A reservoir flood control optimization scheduling method considering forecast uncertainty based on digital twin

Abstract

The present invention discloses a reservoir flood control optimization scheduling method based on digital twins considering forecast uncertainty, the method comprising: constructing a digital twin scene of a reservoir flood control scheduling system; obtaining physical data of the flood control scheduling system and virtual data under the digital twin scene, and updating the digital twin scene in real time based on the physical data; constructing a flood forecasting model based on the real-time updated digital twin scene of the flood control scheduling system in combination with a variety of machine learning algorithms; constructing a reservoir flood control optimization scheduling model in combination with the uncertainty of the prediction results of each flood forecasting model, and solving the flood control optimization scheduling strategy; constructing a flood risk assessment system, analyzing and assessing the flood control risk according to the forecast results and the flood control optimization scheduling strategy, and guiding the actual flood control scheduling. The present invention reduces the risk brought by the scheduling strategy generated under uncertain flood forecast conditions, ensures the safety of the upstream and downstream of the reservoir, and can effectively reduce the flood risk.

Description

Technical Field

[0001] The present invention belongs to the field of reservoir flood control scheduling, and specifically relates to a reservoir flood control optimization scheduling method based on digital twins and considering forecast uncertainty. Background Art

[0002] As one of the most dangerous natural disasters, floods often affect the normal social life of human beings. In recent years, under the influence of factors such as the global monsoon climate, the frequency and intensity of flood disasters in China have continued to increase, seriously affecting economic development and threatening human survival. People build reservoirs and other hydraulic structures to intercept floods, protect the safety of upstream and downstream, and reduce the losses caused by floods. Reservoir flood control dispatch, as a non-engineering measure, uses the regulation and control capabilities of reservoirs to control and regulate floods in a planned manner, and constantly seeks a balance between risk control and water resource utilization. In order to better avoid flood risks, forecast dispatch can use forecast information to judge the magnitude of floods in advance, determine the pre-discharge time of reservoirs, free up flood control storage capacity, and effectively intercept floods, so as to achieve the purpose of improving the utilization rate of flood resources. However, due to the limitations of the current technical level, flood forecasting has a high degree of uncertainty, which in turn brings potential risks to reservoir flood control dispatching. Therefore, it is very necessary and of practical significance to study reservoir flood control dispatching under uncertain flood forecasts.

[0003] Traditional dispatching systems have poor compatibility and sometimes experience delays, making it difficult to collect real-time rainfall, water level, flood volume and other data. In addition, the rehearsal of the dispatching plan cannot obtain the operating conditions of various water conservancy projects in the current river basin, and therefore cannot simulate actual dispatching and lacks true reliability, affecting the accuracy of flood forecasts and risk analysis of dispatching plans. Summary of the invention

[0004] In view of the shortcomings of the existing technology, the present invention proposes a reservoir flood control optimization scheduling method based on digital twins and taking into account forecast uncertainty.

[0005] To achieve the above technical problems, the technical solution of the present invention is:

[0006] A reservoir flood control optimization dispatching method based on digital twins considering forecast uncertainty includes the following steps:

[0007] S1. Construct a digital twin scenario of the reservoir flood control and dispatching system.

[0008] S2. Obtain the physical data of the reservoir flood control and dispatching system and the virtual data in the digital twin scenario, and update the digital twin scenario of the reservoir flood control and dispatching system in real time based on the physical data.

[0009] S3. Based on the real-time updated digital twin scenario of the reservoir flood control and dispatching system, various flood forecasting models are constructed in combination with a variety of machine learning algorithms.

[0010] S4. Considering the uncertainty of the prediction results of various flood forecasting models, a reservoir flood control optimization scheduling model is constructed, and a multi-objective genetic algorithm is used to solve it to obtain the optimized flood control scheduling strategy.

[0011] S5. Build a flood risk assessment system, analyze and evaluate flood control risks based on prediction results and optimized flood control scheduling strategies, and guide actual flood control scheduling.

[0012] Furthermore, the digital twin scenario of constructing the reservoir flood control dispatching system described in step S1 includes the following contents:

[0013] S11. Establish the physical module of the reservoir flood control and dispatching system.

[0014] S12. Based on the physical module of the reservoir flood control dispatching system and the existing reservoir flood control dispatching control operation plan, the dispatching rules are extracted and the operation module of the reservoir flood control dispatching system is established.

[0015] S13. Construct a digital twin scenario of the reservoir flood control and dispatching system, including the operating status of the reservoir before and after the flood season and during the flood, specifically including the pre-discharge process, flood discharge process, water level operation change process, gate or flood discharge tunnel opening and closing status, and the flow process of important flood control sections downstream of the reservoir.

[0016] Furthermore, the physical module of the reservoir flood control and dispatching system in step S11 is specifically constructed based on the meteorological data, hydrological data, topographic data of the basin where the reservoir is located, and the engineering properties and design standards of the reservoir itself, to construct a three-dimensional physical module.

[0017] Furthermore, the extracted dispatching rules described in step S12 refer to dispatching rules for extracting reservoirs when dealing with small floods or super-standard floods.

[0018] Furthermore, the real-time updating of the digital twin scene of the reservoir flood control and dispatching system described in step S2 includes real-time updating of the data of the physical module of the river basin where the target reservoir is located and the virtual status data of the reservoir dispatching system according to the monitoring data of the physical entity. The data of the physical module include rainfall data, evaporation data, upstream and downstream river siltation data, dam safety data of the reservoir itself, downstream flow data of the upstream reservoir group, and runoff data of the upstream main and tributary rivers; the virtual status data include real-time water level status data of the reservoir, runoff status data, discharge status of the spillway tunnel, opening and closing status of the spillway gate, upstream river water level status data, and water level status data and flow status data of important downstream flood control sections.

[0019] Furthermore, the machine learning algorithm described in step S3 includes a long short-term memory neural network algorithm (LSTM), a recurrent neural network (RNN), a convolutional neural network (CNN), Boosting, GBDT, DeepAR, Transformer, Informer, etc.

[0020] Furthermore, the construction of multiple flood forecasting models in step S3 includes the following steps:

[0021] S31. Use autocorrelation analysis and cross-correlation analysis methods to determine the relationship between historical rainfall data, historical runoff data, historical evaporation data, current rainfall data, current evaporation data and current runoff sequence, and select the driving factors of the model; use Gamma Test to reduce noise on the data, arrange and combine the selected driving factors and estimate the noise separately, and select the combination with the smallest noise as the driving factor; establish a mapping relationship between the driving factor and the prediction factor.

[0022] S32. Collect historical flood data and divide them into training sets and test sets; construct a reservoir flood forecasting model based on various machine learning algorithms; in determining the model parameters, use intelligent algorithms to optimize and find the optimal parameter combination within the feasible domain to obtain a flood forecasting model.

[0023] S33. Use the trained flood forecasting model to make predictions and obtain the forecast results under different models, which are used as the inflow runoff sequence under different scenarios of the scheduling model, thereby simulating the forecast error caused by using different forecasting models.

[0024] Furthermore, the optimization using intelligent algorithm described in step S32 refers to genetic algorithm (NSGA), grey wolf algorithm, particle swarm algorithm, frog leaping algorithm, etc.

[0025] Furthermore, the step S4 described in which the uncertainty of the prediction results of each flood forecasting model is combined to construct a reservoir flood control optimization scheduling model includes the following steps:

[0026] S41. Select the discharge flow of the reservoir in a unit time period as the decision variable, and use the runoff sequence predicted by multiple forecasting models as input.

[0027] S42, considering the upstream and downstream safety of the reservoir, determine the objective function and constraints of the optimization scheduling model.

[0028] S43. The NSGA-Ⅲ multi-objective genetic algorithm is used to solve the optimization scheduling model. In the iterative process, the "Min-Max" concept in robust optimization is combined to screen the candidate sequences of decision variables generated for each individual in the population under different forecast scenarios, and obtain the decision variable sequence representing the individual, and then carry out the next optimization iteration to finally obtain a non-inferior solution set.

[0029] Furthermore, the objective function described in step S42 includes:

[0030] (1) The objective function is to minimize the probability that the maximum water level of the reservoir exceeds the safety threshold during the operation process:

[0031]

[0032] (2) The objective function is to minimize the probability that the maximum discharge flow of the reservoir exceeds the safe discharge threshold during the operation process:

[0033]

[0034] (3) The objective function is to minimize the proportion of the duration of the reservoir water level exceeding the safety threshold during the operation process to the total operation time:

[0035]

[0036] (4) The objective function is to minimize the proportion of the duration of the reservoir discharge exceeding the safe discharge threshold during the dispatching process to the total dispatching time:

[0037]

[0038] Among them, Z n (t) represents the reservoir water level at time t under forecast scenario n (m); Z up is the upper safety threshold of water level; q n (t) represents the reservoir discharge at time t under forecast scenario n (m 3 / s); Q up is the safe discharge threshold of the reservoir; N is the total number of forecast scenarios; T is the total dispatching time (h); Count(·) is the statistical counting function;

[0039] The constraints include water balance constraints, upper and lower limit constraints of reservoir water level, discharge capacity constraints, scheduling process fluctuation constraints and non-negative constraints.

[0040] Furthermore, in step S43, the “Min-Max” concept in robust optimization is combined in the iterative process to screen the candidate decision variable sequences generated for each individual in the population under different forecast scenarios; the “STP” method is used in the screening process, and the specific steps are as follows:

[0041] S431. Set the population size to M and generate an initial population.

[0042] S432, each individual in the population represents a set of decision variable sequences with a length of scheduling duration T. When faced with N different forecast scenarios, each decision variable needs to be adjusted within the acceptable value range under each forecast scenario so that the entire sequence satisfies all constraints. After adjustment, the final generated (x1 1 , x2 1 , x3 1 , ...x T 1 ), (x1 2 , x2 2 , x3 2 , ...x T 2 ), ..., (x1 N , x2 N , x3 N , ...x T N ) There are N groups of sequences to be selected.

[0043] S433, calculating the four objective function values ​​corresponding to each group of candidate sequences, and then obtaining the regret value RV of each group of candidate sequences, as shown in formula (5):

[0044] RV=(F1-0)+(F2-0)+(F3-0)+(F4-0) (5)

[0045] S434, selecting the decision sequence with the largest RV value as the representative sequence of the individual, entering the next iteration process, and further completing the optimization process of the genetic algorithm.

[0046] S435. Repeat the process of S432-S434 in each iteration. After the number of iterations reaches a preset value, stop the iteration to obtain a non-inferior solution set.

[0047] Furthermore, step S5 is specifically as follows: after the optimization calculation, according to the current flood control requirements, the upstream and downstream risk indicators of the scheduling schemes in the non-inferior solution set are calculated respectively; the optimized flood control scheduling strategy and the corresponding risk indicators are provided to decision makers as a scientific reference basis, so as to reasonably call upon the flood control capacity of the reservoir and maximize the potential of the flood control engineering system.

[0048] The present invention adopts the above technical solution, which has the following advantages and beneficial effects:

[0049] (1) Taking into account the uncertainty of flood forecasting, a flood forecasting model is constructed based on a machine learning algorithm. The model parameters are optimized using an intelligent algorithm, and the forecast characteristics of each model are retained as the source of forecast error, which is used as the uncertain forecast input of the flood control scheduling model.

[0050] (2) In combination with uncertain flood forecasts, a reservoir flood control optimization scheduling model is constructed, and the upstream and downstream risks caused by flood control scheduling strategies in the face of different forecast scenarios are taken as optimization targets, which effectively improves the efficiency of reservoir flood control scheduling and ensures the safety of reservoirs and upstream and downstream areas.

[0051] (3) A genetic evolutionary algorithm was used to solve the scheduling model. It was proposed that in the iterative optimization process, the "STP" robust optimization method be used to convert the values ​​of each objective function under different scenarios into a specific numerical value. Combined with the "Min-Max" concept, a robust non-inferior solution set was obtained, so that the optimized scheduling strategy can effectively reduce flood risks even when facing the most extreme forecast floods.

[0052] (4) Construct a digital twin scenario of the reservoir flood control and dispatching system to simulate the actual operation process of flood dispatching, which will facilitate the transmission and sharing of real-time data, integrate the topographic data of the reservoir and surrounding basins, and construct a three-dimensional model. By updating real-time data, the accuracy of the forecast can be improved, making the flood control and dispatching results more reliable.

[0053] (5) Relying on risk indicators, a flood risk assessment system is constructed. By combining the virtual-reality mapping of digital twin scenarios and reservoir flood control and dispatching systems, the risks of optimized dispatching plans are analyzed and predicted to assist scientific decision-making and improve the benefits of reservoir flood control. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of the method of the present invention;

[0055] Figure 2 Schematic diagram of the process of "STP" method. DETAILED DESCRIPTION

[0056] The present invention will be further described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0057] Figure 1 The flowchart of the method of the present invention is a reservoir flood control optimization scheduling method based on digital twins considering forecast uncertainty, which specifically includes the following steps:

[0058] S1. Constructing the digital twin scenario of the reservoir flood control dispatching system, which specifically includes the following steps:

[0059] S11. Establish the physical module of the reservoir flood control and dispatching system.

[0060] The physical module for establishing the physical entity of reservoir flood control dispatching is specifically based on the meteorological data, hydrological data, topographic data of the basin where the reservoir is located, as well as the engineering properties and design standards of the reservoir itself, to construct a three-dimensional physical module.

[0061] S12. Based on the physical module of the reservoir flood control dispatching system and the existing reservoir flood control dispatching control operation plan, the dispatching rules are extracted and the operation module of the reservoir flood control dispatching system is established.

[0062] Based on the properties and design standards of reservoirs and other construction projects, as well as the flood control policies and development plans of the region, the dispatching rules for reservoirs when responding to small floods or super-standard floods are extracted from the current flood control and operation plans of the reservoirs.

[0063] S13. Construct a digital twin scenario of the reservoir flood control and dispatching system, including the operating status of the reservoir before and after the flood season and during the flood, specifically including the pre-discharge process, flood discharge process, water level operation change process, gate or flood discharge tunnel opening and closing status, and the flow process of important flood control sections downstream of the reservoir.

[0064] S2. Obtain the physical data of the reservoir flood control and dispatching system and the virtual data in the digital twin scenario, and update the digital twin scenario of the reservoir flood control and dispatching system in real time based on the physical data.

[0065] Real-time updating includes updating the data of the physical module of the basin where the target reservoir is located and the virtual status data of the reservoir dispatching system in real time according to the monitoring data of the physical entity. The data of the physical module include rainfall data, evaporation data, upstream and downstream river siltation data, dam safety data of the reservoir itself, downstream flow data of the upstream reservoir group, and runoff data of the upstream main and tributary rivers; the virtual status data include real-time reservoir water level status data, runoff status data, spillway tunnel discharge status, spillway gate opening and closing status, upstream river water level status data, and water level status data and flow status data of important downstream flood control sections.

[0066] S3. Based on the real-time updated digital twin scenario of the reservoir flood control dispatching system, combined with multiple machine learning algorithms, multiple flood forecasting models are constructed, which specifically include the following steps:

[0067] S31. Use autocorrelation analysis and cross-correlation analysis methods to determine the relationship between historical rainfall data, historical runoff data, historical evaporation data, current rainfall data, current evaporation data and current runoff sequence, and select the driving factors of the model; use Gamma Test to reduce the noise of the data, arrange and combine the selected driving factors and estimate the noise respectively, and select the combination with the smallest noise as the driving factor; establish a mapping relationship between the driving factor and the prediction factor;

[0068] S32, collect historical flood data and divide them into training sets and test sets; construct a reservoir flood forecasting model based on various machine learning algorithms; in determining the model parameters, use intelligent algorithms to find the optimal parameter combination solution within the feasible domain to obtain a flood forecasting model;

[0069] Machine learning algorithms include long short-term memory neural network algorithm (LSTM), recurrent neural network (RNN), convolutional neural network (CNN), Boosting, GBDT, DeepAR, Transformer, Informer, etc.

[0070] Intelligent algorithms for optimization include genetic algorithm (NSGA), grey wolf algorithm, particle swarm algorithm, frog leaping algorithm, etc.

[0071] S33. Use the trained flood forecasting model to make predictions and obtain the forecast results under different models, which are used as the inflow runoff sequence under different scenarios of the scheduling model, thereby simulating the forecast error caused by using different forecasting models.

[0072] S4. Combined with the uncertainty of the prediction results of various flood forecasting models, a reservoir flood control optimization scheduling model is constructed, and a multi-objective genetic algorithm is used to solve it to obtain the optimized flood control scheduling strategy, which specifically includes the following steps:

[0073] S41. Select the discharge flow of the reservoir in a unit time period as the decision variable, and use the runoff sequence predicted by multiple forecasting models as input.

[0074] S42, considering the upstream and downstream safety of the reservoir, determine the objective function and constraints of the optimization scheduling model;

[0075] The objective function of the optimization scheduling model includes:

[0076] (1) The objective function is to minimize the probability that the maximum water level of the reservoir exceeds the safety threshold during the operation process:

[0077]

[0078] (2) The objective function is to minimize the probability that the maximum discharge flow of the reservoir exceeds the safe discharge threshold during the operation process:

[0079]

[0080] (3) The objective function is to minimize the proportion of the duration of the reservoir water level exceeding the safety threshold during the operation process to the total operation time:

[0081]

[0082] (4) The objective function is to minimize the proportion of the duration of the reservoir discharge exceeding the safe discharge threshold during the dispatching process to the total dispatching time:

[0083]

[0084] Among them, Z n (t) represents the reservoir water level at time t under forecast scenario n (m); Z up is the upper safety threshold of water level; q n (t) represents the reservoir discharge at time t under forecast scenario n (m 3 / s); Q up is the safe discharge threshold of the reservoir; N is the total number of forecast scenarios; T is the total dispatching time (h); Count(·) is the statistical counting function;

[0085] The constraints of the optimization scheduling model include:

[0086] Water balance constraints, reservoir water level upper and lower limit constraints, discharge capacity constraints, scheduling process fluctuation constraints, and non-negative constraints.

[0087] S43. The NSGA-Ⅲ multi-objective genetic algorithm is used to solve the optimization scheduling model. In the iterative process, the "Min-Max" concept in robust optimization is combined, and the "STP" method is used to screen the candidate sequences of decision variables generated for each individual in the population under different forecast scenarios to obtain the decision variable sequence representing the individual, and then the next optimization iteration is carried out to finally obtain a non-inferior solution set.

[0088] The specific steps of the NSGA-Ⅲ algorithm are described as follows:

[0089] (1) Randomly generate an initial population P0 consisting of N individuals;

[0090] (2) Generate reference points;

[0091] (3) Select individuals from the current population for genetic operations to generate offspring individuals to form a population Q t ;

[0092] (4) Merge the parent and offspring populations, R t =P t ∪Q t ;

[0093] (5) Use the fast non-dominated sorting method to sort R t Divide into several non-dominated layers such as F1, F2 and F3, and determine the non-dominated layer where each individual is located;

[0094] (6) Starting from layer F1, L The individuals of the layer are placed into P t+1 In the next generation of population, the L value satisfies

[0095] (7) For F L+1 Individuals in the layer perform niche-based selection strategies, choosing Individuals enter P t+1 In, P t+1 The total number of individuals in is N;

[0096] (8) Determine whether the algorithm has ended. If the end condition is not met, go to step (2); otherwise, end the algorithm.

[0097] The application of the "STP" method is nested in the iteration process of the NSGA-Ⅲ algorithm. Its main function is to select the scheduling sequence that can represent the individual from the candidate decision variable sequence generated under multiple uncertain forecast scenarios and enter the next iterative optimization process. The principle flow chart is as follows Figure 2 As shown, the specific steps include:

[0098] S431, setting the population size to M and generating an initial population;

[0099] S432, each individual in the population represents a set of decision variable sequences with a length of scheduling duration T. When faced with N different forecast scenarios, each decision variable needs to be adjusted within the acceptable value range under each forecast scenario so that the entire sequence satisfies all constraints. After adjustment, the final generated (x1 1 , x2 1 , x3 1 , ...x T 1 ), (x1 2 , x2 2 , x3 2 , ...x T 2 ), ..., (x1 N , x2 N , x3 N , ...x T N) There are N groups of sequences to be selected.

[0100] S433, calculating the four objective function values ​​corresponding to each group of candidate sequences, and then obtaining the regret value (RV) of each group of candidate sequences, so that the optimization of multiple objectives can be converted into a specific numerical embodiment, as shown in formula (5):

[0101] RV=(F1-0)+(F2-0)+(F3-0)+(F4-0) (5)

[0102] S434, selecting the decision sequence with the largest RV value as the representative sequence of the individual, entering the next iteration process, and further completing the optimization process of the genetic algorithm.

[0103] S435. Repeat the process of S432-S434 in each iteration. After the number of iterations reaches a preset value, stop the iteration to obtain a non-inferior solution set.

[0104] S5. Construct a flood risk assessment system, analyze and assess flood control risks based on forecast situations and dispatch strategies, and guide actual flood control dispatch. The specific methods are as follows:

[0105] After optimization calculation, based on the current flood control requirements, the upstream and downstream risk indicators of the scheduling schemes in the non-inferior solution set are calculated respectively; the optimized flood control scheduling schemes and the corresponding risk indicators are provided to decision makers as a scientific reference, so as to reasonably call on the flood control capacity of the reservoir and maximize the potential of the flood control engineering system.

[0106] In summary, the example scheduling process can generate a flood control optimization scheduling strategy based on the actual situation and taking into account the uncertain flood forecast. It can also obtain the specific response and risk status of the flood control scheduling system if this strategy is implemented, providing a scientific reference for decision makers.

[0107] The present invention provides a reservoir flood control optimization scheduling method based on digital twins taking into account forecast uncertainty, applies digital twin technology to flood control optimization scheduling, realizes dynamic interaction between twin scenes and real scenes, provides ideas for the digitalization and intelligence of reservoir flood control optimization scheduling, optimizes the adverse effects of flood forecast uncertainty on reservoir flood control scheduling, and improves the comprehensive flood control benefits of reservoirs.

[0108] The present invention can reflect the real-time information of flood control scheduling, reduce the risks brought by scheduling strategies generated under uncertain flood forecast conditions, make the scheduling process more intelligent, ensure the safety of upstream and downstream of the reservoir, improve the reliability of scheduling, and reduce operational risks.

[0109] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

[0110] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A reservoir flood control optimization scheduling method based on digital twins considering forecast uncertainty, characterized in that: The following steps are involved: S1. Construct a digital twin scenario of the reservoir flood control and dispatching system; S2. Obtain the physical data of the reservoir flood control dispatching system and the virtual data in the digital twin scenario, and update the digital twin scenario of the reservoir flood control dispatching system in real time based on the physical data; S3. Based on the real-time updated digital twin scenario of the reservoir flood control dispatching system, multiple machine learning algorithms are combined to build multiple flood forecasting models; S4. Combined with the uncertainty of the prediction results of various flood forecasting models, a reservoir flood control optimization scheduling model is constructed, and a multi-objective genetic algorithm is used to solve it to obtain the optimized flood control scheduling strategy; S5. Build a flood risk assessment system to analyze and assess flood control risks based on forecast results and optimized flood control dispatch strategies, and guide actual flood control dispatch; The step S4 combines the uncertainty of the prediction results of each flood forecasting model to construct a reservoir flood control optimization scheduling model, adopts a multi-objective genetic algorithm to solve it, and obtains an optimized flood control scheduling strategy, which specifically includes the following steps: S41, selecting the discharge flow of the reservoir in a unit time period as a decision variable, and taking the runoff series predicted by multiple forecasting models as input; S42, considering the upstream and downstream safety of the reservoir, determine the objective function and constraints of the optimization scheduling model; S43. The NSGA-Ⅲ multi-objective genetic algorithm is used to solve the optimization scheduling model. In the iterative process, the "Min-Max" concept in robust optimization is combined to screen the candidate decision variable sequences generated by each individual in the population under different forecast scenarios, and obtain the decision variable sequence representing the individual, and then the next optimization iteration is carried out to finally obtain a non-inferior solution set; In step S43, the "Min-Max" concept in robust optimization is combined in the iterative process to screen the candidate decision variable sequences generated for each individual in the population under different forecast scenarios; the "STP" method is used in the screening process, and the specific steps are as follows: S431, setting the population size to M and generating an initial population; S432, each individual in the population represents a set of decision variable sequences with a length of scheduling duration T. When faced with N different forecast scenarios, each decision variable needs to be adjusted within the acceptable value range under each forecast scenario so that the entire sequence satisfies all constraints. After adjustment, the final generated (x1 1 , x2 1 , x3 1 , ...x T 1 ), (x1 2 , x2 2 , x3 2 , ...x T 2 ), ..., (x1 N , x2 N , x3 N , ...x T N ) A total of N groups of sequences to be selected; S433, calculating the four objective function values ​​corresponding to each group of candidate sequences, and then obtaining the regret value RV of each group of candidate sequences, as shown in formula (5): RV=(F1-0)+(F2-0)+(F3-0)+(F4-0) (5) S434, selecting the decision sequence with the largest RV value as the representative sequence of the individual, entering the next iteration process, and further completing the optimization process of the genetic algorithm; S435. Repeat the process of S432-S434 in each iteration. After the number of iterations reaches a preset value, stop the iteration to obtain a non-inferior solution set.

2. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty according to claim 1 is characterized in that: The digital twin scene of the reservoir flood control dispatching system is constructed in step S1, which specifically includes the following steps: S11. Establish the physical module of reservoir flood control dispatching system; S12, based on the physical module of the reservoir flood control dispatching system and the existing reservoir flood control dispatching control operation plan, extract the dispatching rules and establish the operation module of the reservoir flood control dispatching system; S13. Construct a digital twin scenario of the reservoir flood control and dispatching system, including the operating status of the reservoir before and after the flood season and during the flood, specifically including the pre-discharge process, flood discharge process, water level operation change process, gate or flood discharge tunnel opening and closing status, and the flow process of important flood control sections downstream of the reservoir.

3. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty according to claim 2 is characterized in that: The physical module of the reservoir flood control dispatching system is established in S11, which specifically constructs a three-dimensional physical module based on the meteorological data, hydrological data, topographic data of the basin where the reservoir is located, as well as the engineering properties and design standards of the reservoir itself.

4. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty according to claim 2 is characterized in that: The extraction dispatching rules described in S12 refer to the dispatching rules of the extraction reservoir when dealing with small floods or super-standard floods.

5. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty according to claim 1, characterized in that: The real-time updating of the digital twin scene of the reservoir flood control and dispatching system described in step S2 includes real-time updating of the data of the physical module of the river basin where the target reservoir is located and the virtual status data of the reservoir dispatching system according to the monitoring data of the physical entity. The data of the physical module include rainfall data, evaporation data, upstream and downstream river siltation data, dam safety data of the reservoir itself, downstream flow data of the upstream reservoir group, and runoff data of the upstream main and tributary rivers; the virtual status data include real-time water level status data of the reservoir, runoff status data, discharge status of the spillway tunnel, opening and closing status of the spillway gate, upstream river water level status data, and water level status data and flow status data of important downstream flood control sections.

6. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty according to claim 1 is characterized in that: The construction of multiple flood forecasting models in step S3 comprises the following steps: S31. Use autocorrelation analysis and cross-correlation analysis methods to determine the relationship between historical rainfall data, historical runoff data, historical evaporation data, current rainfall data, current evaporation data and current runoff sequence, and select the driving factors of the model; use Gamma Test to reduce the noise of the data, arrange and combine the selected driving factors and estimate the noise respectively, and select the combination with the smallest noise as the driving factor; establish a mapping relationship between the driving factor and the prediction factor; S32, collect historical flood data and divide them into training sets and test sets; construct a reservoir flood forecasting model based on various machine learning algorithms; in determining the model parameters, use intelligent algorithms to find the optimal parameter combination solution within the feasible domain to obtain a flood forecasting model; S33. Use the trained flood forecasting model to make predictions and obtain the forecast results under different models, which are used as the inflow runoff sequence under different scenarios of the scheduling model, thereby simulating the forecast error caused by using different forecasting models.

7. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty as claimed in claim 1, constructing a reservoir flood control optimization scheduling model, characterized in that: In step S42, the objective function includes: (1) The objective function is to minimize the probability that the maximum water level of the reservoir exceeds the safety threshold during the operation process: (2) The objective function is to minimize the probability that the maximum discharge flow of the reservoir exceeds the safe discharge threshold during the operation process: (3) The objective function is to minimize the proportion of the duration of the reservoir water level exceeding the safety threshold during the operation process to the total operation time: (4) The objective function is to minimize the proportion of the duration of the reservoir discharge exceeding the safe discharge threshold during the dispatching process to the total dispatching time: Among them, Z n (t) represents the reservoir water level at time t under forecast scenario n (m); Z up is the upper safety threshold of water level; q n (t) represents the reservoir discharge at time t under forecast scenario n (m 3 / s); Q up is the safe discharge threshold of the reservoir; N is the total number of forecast scenarios; T is the total dispatching time (h); Count(·) is the statistical counting function; The constraints include water balance constraints, upper and lower limit constraints of reservoir water level, discharge capacity constraints, scheduling process fluctuation constraints and non-negative constraints.

8. The reservoir flood control optimization scheduling method based on digital twin considering forecast uncertainty according to claim 1, characterized in that: Specifically, the step S5 is as follows: after the optimization calculation, according to the current flood control requirements, the upstream and downstream risk indicators of the scheduling schemes in the non-inferior solution set are calculated respectively; the optimized flood control scheduling strategy and the corresponding risk indicators are provided to decision makers as a scientific reference basis, so as to reasonably call on the flood control capacity of the reservoir and maximize the potential of the flood control engineering system.

Images