Reservoir operation decision method for flood risk assessment

Abstract

This invention discloses a reservoir scheduling decision-making method for flood season risk assessment, belonging to the field of scheduling decision-making technology. The method first acquires real-time reservoir status data and ensemble forecast data, calculates the comprehensive risk index for the current moment, which quantifies the comprehensive risk level of the forecasted flood threat, current water level, and the degree of tension in remaining flood control capacity. Then, it constructs a rule pool containing multiple basic scheduling rules, with each rule outputting a suggested outflow. Next, it employs a recursive risk memory weight synthesis mechanism to establish short-term risk memory variables and long-term risk pressure variables, calculates the tendency weights of each rule through nonlinear mapping, and then recursively obtains the actual weights through first-order lag filtering. Finally, it synthesizes the outflow decision value. This invention integrates real-time risk assessment with scheduling rules, achieving recursive adaptive adjustment of weights, resulting in refined risk perception, continuous and stable decision-making, and accelerated response to persistent risks. It can be applied to the scheduling decisions of small and medium-sized reservoirs during the flood season.

Description

Technical Field

[0001] This invention relates to the field of scheduling decision-making technology, and more specifically, to a reservoir scheduling decision-making method for flood season risk assessment. Background Technology

[0002] Reservoirs are an important means for humans to redistribute water resources in time and space, undertaking multiple functions and tasks such as flood control, power generation, navigation, and water supply. Reservoir scheduling technology is one of the essential means to ensure the normal operation of reservoirs. Among them, flood season scheduling is directly related to dam safety and downstream flood control safety, and is the key to scheduling work.

[0003] For reservoir flood season optimization scheduling problems, scheduling functions (such as linear functions, neural networks, fuzzy methods, decision trees, etc.) and scheduling diagrams are commonly used as reservoir scheduling rule forms. In recent years, scheduling rule synthesis methods have attracted attention. For example, Chinese invention patent CN104462861A discloses a reservoir scheduling decision-making method based on reservoir scheduling rule synthesis, which synthesizes multiple single scheduling rules through a linear weighted model and uses objective weighting methods such as Bayesian model averaging to determine the weights. This method can integrate the advantages of multiple rules, making the scheduling decision more robust.

[0004] However, the aforementioned prior art has the following drawbacks: (1) The weights are fixed values ​​obtained through offline training using historical optimal scheduling data, and once determined, they will not be changed. However, the runoff during the flood season has a high degree of uncertainty and dynamic change characteristics, and the synthesis rules with fixed weights cannot respond to changes in real-time risk status (such as the size of the forecast flood, the current water level, and the degree of tension of the remaining flood control capacity).

[0005] (2) When a major flood is predicted, the flood control safety orientation rule cannot obtain the dominant weight in time, which may result in untimely flood discharge or reservoir water level exceeding the limit; when a minor flood is predicted, the benefit orientation rule cannot obtain sufficient weight, resulting in waste of flood resources.

[0006] (3) Risk analysis is only used as a pre-schedule evaluation indicator, rather than a real-time driving variable for scheduling decisions. There is a lack of dynamic coupling mechanism between risk information and scheduling rules.

[0007] (4) The weight update lacks memory and smoothness. A single forecast error or water level fluctuation may lead to sudden changes in decision-making, affecting the continuity and stability of scheduling. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a reservoir scheduling decision-making method for flood season risk assessment, thereby solving the problems of statically fixed weights in existing reservoir scheduling rule synthesis methods, which cannot be dynamically adjusted according to real-time risks, and the separation of risk analysis and scheduling decision-making.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A reservoir scheduling decision-making method for flood season risk assessment includes the following steps: S1: Obtain real-time status data and ensemble forecast data of inflow from the reservoir, and calculate the comprehensive risk index at the current moment. The comprehensive risk index is used to quantify the overall risk level of forecasting flood threats, current water levels, and the urgency of remaining flood control capacity.

[0010] As a further improvement, the comprehensive risk index is calculated using a weighted summation method. The forecast flood threat index is determined based on the ratio of the upper limit of the confidence interval of the forecast inflow to the downstream safe discharge. The current water level index is determined based on the relative position of the current water level between the dead water level and the flood control high water level. The remaining reservoir capacity index is determined based on the ratio of the occupied flood control capacity to the maximum flood control capacity.

[0011] As a further improvement, the ensemble forecast data is the probability distribution of inflow and its confidence interval for each period in the future; the real-time status data includes the current water level and the current reservoir capacity.

[0012] S2: Constructing a system containing A rule pool for basic scheduling rules, where each basic scheduling rule suggests outbound traffic based on the current input and output. , .

[0013] As a further improvement, the basic scheduling rules include at least three types: flood control safety-oriented rules, benefit-oriented rules, and balanced rules.

[0014] S3: Employs a recursive risk memory weighting synthesis mechanism, based on the current comprehensive risk index. Based on historical risk information, the dynamic weight of each basic scheduling rule is calculated recursively. The mechanism specifically includes: S31: Establish short-term risk memory variables to characterize recent risk trends. and long-term risk stress variables used to characterize the cumulative effect of persistent high risk. And updated respectively using recursive formulas. Among them, the short-term risk memory variable The update formula is , For memory decay coefficient; long-term risk stress variable The update method is as follows: when the comprehensive risk index exceeds the preset threshold, a fixed increment is added; when it is below the threshold, it decays at a preset fixed rate and is respectively limited to the upper boundary value 1 and the lower boundary value 0.

[0015] S32: Based on the short-term risk memory variable and the long-term risk pressure variable, calculate the tendency weight of each basic scheduling rule through nonlinear mapping. The nonlinear mapping causes the tendency weight of the flood control-oriented rule to increase relatively under high-risk conditions, and the tendency weight of the benefit-oriented rule to decrease relatively.

[0016] As a further improvement, the nonlinear mapping adopts an exponential weighted normalization method: the risk response value of each basic scheduling rule is calculated separately. The risk response value is a linear combination of the short-term risk memory variable and the long-term risk pressure variable. Each rule has a preset first sensitivity coefficient and a second sensitivity coefficient. The risk response values ​​of each rule are exponentially summed and normalized to obtain the tendency weight of each rule.

[0017] As a further improvement, when the basic scheduling rules include at least flood control safety-oriented rules, benefit-oriented rules, and balance rules, the signs of the sensitivity coefficients are set as follows: the first and second sensitivity coefficients of the flood control safety-oriented rules are both positive; the first sensitivity coefficient of the benefit-oriented rules is negative and the second sensitivity coefficient is non-positive; and both sensitivity coefficients of the balance rules are zero.

[0018] S33: The actual weights from the previous time step and the tendency weights from the current time step are weighted and smoothed to obtain the actual weights for the current time step. ,in The initial weights are positive numbers less than 1. .

[0019] As a further improvement, after calculating the tendency weight and before performing weighted smoothing, an inertia retention constraint is introduced: the tendency weight at the current moment is compared with the actual weight at the previous moment. If the difference between the two is less than a preset threshold, the tendency weight at the current moment is corrected to the actual weight value at the previous moment.

[0020] As a further improvement, after obtaining the actual weights at the current moment through weighted smoothing, boundary clamping is performed: each actual weight is restricted to between a preset lower limit and an upper limit, and all clamped weights are normalized so that the sum of the weights of each basic scheduling rule is 1.

[0021] As a further improvement, the memory decay coefficient The value range of the weighted smoothing coefficient is from 0.5 to 0.9. The value range is from 0.1 to 0.5.

[0022] S4: Based on the dynamic weights and the outputs of each basic scheduling rule, synthesize the outbound flow decision value for the current moment. .

[0023] As a further improvement, after S4, the method further includes: using Monte Carlo combined sampling to generate the uncertainty interval of the scheduling decision at the current moment, specifically: randomly selecting a rule as the probability distribution using the dynamic weights of each basic scheduling rule, sampling a flow value from the outbound flow probability distribution corresponding to the rule, repeating the sampling multiple times to obtain a sample sequence, and taking the predetermined quantile of the sample sequence as the upper and lower bounds of the decision interval.

[0024] As a further improvement, the method also includes: outputting the current moment's synthetic outflow decision value, as well as the short-term risk memory variable and the long-term risk pressure variable as auxiliary information.

[0025] The technical effects and advantages of this invention are as follows: Compared to existing technologies, this invention establishes short-term risk memory variables. and long-term risk stress variables This creates a risk characterization mechanism with two time scales. Short-term variables rapidly track changes in risk trends, while long-term variables accumulate pressure from persistently high risks. The combination of these two allows the system to distinguish between instantaneous fluctuations and continuous deterioration, thereby avoiding misjudgments caused by single forecast anomalies.

[0026] Compared to existing technologies, this invention updates the actual weights recursively using a first-order lag filter. This allows for a smooth transition of weights rather than abrupt changes. Furthermore, the inertial retention constraint filters out minor weight drifts. The resulting technical effect is that scheduling decisions maintain continuity and stability over time, avoiding the operational burden and safety hazards caused by frequent gate adjustments.

[0027] Compared to existing technologies, this invention utilizes long-term risk stress variables. Under sustained high-risk conditions, the pressure gradually increases, and the sensitivity coefficient is used to amplify the tendency weight of the flood control guidance rule, realizing the response characteristic that "the longer the risk lasts, the faster the flood control weight increases"; at the same time, the pressure variable slowly decays after the risk subsides, avoiding premature relaxation of vigilance. It can be applied to small and medium-sized reservoirs with small regulation capacity and short decision-making time.

[0028] Compared to existing technologies, this invention ensures that each basic scheduling rule retains the lowest weight under any risk state through boundary clamping processing, avoiding the system from being completely biased towards a certain rule and losing decision diversity, thus enhancing the robustness of the scheduling strategy; by outputting a risk situation vector as auxiliary information, scheduling personnel can intuitively understand the basis for weight allocation, improving the interpretability of the method.

[0029] Compared to existing technologies, this invention generates decision intervals through Monte Carlo combined sampling, quantifying forecast uncertainty and rule diversity into actionable confidence intervals, providing schedulers with a visual reference for decision-making risks. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the method execution of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1 As attached Figure 1 The reservoir scheduling decision-making method shown for flood season risk assessment has the following specific implementation details: The reservoir scheduling decision-making method provided in this paper for flood season risk assessment cyclically executes the following steps in each scheduling period (e.g., every hour, every 3 hours, or every 6 hours, the specific duration of which is determined based on the reservoir size, catchment area, and forecast period): Data Acquisition and Risk Quantification: Obtain real-time status data and ensemble forecast data, and calculate the comprehensive risk index for the current period.

[0033] Basic rule calculation: Invokes various pre-built basic scheduling rules, each of which outputs suggested outbound traffic.

[0034] Recursive weighting: The weights of each rule are dynamically calculated using historical risk memory (short-term memory variables and long-term stress variables) and the current risk index.

[0035] Decision output: Synthesize the final outbound flow decision value, and optionally generate an uncertainty range.

[0036] State update and loop: Save the weight and memory variables of the current time period as the initial values ​​for the next time period, and wait for the next scheduling time period to be triggered.

[0037] The following is a detailed description of each step.

[0038] Step S1: Real-time Risk Situation Awareness and Quantification The specific data acquisition process is as follows: At the start of each scheduling period, the following data is collected: Real-time status data: Current water level (Unit: meter): Read directly from the reservoir water level gauge, with an accuracy of 0.01 meters.

[0039] Current storage capacity (Unit: 10,000 cubic meters): The value is obtained by interpolation based on the water level value using the water level-reservoir capacity relationship curve (usually stored in tabular or polynomial form) pre-calibrated by the reservoir management department.

[0040] Ensemble forecast data: provided by the hydrological forecasting department for the future Each time period (e.g., the next 24 hours, divided into 3-hour intervals) The inflow probability forecast for the reservoir is as follows. The forecast results for each time period are given in the form of a normal distribution: ,in Mean (unit: m) 3 / s), Standard deviation (unit: m) 3 / s). Based on this, calculate the 90% confidence interval: This step requires the maximum value among the upper bounds of the confidence intervals for all time periods, denoted as . .

[0041] In addition, the following reservoir characteristic parameters (obtained from reservoir design documents or operating procedures) should be pre-stored: Dead water level (The minimum water level that is allowed to drop during normal operation of the reservoir, in meters).

[0042] High flood level (The highest water level reached in front of the dam when the reservoir encounters the design flood of the downstream protected object, in meters).

[0043] Maximum flood control capacity (Reservoir capacity between flood control high water level and dead water level, unit: 10,000 cubic meters).

[0044] Reservoir capacity corresponding to dead water level (Found through the water level-reservoir capacity curve, unit: 10,000 cubic meters).

[0045] Downstream safe discharge (Maximum safe flow rate allowed to pass through the downstream channel, unit: m³) 3 / s, as determined by the water resources department).

[0046] Overall Risk Index The calculation is as follows: Overall Risk Index This is a weighted sum of three dimensionless indicators, ranging from [0, 1]. The three indicators respectively reflect the forecast flood threat, the current water level, and the degree of tension in the remaining flood control capacity.

[0047] Indicator 1: Flood Threat Forecasting Indicator ; In the specific calculation, first calculate and The ratio is taken as 1 if it is greater than 1 (i.e., clamping operation); otherwise, the ratio is taken. The larger the value, the higher the risk that the predicted flood will exceed the downstream safe discharge capacity.

[0048] Indicator 2: Current water level ; This indicates the relative position of the current water level between the dead water level and the flood control high water level. When the index is 0, The time index is 1. If If the water level is below the dead water level (extreme case), then a value of 0 will be forced; if If the water level is higher than the flood control high water level, then a value of 1 will be forcibly applied.

[0049] Indicator 3: Remaining Storage Capacity Tension Indicator ; The numerator represents the occupied flood control capacity (the portion of the current capacity exceeding the capacity corresponding to the dead water level). The numerator is set to 0; the denominator is the maximum flood control capacity. This ratio represents the proportion of flood control capacity that has been utilized; the larger the value, the more strained the remaining flood control capacity. If (i.e., the water level reaches or exceeds the flood control high water level), then the index is 1.

[0050] Weighted summation: ; Where a, b, and c are preset weight coefficients, satisfying The weighting should reflect the reservoir management's emphasis on different sources of risk: if the flood control task is extremely heavy, a weighting can be set... If the water level has a significant impact on dispatching decisions (e.g., increased dam safety risk at high water levels), the amount can be increased. In this embodiment, we take... .

[0051] Calculated This will be used in subsequent steps. For easier understanding by dispatchers, a risk level can be added: low risk. Medium risk High risk [0.7, 1], this level is only for display and does not affect the weight calculation.

[0052] Step S2: Construct a basic scheduling rule pool Before implementing the method, a set of basic scheduling rules needs to be constructed based on reservoir scheduling regulations and operational experience. Each rule is an executable function that takes the current water level as input. Forecast inflow characteristics (such as mean) (or peak value), etc., output suggested outbound flow rate (Unit: m) 3 / s). This implementation uses three typical rules as examples; in actual applications, rules can be added or replaced.

[0053] Flood control safety guidelines Specifically as follows: The primary goal is to ensure the safety of the dam and downstream flood control. A simple and practical implementation is as follows: ; The discharge volume increases linearly with the water level, but does not exceed [a certain limit]. In the specific calculation, first calculate... If the result is negative, take 0; if it is greater than 1, take 1, then multiply by... More refined segmented discharge control curves can also be used; this implementation method is not limited to any particular approach. The output of this rule increases monotonically with rising water level, and tends to increase the discharge volume during high-risk periods.

[0054] Profit-oriented rules Specifically as follows: The goal is to maximize power generation or ensure the highest water supply guarantee rate. During the flood season, this rule tends to maintain a higher water level to facilitate later benefits, but must ensure that it does not exceed the flood control high water level. One implementation method is: ; in The mean value of the predicted peak flow (which can be obtained from ensemble forecasts, for example, by taking...) ), The full-capacity flow rate or basic water consumption rate of the generating unit (given by the operating procedures, unit: m³). 3 / s). When the predicted flood peak is less than the basic water consumption flow rate. This indicates that water is not actively released (it is only released through generator power generation). It can also be set to a smaller constant value (e.g., ...). The rule indicates that water should be stored as much as possible when the risk is low. A smaller value should be output when the risk is high.

[0055] Balanced rules Specifically as follows: Typically, based on a standard dispatching chart, the reservoir's water level is divided into several zones (such as flood control zones, transition zones, and beneficial zones), each corresponding to a fixed discharge flow or a linear function. For example: like ,but ; like ,but ; like ,but ; like ,but .

[0056] Specific partition thresholds (e.g.) ) and flow value ( The decision is based on the reservoir scheduling map and is provided by the reservoir management unit. This rule serves as a compromise for medium-risk situations.

[0057] In this implementation, it is assumed that the three rules have been predefined and can be obtained by calling them in each scheduling period. .

[0058] Step S3: Recursive Risk Memory Weight Synthesis This step dynamically updates the weights of each rule using historical risk memory and the current risk index. This step requires maintaining the state variables from the previous time period: , , Initial moment (first scheduling period) , , (i.e., equal weight).

[0059] Sub-step S31: Dual-risk memory retrieval Short-term risk memory variables Using exponentially weighted moving average recursion: ; in This is the memory decay coefficient, with a value ranging from 0.5 to 0.9, and a typical value of 0.7. The larger the value, the smoother the short-term memory, and the slower the response to near-term risks; The smaller the size, the more sensitive it is to current risks. For large reservoirs with large regulating capacity and slow flood evolution, a larger size can be adopted. (e.g., 0.85); for small and medium-sized reservoirs with small regulating capacity and requiring rapid response, a smaller value can be used. (e.g., 0.6).

[0060] Long-term risk stress variables Using an integral recursion with a threshold: Preset risk trigger threshold (Typical value 0.6, range 0.5–0.7), cumulative increment step size (Typical value 0.15, range 0.1–0.2), decay step size (Typical value 0.05, range 0.03 to 0.08).

[0061] like ,but .

[0062] like ,but .

[0063] This variable reflects the cumulative "stress" of sustained high risk: repeated high risks lead to... It gradually increases, and then slowly decreases after the risk subsides.

[0064] Sub-step S32: Calculation of propensity score weights Preset sensitivity coefficients for each rule . Control rules for short-term risk memory Sensitivity Controlling long-term stress The sensitivity of the risk. The absolute value determines the strength of the impact of risk changes on the weights, with a typical range of values: Between 1 and 5, The value is between 0.5 and 3. The three rules for this implementation are set as follows: Flood control safety guidelines: , (Positive value).

[0065] Profit-oriented principles: , (Negative value).

[0066] Balanced rules: , (neutral).

[0067] Calculate the risk response value for each rule: ; Using the softmax function to Convert to propensity score weights: ; This transformation guarantees that all tendency weights are positive and sum to 1, and when When it is larger, It amplifies exponentially.

[0068] Inertia Preservation Constraint: To filter out weight jitter caused by minor fluctuations in input data, a change-aware threshold is set. (Typical value 0.02, range 0.01–0.03). Calculate the absolute value of the difference between the tendency weight of each rule and the actual weight of the previous period. If all The corresponding differences are all less than If the fluctuation is small, the tendency weight will be adjusted to the actual weight of the previous period. Otherwise, the bias weight remains unchanged, i.e. .

[0069] Sub-step S33: Recursive smooth update and boundary clamping First-order lag filtering is used to smoothly transition the bias weights to the actual weights: ; in This is the weighted smoothing coefficient, with a value ranging from 0.1 to 0.5, and a typical value of 0.3. The smaller the value, the slower the actual weight changes, and the greater the decision inertia. For reservoirs that need to respond quickly to changes in risk, a larger value (e.g., 0.4–0.5) can be used; for reservoirs that want to maintain stable scheduling, a smaller value (e.g., 0.1–0.2) can be used.

[0070] Boundary clamping: Preset weight lower limit (For example, 0.05, with a value range of 0.05 to 0.10) and upper limit. (For example, 0.90, with a value range of 0.85 to 0.95). For each Perform clamping: ; Normalization: The sum of the weights after clamping may not be 1, requiring renormalization. Calculate the sum of the weights after clamping. Then scale each rule: ; The final weights satisfy And the sum is 1.

[0071] Step S4: Synthesize scheduling decision output Based on the dynamic weights and the suggested outbound flow rates of each rule, the final outbound flow decision value is synthesized: ; This value represents the discharge flow rate (unit: cubic meters per second) that the reservoir should execute during the current scheduling period. Dispatchers can use this value to operate the gates or adjust the generator output.

[0072] Uncertainty interval estimation (optional) To provide quantitative information on decision-making risk, the Monte Carlo combined sampling method can be used to generate confidence intervals. The specific steps are as follows: Determine the probability distribution of outbound flow for each rule. Assume the... The outbound flow rate of this rule follows a normal distribution, and the mean is... Standard deviation It is necessary to pre-calculate the following: collect a sequence of differences between the suggested outbound flow and the actual optimal scheduled outbound flow (e.g., obtained through dynamic programming or practical operational experience) from historical operations of this rule, and calculate the standard deviation of this sequence. If historical data is unavailable, a relative error ratio can be set based on expert experience, for example... .

[0073] Set the number of samples (For example (Generally, 500 to 2000 times are taken).

[0074] Cyclic sampling: for arrive : According to weight distribution Randomly select a rule Specific method: Generate a random number that is uniformly distributed in the interval [0, 1]. Calculate the cumulative probability , Find the smallest Make Then select the rule. .

[0075] From normal distribution Randomly select a traffic value It can be generated using a standard normal random number generator. Then calculate .

[0076] Sort and quantiles: The obtained Traffic value Sort in ascending order to obtain the sequence. .

[0077] Output confidence interval: take the first... The and the first The individual values ​​serve as the lower bound of the 90% confidence interval. and the Upper Realm If the index is not an integer, the average of the values ​​of two adjacent indices can be used.

[0078] Output the interval This is for dispatchers' reference.

[0079] To further enhance decision-making transparency, the following supplementary information can be output (for dispatchers to review or record in the dispatch log): Current overall risk index ; Risk Situation Vector ; Dynamic weights of each rule .

[0080] The specific details of the state update and the next time period cycle are as follows: After making decisions for the current time period, update the state variables for use in the next time period: Will Assign to (Used for memory updates in the next period); Will Assign to ; Will Assign to .

[0081] Then it waits for the next scheduling period (e.g., after 3 hours) to trigger, and then restarts execution from step S1. This forms a closed-loop dynamic adaptive scheduling system. The initial time (the first scheduling period) is... Set as follows If there are existing historical risk records, initialization can also be performed based on previous data. and .

[0082] To facilitate those skilled in the art in adjusting parameters according to the specific characteristics of the reservoir, the meaning of each parameter, typical value range, and adjustment suggestions are given below.

[0083] The risk index weights a, b, and c sum to 1, corresponding to the weights of the predicted flood threat, current water level, and remaining reservoir capacity strain, respectively. If the reservoir's flood control workload is extremely heavy, the weights can be increased. (For example, take 0.6); if the water level has a significant impact on the scheduling decision (e.g., the safety risk of the dam increases at high water levels), the value can be increased. If the tightness of storage capacity is a key constraint, it can be increased. Typical configuration is .

[0084] Memory decay coefficient To control the smoothness of short-term risk memory, the value should range from 0.5 to 0.9. Smaller values ​​(e.g., 0.6) should be used for reservoirs with weak regulation capacity and requiring rapid response; larger values ​​(e.g., 0.8) should be used for reservoirs with large capacity and slow flood evolution.

[0085] Long-term stress parameters: trigger threshold The value ranges from 0.5 to 0.7. A smaller value (e.g., 0.5) is used to accumulate pressure earlier, while a larger value (e.g., 0.7) is used only for high-risk accumulation. Accumulation increment step size. The value ranges from 0.1 to 0.2. For rapid response and sustained risk, a larger value (e.g., 0.2) is used; for a conservative approach, 0.1 is used. Decay step size. The value range is 0.03 to 0.08. Take a larger value (e.g., 0.08) to quickly relax after the risk has subsided, and take a smaller value (e.g., 0.03) to remain vigilant.

[0086] Weight update smoothing coefficient The value ranges from 0.1 to 0.5; smaller values ​​(e.g., 0.2) are used for stable decisions, while larger values ​​(e.g., 0.4) are used for rapid risk tracking. Inertia retention threshold. The value ranges from 0.01 to 0.03, with 0.02 being the most common value. It can be adjusted according to the fluctuation range of the weight.

[0087] Weight Boundaries and Used for clamping, ensuring each rule has at least 5%–10% participation in the decision-making process, while avoiding any single rule having a weight exceeding 85%–95%. Typical value , .

[0088] Sensitivity coefficient The absolute value of the risk factor determines the strength of its impact on the weight, and can be determined through historical data calibration or expert experience. The sensitivity coefficient for flood control rules should be positive, for beneficial rules it should be negative, and for balanced rules it should be zero. Typical example: Flood control rules. ; Profit-promoting rules Balanced type .

[0089] Through the detailed embodiments described above, those skilled in the art should be able to fully understand and implement the method described in this invention. This invention, through a recursive risk memory weight synthesis mechanism, dynamically integrates real-time risk perception with scheduling rules, solving the problems of statically fixed weights and separation of risk and scheduling in existing methods. In practical applications, parameters can be adjusted according to the specific characteristics of the reservoir and scheduling requirements, all without departing from the protection scope of this invention.

[0090] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any person skilled in the art can make some modifications or alterations to the disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A reservoir scheduling decision-making method for flood season risk assessment, characterized in that, Includes the following steps: S1: Obtain real-time status data and ensemble forecast data of inflow from the reservoir, and calculate the comprehensive risk index for the current moment. The comprehensive risk index is used to quantify the overall risk level of the forecasted flood threat, current water level, and remaining flood control capacity. S2, constructing a system containing A rule pool for basic scheduling rules, where each basic scheduling rule suggests outbound traffic based on the current input and output. , ; S3 employs a recursive risk memory weighting synthesis mechanism, based on the comprehensive risk index at the current moment. Based on historical risk information, the dynamic weight of each basic scheduling rule is calculated recursively. ; S4. Based on the dynamic weights and the outputs of each basic scheduling rule, the decision value for the outbound flow at the current moment is synthesized. ; The characteristic feature is that the recursive risk memory weight synthesis mechanism in step S3 includes: S31, Establish short-term risk memory variables to characterize recent risk trends. and long-term risk stress variables used to characterize the cumulative effect of persistent high risk. And updated respectively using recursive formulas, wherein the short-term risk memory variable The update formula is , The memory decay coefficient is the long-term risk stress variable. The update method is as follows: when the comprehensive risk index exceeds the preset threshold, a fixed increment is added; when it is below the threshold, it decays at a preset fixed rate and is respectively limited to the upper boundary value 1 and the lower boundary value 0. S32, Based on the short-term risk memory variable and the long-term risk pressure variable, calculate the tendency weight of each basic scheduling rule through nonlinear mapping. The nonlinear mapping causes the tendency weight of the flood control-oriented rule to increase relatively under high-risk conditions, and the tendency weight of the benefit-oriented rule to decrease relatively. S33, the actual weight of the previous time step is weighted and smoothed with the tendency weight of the current time step to obtain the actual weight of the current time step: ,in The initial weights are positive numbers less than 1. .

2. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, The nonlinear mapping in step S32 adopts an exponential weighted normalization method: the risk response value of each basic scheduling rule is calculated respectively. The risk response value is a linear combination of the short-term risk memory variable and the long-term risk pressure variable. Each rule has a preset first sensitivity coefficient and second sensitivity coefficient. The risk response values ​​of each rule are exponentially summed and normalized to obtain the tendency weight of each rule.

3. The reservoir scheduling decision-making method for flood season risk assessment according to claim 2, characterized in that, The basic scheduling rules include at least flood control safety-oriented rules, benefit-oriented rules, and balance rules; wherein, the first and second sensitivity coefficients of the flood control safety-oriented rules are both positive, the first sensitivity coefficient of the benefit-oriented rules is negative and the second sensitivity coefficient is non-positive, and both sensitivity coefficients of the balance rules are zero.

4. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, In step S3, after calculating the tendency weight and before performing weighted smoothing, an inertia retention constraint is introduced: the tendency weight at the current moment is compared with the actual weight at the previous moment. If the difference between the two is less than a preset threshold, the tendency weight at the current moment is corrected to the actual weight value at the previous moment.

5. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, In step S3, after obtaining the actual weights at the current moment through weighted smoothing, boundary clamping is performed: each actual weight is restricted to between a preset lower limit and an upper limit, and all clamped weights are normalized so that the sum of the weights of each basic scheduling rule is 1.

6. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, Step S4 is followed by: generating the uncertainty interval of the current scheduling decision using the Monte Carlo combined sampling method. Specifically, a rule is randomly selected using the dynamic weights of each basic scheduling rule as the probability distribution, a flow value is sampled from the outbound flow probability distribution corresponding to the rule, the sampling is repeated multiple times to obtain a sample sequence, and the predetermined quantile of the sample sequence is taken as the upper and lower bounds of the decision interval.

7. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, In step S1, the comprehensive risk index is calculated. The method is as follows: calculate the forecasted flood threat index, the current water level index, and the remaining reservoir capacity index separately, and then sum the three in a weighted manner; wherein, the forecasted flood threat index is determined based on the ratio of the upper limit of the confidence interval of the forecasted inflow to the downstream safe discharge, the current water level index is determined based on the relative position of the current water level between the dead water level and the flood control high water level, and the remaining reservoir capacity index is determined based on the ratio of the occupied flood control capacity to the maximum flood control capacity.

8. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, The ensemble forecast data in step S1 is the probability distribution of inflow and its confidence interval for each time period in the future; the real-time status data includes the current water level and the current reservoir capacity.

9. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, The memory decay coefficient The value range of the weighted smoothing coefficient is from 0.5 to 0.

9. The value range is from 0.1 to 0.

5.

10. The reservoir scheduling decision-making method for flood season risk assessment according to claim 1, characterized in that, The method further includes: outputting the current moment's synthetic outflow decision value, as well as the short-term risk memory variable and the long-term risk pressure variable as auxiliary information.

Images