A method, system, device and storage medium for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system

Abstract

The present invention discloses a method, system, device and storage medium for simulating extreme floods and assessing flood control risks in a reservoir-lake-river system. The method first identifies the risk source factors of historical extreme flood events, uses the Vine Copula function to characterize the uncertainty of their spatial distribution, and generates risk source factor samples through Latin hypercube sampling. Subsequently, based on the Neural Gas algorithm, the samples are reduced to construct a risk source factor scenario tree. Combining historical flood data and using the same ratio scaling method, extreme flood hydrograph scenarios are generated. Then, a joint flood control optimal operation model is constructed to evaluate the flood control risks of reservoirs and lakes and analyze the evaluation effect. The structure of the present invention is simple, and the risk assessment is accurate and stable. It not only helps with the refined risk assessment in flood-prone areas but also provides theoretical support for extreme flood risk assessment in data-deficient areas.

Description

Technical Field

[0001] The present invention belongs to the technical field of flood control scheduling, and particularly relates to a method, system, device and storage medium for simulating extreme floods and evaluating flood control risks in a reservoir-lake-river system. Background Art

[0002] Extreme flood events have extremely high destructive potential. Their scale and intensity often exceed the design standards of flood control projects, which may lead to widespread damage, casualties, and disruptions to the economy and society. The occurrence of such events is usually triggered by the accumulation of multiple risk source factors, and these risk source factors exhibit significant spatio-temporal distribution uncertainties, making it difficult to accurately predict extreme flood events. In addition, due to the scarcity of extreme flood events, relying solely on historical flood data for flood control risk assessment has certain limitations. Traditional flood prediction methods often struggle to accurately forecast these uncertain extreme flood events, further increasing the difficulty of flood control risk assessment. Although it is impossible to completely prevent the occurrence of extreme flood events, effective risk assessment methods can identify high-risk factors and take flood control optimization scheduling measures, thereby effectively reducing flood disasters.

[0003] In existing research, analytical methods and stochastic simulation methods have been widely applied to the flood control risk assessment of reservoir systems. Analytical methods rely on simplifying risk characteristics and assuming model conditions, and evaluate flood control risks through integration or differentiation methods. However, in a complex reservoir-lake-river system, due to the complex characteristics of its hydraulic connection and hydrological features, it is difficult to apply analytical methods to the simulation of extreme floods and flood control risk assessment in a reservoir-lake-river system. The stochastic simulation method first fits the marginal distribution function of risk source factors, then randomly samples from the marginal distribution function to generate risk source factor samples, and finally inputs the risk source factor samples into the joint flood control optimization scheduling model, which can randomly simulate the flood scheduling process and conduct flood control risk assessment.

[0004] However, there are still some challenges in randomly simulating extreme flood events and flood control risk assessment. First, in the stage of characterizing and describing extreme flood events, due to the small sample size of historical extreme flood events, the historically occurring extreme flood events cannot fully represent all possible extreme flood events. Therefore, numerical simulation methods need to be used to enrich the sample library of extreme flood events. Second, the occurrence of extreme flood events is usually the result of the combined accumulation of multiple risk source factors, and there are significant spatio-temporal distribution uncertainties in these risk source factors. Therefore, flood control risk assessment needs to consider the uncertain relationships between multiple risk source factors that are interrelated. Third, flood control scheduling requires a rapid and accurate response to extreme flood events. Therefore, the calculation time of flood control risk assessment must be controlled within a small range to meet the needs of actual flood control scheduling. Therefore, in view of the characteristics of the spatio-temporal combination uncertainty of multiple risk source factors in areas lacking historical flood data, how to construct an accurate comprehensive flood control risk assessment method for extreme flood events is a scientific problem that urgently needs to be solved. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide a method, system, device and storage medium for simulating extreme floods and assessing flood control risks in a reservoir-lake-river system, to improve the accuracy and stability of flood control risk assessment for extreme flood events, and to provide a scientific decision-making reference for flood control scheduling.

[0006] Technical Solution: A method for simulating extreme floods and assessing flood control risks in a reservoir-lake-river system according to the present invention includes the following steps:

[0007] S1. Collect historical flood data and flood control project data of the reservoir-lake-river system, count historical extreme flood events, identify risk source factors that cause historical extreme flood events, and the risk source factors include upstream inflow, inflow into the lake from the interval, and the highest water level of the external river. Calculate each type of risk source factor in the historical flood data of each year using the annual maximum value method;

[0008] S2. Fit marginal distribution functions for each type of risk source factor, construct corresponding two-dimensional Copula functions and conditional Copula functions, generate a Vine Copula function, generate probability samples of risk source factors through the Latin hypercube sampling method, calculate the joint probability of each group of risk source factor probability samples, screen out the risk source factor probability samples with a joint probability lower than the extreme flood event probability threshold, and use the marginal distribution function to perform inverse transformation on them to generate risk source factor samples;

[0009] S3. Use the K-Means method to cluster the risk source factor samples into different groups, use the Neural Gas algorithm to construct a risk source factor scenario tree for each group, calculate the extreme flood scenario probability, and use the same ratio scaling method to couple historical flood data to generate extreme flood hydrograph scenarios;

[0010] S4. Construct a joint flood control optimization scheduling model for the reservoir-lake-river system with the objective of minimizing the average excess flood volume of the lake during the entire scheduling period. Take the extreme flood hydrograph scenario as the input data of the model, simulate the joint flood control scheduling plan for extreme floods, analyze the situations exceeding the safety storage thresholds of the reservoir and the lake in the joint flood control scheduling plan for extreme floods, evaluate the flood control risks of the reservoir and the lake, and use the continuous ranked probability score (CRPS) to evaluate the results of the flood control risks of the reservoir and the lake.

[0011] Further, in step S1, the annual maximum value method is used to calculate each type of risk source factor in the historical flood data of each year, including:

[0012] Based on the historical flood data, extract the maximum flood event that occurred in each year, and calculate its corresponding upstream inflow into the reservoir, inflow into the lake from the intermediate area, and the highest water level of the outer river. The calculation formulas are as follows:

[0013] ,

[0014] where, is the upstream inflow into the reservoir in the historical flood data of the th year, , is the year index corresponding to the historical flood data, is the total number of years of the historical flood data, is the th year, and is the historical upstream inflow rate of the rd reservoir at the th time period, is the total number of reservoirs,

[0015] ,

[0016] where, is the inflow into the lake from the intermediate area in the historical flood data of the th year, is the th year, and is the historical inflow into the lake from the intermediate area at the

[0017] ,

[0018] where, is the highest water level of the outer river in the historical flood data of the th year, is the th year, and is the historical water level of the outer river at the

[0019] Further, the method for generating the Vine Copula function in step S2 is as follows:

[0020] (1) Based on the three types of risk source factors, namely the upstream inflow flood volume, the in-lake inflow flood volume from the intermediate area, and the highest water level of the outer river, the maximum likelihood estimation method is used to fit the marginal distribution function for each type of risk source factor; the expression of the marginal distribution function of the risk source factor is:

[0021] ,

[0022] where is the vector of the th type of risk source factor, , , is the th type of risk source factor in the historical flood data of the th year, , is the year index corresponding to the historical flood data, is the total number of years of the historical flood data, is 's corresponding marginal distribution function, is the marginal distribution function value of the th type of risk source factor;

[0023] (2) Calculate the Kendall rank correlation coefficient between various risk source factors, and construct a two-dimensional Copula function based on the marginal distribution function values of the risk source factors, the Kendall rank correlation coefficient, and the Bayesian information criterion BIC. The expression is:

[0024] ,

[0025] where represents the two-dimensional Copula function of the joint distribution between the th type and the th type of risk source factors, , is the vector of the th type of risk source factor, and its marginal distribution function is , is the marginal distribution function value of the th type of risk source factor;

[0026] (3) Based on the marginal distribution function of the risk source factor and the two-dimensional Copula function, construct a conditional Copula function. The conditional Copula function is used to calculate the conditional joint distribution function of the th type of risk source factor under the condition of given the remaining th type of risk source factor. The conditional Copula function The calculation formula is:

[0027] ,

[0028] where, is the joint Copula function of risk source factors from the 1st to the th class, , , are the marginal distribution function values of the 1st, 2nd and the th class risk source factors respectively;

[0029] (4) Construct a Vine Copula structure based on two-dimensional Copula function and conditional Copula function, and use the Akaike information criterion AIC and maximum likelihood estimation method for structure selection and parameter estimation. Build a Vine Copula structure by nesting two-dimensional Copula function and conditional Copula function layer by layer. The first layer of the Vine Copula structure consists of two-dimensional Copula functions, and the second layer is connected by conditional Copula functions. The calculation formula of the Vine Copula function is:

[0030] ,

[0031] where, is the -dimensional Vine Copula function, used to describe the class risk source factors' dependence relationship, represents the iteration of the hierarchical structure, indicates the conditional Copula function of the th and the th class risk source factors under the given conditional risk source factor set , is the total number of classes of risk source factors.

[0032] Furthermore, the method for generating risk source factor samples in step S2 is:

[0033] (1) Based on the marginal distribution function of risk source factors, combined with two-dimensional Copula function, conditional Copula function and Vine Copula structure, use the Latin hypercube sampling method to generate risk source factor probability samples , , , is the probability sample of the 1st class risk source factor, is the probability sample of the 2nd class risk source factor, is the probability sample of the 3rd class risk source factor;

[0034] (2) According to the Vine Copula structure and the principle of probability mutual exclusion, the joint probability calculation formula for the probability samples of risk source factors is:

[0035] ,

[0036] where, represents , , joint probability, represents the Vine Copula function, and the input is the marginal distribution function values of the corresponding risk source factors, represents that the marginal distribution function value of the first type of risk source factor is , and the marginal distribution function values of the second and third types are 1 (the maximum value) of the Vine Copula function, represents that the marginal distribution function value of the second type of risk source factor is , and the marginal distribution function values of the first and third types are 1 of the Vine Copula function, represents that the marginal distribution function value of the third type of risk source factor is , and the marginal distribution function values of the first and second types are 1 of the Vine Copula function, represents that the marginal distribution function values of the first and second types of risk source factors are and , and the marginal distribution function value of the third type is 1 of the Vine Copula function, represents that the marginal distribution function values of the first and third types of risk source factors are and , and the marginal distribution function value of the second type is 1 of the Vine Copula function, represents that the marginal distribution function values of the second and third types of risk source factors are and , and the marginal distribution function value of the first type is 1 of the Vine Copula function, represents that the marginal distribution function values of the first, second, and third types of risk source factors are , and of the Vine Copula function;

[0037] (3) Based on the probability threshold of the extreme flood event and the joint probability of the probability samples of the risk source factors, screen the probability samples that meet the following conditions:

[0038] ,

[0039] Take the risk source factor probability samples that meet the above conditions as the risk source factor probability samples corresponding to extreme flood events;

[0040] Based on the marginal distribution function, perform inverse transformation on the risk source factor probability samples corresponding to extreme flood events to generate risk source factor samples.

[0041] Further, step S3 includes the following steps:

[0042] S31. Based on the generated risk source factor samples, determine the optimal number of clusters 𝑅 according to the elbow method, and use the K-Means method to divide the risk source factor samples;

[0043] S32. According to the risk source factors of historical flood data, calculate the variance of each type of risk source factor. Take the risk source factor with the smallest variance as the root node of the first stage, and the risk source factor with the largest variance as the leaf node of the final stage, so as to determine the position of the risk source factor in the scenario tree, predefine the structure of the scenario tree, then randomly select risk source factor samples as the initial node values of the scenario tree, and calculate the initial Euclidean distance between the risk source factor samples and the nodes of the scenario tree. Through the Neural Gas algorithm iteration optimization, dynamically adjust the node values of the scenario tree to minimize the overall Euclidean distance from the risk source factor samples to the nodes of the scenario tree. Finally, based on the optimized scenario tree, calculate the probabilities of different scenarios;

[0044] S33. Based on the divided groups of risk source factor samples, repeat step S32 to construct a scenario tree for each group separately; the structure of the scenario tree and the probability calculation of each group are based on the risk source factor sample data of that group, calculate the scenarios and corresponding probabilities in all scenario trees, and these scenarios constitute extreme flood scenarios;

[0045] S34. According to the probability threshold of extreme flood events , the divided groups, and the scenario trees constructed by each group, calculate that the probability of each scenario tree is , then under the condition that the extreme flood event occurs, the probability of the th scenario tree is ; each scenario tree contains extreme flood scenarios, then under the condition that the extreme flood event and the group occur, the probability that the extreme flood scenario occurs is . Based on the conditional probability formula, the probability of the extreme flood scenario is calculated as follows:

[0046] ,

[0047] S35. Adopt the same multiple ratio scaling method to couple the extreme flood scenario with historical flood data and construct an extreme flood hydrograph scenario; the extreme flood hydrograph scenario is used as the input data of the joint flood control optimization scheduling model for the reservoir-lake-river system to simulate the joint flood control scheduling plan under extreme flood conditions.

[0048] Furthermore, the objective function expression in step S4 is:

[0049] ,

[0050] where is the objective function of the joint flood control optimization scheduling model, is the initial storage volume of the lake at the beginning of the period, is the total scheduling duration, is the safety storage volume threshold of the lake;

[0051] The model constraint conditions include: reservoir water balance constraint, reservoir characteristic water level constraint, initial / final water level constraint, reservoir discharge capacity, outlet flow variation constraint, reservoir-lake water balance constraint and outer river water level constraint;

[0052] Input the generated extreme flood hydrograph scenario into the joint flood control optimization scheduling model of the reservoir-lake-river system, use LINGO software to calculate and solve, simulate the joint flood control scheduling plan for extreme floods, and calculate the maximum storage volume of the reservoir corresponding to each scheduling plan and the maximum storage volume of the lake .

[0053] Furthermore, the methods for evaluating the flood control risk of the reservoir and the flood control risk of the lake in step S4 are:

[0054] Evaluate the flood control risk of the reservoir:

[0055] The flood control risk of the reservoir is defined as: the weighted probability sum that the maximum storage volume of the reservoir exceeds its flood control safety threshold in all joint flood control scheduling plans for extreme floods, and its calculation formula is:

[0056] ,

[0057] where [[ID=�2]]is the flood control risk of the reservoir, is the number of scenarios, is the number of scenario trees, is the th extreme flood scenario of the The storage volume of a reservoir at the beginning of the time period, is the flood control safety threshold corresponding to the th reservoir, where is the total number of reservoirs, is the total scheduling duration, is the indicator function, and is the probability of the th extreme flood scenario of the

[0058] th scenario tree;

[0059] Evaluating the flood control risk of a lake:

[0060] The flood control risk of a lake is defined as: the weighted sum of the probabilities that the maximum storage volume of the lake exceeds its flood control safety threshold in all extreme flood joint flood control scheduling schemes, and its calculation formula is:

[0061] where is the flood control risk of the lake, is the storage volume of the lake at the beginning of the th time period under the scenario condition of the extreme flood hydrograph calculated by the th extreme flood scenario of the th scenario tree through the joint flood control optimization scheduling model, is the flood control safety threshold corresponding to the safe water level of the lake;

[0062] Evaluating the effect of risk assessment:

[0063] Based on the fitted marginal distribution function of risk source factors, Monte Carlo random sampling is used to generate random probability samples of risk source factors, and random samples of risk source factors are generated based on the inverse transformation of the marginal distribution function of risk source factors. Then, historical flood data is coupled to generate extreme flood random hydrograph scenarios, which are input into the joint flood control optimization scheduling model of the reservoir-lake-river system to calculate the Monte Carlo joint flood control scheduling scheme. Based on this scheme, reservoir flood control risk assessment and lake flood control risk assessment are carried out to obtain the Monte Carlo values of reservoir flood control risk and lake flood control risk, which are used as risk true values; the continuous ranked probability score is used to evaluate the effect of risk assessment, and the lower the value of CRPS, the better the effect of risk assessment.

[0064] The system corresponding to the method includes:

[0065] A risk source factor calculation unit, which is used to collect historical flood data and flood control project data of the reservoir-lake-river system, count historical extreme flood events, identify risk source factors that cause historical extreme flood events. The risk source factors include upstream inflow, inflow into the lake in the interval, and the highest water level of the external river, and calculate each type of risk source factor in the historical flood data of each year using the annual maximum value method;

[0066] A risk source factor sample generation unit, which is used to fit marginal distribution functions for each type of risk source factor, construct corresponding two-dimensional Copula functions and conditional Copula functions, generate a Vine Copula function, generate risk source factor probability samples through the Latin hypercube sampling method, calculate the joint probability of each group of risk source factor probability samples, screen out risk source factor probability samples with joint probabilities lower than the extreme flood event probability threshold, and use the marginal distribution function to perform inverse transformation to generate risk source factor samples;

[0067] An extreme flood hydrograph scenario generation unit, which is used to cluster risk source factor samples into different groups using the K-Means method, construct a risk source factor scenario tree for each group using the Neural Gas algorithm, calculate the extreme flood scenario probability, and generate an extreme flood hydrograph scenario by coupling historical flood data using the same ratio scaling method;

[0068] A scheduling plan generation and evaluation unit, which is used to construct a joint flood control optimization scheduling model for the reservoir-lake-river system with minimizing the average excess flood volume of the lake during the entire scheduling period as the objective function, use the extreme flood hydrograph scenario as the input data of this model, simulate the extreme flood joint flood control scheduling plan, analyze the situation where the safety storage thresholds of the reservoir and the lake are exceeded in the extreme flood joint flood control scheduling plan, evaluate the flood control risks of the reservoir and the lake, and use the continuous ranked probability score CRPS to evaluate the results of the flood control risks of the reservoir and the lake.

[0069] An electronic device for storing and executing the above method, including a memory, a processor, and a computer program / instructions stored on the memory and executable on the processor. When the computer program / instructions are executed by the processor, the steps of the extreme flood simulation and flood control risk assessment method for the reservoir-lake-river system are implemented.

[0070] A computer-readable storage medium for storing and executing the above method. The computer-readable storage medium stores computer instructions, and when the computer instructions are called, they are used to execute the steps of the extreme flood simulation and flood control risk assessment method for the reservoir-lake-river system.

[0071] Beneficial effects: Compared with the prior art, the remarkable technical effects of the present invention are as follows: By using Vine Copula to characterize the joint distribution of multi-risk source factors, the accuracy of extreme flood event simulation is significantly improved. The scenario tree is used to effectively reduce the samples of risk source factors, improving the computational efficiency of flood control risk assessment, meeting the timeliness requirements of actual scheduling, and providing reliable data support and decision-making basis for formulating scientific and efficient flood control joint scheduling strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 is a flowchart of the method of the present invention;

[0073] Figure 2 is a flowchart of the generation and structure construction of extreme flood scenarios based on Vine Copula and Neural Gas algorithm; where (a) is the Vine Copula structure, (b) is the samples of risk source factors, (c) is the structure of the predefined scenario tree, and (d) is R a scenario tree. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, some well-known technical features in the art are not described in order to avoid confusion with the present invention. The order of the relevant steps in the present invention is not restrictive, that is, those skilled in the art can adjust it. The order in the present invention is a case-by-case writing method, rather than a restrictive description.

[0075] The present invention proposes a method for simulating extreme floods and assessing flood control risks in a reservoir-lake-river system under uncertain conditions. The method first statistically analyzes historical extreme flood events, identifies and extracts risk source factors, and calculates the risk source factors in historical flood data. Subsequently, the marginal distribution functions of the risk source factors are respectively fitted, two-dimensional Copula functions and conditional Copula functions are constructed, and then the Vine Copula structure is deduced, thereby realizing the modeling of the dependence structure between multiple risk source factors, calculating the Vine Copula function and generating probability samples of the risk source factors based on this. Through joint probability calculation and inverse transformation of the marginal distribution function, risk source factor samples conforming to the characteristics of historical extreme flood events are obtained. Then, the generated risk source factor samples are clustered and divided, and the Neural Gas algorithm is used to construct a risk source factor scenario tree, and the probability of extreme flood scenarios is calculated. On this basis, historical flood data is coupled to generate extreme flood hydrograph scenarios. Finally, a joint flood control optimization scheduling model for the reservoir-lake-river system is constructed, and the joint flood control scheduling scheme for extreme floods is simulated. The flood control risks of the reservoir and the lake are evaluated through the simulation results, and the effectiveness of flood control risk assessment is analyzed. The method of the present invention has a clear structure and reasonable steps, can stably and accurately evaluate flood risks, and is particularly suitable for flood control and disaster reduction work in flood-prone areas. At the same time, the method has strong applicability, can be applied to various hydrological conditions and basins with different terrain structures, and can also provide reliable references for extreme flood simulation and risk assessment in data-deficient areas, having important engineering application value and theoretical significance.

[0076] As Figure 1 shown, a method for simulating extreme floods and assessing flood control risks in a reservoir-lake-river system of the present invention includes the following steps:

[0077] S1. Identification of risk source factors;

[0078] Collect historical flood data and flood control project data of the reservoir-lake-river system, statistically analyze historical extreme flood events, and identify the risk source factors that cause historical extreme flood events. Among them, the risk source factors include three categories: upstream inflow, inflow into the lake in the interval, and the highest water level of the external river. The annual maximum value method is used to calculate each type of risk source factor in the historical flood data of each year.

[0079] Specifically, it includes the following steps:

[0080] S11. Collection and collation of data;

[0081] Determine the research area, and collect historical flood data and flood control project data of the reservoir-lake-river system through channels such as hydrological monitoring stations, basin management agencies, and historical documents.

[0082] S12. Statistical analysis of historical extreme flood events;

[0083] Based on historical flood data, set the recurrence period according to flood control standards or risk management requirements, and use the recurrence period analysis method to calculate the probability threshold of extreme flood events. The calculation formula is as follows:

[0084] (1),

[0085] Where, is the probability threshold of extreme flood events, is the recurrence period set based on historical flood data.

[0086] According to the calculated probability threshold, count the historical extreme flood events with occurrence probabilities less than this probability threshold.

[0087] S13. Identify the risk source factors that led to the occurrence of historical extreme flood events;

[0088] Based on historical extreme flood events, identify the key factors (i.e., risk source factors) that led to the occurrence of historical extreme flood events. The identified risk source factors include: upstream reservoir inflow (referring to the total amount of flood entering the reservoir, which determines the reservoir operation pressure), flood inflow into the lake from the intermediate area (referring to the amount of flood flowing into the lake within the lake catchment area, which affects the lake water level change), and the highest water level of the external river (referring to the peak water level of the external river during the flood period, which affects the flood discharge capacity).

[0089] S14. Use the annual maximum value method to calculate the risk source factors of historical flood data;

[0090] Based on historical flood data, extract the maximum flood event that occurred each year, and calculate its corresponding upstream reservoir inflow, flood inflow into the lake from the intermediate area, and the highest water level of the external river. The calculation formulas are as follows:

[0091] (2),

[0092] Where, is the upstream reservoir inflow in the historical flood data of the th year, , is the year index corresponding to the historical flood data, is the total number of years of historical flood data, is the th reservoir's historical upstream inflow at the time period in the th year, is the total number of reservoirs, is the total operation duration, ;

[0093] (3),

[0094] Among them, is the flood volume entering the lake in the interval in the historical flood data of the th year, is the historical flow rate of the interval entering the lake in the th year at the th time period, is the vector of the flood volume entering the lake in the interval, ;

[0095] (4),

[0096] Among them, is the highest water level of the outer river in the historical flood data of the th year, is the historical water level of the outer river in the th year at the th time period, is the vector of the highest water level of the outer river, .

[0097] S2. Generate samples of risk source factors based on the Vine Copula function;

[0098] For each type of risk source factor, fit the marginal distribution function, construct the corresponding two-dimensional Copula function and conditional Copula function, generate the Vine Copula function, generate the probability samples of risk source factors through the Latin hypercube sampling method, calculate the joint probability of each group of samples, screen out the probability samples of risk source factors with joint probability lower than the probability threshold of extreme flood events, and use the marginal distribution function to perform inverse transformation on them to generate risk source factor samples.

[0099] Specifically, it includes the following steps:

[0100] S21. Fit the marginal distribution function of each type of risk source factor;

[0101] Based on the three types of risk source factors (inflow flood volume from the upper reaches, flood volume entering the lake in the interval, highest water level of the outer river) calculated in step S14, use the maximum likelihood estimation method to fit the marginal distribution function for each type of risk source factor. The candidate marginal distribution functions include the generalized extreme value distribution function, normal distribution function, exponential distribution function, gamma distribution function, and lognormal distribution function. Based on the candidate marginal distribution functions, perform optimization through the Kolmogorov-Smirnov (KS) test and Akaike information criterion (AIC), and select the candidate marginal distribution function that passes the KS test and has the minimum AIC value as the marginal distribution function of this risk source factor. The expression of the marginal distribution function of the risk source factor is as follows:

[0102] (5),

[0103] Among them, is the vector of the type risk source factors, , , is the type risk source factor in the year historical flood data, is corresponding marginal distribution function, is the marginal distribution function value of the type risk source factor, and its value range is [0, 1].

[0104] S22: Construct a two-dimensional Copula function;

[0105] Calculate the Kendall rank correlation coefficient between various risk source factors, and construct a candidate set of two-dimensional Copula functions based on the marginal distribution function values of the risk source factors and the Kendall rank correlation coefficient. The candidate set of two-dimensional Copula functions includes Clayton Copula function, Frank Copula function and Gumbel Copula function. Based on the candidate set of two-dimensional Copula functions, use the Bayesian Information Criterion (BIC) to evaluate the fitting effect of each candidate two-dimensional Copula function, and finally select the two-dimensional Copula function with the smallest BIC value as the n type and the k type two-dimensional Copula function between risk source factors. The expression of the two-dimensional Copula function is as follows:

[0106] (6),

[0107] Among them, represents the two-dimensional Copula function of the joint distribution between the type and the type risk source factors, , is the vector of the type risk source factors, and its marginal distribution function is , is the type risk source factor marginal distribution function value.

[0108] S23: Construct a conditional Copula function;

[0109] Based on the marginal distribution function of the risk source factors and the two-dimensional Copula function, construct a conditional Copula function, which is used to calculate the conditional probability of the type risk source factors given the remaining Conditional joint distribution function of class risk source factors, conditional Copula function The calculation formula is as follows:

[0110] (7),

[0111] where is the joint Copula function of the 1st to the class risk source factors, , , are the marginal distribution function values of the 1st, 2nd and the class risk source factors respectively, and the partial derivative is used to calculate the conditional Copula function value corresponding to the class risk source factor under the condition of given .

[0112] S24. Calculate the Vine Copula function;

[0113] To characterize the high-dimensional correlation structure of the three types of risk source factors, a Vine Copula structure is constructed based on two-dimensional Copula functions and conditional Copula functions. The Akaike information criterion (AIC) and the maximum likelihood estimation method are used for structure selection and parameter estimation. The Vine Copula structure is constructed by nesting two-dimensional Copula functions and conditional Copula functions layer by layer. The Vine Copula structure is as shown in Figure 2 (a) in. The first layer consists of two-dimensional Copula functions, and the second layer is connected by conditional Copula functions. The calculation formula of the Vine Copula function is as follows:

[0114] (8),

[0115] where is the -dimensional Vine Copula function, which is used to describe the dependence relationship between class risk source factors, represents the iteration of the hierarchical structure, represents the conditional Copula function of the th and the th class risk source factors under the condition of the given conditional risk source factor set , is the total number of categories of risk source factors.

[0116] S25. Generate probability samples of risk source factors;

[0117] Based on the marginal distribution function of risk source factors, combined with two-dimensional Copula function, conditional Copula function and Vine Copula structure, the Latin hypercube sampling method is used to generate probability samples of risk source factors.

[0118] Specifically, it includes the following sub-steps:

[0119] S25a. On the value range [0, 1] of the marginal distribution function value of the first risk source factor, the Latin hypercube sampling is used to generate the sample value of the marginal distribution function of the first risk source factor ;

[0120] S25b. Based on , through the two-dimensional Copula function and its corresponding conditional Copula function , calculate the sample value of the marginal distribution function of the second risk source factor ;

[0121] S25c. Given , , under the condition of , according to the corresponding conditional Copula function in the Vine Copula structure, calculate the sample value of the marginal distribution function of the third risk source factor

[0122] S25d. And so on, generate the sample values of the marginal distribution functions of risk source factors layer by layer from top to bottom in the Vine Copula structure , , …, , and the sample values of the marginal distribution functions of these risk source factors form the probability samples of risk source factors.

[0123] S26. Calculate the joint probability of the probability samples of risk source factors;

[0124] According to the Vine Copula structure and the principle of probability mutual exclusion, the joint probability calculation formula of the probability samples of risk source factors is as follows:

[0125] (9),

[0126] Where, represents the joint probability of , , , represents the Vine Copula function, and the input is the marginal distribution function value of the corresponding risk source factor, represents that the marginal distribution function value of the first type of risk source factor is For the Vine Copula function with marginal distribution function values of 1 (the maximum value) for the second and third categories, indicating that the marginal distribution function value of the risk source factor of the second category is For the Vine Copula function with marginal distribution function values of 1 for the first and third categories, indicating that the marginal distribution function value of the risk source factor of the third category is For the Vine Copula function with marginal distribution function values of 1 for the first and second categories, indicating that the marginal distribution function values of the risk source factors of the first and second categories are and For the Vine Copula function with marginal distribution function value of 1 for the third category, indicating that the marginal distribution function values of the risk source factors of the first and third categories are and For the Vine Copula function with marginal distribution function value of 1 for the second category, indicating that the marginal distribution function values of the risk source factors of the second and third categories are and For the Vine Copula function with marginal distribution function value of 1 for the first category, indicating that the marginal distribution function values of the risk source factors of the first, second, and third categories are , and of the Vine Copula function.

[0127] S27. Generate risk source factor samples based on the inverse transformation of the marginal distribution function;

[0128] Based on the probability threshold of the extreme flood event in step S12 and the joint probability of the risk source factor probability samples calculated in step S26 , screen the probability samples that meet the following conditions:

[0129] (10),

[0130] Take the risk source factor probability samples that meet the above conditions as the risk source factor probability samples corresponding to the extreme flood event.

[0131] Based on the marginal distribution function calculated in step S21, perform inverse transformation on the risk source factor probability samples corresponding to the extreme flood event to generate risk source factor samples, as shown in (b) in Figure 2 . The inverse transformation formula is:

[0132] (11),

[0133] where is the inverse function of the marginal distribution function of the -th risk source factor, is the vector of risk source factor samples, , is the -th sample of the -th type of risk source factor, , and

[0134] S3. Generate a set of extreme flood hydrograph scenarios based on the Neural Gas algorithm;

[0135] Use the K-Means method to cluster the risk source factor samples into different groups, use the Neural Gas algorithm to construct a risk source factor scenario tree for each group, calculate the extreme flood scenario probability, and use the same ratio scaling method to couple historical flood data to generate extreme flood hydrograph scenarios.

[0136] Specifically, it includes the following steps:

[0137] S31. Divide the risk source factor samples into different groups;

[0138] Based on the risk source factor samples generated in step S27, in order to facilitate subsequent modeling of the scenario tree, these samples need to be clustered and divided into different groups.

[0139] Determine the optimal number of groups 𝑅 for clustering according to the elbow method. The elbow method plots the relationship curve between the number of clusters and the within-cluster sum of squares (WCSS), and selects the number of groups corresponding to the point where the decline of WCSS begins to flatten as the optimal number of clusters. After determining 𝑅, use the K-Means method to divide the risk source factor samples. This algorithm iterates through the following sub-steps:

[0140] S31a. Initialization: Randomly select risk source factor samples as the initial group centers;

[0141] S31b. Assign samples: Assign each risk source factor sample to the group center with the closest Euclidean distance;

[0142] S31c. Update centers: Calculate the mean of the samples in each group and use this mean as the new group center;

[0143] S31d. Determine convergence: If the group centers do not change significantly, or the preset maximum number of iterations is reached, the algorithm terminates; otherwise, return to step S31b to continue iterating.

[0144] This process aims to minimize the sum of squared errors of all risk source factor samples within each group to their group centers, ensuring that the clustering results have good compactness and separation. Then, the number of samples in all groups should satisfy the following relationship:

[0145] (12),

[0146] where, is the number of risk source factor samples included in the th group.

[0147] S32. Construct a risk source factor scenario tree using the Neural Gas algorithm;

[0148] The scenario tree consists of nodes, starting from the root node in the first stage, branching backward step by step, and reaching the leaf nodes in the final stage. Each leaf node corresponds to a scenario. Based on the risk source factors of the historical flood data calculated in step S14, calculate the variance of each type of risk source factor. Take the risk source factor with the smallest variance as the root node in the first stage and the risk source factor with the largest variance as the leaf node in the final stage, thereby determining the position of the risk source factors in the scenario tree, predefined the structure of the scenario tree, then randomly select risk source factor samples as the initial node values of the scenario tree, and calculate the initial Euclidean distance between the risk source factor samples and the nodes of the scenario tree. Through iterative optimization using the Neural Gas algorithm, dynamically adjust the node values of the scenario tree to minimize the overall Euclidean distance from the risk source factor samples to the nodes of the scenario tree. Finally, based on the optimized scenario tree, calculate the probabilities of different scenarios.

[0149] Specifically, it includes the following sub-steps:

[0150] S32a. Predefine the structure of the scenario tree;

[0151] According to the Neural Gas algorithm, predefine the structure of the scenario tree. As shown in (c) in Figure 2 , the root node, intermediate nodes, and leaf nodes of the scenario tree are represented by matrices. Then, the th scenario tree generated by the th group can be represented as:

[0152] (13),

[0153] where, the columns of the matrix correspond to different nodes (i.e., the root node, intermediate nodes, and leaf nodes), each column of nodes corresponds to a type of risk source factor, and the number of rows of the matrix corresponds to the number of leaf nodes of the scenario tree, that is, each row represents a scenario. represents the th scenario tree, represents the The th scenario of a scenario tree, represents the node value of the th scenario tree, is the number of scenarios in the scenario tree, is the number of nodes included in each scenario.

[0154] S32b. Initialization of scenario tree node values;

[0155] In the th group, randomly select a risk source factor sample as the initial node value of the scenario tree, and its calculation formula is as follows:

[0156] (14),

[0157] where, is the node value of the th scenario tree, is a sample randomly selected from the risk source factor samples of the th group , , .

[0158] Step S32c. Euclidean distance calculation;

[0159] Calculate the Euclidean distance between the node value of the scenario tree and the risk source factor sample , and the calculation formula is as follows:

[0160] (15),

[0161] where, is the Euclidean distance between the node value of the scenario tree and the risk source factor sample , after calculating the Euclidean distance, sort the Euclidean distance in descending order, and record the sorted Euclidean distance sequence in the array for subsequent node value updates.

[0162] Step S32d. Iteratively update the node values of the scenario tree based on the Neural Gas algorithm;

[0163] To minimize the distance between the risk source factor sample and the node values of the constructed scenario tree, adjust the node values of the scenario tree to make it gradually approach the risk source factor sample. By iteratively updating the node values of the scenario tree, correct according to the Euclidean distance sorting of each node value and the risk source factor sample and the number of iterations. The adjustment process follows the following formula:

[0164] (16),

[0165] (17),

[0166] (18),

[0167] (19),

[0168] Wherein, is the node value of the scenario tree is the adjustment amount during one iteration process, is the step size function, which determines the overall step size of the change of the node values of all scenario trees in each iteration, and are step size parameters (the initial step size is set to 5, and the final step size is set to 1), is the current iteration number, , and the step size decreases as the iteration number increases, is the maximum iteration number ( is set to 10000), is the local step size function, which determines the step size of the change of the node value of a single scenario tree, and are adaptation parameters, is the adaptation function, which provides an adaptation value for the node value of each scenario tree according to the magnitude order of the Euclidean distance between the node value of the scenario tree and the risk source factor sample .

[0169] At the iteration number of , the node value of the scenario tree is updated according to , and the update formula is as follows:

[0170] (20),

[0171] Wherein, is the node value of the -th iteration of the -th scenario tree, is the node value of the -th iteration of the -th scenario tree.

[0172] Repeat step S32c and step S32d. In each iteration, the node values of the scenario tree need to be adjusted according to these formulas until the maximum iteration number is reached.

[0173] Step S32e: Calculate the probability of each scenario in the scenario tree;

[0174] Calculate the probability of each scenario in the scenario tree based on the ratio of the number of risk source factor samples closest to each scenario to the total number of risk source factor samples in the group. The calculation formula is as follows:

[0175] (21),

[0176] where, is the probability of the th scenario in the th scenario tree, is a counting function that calculates the number of that meet the conditions, , is the index of the risk source factor sample, that is, there are risk source factor samples in the th group, is the Euclidean distance, is the total number of scenarios in the scenario tree.

[0177] S33: Generate extreme flood scenarios;

[0178] Based on the risk source factor sample groups divided in step S31, repeat step S32 to construct a scenario tree for each group separately, as shown in (d) of Figure 2 . The scenario tree structure and the calculation of scenario probabilities for each group are based on the risk source factor sample data of that group. Calculate the scenarios and their corresponding probabilities in all scenario trees. These scenarios constitute extreme flood scenarios.

[0179] S34: Calculate the probability of extreme flood scenarios;

[0180] According to the probability threshold of extreme flood events in step S12, the groups divided in step S31, and the scenario trees constructed according to each group in step S33, calculate the probability of each scenario tree as . Then, under the condition that an extreme flood event occurs, the probability of the th scenario tree is . Each scenario tree contains extreme flood scenarios. Then, under the condition that extreme flood event and group occur, the probability of extreme flood scenario occurring is . Based on the conditional probability formula, the probability of extreme flood scenario is calculated as follows:

[0181] (twenty two),

[0182] S35, generating extreme flood process line scenarios;

[0183] The spatiotemporal distribution of extreme flood events is subject to significant uncertainty. Due to the scarcity of extreme flood events in historical flood data, it is difficult to accurately characterize and calibrate the uncertainty in their spatiotemporal distribution. Extreme flood scenarios can only reflect the uncertainty in their spatial distribution, but fail to capture the temporal evolution of extreme flood events. This also makes it difficult to meet the flood hydrograph input data requirements of the joint flood control optimization scheduling model. Therefore, a scaling method was used to couple extreme flood scenarios with historical flood data to construct extreme flood hydrograph scenarios. This method comprehensively characterizes the uncertainty in the temporal and spatial distribution of extreme flood events, providing reliable support for subsequent extreme flood simulations and flood control risk assessments.

[0184] It includes the following sub-steps:

[0185] Step S35a: Based on extreme flood scenarios The root node value (i.e. the highest water level of the outer river) and step S14 In the Historical outer river water level during the period , using the same-ratio scaling method to calculate extreme flood scenarios No. In the Water level of outer river during the period , the calculation formula is as follows:

[0186] (twenty three),

[0187] Step S35b: Based on extreme flood scenarios The intermediate node value (i.e. upstream inflow volume) and step S14 The year The reservoir is in Historical upstream inflow flow during the period , using the same-ratio scaling method to calculate extreme flood scenarios No. The year The reservoir is in Upstream inflow during the period , the formula is as follows:

[0188] (twenty four),

[0189] in, The unit conversion coefficient for flow to flood volume.

[0190] Step S35c: Based on the extreme flood scenario of the leaf node value (i.e., the inflow volume into the lake in the interval) and the historical inflow into the lake in the interval in the th year at the time period in step S14 , using the same multiple scaling method, calculate the inflow into the lake in the interval in the th year at the th time period of the extreme flood scenario . The calculation formula is as follows:

[0191] (25),

[0192] Step S35d: According to the upstream reservoir inflow, the inflow in the interval, and the river channel water level obtained in steps S34a, S34b, and S34c, arranging them in chronological order can form the process line data under the extreme flood scenario, thus forming the extreme flood process line scenario. Since the extreme flood process line scenario is calculated based on the scenario and historical flood data, the generated extreme flood process line scenario not only conforms to the statistical characteristics of historical flood data but also inherits the probability of the extreme flood scenario . These extreme flood process line scenarios are used as the input data for the joint flood control optimization scheduling model of the reservoir-lake-river system to simulate the joint flood control scheduling plan under extreme flood conditions.

[0193] S4. Simulation of extreme flood events and flood control risk assessment;

[0194] Construct a joint flood control optimization scheduling model for the reservoir-lake-river system, use the extreme flood process line scenario as the input data for this model to simulate the joint flood control scheduling plan for extreme floods, analyze the situations exceeding the safety storage thresholds of the reservoir and the lake in the joint flood control scheduling plan for extreme floods, evaluate the flood control risks of the reservoir and the lake, and use the continuous ranked probability score (CRPS) to evaluate the results of the flood control risks of the reservoir and the lake.

[0195] Specifically, it includes the following steps:

[0196] S41. Construct a joint flood control optimization scheduling model for the reservoir-lake-river system;

[0197] The objective function of the joint flood control optimization scheduling model for the reservoir-lake-river system is to minimize the average excess flood volume of the lake during the entire scheduling period. The calculation formula is:

[0198] (26),

[0199] Among them, is the objective function of the joint flood control optimal operation model, is the storage volume of the lake at the beginning of time period, is the total operation duration, is the safety storage threshold of the lake.

[0200] The constraint conditions of the joint flood control optimal operation model of the reservoir-lake-river system are as follows:

[0201] (a) Reservoir water balance constraint:

[0202] (27),

[0203] Among them, is the storage volume of the th reservoir at the beginning of time period, is the storage volume of the th reservoir at the beginning of time period, and are the inflow and outflow discharges of the th reservoir during time period respectively, is the time step.

[0204] (b) Reservoir characteristic water level constraint:

[0205] (28),

[0206] Among them, is the water level of the th reservoir during time period, and are the maximum and minimum restricted water levels of the th reservoir during time period respectively.

[0207] (c) Initial / final water level constraint:

[0208] (29),

[0209] Among them, is the water level of the th reservoir at the initial time period , is the water level of the th reservoir at the final time period , and are the The initial water level and the final water level of the reservoir set according to the flood control operation requirements.

[0210] (d)Discharge capacity of the reservoir:

[0211] (30),

[0212] where, is the water level of the th reservoir at the time period corresponding to the discharging capacity.

[0213] (e)Constraint on the variation range of the outflow variable:

[0214] (31),

[0215] where, is the limit value of the variation range of the outflow between adjacent time periods of the th reservoir, is the outflow of the th reservoir at the time period.

[0216] (f)Constraint on the water volume balance between the reservoir and the lake:

[0217] (32),

[0218] where, is the storage volume of the lake at the beginning of the time period, is the storage volume of the lake at the beginning of the time period, is the outflow of the th reservoir at the time period, and are respectively the inflow into the lake from the watershed and the outflow from the lake during the time period.

[0219] (g)Constraint on the water level of the external river:

[0220] (33),

[0221] where, is the outflow of the lake during the time period, is the water level of the lake during the time period, is the water level of the external river during the , when the water level of the outer river is higher than that of the lake, the outflow of the lake is equal to 0.

[0222] S42: Simulate the joint flood control operation plan for extreme floods;

[0223] Input the extreme flood hydrograph scenarios generated in step S35 into the joint flood control optimization model of the reservoir-lake-river system, calculate and solve using LINGO software, simulate the joint flood control operation plan for extreme floods, and calculate the maximum storage volume of the reservoir corresponding to each operation plan and the maximum storage volume of the lake . Since the generation of these joint flood control operation plans is based on extreme flood hydrograph scenarios, each operation plan not only reflects the operation characteristics of different extreme flood hydrograph scenarios but also inherits the probability attributes of the extreme flood hydrograph scenarios.

[0224] S43: Evaluate the flood control risk of the reservoir;

[0225] The flood control risk of the reservoir is measured by the probability that the maximum storage volume of the reservoir during the operation period exceeds its corresponding flood control safety threshold. Therefore, the flood control risk of the reservoir can be defined as: the weighted sum of probabilities that the maximum storage volume of the reservoir exceeds its flood control safety threshold among all joint flood control operation plans for extreme floods. The calculation formula is:

[0226] (34),

[0227] where, is the flood control risk of the reservoir, is the number of scenarios, is the number of scenario trees, is the th scenario tree, the th extreme flood scenario, the storage volume of the th reservoir at the beginning of the period calculated by the joint flood control optimization model for the extreme flood joint flood control operation plan, is the th reservoir's corresponding flood control safety threshold, , is the total number of reservoirs, is the th scenario tree, the th extreme flood scenario's probability, is the indicator function (equal to 1 when the evaluated event is true, otherwise equal to 0).

[0228] S44: Evaluate the flood control risk of the lake;

[0229] The flood control risk of a lake is measured by the probability that the maximum water storage of the lake during the scheduling period exceeds its corresponding flood control safety threshold. Therefore, the flood control risk of a lake can be defined as: the sum of the weighted probabilities that the maximum water storage of the lake exceeds its flood control safety threshold in all extreme flood joint flood control scheduling schemes. The calculation formula is as follows:

[0230] (35),

[0231] where, is the flood control risk of the lake, is the th scenario tree, the water storage of the lake at the beginning of the th extreme flood scenario calculated by the joint flood control optimization scheduling model under the extreme flood hydrograph scenario of the th period, is the flood control safety threshold corresponding to the safety water level of the lake.

[0232] S45: Evaluate the effect of risk assessment;

[0233] To verify the stability and reliability of the risk assessment results, the risk assessment value is compared with the risk true value. Since it is difficult to directly obtain the true values of the flood control risks of reservoirs and lakes, based on the marginal distribution function of risk source factors fitted in step S21, Monte Carlo random sampling is used to generate random probability samples of risk source factors, and random samples of risk source factors are generated based on the inverse transformation of the marginal distribution function of risk source factors. Then, historical flood data is coupled to generate extreme flood random hydrograph scenarios, which are input into the joint flood control optimization scheduling model of the reservoir-lake-river system to calculate the Monte Carlo joint flood control scheduling scheme. Based on this scheme, the flood control risk assessment of the reservoir and the flood control risk assessment of the lake are carried out to obtain the Monte Carlo value of the flood control risk of the reservoir ( ) and the Monte Carlo value of the flood control risk of the lake ( ), which are used as the risk true values. The continuous ranked probability score (CRPS) is used to evaluate the effect of risk assessment. The calculation formula is as follows:

[0234] (36),

[0235] (37),

[0236] where, is the continuous ranking score function, is the cumulative probability distribution function, is the Monte Carlo value of risk assessment, is the numerical value of risk assessment, is the cumulative probability distribution function of the risk assessment value, is the The Heaviside function of Class is the th assessment of the class risk. is the th assessment of the class risk Heaviside function. The lower the CRPS value, the better the risk assessment effect.

[0237] The system described in this invention includes:

[0238] A risk source factor calculation unit, which is used to collect historical flood data and flood control project data of the reservoir-lake-river system, count historical extreme flood events, identify risk source factors that cause historical extreme flood events. The risk source factors include upstream inflow, inflow into the lake in the interval, and the highest water level of the outer river, and calculate each type of risk source factor in the historical flood data of each year using the annual maximum value method;

[0239] A risk source factor sample generation unit, which is used to fit marginal distribution functions for each type of risk source factor, construct corresponding two-dimensional Copula functions and conditional Copula functions, generate a Vine Copula function, generate risk source factor probability samples through the Latin hypercube sampling method, calculate the joint probability of each group of risk source factor probability samples, screen out risk source factor probability samples with a joint probability lower than the extreme flood event probability threshold, and use the marginal distribution function to perform inverse transformation to generate risk source factor samples;

[0240] An extreme flood hydrograph scenario generation unit, which is used to cluster risk source factor samples into different groups using the K-Means method, construct a risk source factor scenario tree for each group using the Neural Gas algorithm, calculate the extreme flood scenario probability, and generate an extreme flood hydrograph scenario by coupling historical flood data using the same ratio scaling method;

[0241] A scheduling plan generation and evaluation unit, which is used to construct a joint flood control optimization scheduling model for the reservoir-lake-river system with minimizing the average excess flood volume of the lake during the entire scheduling period as the objective function, use the extreme flood hydrograph scenario as the input data of this model, simulate the extreme flood joint flood control scheduling plan, analyze the situation where the safety storage thresholds of the reservoir and the lake are exceeded in the extreme flood joint flood control scheduling plan, evaluate the reservoir flood risk and the lake flood risk, and use the continuous ranked probability score CRPS to evaluate the results of the reservoir flood risk and the lake flood risk.

[0242] The electronic device according to the present invention includes a memory, a processor, and a computer program / instructions stored on the memory and executable on the processor. When the computer program / instructions are executed by the processor, the steps of the extreme flood simulation and flood control risk assessment method for the reservoir-lake-river system are implemented.

[0243] The computer-readable storage medium according to the present invention stores computer instructions. When the computer instructions are called, they are used to execute the steps of the extreme flood simulation and flood control risk assessment method for the reservoir-lake-river system.

Claims

1. A method for extreme flood simulation and flood control risk assessment in a reservoir-lake-river system, characterized by: The following steps are involved: S1. Collect historical flood data and flood control project data for the reservoir-lake-river system, count historical extreme flood events, and identify the risk factors that lead to historical extreme flood events. Risk factors include upstream flood inflows, intermediate flood inflows into lakes, and the highest water level of external rivers. Use the annual maximum value method to calculate each risk factor in the historical flood data for each year. S2. Fit the marginal distribution function for each type of risk source factor, construct the corresponding two-dimensional Copula function and conditional Copula function, generate the Vine Copula function, generate the risk source factor probability sample through the Latin hypercube sampling method, calculate the joint probability of each group of risk source factor probability samples, screen out the risk source factor probability samples whose joint probability is lower than the extreme flood event probability threshold, and use the marginal distribution function to perform the inverse transformation to generate the risk source factor sample; S3. Use the K-Means method to cluster the risk source factor samples into different groups. Use the Neural Gas algorithm to construct a risk source factor scenario tree for each group, calculate the probability of extreme flood scenarios, and use the same-ratio scaling method to couple historical flood data to generate extreme flood process line scenarios. The specific method of using the Neural Gas algorithm to construct the risk source factor scenario tree is as follows: S32a, according to the Neural Gas algorithm, predefine the structure of the scene tree, the root node, the intermediate node and the leaf node of the scene tree are represented by a matrix, then The first group generated A scene tree is represented as: ， The columns of the matrix correspond to different nodes, each column node corresponds to a type of risk source factor, and the number of rows of the matrix corresponds to the number of scenario tree child nodes, that is, each row represents a scenario. Representative A scene tree, Representative The first scene tree A scene, For the The node value of a scene tree, is the number of scenes in the scene tree, The number of nodes included in each scene; S32b, in In the groups, randomly select risk source factor samples as the initial node values of the scenario tree. The formula is: ， in, It is from Risk factor samples for each group A randomly selected sample from , , For the The number of risk factor samples included in each group; Step S32c: Calculate the node value of the scene tree To the risk source factor sample Euclidean distance , the formula is: ， in, is the index of the risk source factor sample, , the Euclidean distance Arrange in descending order and record the sorted Euclidean distance sequence in an array In, it is used for subsequent node value updates; Step S32d, iteratively updating the node values of the scene tree based on the Neural Gas algorithm; ， in, Node value of the scene tree The amount of adjustment during an iteration, is the step size function, and is the step size parameter, is the current iteration number, , is the maximum number of iterations, It is a local step function that determines the step size of the node value change of a single scene tree. and To adapt the parameters, is the adaptation function, is the total number of scenes in the scene tree; At the number of iterations When the node value of the scene tree according to Update, the update formula is: ， in, For the The first iteration The node value of a scene tree, For the The first iteration The node value of a scene tree; Repeat steps S32c and S32d, and in each iteration, adjust the node values of the scene tree according to these formulas until the maximum number of iterations is reached; Step S32e: Calculate the probability of each scenario in the scenario tree based on the ratio of the number of risk source factor samples closest to each scenario to the total number of risk source factor samples in the group. The calculation formula is: ， in, For the A scene tree scene The probability of is a counting function that calculates the number of the number of S4. A reservoir-lake-river system joint flood control optimization scheduling model is constructed with the objective function of minimizing the average excess flood volume of lakes during the entire scheduling period. Extreme flood process line scenarios are used as input data for the model to simulate extreme flood joint flood control scheduling plans. The situations in which the safety storage capacity of reservoirs and lakes is exceeded in the extreme flood joint flood control scheduling plans are analyzed. The reservoir flood control risk and lake flood control risk are assessed, and the continuous ranking probability score (CRPS) is used to evaluate the results of reservoir flood control risk and lake flood control risk.

2. The method for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system according to claim 1, characterized in that: In step S1, the annual maximum value method is used to calculate each type of risk source factor in the historical flood data of each year, including: Based on historical flood data, the largest flood event that occurred in each year was extracted, and the corresponding upstream flood volume, interval flood volume into the lake, and maximum water level of the outer river were calculated. The calculation formula is as follows: ， in, For the The flood volume entering the reservoir from the upper reaches in the historical flood data of 2017 is: For the The year The reservoir is in Historical upstream inflow flow during the period, is the total number of reservoirs, is the total dispatching time; ， in, For the The flood volume entering the lake in the middle interval of the historical flood data is: For the In the Historical interval inflow to the lake during the time period; ， in, For the The highest water level of the Waihe River in the historical flood data of 2017, For the In the The historical outer river water level during the period.

3. The method for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system according to claim 1, characterized in that: The Vine Copula function generation method in step S2 is: (1) Based on the three risk source factors, namely, upstream flood volume, interval flood volume and highest water level of the outer river, the maximum likelihood estimation method is used to fit the marginal distribution function of each risk source factor. The marginal distribution function expression of the risk source factor is: ， in, For the A vector of risk-source factors, , , For the The historical flood data of Risk source factors, , is the year index corresponding to the historical flood data, is the total number of years of historical flood data, for The corresponding marginal distribution function is, For the The marginal distribution function value of the risk source factor; (2) Calculate the Kendall rank correlation coefficient between various risk source factors, and construct a two-dimensional Copula function based on the marginal distribution function value, Kendall rank correlation coefficient and Bayesian information criterion BIC of the risk source factors. The expression is: ， in, Indicates the Class and Two-dimensional Copula function of the joint distribution between risk source factors, , For the The vector of risk source factors, whose marginal distribution function is , For the The marginal distribution function value of the risk source factor; (3) Based on the marginal distribution function of the risk source factor and the two-dimensional Copula function, a conditional Copula function is constructed. The conditional Copula function is used to calculate the Under the conditions of risk source factors, Conditional joint distribution function of risk-like factors, conditional Copula function The calculation formula is: ， in, Category 1 to Joint Copula function of risk source factors, 、 、 They are classified into Class 1, Class 2 and Class The marginal distribution function value of the risk source factor; (4) The Vine Copula structure is constructed based on the two-dimensional Copula function and the conditional Copula function. The Akaike information criterion (AIC) and the maximum likelihood estimation method are used for structure selection and parameter estimation. The Vine Copula structure is constructed by nesting the two-dimensional Copula function and the conditional Copula function layer by layer. The first layer of the Vine Copula structure is composed of the two-dimensional Copula function, and the second layer is connected by the conditional Copula function. The calculation formula of the Vine Copula function is: ， in, for dimensional Vine Copula function, used to describe The dependencies between risk source factors, represents an iteration of a hierarchical structure, Represents the risk source factor set under given conditions Next, Class and Conditional Copula function of risk source factor, is the total number of categories of risk source factors.

4. The method for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system according to claim 1, characterized in that: The method for generating risk source factor samples in step S2 is: (1) Based on the marginal distribution function of risk source factors, combined with the two-dimensional Copula function, conditional Copula function and VineCopula structure, the Latin hypercube sampling method is used to generate the probability sample of risk source factors. 、 、 , is the probability sample of the first type of risk source factor, is the probability sample of the second type of risk source factor, It is the probability sample of the third type of risk source factors; (2) According to the Vine Copula structure and the probability mutual exclusion principle, the formula for calculating the joint probability of the risk source factor probability sample is: ， in, express 、 、 The joint probability of Represents the Vine Copula function, and the input is the marginal distribution function value of the corresponding risk source factor. The marginal distribution function value of the first type of risk source factor is , Vine Copula function with marginal distribution function value of 1 for category 2 and category 3, The marginal distribution function value of the second type of risk source factor is , Vine Copula function with marginal distribution function value of 1 for category 1 and category 3, The marginal distribution function value of the third type of risk source factor is , VineCopula function with marginal distribution function value of 1 for category 1 and category 2, The marginal distribution function values of the risk source factors of category 1 and category 2 are and , the Vine Copula function with a marginal distribution function value of 1 for the third category, The marginal distribution function values of the risk source factors of category 1 and category 3 are and , the Vine Copula function with a marginal distribution function value of 1 for the second category, The marginal distribution function values of the risk source factors of category 2 and category 3 are and , the VineCopula function with a marginal distribution function value of 1 for the first category, The marginal distribution function values of the risk source factors of category 1, category 2 and category 3 are 、 and Vine Copula function; (3) Probability threshold based on extreme flood events and the joint probability of the risk source factor probability sample , filter the probability samples that meet the following conditions: ， The probability samples of risk source factors that meet the above conditions are used as the probability samples of risk source factors corresponding to extreme flood events; Based on the marginal distribution function, the risk source factor probability samples corresponding to extreme flood events are inversely transformed to generate risk source factor samples.

5. The method for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system according to claim 1, characterized in that: Step S3 includes the following steps: S31. Based on the generated risk source factor samples, the optimal number of clustering groups 𝑅 is determined according to the elbow rule, and the risk source factor samples are divided using the K-Means method; S32. Repeat the steps of constructing a risk source factor scenario tree using the Neural Gas algorithm based on the divided risk source factor sample groups, and construct a separate scenario tree for each group. The scenario tree structure and scenario probability calculation for each group are performed based on the risk source factor sample data of the group. All scenarios in the scenario tree and their corresponding probabilities are calculated. These scenarios constitute an extreme flood scenario. S33. Based on the probability threshold of extreme flood events , divided groups and scene trees constructed by each group, calculate the probability of each scene tree as , then under the conditions of extreme flood events The probability of a scene tree is ; Each scene tree contains extreme flood scenarios, then extreme flood events and groups Under the conditions of occurrence, extreme flood scenarios The probability of occurrence is , based on the conditional probability formula, the probability expression of the extreme flood scenario is: ， in, is the probability of extreme flood scenarios; S34. Use the same-ratio scaling method to couple extreme flood scenarios with historical flood data to construct extreme flood process line scenarios. Extreme flood process line scenarios serve as input data for the joint flood control optimization scheduling model of the reservoir-lake-river system to simulate the joint flood control scheduling plan under extreme flood conditions.

6. The method for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system according to claim 1, characterized in that: The objective function expression in step S4 is: ， in, is the objective function of the joint flood control optimization scheduling model, For the lake The storage capacity at the beginning of the period, is the total scheduling time, is the safe storage threshold of the lake; The model constraints include: reservoir water balance constraint, reservoir characteristic water level constraint, initial / final water level constraint, reservoir discharge capacity, outflow amplitude constraint, reservoir-lake water balance constraint, and external river water level constraint; The generated extreme flood process line scenario is input into the joint flood control optimization scheduling model of the reservoir-lake-river system. The LINGO software is used to calculate and solve the extreme flood joint flood control scheduling scheme, and the maximum storage capacity of the reservoir corresponding to each scheduling scheme is calculated. and the maximum storage capacity of the lake .

7. The method for extreme flood simulation and flood control risk assessment of a reservoir-lake-river system according to claim 1, characterized in that: The reservoir flood control risk and lake flood control risk assessment method in step S4 is: Assessing reservoir flood risk: The flood control risk of a reservoir is defined as the sum of the weighted probabilities that the maximum storage capacity of a reservoir exceeds its flood control safety threshold in all extreme flood joint flood control scheduling schemes. The calculation formula is: ， in, To prevent flood risks in reservoirs, is the number of scenes, is the number of scene trees, For the Scene Tree The first extreme flood joint flood control scheduling scheme is calculated by the joint flood control optimization scheduling model for the extreme flood scenario. Reservoirs in The storage capacity at the beginning of the period, For the The flood control safety threshold corresponding to the reservoir, , is the total number of reservoirs, is the total scheduling time, is the indicator function, For the Scene Tree the probability of an extreme flood scenario; Assessing lake flood risk: The flood control risk of a lake is defined as the sum of the weighted probabilities that the maximum storage capacity of a lake exceeds its flood control safety threshold in all extreme flood joint flood control scheduling schemes. The calculation formula is: ， in, To prevent lake flooding risks, For the Scene Tree The extreme flood scenario is calculated by the joint flood control optimization scheduling model under the extreme flood process line scenario conditions. The lake storage at the beginning of the period, is the flood control safety threshold corresponding to the safe water level of the lake; Evaluate the effectiveness of risk assessment: Based on the fitted marginal distribution function of risk source factors, Monte Carlo random sampling is used to generate random probability samples of risk source factors, and random samples of risk source factors are generated based on the inverse transformation of the marginal distribution function of risk source factors. Then, historical flood data are coupled to generate extreme flood random process line scenarios, which are input into the joint flood control optimization scheduling model of the reservoir-lake-river system, and the Monte Carlo joint flood control scheduling scheme is calculated. Based on the scheme, reservoir flood control risk assessment and lake flood control risk assessment are carried out, and the Monte Carlo values of reservoir flood control risk and lake flood control risk are obtained, which are used as the true value of risk. The continuous ranking probability score is used to evaluate the effect of risk assessment. The lower the CRPS value, the better the effect of risk assessment.

8. A reservoir-lake-river system extreme flood simulation and flood control risk assessment system, characterized by: include: The risk source factor calculation unit is used to collect historical flood data and flood control project data of the reservoir-lake-river system, count historical extreme flood events, and identify the risk source factors that lead to historical extreme flood events. Risk source factors include upstream flood inflow, interval flood inflow into the lake, and the highest water level of the external river. The annual maximum value method is used to calculate each type of risk source factor in the historical flood data each year; The risk source factor sample generation unit is used to fit the marginal distribution function for each type of risk source factor, construct the corresponding two-dimensional Copula function and conditional Copula function, generate the Vine Copula function, generate the risk source factor probability sample through the Latin hypercube sampling method, calculate the joint probability of each group of risk source factor probability samples, screen out the risk source factor probability samples whose joint probability is lower than the extreme flood event probability threshold, and use the marginal distribution function to perform the inverse transformation to generate the risk source factor sample; The extreme flood process line scenario generation unit is used to cluster risk source factor samples into different groups using the K-Means method, construct a risk source factor scenario tree for each group using the Neural Gas algorithm, calculate the probability of extreme flood scenarios, and use the same-ratio scaling method to couple historical flood data to generate extreme flood process line scenarios; The Neural Gas algorithm is used to construct a risk source factor scenario tree, including: The scene tree structure pre-defined module is used to pre-define the structure of the scene tree according to the Neural Gas algorithm. The root node, intermediate node and leaf node of the scene tree are represented by a matrix. The first group generated A scene tree is represented as: ， The columns of the matrix correspond to different nodes, each column node corresponds to a type of risk source factor, and the number of rows of the matrix corresponds to the number of scenario tree child nodes, that is, each row represents a scenario. Representative A scene tree, Representative The first scene tree A scene, For the The node value of a scene tree, is the number of scenes in the scene tree, The number of nodes included in each scene; Scene tree node value initialization module, used in the In the groups, randomly select risk source factor samples as the initial node values of the scenario tree. The formula is: ， in, It is from Risk factor samples for each group A randomly selected sample from , , For the The number of risk factor samples included in each group; Euclidean distance calculation module, used to calculate the node value of the scene tree To the risk source factor sample Euclidean distance , the formula is: ， in, is the index of the risk source factor sample, , the Euclidean distance Arrange in descending order and record the sorted Euclidean distance sequence in an array In, it is used for subsequent node value updates; The node value update module of the scene tree is used to iteratively update the node values of the scene tree based on the Neural Gas algorithm; ， in, Node value of the scene tree The amount of adjustment during an iteration, is the step size function, and is the step size parameter, is the current iteration number, , is the maximum number of iterations, It is a local step function that determines the step size of the node value change of a single scene tree. and To adapt the parameters, is the adaptation function, is the total number of scenes in the scene tree; At the number of iterations When the node value of the scene tree according to Update, the update formula is: ， in, For the The first iteration The node value of a scene tree, For the The first iteration The node value of a scene tree; Repeat the Euclidean distance calculation module and the scene tree node value update module. In each iteration, adjust the node value of the scene tree according to these formulas until the maximum number of iterations is reached; The scenario probability calculation module is used to calculate the probability of each scenario in the scenario tree based on the ratio of the number of risk source factor samples closest to each scenario to the total number of risk source factor samples in the group. The calculation formula is: ， in, For the A scene tree scene The probability of is a counting function that calculates the number of the number of The scheduling scheme generation and evaluation unit is used to construct a reservoir-lake-river system joint flood control optimization scheduling model with the objective function of minimizing the average excess flood volume of lakes during the entire scheduling period. The extreme flood process line scenario is used as the input data of the model to simulate the extreme flood joint flood control scheduling scheme, analyze the situation where the extreme flood joint flood control scheduling scheme exceeds the safe storage capacity threshold of reservoirs and lakes, evaluate the reservoir flood control risk and lake flood control risk, and use the continuous ranking probability score CRPS to evaluate the results of reservoir flood control risk and lake flood control risk.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program / instruction stored in the memory and executable on the processor, wherein when the computer program / instruction is executed by the processor, the steps of the method for extreme flood simulation and flood risk assessment of a reservoir-lake-river system are implemented according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, which, when called, are used to execute the steps of the reservoir-lake-river system extreme flood simulation and flood control risk assessment method according to any one of claims 1 to 7.

Images