Methods for Flood Control Operation of Multi-hole Reservoirs under Multiple Rounds of Heavy Rainfall
Patent Information
- Application Number
- CN202610667187.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-15
- Publication Date
- 2026-08-04
AI Technical Summary
提前泄洪腾出库容,则容易带来发电损失或蓄水不足
[0022]通过采用上述的技术方案,本发明的有益效果是:本发明将梅雨期内每一轮有效强降水过程视为一次对水库防洪系统的随机“冲击”,并将水库防洪调度过程刻画为由初始运行阶段与若干分级泄洪阶段构成的多阶段动态决策过程,考虑分级泄洪本身带来的下游撤离、行洪扰动及发电损失等综合代价, 构建风险—代价权衡下的最优调度。
Smart Images

Figure CN122509591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of reservoir flood control, and in particular to a method for flood control scheduling of multi-hole reservoirs under multiple rounds of heavy rainfall. Background Technology
[0002] A large flood control reservoir equipped with multi-stage flood discharge capacity faces its core flood risk concentrated in the continuous flood events triggered by multiple rounds of heavy rainfall during the Meiyu season. Heavy rainfall during the Meiyu season exhibits significant clustering and self-excitation characteristics, meaning that after one round of heavy rainfall, the probability of another effective round of heavy rainfall occurring within the basin in a short period increases significantly. The cumulative effect of flood peaks caused by multiple rounds of rainfall poses a risk to the reservoir's operation.
[0003] While tiered flood discharge can mitigate risks, improper tiered discharge can lead to operational costs such as downstream evacuation and flood disturbance, with these costs increasing non-linearly with the number of orifices. Premature flood discharge to free up reservoir capacity can easily result in power generation losses or insufficient water storage. Therefore, optimizing the flood control scheduling of multi-orifice reservoirs during periods of multiple rounds of heavy rainfall in the plum rain season has become a pressing need. Summary of the Invention
[0004] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for flood control scheduling of multi-hole reservoirs under multiple rounds of heavy rainfall that takes into account the comprehensive costs of downstream evacuation, flood disturbance and power generation loss caused by the tiered flood discharge itself, and constructs an optimal scheduling method under risk-cost trade-off.
[0005] To achieve the above objectives, the technical solution of the present invention is: a method for flood control scheduling of multi-hole reservoirs under multiple rounds of heavy rainfall, comprising the following steps: S1: Perform independent simulations for each m from 1 to M (representing the number of simulations). The generation, dynamic scheduling, and total cost calculation of a single sample path are performed as follows: S1.1 is based on a labeled Hawkes process, during the study period Endogenous heavy precipitation arrival time sequence Generate a corresponding label vector for each round of precipitation. ; It represents the number of rounds of heavy rainfall generated; For the effective arrival time of heavy precipitation processes in the watershed, it must meet the following requirements.
[0006] For the research period Total number of effective heavy precipitation events occurring within the region; The label vector is a label vector defined to fully characterize the runoff generation and confluence features of each round of precipitation.
[0007] As an indicator of precipitation intensity, For the duration of precipitation, As for the spatial coverage characteristics of precipitation, This is an indicator of the watershed's moisture state before precipitation occurs; S1.2: For each precipitation event generated, the inflow at each discrete time point is calculated based on the Gamma response kernel, and the discrete inflow process over the entire time period is obtained by superimposing the baseflow. ; The first Rainfall time The contribution of inbound traffic is recorded as follows:
[0008] The effective runoff coefficient is determined by precipitation intensity, spatial coverage, and the pre-wet state of the watershed. For the response kernel parameter vector; For characteristic functions, only if Later than the time of precipitation arrival Only then will this round of precipitation contribute to the inflow into the reservoir; The Gamma response kernel is in the form of
[0009] Among them, scale parameter satisfy
[0010] In the formula, The shape parameter of the Gamma distribution. and The parameters to be calibrated; Inbound flow:
[0011] S1.3 Calculates the real-time failure probability based on the current precipitation history, water storage status, and orifice status:
[0012] According to the threshold group The current probability of non-failure Mapped to target aperture level ; like ,but ; Update the aperture state according to the threshold-driven state transition rule. ; Calculate the total discharge flow rate based on the current orifice status. ; The total outflow from the reservoir can be expressed as:
[0013] Reservoir Time The water storage capacity; Opening status ; This represents L progressively increasing flood discharge stages; The discharge capacity of the flood discharge gate is determined by the current reservoir water storage capacity and the gate opening level. This refers to the discharge flow rate of the generator set; According to the principle of water balance, the dynamic evolution of reservoir storage satisfies:
[0014] Update the water storage volume for the next moment. ; Accumulate the running cost of this time step ;like Then let fail = true; S1.4 Calculate the total cost of the sample path
[0015] S1.5 Update Cumulative Amount: sum_total = sum_total + ; sum_run = sum_run + ; sum_fail = sum_fail + ; fail_count = fail_count +
[0016] S2 calculation index: Expected total cost estimate ; Average operating cost
[0017] Average Risk Cost
[0018] Scheduling failure probability
[0019] S3 Based on the optimal threshold group that minimizes the expected total cost Flood control scheduling for multi-hole reservoirs.
[0020] Preferably, the following steps are also included: S4 calculates the estimated probability of conditional non-failure at the current time. ,according to Adjust the opening level of the reservoir; Estimated probability of condition not failing at the current time The calculation is as follows: S4.1 Initialize the success counter success_count=0; S4.2 for m from 1 to Perform independent simulations one after another; S4.3 Given historical rainfall information Under these conditions, based on the labeled Hawkes process to generate Rainfall event sequence within; S4.4 superimposes the residual confluence contribution of historical rainfall events to future periods with the confluence contribution of newly generated future rainfall events to obtain the inflow path. ; S4.5 sets the initial value of the test water storage. ; S4.6 in Maintain the current aperture level. Unchanged, gradually recursively ; S4.7 If at any future time the condition is met If the simulation fails, the inner simulation is recorded as a failure; otherwise, it is recorded as a success. S4.8 If this simulation is successful, then success_count = success_count + 1 S4.9 Conditional non-failure probability estimate .
[0021] Preferably, the following steps are also included: Constructing a framework for solving optimal threshold sets based on a nested approach of intelligent optimization algorithms and Monte Carlo simulation: The first step is to optimize the algorithm initialization by setting the algorithm parameters, including population size, maximum number of iterations, feasible region of the threshold set, threshold monotonicity constraint, and convergence criterion. The Particle Swarm Optimization (PSO) algorithm is used, with each particle corresponding to a set of feasible thresholds. ; The second step is to evaluate the objective function value; this involves setting thresholds for each particle. Call step S1 to calculate its expected total cost estimate. And use this estimate as the fitness function value of the particle; The third step is population iterative update; according to the optimization algorithm update rules, the particle position and velocity are iteratively adjusted to generate a new generation of threshold group population, and Monte Carlo simulation is continuously called to estimate its total cost; The fourth step is convergence judgment; if the maximum number of iterations is reached, or the change in the global optimal fitness value satisfies the preset convergence criterion, the iteration is terminated; otherwise, the objective function evaluation and population update process continues. Fifth step: Output the optimal solution; after the iteration, output the threshold set corresponding to the particle with the lowest estimated total expected cost in the population, which is the optimal scheduling threshold set. .
[0022] By adopting the above technical solution, the beneficial effects of the present invention are as follows: the present invention regards each round of effective heavy rainfall during the plum rain season as a random "impact" on the reservoir flood control system, and portrays the reservoir flood control scheduling process as a multi-stage dynamic decision-making process consisting of an initial operation stage and several tiered flood discharge stages. Considering the comprehensive costs of downstream evacuation, flood disturbance and power generation loss caused by the tiered flood discharge itself, the optimal scheduling under the risk-cost trade-off is constructed.
[0023] The dynamic aperture path formed by the optimized threshold scheme has the following distinct characteristics: First, when there are short intervals between consecutive heavy rainfalls, the inflow into the reservoir is rising rapidly, and the probability of a strong impact in the future is significantly increasing, the system tends to raise the opening level in advance, actively release some reservoir capacity, and quickly lower the water level to cope with the pressure of possible subsequent flood peaks.
[0024] Second, during periods of rainfall respite, when inflow to the reservoir decreases, or when the probability of future failure increases significantly, the system will appropriately reduce the opening level or even close the flood discharge outlets to avoid the high operating costs caused by maintaining a high flood discharge state for a long time, thus achieving a balance between cost and risk.
[0025] Third, when the risk of subsequent heavy rainfall rises again and the inflow increases again, the system can quickly upgrade the opening level again, demonstrating its dynamic adaptability to the rainfall time sequence structure, inflow process and risk evolution.
[0026] Fourth, compared with empirical rules, the optimized scheme is more sensitive to the "risk of multiple future shocks," therefore its orifice path is a forward-looking risk-response scheduling path. The essence of the optimized path is to minimize the duration of unnecessary large-flow flood discharge while ensuring that the probability of triggering high failure penalties is low. Detailed Implementation
[0027] The present invention will be further illustrated below with reference to specific embodiments.
[0028] The present invention provides a method for flood control scheduling of multi-hole reservoirs under multiple rounds of heavy rainfall, comprising the following steps: S1: Perform independent simulations for m from 1 to M. The generation, dynamic scheduling, and total cost calculation of a single sample path are performed as follows: S1.1 is based on a labeled Hawkes process, during the study period Endogenous heavy precipitation arrival time sequence Generate a corresponding label vector for each round of precipitation. ; For the effective arrival time of heavy precipitation processes in the watershed, it must meet the following requirements.
[0029] For the research period Total number of effective heavy precipitation events occurring within the region; The label vector is a label vector defined to fully characterize the runoff generation and confluence features of each round of precipitation.
[0030] As an indicator of precipitation intensity, For the duration of precipitation, As for the spatial coverage characteristics of precipitation, This is an indicator of the watershed's moisture state before precipitation occurs; S1.2: For each precipitation event generated, the inflow at each discrete time point is calculated based on the Gamma response kernel, and the discrete inflow process over the entire time period is obtained by superimposing the baseflow. ; The first Rainfall time The contribution of inbound traffic is recorded as follows:
[0031] The effective runoff coefficient is determined by precipitation intensity, spatial coverage, and the pre-wet state of the watershed. For the response kernel parameter vector; For characteristic functions, only if Later than the time of precipitation arrival Only then will this round of precipitation contribute to the inflow into the reservoir; The Gamma response kernel is in the form of
[0032] Among them, scale parameter satisfy
[0033] In the formula, The shape parameter of the Gamma distribution. and The parameters to be calibrated; Inbound flow:
[0034] S1.3 Calculates the real-time failure probability based on the current precipitation history, water storage status, and orifice status:
[0035] According to the threshold group The current probability of non-failure Mapped to target aperture level ; like ,but ; Update the aperture state according to the threshold-driven state transition rule. ; Calculate the total discharge flow rate based on the current orifice status. ; The total outflow from the reservoir can be expressed as:
[0036] Reservoir Time The water storage capacity; Opening status ; This represents L progressively increasing flood discharge stages; The discharge capacity of the flood discharge gate is determined by the current reservoir water storage capacity and the gate opening level. This refers to the discharge flow rate of the generator set; According to the principle of water balance, the dynamic evolution of reservoir storage satisfies:
[0037] Update the water storage volume for the next moment. ; Accumulate the running cost of this time step ;like Then let fail = true; S1.4 Calculate the total cost of the sample path
[0038] S1.5 Update Cumulative Amount: sum_total = sum_total + ; sum_run = sum_run + ; sum_fail = sum_fail + ; fail_count = fail_count +
[0039] S2 calculation index: Expected total cost estimate ; Average operating cost
[0040] Average Risk Cost
[0041] Scheduling failure probability
[0042] S3 Based on the optimal threshold group that minimizes the expected total cost Flood control scheduling for multi-hole reservoirs.
[0043] S4 calculates the estimated probability of conditional non-failure at the current time. ,according to Adjust the opening level of the reservoir.
[0044] Estimated probability of condition not failing at the current time The calculation is as follows: S4.1 Initialize the success counter success_count=0; S4.2 for m from 1 to Perform independent simulations one after another; S4.3 Given historical rainfall information Under these conditions, based on the labeled Hawkes process to generate Rainfall event sequence within; S4.4 superimposes the residual confluence contribution of historical rainfall events to future periods with the confluence contribution of newly generated future rainfall events to obtain the inflow path. ; S4.5 sets the initial value of the test water storage. ; S4.6 in Maintain the current aperture level. Unchanged, gradually recursively ; S4.7 If at any future time the condition is met If the simulation fails, the inner simulation is recorded as a failure; otherwise, it is recorded as a success. S4.8 If this simulation is successful, then success_count = success_count + 1 S4.9 Conditional non-failure probability estimate .
[0045] Constructing a framework for solving optimal threshold sets based on a nested approach of intelligent optimization algorithms and Monte Carlo simulation: The first step is to optimize the algorithm initialization by setting the algorithm parameters, including population size, maximum number of iterations, feasible region of the threshold set, threshold monotonicity constraint, and convergence criterion. The Particle Swarm Optimization (PSO) algorithm is used, with each particle corresponding to a set of feasible thresholds. ; The second step is to evaluate the objective function value; this involves setting thresholds for each particle. Call step S1 to calculate its expected total cost estimate. And use this estimate as the fitness function value of the particle; The third step is population iterative update; according to the optimization algorithm update rules, the particle position and velocity are iteratively adjusted to generate a new generation of threshold group population, and Monte Carlo simulation is continuously called to estimate its total cost; The fourth step is convergence judgment; if the maximum number of iterations is reached, or the change in the global optimal fitness value satisfies the preset convergence criterion, the iteration is terminated; otherwise, the objective function evaluation and population update process continues. Fifth step: Output the optimal solution; after the iteration, output the threshold set corresponding to the particle with the lowest estimated total expected cost in the population, which is the optimal scheduling threshold set. .
[0046] Let the set of reservoir flood discharge operation states be .
[0047] in, This indicates a normal operating state where the flood discharge outlets are closed and water is discharged only through the generator units. This represents L progressively increasing flood discharge stages. For discretized orifice level labels, the actual discharge capacity differences between different levels are characterized by the corresponding discharge function.
[0048] Under given scheduling rules, the reservoir during the study period The internal aperture state path evolves dynamically over time, denoted as
[0049] in, This refers to the scheduling threshold groups corresponding to each flood discharge level. (Opening path) In the threshold group Given a dynamic random path, it is jointly induced by the rainfall impact process, reservoir state evolution, and threshold-driven switching rules.
[0050] Modeling the arrival process of multiple rounds of rainfall impacts: Heavy rainfall during the Meiyu season exhibits significant clustering and self-excitation characteristics, meaning that the probability of another round of effective heavy rainfall occurring within the basin shortly after the previous round of heavy rainfall is significantly increased. To describe this historical dependence, a labeled Hawkes process is used to characterize the arrival process of effective heavy rainfall cycles.
[0051] Let the first The effective heavy precipitation process at time Reaching the watershed, satisfying
[0052] in For the research period The total number of effective heavy precipitation events occurring within the region. To fully characterize the runoff generation and confluence characteristics of each precipitation event, for the first... Defining the marker vector for round precipitation
[0053] in, As an indicator of precipitation intensity, For the duration of precipitation, As for the spatial coverage characteristics of precipitation, This is an indicator of the watershed's moisture state before precipitation occurs.
[0054] The conditional strength function of a labeled Hawkes process is expressed as follows:
[0055] in, For a moment All previous historical precipitation information; The baseline arrival intensity is used to characterize the background precipitation activity at different stages of the plum rain season. For the first The excitation coefficient determined by the precipitation marker vector reflects the excitation effect of that round of precipitation on subsequent precipitation events; The memory decay parameter characterizes the decay rate of the influence of historical precipitation events on the intensity of future precipitation. This function expresses that the probability of heavy precipitation arrival is affected not only by the seasonal background intensity but also by the persistent effect of historical precipitation events, and can effectively characterize the clustering characteristics of heavy precipitation processes during the plum rain season. For any given time... The conditional distribution of heavy precipitation events within the foreseeable future period can be derived from the above formula.
[0056] Modeling of the inflow flood peak process: Based on the given multiple rounds of precipitation arrival processes, it is necessary to further establish the mapping relationship between the precipitation process and the reservoir inflow process to characterize the superposition effect of flood peaks caused by multiple rounds of precipitation. Let time... The inflow of water into the reservoir is A Gamma response kernel with duration correction is used to characterize the watershed runoff response to a single round of precipitation. This form is consistent with the Nash instantaneous unit hydrograph, has clear parameter meanings and good resolvability, and can describe the hysteretic runoff process caused by a single round of heavy precipitation.
[0057] No. Rainfall time The contribution of inbound traffic is denoted as
[0058] in, The effective runoff coefficient is determined by precipitation intensity, spatial coverage, and the pre-wet state of the watershed. For the response kernel parameter vector; For characteristic functions, only if Later than the time of precipitation arrival Only then will the rainfall contribute to the inflow into the reservoir.
[0059] The Gamma response kernel used is in the form of
[0060] scale parameter satisfy
[0061] In the formula, The shape parameter of the Gamma distribution. and The parameters to be calibrated are as follows. The above response kernel form shows that the longer the precipitation duration, the longer the lag time of the confluence response and the smoother the peak flow process, thus reflecting the confluence differences of precipitation processes with different durations.
[0062] Based on the contribution of a single round of precipitation to the reservoir inflow, the total inflow during the study period can be expressed as a linear superposition of the contributions from multiple rounds of precipitation, i.e.
[0063] in This represents the base flow of the watershed, i.e., the background inflow component not formed by rainstorms. This formula can fully reflect the temporal superposition effect of multiple rounds of continuous rainfall during the Meiyu season, which trigger flood peak processes.
[0064] Reservoir state evolution and scheduling failure criteria: Set time The reservoir's water storage capacity is For any given opening state The total outflow from the reservoir can be expressed as:
[0065] in, The discharge capacity of the flood discharge gate is determined by the current reservoir water storage capacity and the gate opening level. This refers to the discharge flow rate of the generator set.
[0066] According to the principle of water balance, the dynamic evolution of reservoir storage satisfies
[0067] Set verification capacity This serves as a hard constraint boundary for flood control scheduling. If at any point within the study period, the following conditions are met...
[0068] If this occurs, the flood control operation is deemed a failure.
[0069] Let the first The total inflow of water caused by the round of heavy rainfall was , continuously experienced before The cumulative inflow of water into the reservoir after the round of heavy rainfall was
[0070] Correspondingly, in the threshold group Under the induced opening path, the front The cumulative discharge of the floodgates during the period of heavy rainfall is recorded as follows: Redefining continuous experience before The net increase in water storage after the round of heavy rainfall is
[0071] and with Indicates as of the date End of heavy rainfall The cumulative flow generated by background factors such as base current and power generation discharge is continuously experienced before The reservoir storage capacity after a round of heavy rainfall can be expressed as:
[0072] Therefore, it will experience continuously The probability that a reservoir will not experience operational failure after a round of heavy rainfall is defined as follows:
[0073] Tiered flood discharge thresholds and dynamic switching rules for each stage: The scheduling threshold corresponding to each aperture level is defined as the failure probability threshold.
[0074] and satisfy
[0075] For any time Based on the current precipitation events, reservoir water level, and open-gate operation, the conditional probability that the reservoir will not experience operational failure in the subsequent period can be assessed.
[0076] in, For a moment All previously known information.
[0077] Set time The current opening status of the system is as follows: If the real-time probability If the flow drops below the threshold corresponding to a higher orifice level, the system switches to a higher orifice level to enhance leakage; if If the threshold rises above the level corresponding to a lower orifice level, the system switches to the lower orifice level to reduce unnecessary flood discharge costs. Therefore, the dynamic update rule for orifice status can be written as follows:
[0078] in This is a threshold-driven state transition function.
[0079] As can be seen from the above formula, as long as a set of scheduling thresholds is given... The entire process of hole opening path during the research period It can then be automatically induced and generated by the random process of precipitation and the evolution of reservoir state.
[0080] Staged flood discharge operations incur operational costs such as downstream evacuation and flood discharge disturbance, and these costs increase non-linearly with the number of orifices. Let the number of orifices be... The cost of running per unit time is ,satisfy
[0081] During the research period Within, by threshold group The cumulative flood discharge operation cost resulting from the induced dynamic orifice path can be expressed as:
[0082] To characterize the power generation loss caused by low reservoir capacity, the cost of low reservoir capacity is defined as... . in This is the minimum storage capacity required by the power generation department to ensure supply.
[0083] To depict the excess storage capacity The resulting severe flood control risks are defined as the cost of scheduling failure risk as...
[0084] in This is the penalty coefficient for scheduling failure, and its magnitude is significantly higher than the cost of any normal flood discharge operation.
[0085] Therefore, given a set of thresholds The total scheduling cost is defined as
[0086] Accordingly, the core optimization objective is to find the optimal threshold set that minimizes the expected total cost among all feasible threshold sets. ,Right now
[0087] in The feasible region to satisfy the threshold monotonicity constraint.
[0088] This optimization objective penalizes both "excessive flood discharge" and "insufficient flood discharge"—two unreasonable strategies: the former incurs higher operational costs, while the latter is penalized for exceeding limits. However, this incurs extremely high risks of failure. Therefore, the optimal scheduling strategy reflects the best balance between flood control safety and operating costs.
[0089] For any given set of thresholds The corresponding dynamic scheduling process will induce the formation of random opening paths. Random inbound traffic process and random water storage trajectory Therefore, the total scheduling cost defined in Section 2...
[0090] Essentially, it is a composite random variable driven by multiple stochastic processes, and its mathematical expectation is...
[0091] objective function It is simultaneously affected by the following three random dimensions.
[0092] First, labeled Hawkes processes have significant historical dependencies. Both the arrival time of precipitation and the label vector are random variables, and previous precipitation events continue to affect the arrival intensity of future precipitation, resulting in different rounds of precipitation not being independent and identically distributed, making it difficult to process them through simple probabilistic convolution.
[0093] Second, the confluence response of a single round of precipitation is determined by a random labeled vector. The flood peaks of multiple rounds of precipitation overlap in time, eventually forming a non-stationary, strongly path-dependent random inflow process that cannot be characterized by a fixed distribution closed form.
[0094] Third, the switching of the orifice state depends on the real-time failure probability assessment, which is closely related to the current water storage state, historical precipitation processes, and future impact distribution. Therefore, the orifice path... It is a random dynamic path. The operational cost is the integral over this random path, while the failure risk cost depends on whether the entire water storage trajectory is crossed. Therefore, the objective function itself cannot be analytically expressed through simple phased expansion.
[0095] Because of the aforementioned multi-coupling randomness, the expected total cost is... It is difficult to obtain explicit closed-form solutions using traditional analytical methods. Therefore, the Monte Carlo numerical simulation method is employed. This involves generating a large number of independently repeated, fully coupled sample paths of "precipitation arrival—reservoir flood peak—dynamic scheduling—cost accumulation" to analyze the results. Perform numerical estimation.
[0096] To achieve numerical simulation, the continuous-time state evolution equation needs to be discretized, and a recursive scheme that can be iterated step by step needs to be constructed. Let the discrete time step be... The research period Discretized Each time step has an equal time interval, and the discrete time is denoted as .
[0097] in .
[0098] Based on the water balance equation, the discrete recursive form of water storage can be expressed as:
[0099]
[0100] in, Indicates time The water storage capacity For a moment The opening status, For a moment Discrete inbound flow rate.
[0101] For any discrete time Based on the currently known information, the conditional probability that the system will remain unfailed during the future research period can be assessed.
[0102] And update the opening status based on threshold rules:
[0103] Therefore, given a set of thresholds Then, the entire dynamic operation path can be generated step by step through a recursive method of "probability assessment - state switching - water storage update".
[0104] In the discrete-time frame, the cumulative running cost corresponding to the sample path can be expressed as:
[0105] superscript Indicates the first A Monte Carlo sample path.
[0106] Accordingly, the first The failure risk cost of each sample path is
[0107] Therefore, the first The total scheduling cost of the sample paths is
[0108] Algorithm 1: Monte Carlo estimation of expected total cost given a threshold set Algorithm input: Research period Time step Hawkes rainfall arrival process parameters (marked) Rainfall-labeled distribution parameters, Gamma confluence response kernel parameters Production flow mapping parameters base current Unit leak Check the storage capacity Initial water storage , discharge capacity function , running cost function Failure penalty coefficient Threshold group Number of outer layer simulations
[0109] Algorithm output: Estimated total expected cost Average operating cost, average risk cost, probability of failure Algorithm flow: 1: Initialization: sum_total sum_run sum_fail fail_count . 2: Perform independent simulations for m from 1 to M one by one. The generation, dynamic scheduling and total cost calculation of a single sample path are carried out in the following steps.
[0110] 2.1: Based on labeled Hawkes processes, during the research period Endogenous heavy precipitation arrival time sequence And generate a corresponding label vector for each round of precipitation.
[0111] 2.2: For each precipitation event, the inflow at each discrete time point is calculated based on the Gamma response kernel, and the discrete inflow process over the entire time period is obtained by superimposing the baseflow.
[0112] 2.3: Initialize system state and cost variables.
[0113] Set the initial system state And initialize the cumulative running cost. At the same time, set the failure flag variable fail=false.
[0114] 2.4: Dynamic scheduling and state updates for k, executed step-by-step from 0 to K-1. That is, for each discrete time step... If the system has not yet failed, perform the following operations in sequence: 2.4.1. Based on the current precipitation history, water storage status, and orifice status, Algorithm 2 is invoked to calculate the real-time failure probability.
[0115] 2.4.2. Based on threshold groups The current probability of non-failure Mapped to target aperture level : like ,but ;like ,but ;…;like ,but
[0116] 2.4.3. Update the aperture state according to the threshold-driven state transition rule. ; 2.4.4. Calculate the total discharge flow rate based on the current orifice status. ; 2.4.5. Update the water storage at the next time step based on the discrete water balance equation. ; 2.4.6. Accumulate the runtime cost of this time step.
[0117] 2.4.7. If If so, then let fail = true.
[0118] 2.5. Calculate the total cost of the sample path.
[0119] 2.6. Update cumulative total: sum_total = sum_total + sum_run = sum_run + sum_fail = sum_fail + fail_count = fail_count +
[0120] 3. Calculation indicator: Estimated expected total cost Average operating cost Average Risk Cost Scheduling failure probability
[0121] 4. Output Indicators
[0122] Algorithm 2: Conditional Monte Carlo estimation of real-time failure probability Algorithm input: Current time step Current water storage Current aperture level As of Historical rainfall event information Forecast period length Inner layer simulation times
[0123] Algorithm output: Estimated probability of the condition not failing at the current time.
[0124] Algorithm flow: 1. Initialize the success counter success_count = 0 2: For m from 1 to Perform independent simulations one at a time.
[0125] 2.1. Given historical rainfall information Under these conditions, based on the labeled Hawkes process to generate Rainfall event sequence within the region.
[0126] 2.2. The inflow path is obtained by superimposing the residual confluence contribution of historical rainfall events to future periods with the confluence contribution of newly generated future rainfall events.
[0127] 2.3. Set the initial value of the test water storage volume.
[0128] 2.4. In Maintain the current aperture level. Unchanged, gradually recursively
[0129] 2.5. If at any future time the following condition is met... If the simulation fails, the inner simulation is recorded as a failure; otherwise, it is recorded as a success.
[0130] 2.6. If the simulation is successful, then success_count = success_count + 1 3. Output the estimated probability of conditional non-failure. . Framework for solving the optimal threshold set The optimal scheduling problem is formulated as follows:
[0131] Due to the objective function Since this problem lacks an explicit analytical expression and can only be estimated numerically through the aforementioned Monte Carlo simulation, it is essentially a typical black-box stochastic optimization problem. Therefore, a solution framework for the optimal threshold set based on a nested approach of intelligent optimization algorithms and Monte Carlo simulations is constructed, and its basic process is as follows.
[0132] The first step is to optimize the algorithm initialization. This involves setting the algorithm parameters, including population size, maximum number of iterations, feasible region of the threshold set, threshold monotonicity constraints, and convergence criteria. Particle Swarm Optimization (PSO) is used, with each particle corresponding to a set of feasible thresholds. .
[0133] The second step is to evaluate the objective function value. This involves setting thresholds for each particle. Call Algorithm 1 to calculate its expected total cost estimate. The estimated value is then used as the fitness function value for the particle.
[0134] The third step is population iterative update. According to the optimization algorithm update rules, the particle positions and velocities are iteratively adjusted to generate a new generation of threshold group population, and Monte Carlo simulations are continuously called to estimate its total cost.
[0135] The fourth step is convergence testing. If the maximum number of iterations is reached, or the change in the global optimal fitness value satisfies the preset convergence criterion, the iteration is terminated; otherwise, the objective function evaluation and population update process continues.
[0136] Fifth step: Output the optimal solution. After iteration, output the threshold set corresponding to the particle with the lowest estimated total expected cost in the population; this is the optimal scheduling threshold set. .
[0137] Case Analysis: The Xin'anjiang Reservoir, located on the upper reaches of the Qiantang River, is a key control project for flood control, power generation, and comprehensive water resource utilization in the basin. During the plum rain season, its basin is simultaneously affected by continuous frontal rainfall and periods of severe convective weather, often exhibiting typical characteristics such as multiple rounds of heavy rainfall, significant superposition of inflow peaks, and a clear temporal correlation in flood evolution. Compared to single-peak floods, plum rain season floods are more likely to create sustained high inflow pressure under the cumulative effect of multiple rounds of rainfall, making it a typical application scenario for the invented "multi-round random impact—dynamic graded flood discharge" model.
[0138] Based on actual engineering operations, the reservoir's flood discharge operation status is discretized into 6 levels:
[0139] in, This indicates a normal operating state where the flood discharge outlets are closed and water is only discharged through the generating units. This represents five progressively increasing stages of flood discharge. (Study period) A complete Meiyu season flood event was selected to cover multiple rounds of heavy rainfall impacts and the resulting full-process dispatch and response.
[0140] Parameter settings: Based on decades of hydrological and flood discharge scheduling data from the Xin'anjiang Reservoir during the plum rain season, this section performs data fitting on specific functional forms and parameter values related to rainfall arrival process, inflow flood peak response, reservoir operation, and cost function. The results are as follows.
[0141] 1) Hawkes Rainfall Arrival Process Parameters Study period This corresponds to a flood process that lasts approximately 10 days during the plum rain season.
[0142] Benchmark strength The following piecewise constant form can be used to characterize the multi-round fluctuation process under the alternating control of frontal oscillation and convective activity.
[0143]
[0144] Excitation coefficient function Take linear form The parameter value is This parameter setting indicates that the greater the rainfall intensity, the longer the duration, and the more humid the preceding period, the stronger its triggering effect on subsequent rainfall in the short term; simultaneously, through max{ ,0} ensures that the excitation intensity is non-negative.
[0145] Memory decay parameters according to The decay law, when By then, the excitation effect has significantly weakened, thus it can better simulate the characteristics of the plum rain season: "continuous triggering within an active cluster, and separation between different clusters."
[0146] The labeled variables for each round of effective heavy rainfall follow the following distribution: DiscreteUniform in, The interval corresponds to the intensity of moderate to heavy rainfall events. The range can cover everything from short-duration heavy rainfall to long-duration continuous rainfall. Used to reflect differences in spatial coverage This reflects the previous watershed moisture level. The above settings can provide sufficiently rich event heterogeneity without increasing the complexity of distribution calibration.
[0147] 2) Gamma Kernel Response The response of a single rainfall event to inflow was determined using the Gamma response kernel parameter values with duration correction. Therefore, for duration Rainfall events with a scale parameter h Roughly in Within this range, this set of values implies that short-duration rainfall corresponds to a sharper confluence response, while long-duration rainfall corresponds to a wider and more delayed flood peak process. This is consistent with the pattern in Meiyu season floods where "short-duration strong convection forms a faster peak, while long-duration stable rainfall forms a longer tail."
[0148] The effective flow generation coefficient is expressed in the following linear form. and retrieve parameters After substitution, It usually falls within a moderately high positive range. This refers to the initial moist state. The weight is slightly higher than that of spatial coverage. This is because, under the background of continuous rainfall during the plum rain season, the soil moisture in the early stage often has a more direct effect on the amplification of runoff.
[0149] 3) Reservoir capacity and water balance parameters Initial water storage and check reservoir capacity The base flow of the total inflow is taken as a constant. The generator set discharges a constant value.
[0150] For aperture level Using the discharge function in This represents the current water storage level. , The basic discharge capacity at each level is taken as follows
[0151] 4) Setting the cost function and penalty coefficient Take the unit time running cost function as then As can be seen, the cost function grows much faster than linearly: from low to high levels, the cost increases faster and faster, which can better reflect the reality of the accelerated increase in downstream evacuation, flood control pressure and power generation losses corresponding to large-scale flood discharge.
[0152] Paying the cost of power generation losses due to low reservoir capacity in =1320. Taking the risk of failure This value is much higher than the upper bound of the cumulative operating cost over the entire study period under normal circumstances, which ensures that the model conforms to the real-world setting that "the cost of failure is much greater than the cost of normal flood discharge".
[0153] 5) Monte Carlo and Optimization Algorithm Parameters The number of outer Monte Carlo simulations is taken as follows Used to evaluate the expected total cost under a given set of thresholds; Inner layer condition Monte Carlo degree Used to estimate the probability of real-time non-failure; Forecast period length is taken It is sufficient to cover the subsequent re-triggering of a short cluster of rainfall and the main confluence tail, without causing the short-term rolling control to lose sensitivity as much as an excessively long forecast period.
[0154] The optimization algorithm uses particle swarm optimization (PSO), with the number of particles set to... Maximum number of iterations
[0155] Results analysis: To verify the effectiveness of the model and examine the differences under different scheduling philosophies, four comparison schemes were set up: 1. Empirical Rule Scheme: Adopt representative empirical or historical outfall scheduling rules from the current or historical operation of the reservoir; 2. Conservative threshold scheme: This approach tends to upgrade the opening level earlier to prioritize reducing flood risk. 3. Aggressive threshold scheme: This approach tends to delay raising the orifice level in order to minimize the operational costs of early flood discharge. 4. Optimizing the threshold scheme: With the goal of minimizing the expected total cost, the optimal threshold set is obtained using the nested Monte Carlo optimization method proposed in Part III. .
[0156] The study focuses on comparing expected total cost, average operating cost, average failure risk cost, scheduling failure probability, and dynamic orifice path characteristics. These indicators provide a more comprehensive understanding of the trade-offs between "excessive flood discharge" and "insufficient flood discharge" across different threshold groups.
[0157] 1) Comparison of expected total cost for different threshold groups Based on Monte Carlo simulation, the expected total cost, component costs, and failure probabilities under different threshold groups can be obtained, and the results are shown in Table 1.
[0158] Table 1 Comparison of Costs and Failure Probabilities of Different Scheduling Schemes
[0159] The results show that the conservative threshold scheme, by increasing the orifice level in the early stages of the flood process, can effectively prevent the water storage from approaching its target level. The risk is mitigated by keeping the scheduling failure probability at a low level of 0.0167, with a risk cost of only 1333.33. However, this scheme maintains a high flood discharge level for a prolonged period, leading to conservative reservoir utilization and a low reservoir cost of 5538.50, the highest among all schemes, thus failing to achieve the global optimum in total cost. Conversely, while the aggressive threshold scheme minimizes the initial operating cost (operating cost of 4601.90, the lowest among all schemes), the delayed riser gate and smaller discharge make it more prone to causing the water storage to approach or even exceed the limit under multiple rounds of rainfall. This leads to a high failure risk cost. Its failure probability reaches 0.1333, the risk cost is as high as 10666.67, and the final expected total cost significantly increases to 17578.36, the highest among all schemes. The empirical rule scheme and the fixed threshold scheme typically exhibit characteristics between the two: empirical rules have a certain degree of engineering operability, but lack fine-grained response capabilities for the temporal structure of multiple rounds of heavy rainfall; the fixed threshold scheme, although simple and easy to implement, has limited adaptability and suitability to different flood scenarios. Both schemes have a failure probability of 0.0667 and a risk cost of 5333.33, with a total cost significantly higher than the optimized scheme. In comparison, the optimized threshold scheme... The optimized scheme achieves a better overall balance between operational costs, low storage capacity costs, and risk costs. In this simulation, the optimized scheme achieved a zero probability of scheduling failure and a zero risk cost throughout the entire process, while the total cost was only 8243.80, significantly lower than all other comparative schemes.
[0160] 2) Dynamic aperture path characteristics of the optimal threshold group Optimal threshold group Induced aperture path This is not a fixed template, but a dynamic response path that adjusts in real time based on future risk information. A representative sample path was extracted from the optimization search process, and its key process data are shown in Table 2. This sample has a maximum orifice level of 5 and an ending water storage of 1341.26, which is less than the check reservoir capacity. Total operating cost: 6356.00; total low storage cost: 2114.82; total cost: 8470.82.
[0161] Table 2 Representative dynamic paths of the optimized threshold scheme
[0162] Typical sample path analysis shows that the dynamic opening path formed by the optimized threshold scheme has the following distinct characteristics: First, when there are short intervals between consecutive heavy rainfalls, the inflow into the reservoir is rising rapidly, and the probability of a strong impact in the future is significantly increasing, the system tends to raise the opening level in advance, actively release some reservoir capacity, and quickly lower the water level to cope with the pressure of possible subsequent flood peaks.
[0163] Second, during periods of rainfall respite, when inflow to the reservoir decreases, or when the probability of future failure increases significantly, the system will appropriately reduce the opening level or even close the flood discharge outlets to avoid the high operating costs caused by maintaining a high flood discharge state for a long time, thus achieving a balance between cost and risk.
[0164] Third, when the risk of subsequent heavy rainfall rises again and the inflow increases again, the system can quickly upgrade the opening level again, demonstrating its dynamic adaptability to the rainfall time sequence structure, inflow process and risk evolution.
[0165] Fourth, compared with empirical rules, the optimized scheme is more sensitive to the "risk of multiple future shocks," therefore its orifice path is a forward-looking risk-response scheduling path. The essence of the optimized path is to minimize the duration of unnecessary large-flow flood discharge while ensuring that the probability of triggering high failure penalties is low.
[0166] The above description is merely a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent changes and modifications made in accordance with the claims of the present invention should still fall within the scope of the present invention.
Claims
1. A method for flood control scheduling of multi-hole reservoirs under multiple rounds of heavy rainfall, characterized in that, Includes the following steps: S1: Perform independent simulations for m starting from 1, and the generation, dynamic scheduling, and total cost calculation of a single sample path are performed according to the following steps: S1.1 is based on a labeled Hawkes process, during the study period Endogenous heavy precipitation arrival time sequence Generate a corresponding label vector for each round of precipitation. ; For the effective arrival time of heavy precipitation processes in the watershed, it must meet the following requirements. For the research period Total number of effective heavy precipitation events occurring within the region; The label vector is a label vector defined to fully characterize the runoff generation and confluence features of each round of precipitation. As an indicator of precipitation intensity, For the duration of precipitation, As for the spatial coverage characteristics of precipitation, This is an indicator of the watershed's moisture state before precipitation occurs; S1.2: For each precipitation event generated, the inflow at each discrete time point is calculated based on the Gamma response kernel, and the discrete inflow process over the entire time period is obtained by superimposing the baseflow. ; The first Rainfall time The contribution of inbound traffic is recorded as follows: The effective runoff coefficient is determined by precipitation intensity, spatial coverage, and the pre-wet state of the watershed. For the response kernel parameter vector; For characteristic functions, only if Later than the time of precipitation arrival Only then will this round of precipitation contribute to the inflow into the reservoir; The Gamma response kernel is in the form of Among them, scale parameter satisfy In the formula, The shape parameter of the Gamma distribution. and The parameters to be calibrated; Inbound flow: S1.3 Calculates the real-time failure probability based on the current precipitation history, water storage status, and orifice status: According to the threshold group The current probability of non-failure Mapped to target aperture level ; like ,but ; Update the aperture state according to the threshold-driven state transition rule. ; Calculate the total discharge flow rate based on the current orifice status. ; The total outflow from the reservoir can be expressed as: Reservoir Time The water storage capacity; Opening status ; This represents L progressively increasing flood discharge stages; The discharge capacity of the flood discharge gate is determined by the current reservoir water storage capacity and the gate opening level. This refers to the discharge flow rate of the generator set; According to the principle of water balance, the dynamic evolution of reservoir storage satisfies: Update the water storage volume for the next moment. ; Accumulate the running cost of this time step ;like Then let fail = true; S1.4 Calculate the total cost of the sample path S1.5 Update Cumulative Amount: sum_total= sum_total+ ; sum_run= sum_run+ ; sum_fail= sum_fail+ ; fail_count= fail_count+ S2 calculation index: Expected total cost estimate ; Average operating cost Average Risk Cost Scheduling failure probability S3 Based on the optimal threshold group that minimizes the expected total cost Flood control scheduling for multi-hole reservoirs.
2. The flood control scheduling method for multi-hole reservoirs under multiple rounds of heavy rainfall as described in claim 1, characterized in that, It also includes the following steps: S4 calculates the estimated probability of conditional non-failure at the current time. ,according to Adjust the opening level of the reservoir; Estimated probability of condition not failing at the current time The calculation is as follows: S4.1 Initialize the success counter success_count=0; S4.2 for m from 1 to Perform independent simulations one after another; S4.3 Given historical rainfall information Under these conditions, based on the labeled Hawkes process to generate Rainfall event sequence within; S4.4 superimposes the residual confluence contribution of historical rainfall events to future periods with the confluence contribution of newly generated future rainfall events to obtain the inflow path. ; S4.5 sets the initial value of the test water storage. ; S4.6 in Maintain the current aperture level. Unchanged, gradually recursively ; S4.7 If at any future time the condition is met If the simulation fails, the inner simulation is recorded as a failure; otherwise, it is recorded as a success. S4.8 If this simulation is successful, then success_count = success_count + 1 S4.9 Conditional non-failure probability estimate .
3. The flood control scheduling method for multi-hole reservoirs under multiple rounds of heavy rainfall as described in claim 1, characterized in that, It also includes the following steps: Constructing a framework for solving optimal threshold sets based on a nested approach of intelligent optimization algorithms and Monte Carlo simulation: The first step is to optimize the algorithm initialization by setting the algorithm parameters, including population size, maximum number of iterations, feasible region of the threshold set, threshold monotonicity constraint, and convergence criterion. The Particle Swarm Optimization (PSO) algorithm is used, with each particle corresponding to a set of feasible thresholds. ; The second step is to evaluate the objective function value; this involves setting thresholds for each particle. Call step S1 to calculate its expected total cost estimate. And use this estimate as the fitness function value of the particle; The third step is population iterative update; according to the optimization algorithm update rules, the particle position and velocity are iteratively adjusted to generate a new generation of threshold group population, and Monte Carlo simulation is continuously called to estimate its total cost; The fourth step is to determine convergence. If the maximum number of iterations is reached, or the change in the global optimal fitness value satisfies the preset convergence criterion, the iteration is terminated; otherwise, the objective function evaluation and population update process continues. Fifth step: Output the optimal solution; after the iteration, output the threshold set corresponding to the particle with the lowest estimated total expected cost in the population, which is the optimal scheduling threshold set. .
