenario of water storage at the end of the flood season, the risk indicator is defined as the probability that the water level of the reservoir at the end of the flood season is lower than the target water storage level, and the maximum allowable probability of insufficient water storage is set as the upper limit of the risk budget. The risk indicator calculation logic is used to guide the multi-scenario simulation module to count the frequency of events that trigger the above-defined events when calculating risk budget consumption.

6. The method according to claim 3, characterized in that, The disturbance forecast scenario set is input into the scheduling response model for scenario-by-scenario evolution calculation, including: The equivalent inflow impact module of the upstream reservoir group is invoked. The module is configured with the equivalent mapping relationship between the outflow of key upstream reservoirs and the natural inflow of downstream control sections. Using this impact module, each sample sequence in the perturbation forecast scenario set is converted into a perturbation equivalent storage process that includes upstream adjustment effects; The disturbance equivalent data entry process is used as the actual input data and input into the scheduling response model for scenario-by-scenario evolution calculation.

7. The method according to claim 6, characterized in that, The upstream reservoir group equivalent inflow impact module uses a linear superposition function including time delay parameters to calculate the disturbance equivalent inflow process. Its calculation logic is as follows: For each time step, the outflow from each reservoir in the upstream key reservoir set is time-delayed shifted and weighted summed, and then superimposed with the weighted natural inflow forecast data to obtain the equivalent inflow at the control section.

8. The method according to claim 6, characterized in that, The construction of the upstream reservoir group equivalent inflow impact module includes identifying the set of key upstream reservoirs involved in the calculation. This identification specifically involves: Obtain historical operation records of each reservoir in the upstream basin of the target control section and historical measured flow sequences of the control section; Calculate the cross-correlation coefficient and lag response time between the outflow from each reservoir and the historical measured flow sequence at the control section; Based on the significance of cross-correlation coefficients and the scale of reservoir regulation capacity, reservoirs with significant impact on the flow at the control section are selected from the reservoirs in the upstream basin to form a set of key upstream reservoirs.

9. The method according to claim 7, characterized in that, The propagation delay, influence weight, and natural water inflow correction coefficient in the linear superposition function are determined using a parameter calibration method, specifically: Call upon a training sample set containing historical data on outflow from key upstream reservoirs, natural inflow, and measured flow at control sections; Using the training sample set as input, and with the goal of minimizing the sum of squared errors between the calculated value of the linear superposition function and the measured flow at the control section, a parameter estimation algorithm is used to jointly optimize and calibrate the propagation time delay, influence weight, and natural water inflow correction coefficient to obtain the fixed model parameters.

10. The method according to claim 1, characterized in that, Medium- and long-term hydrological and rainfall forecast data are mapped to predefined typical scheduling scenarios and watershed control sections to generate a forecast evaluation dataset containing scenario labels and spatial attributes, including: Obtain the applicable time window and associated watershed spatial range for predefined typical scheduling scenarios; Read the forecast release time, forecast period, and forecast grid coordinate information of medium- and long-term water and rainfall forecast data; Determine whether the forecast period falls within the applicable time window, determine whether the forecast grid coordinates are within the watershed spatial range, and establish a mapping relationship between forecast records and typical scheduling scenarios; Each forecast record that conforms to the mapping relationship is attached with a corresponding scene label to generate a forecast evaluation dataset.

Images