Multi-objective Optimal Scheduling Method and System for Flood Control System of Sluice and Dam Based on Online Data-driven Evolutionary Optimization

Abstract

The present invention discloses a multi-objective optimal scheduling method and system for a sluice and dam flood control system based on online data-driven evolutionary optimization, including: driving a multi-objective evolutionary optimization algorithm online through a machine learning model, using an improved multi-objective expected improvement matrix filling criterion to select potential non-dominated candidate solutions in the offspring population during the evolutionary process as evolutionary guidance, and reducing unnecessary real evaluations by simulating and calculating the candidate solutions through coupling the flood numerical model of the sluice and dam, greatly improving the optimization calculation efficiency, and realizing the direct solution of the multi-objective flood control optimal scheduling of the sluice and dam based on the calculation of the hydrodynamic model. The present invention makes it possible to globally optimize the operation scheduling problem of flood control and drainage projects under the accurate simulation of the hydrodynamic model, leaves sufficient lead time for the emergency scheduling of river sluices and dams, and also provides a scientific basis for flood prevention, disaster reduction and reasonable allocation of resources of rivers, and has important application value for effectively reducing flood risks.

Description

Technical Field

[0001] The present invention belongs to the technical field of optimization and scheduling of flood control and drainage projects, and relates to a multi-objective optimization and scheduling method and system for a dam flood control system based on online data-driven evolutionary optimization. Background Art

[0002] As one of the effective flood control projects, urban river dams can temporarily store floods and regulate flood peaks by operating and dispatching dams, thereby reducing or avoiding flood disasters. Faced with complex and changeable flood processes, the current river flood control system must not only consider the smooth flow of floods downstream, but also the safety of upstream flood control. Therefore, the flood control goals of the upstream and downstream areas of the river usually affect each other and may even conflict with each other. This flood control problem is expressed as a multi-objective optimization problem, which requires multi-objective optimization scheduling to find the optimal scheduling solution that simultaneously meets all flood control goals.

[0003] At present, the first problem in the optimization and scheduling of flood control and drainage projects is the lack of accuracy in representing the flood process. Therefore, it is necessary to use a two-dimensional hydrodynamic numerical model to simulate and calculate the flood process with high precision. For the formulation of the scheduling plan, it is necessary to rely on evolutionary algorithms to find the optimal operation plan for dam flood control. Considering the problem of high optimization calculation cost caused by using hydrodynamic models as solvers for optimization algorithms, it is necessary to use machine learning and data-driven technologies to assist the optimization process in order to achieve the optimal scheduling of dam flood control systems. Data-driven technology has gradually entered people's field of vision, but there are few applications for the optimization and scheduling of dam flood control systems, especially for multi-objective optimization. How to guide and accelerate multi-objective evolutionary algorithms is also a major challenge. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the prior art that flood control pressures in upstream and downstream areas of the river influence each other and even conflict with each other, the insufficient accuracy of flood process representation in the optimization scheduling of flood control and drainage projects, and the existing use of hydrodynamic models as optimization algorithm solvers brings about excessively high optimization calculation costs. A multi-objective optimization scheduling method and system for dam flood control systems based on online data-driven evolutionary optimization is provided.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A multi-objective optimization scheduling method for dam flood control system based on online data-driven evolutionary optimization includes:

[0007] Collect information on the dam dispatching system, express the required dispatching objectives and constraints in the form of functions, form objective functions and constraints, and build a multi-objective flood control optimization dispatching model for dams and gates;

[0008] Construct a multi-objective particle swarm optimization algorithm based on mutation factor improvement;

[0009] Based on the topographic data, hydrological data, the number and construction locations of sluice dams, and the operation data of sluice dams of the sluice dam scheduling system, a two-dimensional hydrodynamic model coupling sluice dams is constructed;

[0010] A machine learning model is constructed, with the optimized decision variables as input data and the hydraulic parameters of the two-dimensional hydrodynamic model obtained under different flood conditions as output data, to establish the mapping relationship between the input and output;

[0011] The machine learning model is coupled into the multi-objective particle swarm optimization algorithm process in an online data-driven manner. Based on the prediction results of the machine learning model, the expected improvement matrix filling criterion applicable to multiple objectives is introduced to select potential non-dominated candidate solutions in the offspring population during the evolution process as the evolution guidance;

[0012] Based on the selected evolution-guided individuals, the two-dimensional hydrodynamic model coupling sluice dams is used for actual calculation to solve the multi-objective flood control optimization scheduling model of sluice dams, realizing the iteration of population evolution and the update of the machine learning model; finally, after reaching the limit filling times, a set of Pareto solutions for the multi-objective flood control optimization scheduling of sluice dams is obtained.

[0013] A further improvement of the present invention lies in:

[0014] Furthermore, the information of the sluice dam scheduling system is collected, and the required scheduling objectives and the restricted conditions are expressed in the form of functions to form the objective function and constraint conditions, specifically:

[0015] The total number of control variables of the sluice dam is where K is the number of different sluice dams on the river, the decision variables are the operating height or gate opening of the sluice dam, and T k represents the number of decision-making scheduling intervals into which the operating duration of each sluice dam in a flood event can be divided, and each sluice dam has a separate operating rule within each scheduling interval;

[0016] The optimized competing objectives are: the lowest water level in front of the upstream dam and the smallest downstream peak flow rate. The competing objectives are described by the following formula:

[0017]

[0018] F2 = min{maxQ down (t) / Q ori} (2)

[0019] where F1 is the upstream flood control objective, minimizing the sum of the highest water levels of each upstream sluice dam; F2 is the downstream flood control objective, minimizing the downstream control point peak flow rate; H k (t) is the upstream water level of the Kth sluice dam at time t; D k,max is the maximum allowable water level of the kth sluice dam; Qdown (t) is the flow rate at the downstream control point of the river channel at time t; Q ori is the flood discharge flow rate without sluice and dam regulation.

[0020] Furthermore, a multi-objective flood control optimal operation model for sluices and dams is constructed, specifically as follows:

[0021] The multi-objective flood control constraint conditions for sluices and dams include water level limits, flow rate limits, and the operation of sluices and dams:

[0022] D k,min <H k (t) < D k,max (3)

[0023] Q k,min <Q k (t) < Q k,max (4)

[0024] B k,min <B k (t) + δ < B k,max δ ∈ S (5)

[0025] Among them, D k,min 、and D k,max are respectively the maximum allowable water depth and the minimum allowable water depth of the sluice and dam k; Q k,min and Q k,max respectively represent the minimum flow rate and the maximum flow rate of the sluice and dam k; Each section of the river channel needs to ensure the requirement of the minimum ecological flow rate, and at the same time, the water level should not be higher than the maximum limit to prevent floods. B k,min and B k,max respectively represent the minimum opening height and the maximum opening height of the gate of the k-th sluice and dam; B k (t) is the operation state of the k-th sluice and dam at time t; δ is the change in the operation of the sluice and dam. Usually, this value should not be too small, and its opening can meet the regulation requirements.

[0026] Furthermore, a multi-objective particle swarm optimization algorithm improved based on mutation factors is constructed, specifically as follows: On the basis of the standard particle swarm algorithm, MOPSO is developed by introducing Pareto solutions; In the l-th iteration, the velocity and position of the i-th particle in the d-th dimensional search space will be updated:

[0027]

[0028]

[0029] Among them, ω is the inertia weight; c 1 and c 2 are respectively the individual learning and social learning factors of the particle; r 1 and r 2is a random number between 0 and 1; pbest i d is the individual optimal solution of the i-th particle; gbest d is the global optimal solution of all particles in the d-dimensional space; x and V are the position and velocity of the particle respectively;

[0030] To ensure that the algorithm has better global search ability in the early stage and has the ability to jump out of the local optimal trap in the later stage, an inertial mutation weight ω is introduced with the idea of mutation:

[0031]

[0032] where ω min and ω max are the minimum and maximum values of the weight; when the change of velocity update in the later stage is very small, that is, V i d (l + 1) - V i d (l) ≤ 0.01V i d (l), the algorithm can jump out of the local optimum through a random mutation value, and its mutation probability is p, σ is the mutation factor, and it takes a random number between (0, 1).

[0033] Furthermore, the hydrodynamic parameters of the two-dimensional hydrodynamic model are specifically as follows: Mark the positions of the sluice dams to be processed in the hydrodynamic model terrain file, regard the water passing gate, that is, the boundary of the gate grid, as a closed boundary, and stop the calculation of the flux by the HLLC Riemann solver on the side where the grid is located; use the water depth h of the grid in front of the water passing gate as the head input parameter of the sluice orifice outflow formula to calculate the flow rate of each upstream grid passing through the gate; for a flat-bottomed sluice gate with free outflow, the sluice orifice outflow formula is as follows:

[0034]

[0035] where Q o is the flow rate passing through a single grid, m 3 / s; B is the grid width, m; e is the gate opening; H o is the head on each grid, m; g is the acceleration due to gravity, m / s 2 ; μ o is the comprehensive discharge coefficient, specifically:

[0036] μ 0 = 0.6 - 0.18×(e / H o )(10)

[0037] Through the flow rate Q through the sluice oCalculate the change in water depth Δh within the same time step for each grid cell at the gate. Subtract the change in water depth Δh from the water levels of each grid cell upstream of the gate as the water depth value at the next moment, while add this change in water depth Δh to the grid cells downstream of the gate as the water depth value at the next moment. Specifically:

[0038]

[0039] In the formula, c l is the length of the grid cell, c k is the width of the grid cell, and dt is the time step of the model calculation.

[0040] Furthermore, based on the prediction results of the machine learning model, introduce the expected improvement matrix filling criterion applicable to multiple objectives to select potential non-dominated candidate solutions in the offspring population during the evolutionary process as the evolutionary guidance. Specifically: In the machine learning model, select a particle individual in the offspring population as the evolutionary guidance through the filling criterion; in order to avoid the larger value of the EI criterion falling into the local optimum, improve this criterion by introducing a weighting term. The improved EI criterion comprehensively considers the predicted mean and standard deviation of the Kriging model and can balance the global optimum and the local optimum. The expression is:

[0041]

[0042] where y min is the optimal response value of the current sample point; the predicted value of any unknown point x follows a normal distribution, that is Φ and φ represent the cumulative distribution function and probability density function of the standard normal distribution respectively; the predicted mean of the Kriging model is smaller, the first term of the WEI criterion is larger, and the predicted standard deviation of the model is larger, the second term of the WEI criterion is larger; w e as the weighting coefficient of this criterion balances the global optimum and the local optimum, and is determined by the following formula:

[0043]

[0044] In the formula, the weighting coefficient w e changes in a non-linear relationship between (0, 1). This weight coefficient has a larger value in the early stage of the algorithm, and the global search ability is better. In the later stage, this value decays rapidly, and the local search ability is stronger.

[0045] Further, introduce the expected improvement matrix filling criterion applicable to multiple objectives to select potential non-dominated candidate solutions in the offspring population during the evolutionary process as the evolutionary guidance, which also includes: With the idea of a matrix, consider the approximate Pareto solution set of the multi-objective optimization problem as the expansion of the single-objective optimal solution in two directions; In terms of dimension, the multi-objective expands from 1D to multi-dimensional; In terms of the number of solutions, the multi-objective expands from 1 optimal solution to multiple Pareto solutions, and finally expands into a two-dimensional matrix; By integrating the single-objective expected improvement of each research point on each objective for each approximate point beyond the Pareto front, a two-dimensional WEI matrix is obtained:

[0046]

[0047] where X is an n-dimensional point, X = [x 1 , x 2 , …, x n ; M is the number of objective functions; J is the number of points on the Pareto front; Each element in the WEIM matrix is a one-dimensional expected improvement function.

[0048] Further, process the WEIM matrix, specifically: Integrate the elements in the WEIM matrix to form a scalar measure of the expected improvement of the prediction point on the overall Pareto front; Considering that the objective function is a multi-dimensional space, the elements of WEIM are combined into a scalar function through the Euler distance improvement function:

[0049]

[0050] According to the evolutionary guidance particles, use the coupled sluice-dam hydrodynamic model for calculation and calculate the true fitness function value; The fitness function value is used for the evolutionary process of the optimization algorithm and to update the machine learning model training database.

[0051] The multi-objective optimization scheduling system for the sluice-dam flood control system based on online data-driven evolutionary optimization includes:

[0052] A scheduling model module, which collects information on the sluice-dam scheduling system, expresses the required scheduling objectives and the constraints received in the form of functions to form an objective function and constraint conditions, and constructs a multi-objective flood control optimization scheduling model for the sluice-dam;

[0053] An optimization algorithm module, which constructs a multi-objective particle swarm optimization algorithm improved based on the mutation factor;

[0054] A hydrodynamic calculation module, which constructs a two-dimensional hydrodynamic model of the coupled sluice-dam based on the topographic data, hydrological data, number and construction location of sluice-dams, and sluice-dam operation data of the sluice-dam scheduling system;

[0055] A machine learning module that constructs a machine learning model, uses optimized decision variables as input data, and hydraulic parameters of a two-dimensional hydrodynamic model obtained under different flood conditions as output data to establish a mapping relationship between the input and the output;

[0056] An online driving module that couples the machine learning model into the multi-objective particle swarm optimization algorithm process in an online data-driven manner. Based on the prediction results of the machine learning model, an expected improvement matrix filling criterion applicable to multiple objectives is introduced to select potential non-dominated candidate solutions in the offspring population during the evolution process as the evolution guidance;

[0057] An iterative update module that, based on the selected evolution guiding individuals, performs real calculations using a two-dimensional hydrodynamic model coupled with sluice gates to solve the multi-objective flood control optimization scheduling model of the sluice gates, realizing the iteration of population evolution and the update of the machine learning model; After finally reaching the limit filling times, a set of Pareto solutions for the multi-objective flood control optimization scheduling of the sluice gates is obtained.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] The present invention drives the multi-objective evolutionary optimization algorithm online through a machine learning model, uses an improved multi-objective expected improvement matrix filling criterion to select potential non-dominated candidate solutions in the offspring population during the evolution process as the optimization guidance, and reduces unnecessary real evaluations by simulating and calculating the candidate solutions using a flood numerical model coupled with sluice gates, greatly improving the optimization calculation efficiency and realizing the direct solution of the multi-objective flood control optimization scheduling of the sluice gates based on the hydrodynamic model calculation. The present invention makes it possible to globally optimize the operation scheduling problem of flood control and drainage projects under the accurate simulation of the hydrodynamic model, leaves sufficient lead time for the emergency scheduling of river sluice gates, provides a scientific basis for flood prevention, disaster reduction and rational allocation of resources of rivers, and then guides decision-makers to formulate operation scheduling rules for sluice gates, which has important application value for effectively reducing flood risks. Description of the Drawings

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0061] Figure 1 It is a schematic flow chart of a multi-objective optimization scheduling method for a sluice gate flood control system based on online data-driven evolution optimization of the present invention;

[0062] Figure 2Schematic structural diagram of the multi-objective optimal scheduling system of the sluice dam flood control system based on online data-driven evolutionary optimization of the present invention;

[0063] Figure 3 Schematic diagram of an algorithm flow of the multi-objective optimal scheduling method of the sluice dam flood control system based on online data-driven evolutionary optimization of the present invention;

[0064] Figure 4 Construction diagram of the hydrodynamic model of the coupled sluice dam of the multi-objective optimal scheduling method of the sluice dam flood control system based on online data-driven evolutionary optimization of the present invention;

[0065] Figure 5 Schematic diagram of the sluice dam coupling method of the multi-objective optimal scheduling method of the sluice dam flood control system based on online data-driven evolutionary optimization of the present invention;

[0066] Figure 6 Pareto result diagram of the actual case optimization solution of the multi-objective optimal scheduling method of the sluice dam flood control system based on online data-driven evolutionary optimization of the present invention;

[0067] Figure 7 Comparison result diagram of the flow rate of the representative solution at the downstream control point of the multi-objective optimal scheduling method of the sluice dam flood control system based on online data-driven evolutionary optimization of the present invention. Detailed implementation manners

[0068] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0069] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0070] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0071] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper", "lower", "horizontal", "inner", etc. are used to indicate the orientation or positional relationship, it is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed during use. This is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, terms such as "first", "second", etc. are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0072] In addition, if the term "horizontal" appears, it does not mean that the component is required to be absolutely horizontal, but it can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and it does not mean that the structure must be completely horizontal, but it can be slightly inclined.

[0073] In the description of the embodiments of the present invention, it should also be noted that unless otherwise clearly specified and limited, if terms such as "set", "installed", "connected", "connected" are used, they should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0074] The following further describes the present invention in detail with reference to the drawings:

[0075] See Figure 1 , the present invention discloses a multi-objective optimal scheduling method for a sluice dam flood control system based on online data-driven evolutionary optimization, including:

[0076] S101: Collect information of the sluice dam scheduling system, express the required scheduling objectives and the restricted conditions in the form of functions to form an objective function and constraint conditions, and construct a multi-objective flood control optimal scheduling model for the sluice dam.

[0077] The total number of control variables of the sluice dam is where K is the number of different sluice dams on the river, the decision variables are the operating height or the opening of the sluice gate, and T k represents the number of decision-making scheduling intervals into which the operating duration of each sluice dam can be divided during a flood event, and the sluice dam has a separate operating rule within each scheduling interval;

[0078] The optimized competing objectives are: the lowest water level in front of the upstream dam and the smallest downstream peak flood discharge. The competing objectives are described by the following formula:

[0079]

[0080] F2 = min{maxQ down (t) / Q ori} (2)

[0081] Among them, F1 is the upstream flood control target, minimizing the sum of the highest water levels of each upstream sluice dam; F2 is the downstream flood control target, minimizing the peak flood flow at the downstream control point; H k (t) is the upstream water level of the Kth sluice dam at time t; D k,max is the maximum allowable water level of the kth sluice dam; Q down (t) is the flow rate at the downstream control point of the river channel at time t; Q ori is the flood discharge flow rate without sluice dam regulation.

[0082] Construct a multi-objective flood control optimal operation model for sluice dams, specifically:

[0083] The multi-objective flood control constraint conditions of sluice dams include water level limit, flow rate limit, and the operation of sluice dams:

[0084] D k,min < H k (t) < D k,max (3)

[0085] Q k,min < Q k (t) < Q k,max (4)

[0086] B k,min < B k (t) + δ < B k,max δ ∈ S (5)

[0087] Among them, D k,min 、and D k,max are respectively the maximum allowable water depth and the minimum allowable water depth of sluice dam k; Q k,min and Q k,max respectively represent the minimum flow rate and the maximum flow rate of sluice dam k; Each section of the river channel needs to ensure the requirement of the minimum ecological flow rate, and at the same time, the water level should not be higher than the maximum limit to prevent floods. B k,min and B k,max respectively represent the minimum opening height and the maximum opening height of the gate of the kth sluice dam; B k (t) is the operation state of the kth sluice dam at time t; δ is the change amount of the sluice dam operation, usually this value should not be too small, and its opening can meet the regulation requirements.

[0088] S102: Construct a multi-objective particle swarm optimization algorithm improved based on mutation factors.

[0089] Based on the standard particle swarm optimization algorithm, MOPSO is developed by introducing Pareto solutions; in the l-th iteration, the velocity and position of the i-th particle in the d-th dimensional search space are updated:

[0090]

[0091]

[0092] Among them, ω is the inertia weight; c 1 and c 2 are the individual learning and social learning factors of the particle respectively; r 1 and r 2 are random numbers between 0 and 1; pbest i d is the individual optimal solution of the i-th particle; gbest d is the global optimal solution of all particles in the d-th dimensional space; x and V are the position and velocity of the particle respectively;

[0093] In order to ensure that the algorithm has better global search ability in the early stage and has the ability to jump out of the local optimal trap in the later stage, an inertial mutation weight ω is introduced with the idea of mutation:

[0094]

[0095] Among them, ω min and ω max are the minimum and maximum values of the weight; when the change of velocity update is very small in the later stage, that is, V i d (l + 1) - V i d (l) ≤ 0.01V i d (l), the algorithm can jump out of the local optimum through a random mutation value, and its mutation probability is p, σ is the mutation factor, and it takes a random number between (0, 1).

[0096] S103: Based on the topographic data, hydrological data, number and construction location of sluice dams, and operation data of sluice dams of the sluice dam scheduling system, a two-dimensional hydrodynamic model coupling sluice dams is constructed;

[0097] S104: Construct a machine learning model, use the optimized decision variables as input data, and use the hydraulic parameters of the two-dimensional hydrodynamic model obtained under different flood conditions as output data to establish a mapping relationship between the input and output;

[0098] The hydraulic parameters of the two-dimensional hydrodynamic model are as follows: Mark the positions of the sluices and dams to be processed in the terrain file of the hydrodynamic model. Consider the water passing gate, i.e., the boundary of the gate grid, as a closed boundary, and stop the calculation of the flux by the HLLC Riemann solver at the side where this grid is located. Use the water depth h of the grid in front of the water passing gate as the head input parameter of the sluice orifice outflow formula to calculate the flow rate of each upstream grid passing through the gate. For a flat-bottomed flat gate with free outflow, the sluice orifice outflow formula is as follows:

[0099]

[0100] Among them, Q o is the flow rate passing through a single grid, m 3 / s; B is the grid width, m; e is the gate opening; H o is the head on each grid, m; g is the acceleration due to gravity, m / s 2 ; μ o is the comprehensive discharge coefficient, specifically:

[0101] μ 0 = 0.6 - 0.18×(e / H o )(10)

[0102] Calculate the change in water depth Δh of each grid cell at the gate within the same time step through the flow rate Q o passing through the gate. Subtract the change in water depth Δh from the water level of each grid upstream of the gate as the water depth value at the next moment, while add this change in water depth Δh to the grid downstream of the gate as the water depth value at the next moment. Specifically:

[0103]

[0104] In the formula, c l is the grid cell length, c k is the grid cell width, and dt is the time step of the model calculation.

[0105] S105: Incorporate the machine learning model into the multi-objective particle swarm optimization algorithm process in an online data-driven manner. Based on the prediction results of the machine learning model, introduce the expected improvement matrix filling criterion applicable to multiple objectives to select potential non-dominated candidate solutions as the evolutionary guidance in the offspring population during the machine learning evolution process;

[0106] In the machine learning model, select a particle individual in the offspring population as the evolutionary guidance through the filling criterion; To avoid the larger value of the EI criterion falling into the local optimum, improve this criterion by introducing a weighted term. The improved EI criterion comprehensively considers the predicted mean and standard deviation of the Kriging model and can balance the global optimum value and the local optimum value. The expression is:

[0107]

[0108] Among them, w e is a random number between 0 and 1, serving as the weighting coefficient of this criterion to balance the global optimal value and the local optimal value; y min is the optimal response value of the current sample point; the predicted value of any unknown point x follows a normal distribution, that is Φ and φ represent the cumulative distribution function and the probability density function of the standard normal distribution respectively; the predicted mean of the Kriging model is smaller, the greater the first term of the WEI criterion, and the predicted standard deviation of the model is larger, the greater the second term of the WEI criterion; w e serves as the weighting coefficient of this criterion to balance the global optimal value and the local optimal value, and is determined by the following formula:

[0109]

[0110] In the formula, the weighting coefficient w e changes in a non-linear relationship between (0, 1). This weight coefficient has a larger value in the early stage of the algorithm, and the global optimal search is better. In the later stage, this value decays rapidly, and the local search ability is stronger.

[0111] Introduce the expected improvement matrix filling criterion applicable to multi-objectives to select evolutionary guidance, which also includes: with the idea of a matrix, regarding the approximate Pareto solution set of the multi-objective optimization problem as the expansion of the single-objective optimal solution in two directions; in terms of dimension, the multi-objective expands from 1 dimension to multiple dimensions; in terms of the number of solutions, the multi-objective expands from 1 optimal solution to multiple Pareto solutions, and finally expands into a two-dimensional matrix; by integrating the expected improvement of each single objective of the research point for each approximate point beyond the Pareto front, a two-dimensional WEI matrix is obtained:

[0112]

[0113] Among them, X is an n-dimensional point, X = [x 1 , x 2 , …, x n ; M is the number of objective functions; J is the number of points on the Pareto front; each element in the WEIM matrix is a one-dimensional expected improvement function.

[0114] It also includes processing the WEIM matrix, specifically: by integrating the elements in the WEIM matrix to form a scalar to measure the expected improvement of the prediction point on the overall Pareto front; considering that the objective function is a multi-dimensional space, the elements of WEIM are combined into a scalar function through the Euler distance improvement function:

[0115]

[0116] According to the evolutionary guiding particles, use the coupled sluice-dam hydrodynamic model for calculation and calculate the true fitness function value; the fitness function value is used to optimize the evolutionary process of the algorithm and update the machine learning model training database.

[0117] S106: Based on the selected evolutionary guiding individuals, use the two-dimensional hydrodynamic model of the coupled sluice-dam for actual calculation to solve the multi-objective flood control optimization scheduling model of the sluice-dam, and realize the iteration of population evolution and the update of the machine learning model; finally, after reaching the limit filling times, obtain a set of Pareto solutions for the multi-objective flood control optimization scheduling of the sluice-dam.

[0118] See Figure 2 , the present invention discloses a multi-objective optimization scheduling system for a sluice-dam flood control system based on online data-driven evolutionary optimization, including:

[0119] A scheduling model module, which collects information on the sluice-dam scheduling system, expresses the required scheduling objectives and the restricted conditions in the form of functions to form an objective function and constraint conditions, and constructs a multi-objective flood control optimization scheduling model for the sluice-dam;

[0120] An optimization algorithm module, which constructs a multi-objective particle swarm optimization algorithm improved based on a mutation factor;

[0121] A hydrodynamic calculation module, which constructs a two-dimensional hydrodynamic model of the coupled sluice-dam based on the topographic data, hydrological data, number and construction location of the sluice-dams, and operation data of the sluice-dam scheduling system;

[0122] A machine learning module, which constructs a machine learning model, uses the optimized decision variables as input data, and uses the hydraulic parameters of the two-dimensional hydrodynamic model obtained under different flood conditions as output data to establish a mapping relationship between the input and output;

[0123] An online driving module, which couples the machine learning model into the multi-objective particle swarm optimization algorithm process in an online data-driven manner, and based on the prediction results of the machine learning model, introduces a desired improvement matrix filling criterion applicable to multiple objectives to select potential non-dominated candidate solutions in the offspring population during the evolutionary process as the evolutionary guidance;

[0124] An iterative update module, which based on the selected evolutionary guiding individuals, uses the two-dimensional hydrodynamic model of the coupled sluice-dam for actual calculation to solve the multi-objective flood control optimization scheduling model of the sluice-dam, and realizes the iteration of population evolution and the update of the machine learning model; finally, after reaching the limit filling times, obtain a set of Pareto solutions for the multi-objective flood control optimization scheduling of the sluice-dam.

[0125] Example: See Figure 3, the present invention discloses a multi-objective optimal scheduling method for a sluice and dam flood control system based on online data-driven evolutionary optimization, including the following steps:

[0126] Step 1, aiming at the conditions where the flood control objectives upstream and downstream of the river are conflicting or even influencing each other, establish a multi-objective flood control optimal scheduling model for sluices and dams, including multiple competing objective functions and various constraint conditions.

[0127] Step 2, construct a model optimization solution framework for directly solving the multi-objective flood control optimal scheduling model. This optimization solution framework is as Figure 3 shown. This optimization framework mainly consists of the following three parts: a two-dimensional hydrodynamic model coupling sluices and dams, a multi-objective evolutionary optimization algorithm, and a machine learning model. Among them, the two-dimensional hydrodynamic model coupling sluices and dams can accurately characterize the flood process, and as a solver for the objective function and constraint conditions in the optimization algorithm, it is of great significance to the accuracy of the optimal scheduling of flood control and drainage projects. Based on the calculation results of the two-dimensional hydrodynamic model, combined with the machine learning model to drive the evolutionary optimization algorithm online, the multi-objective flood control optimal scheduling of sluices and dams can be realized under limited computing resources.

[0128] Step 3, use the PYTHON language to write a multi-objective optimal scheduling program for the sluice and dam flood control system, and write a framework optimization main function according to the objective function and constraint conditions of the optimization model established in Step 1 and the optimization solution framework in Step 2. This main function is the optimization process of the multi-objective particle swarm optimization algorithm;

[0129] Step 4, according to the topographic data, hydrological data, number and construction location of sluices and dams, and operation data of the sluice and dam scheduling system in the basin, construct a river numerical model coupling sluices and dams to solve the hydraulic parameters required for the objective function and constraint conditions in Step 1. The construction of the basin model of the sluice and dam scheduling system in this case is as Figure 4 shown.

[0130] Step 5, use the PYTHON / C++ language to write a dynamic link library (DLL) of the source code of the hydrodynamic model coupling sluices and dams and a parameter calling program, use the hydrodynamic model as a solver for the adaptive multi-objective particle swarm optimization algorithm to realize the automatic import of model input files and the automatic export of calculation results; write a fitness value function according to the objective function and constraint conditions in Step 1.

[0131] Step 6, construct a machine learning model. The machine learning model selects the Kriging model that can describe the prediction variance. The input condition for model training is the decision variable of the sluice and dam operation, and the output condition of the model is the output result of the hydrodynamic model coupling sluices and dams in Step 4.

[0132] Step 7: Incorporate the machine learning model into the optimization algorithm process in an online data-driven manner. Considering that the solution to the multi-objective optimization problem is a set of non-dominated solution sets, it becomes very difficult to sample and select the most potential particles from the offspring population. Therefore, based on the prediction results of the machine learning model in Step 6, evolutionary guidance is selected by introducing an expected improvement matrix filling criterion applicable to multi-objectives, and then real calculations are performed through the hydrodynamic model of the sluice dam, thereby greatly reducing the number of calls to the hydrodynamic model and improving the optimization operation efficiency.

[0133] Step 8: Complete the writing of all functions in the solution framework in Step 2, and set the parameters of the main function of the algorithm program written in Step 3 according to the sluice dam scheduling system settings in Step 4. Run the optimization program. Finally, after the set filling criterion value reaches the set value, a set of Pareto solutions for the multi-objective operation scheduling of the sluice dam flood control system is obtained for the operation operator to select according to the risk preference and conduct flood management.

[0134] Among them, the specific construction steps of the multi-objective flood control optimization scheduling model of the sluice dam in Step 1 are as follows:

[0135] Step 1.1: Establish the multi-objective flood control objective function of the sluice dam. Propose the multi-objective flood control optimization problem of the sluice dam and solve it in the decision space. Assume that there are K different sluice dams on the river, and the decision variables are the operating height or gate opening of the sluice dam. The operating duration of each sluice dam in a flood event can be divided into T k decision scheduling intervals, and each sluice dam can have separate operating rules within each scheduling interval. Therefore, the total number of control variables of the sluice dam is The competing objectives for optimization are: (1) The lowest water level in front of the upstream dam; (2) The minimum downstream peak flood discharge. The competing objectives are described by the following formulas:

[0136]

[0137] F2 = min{maxQ down (t) / Q ori} (2)

[0138] Among them, F1 is the upstream flood control objective, minimizing the sum of the highest water levels of each upstream sluice dam; F2 is the downstream flood control objective, minimizing the peak flood discharge at the downstream control point; H k (t) is the upstream water level of the Kth sluice dam at time t; D k,max is the maximum allowable water level of the kth sluice dam; Q down (t) is the flow rate at the downstream control point of the river at time t; Q ori is the flood discharge when the sluice dam is not scheduled.

[0139] Step 1.2, establish the multi-objective flood control constraints for the sluice dam. Since a hydrodynamic model coupling the sluice dam is used to simulate the water flow propagation, the conservation of the discrete equations is ensured. The setting of constraints usually limits the search space and avoids the occurrence of abnormal situations. The multi-objective flood control constraints for the sluice dam include water level limits, flow rate limits, and the operation of the sluice dam:

[0140] D k,min <H k (t) < D k,max (3)

[0141] Q k,min <Q k (t) < Q k,max (4)

[0142] B k,min <B k (t) + δ < B k,max δ ∈ S (5)

[0143] Among them, D k,min 、and D k,max are the maximum allowable water depth and the minimum allowable water depth of the sluice dam k respectively; Q k,min and Q k,max represent the minimum flow rate and the maximum flow rate of the sluice dam k respectively; The minimum ecological flow rate requirement needs to be ensured for each section of the river channel, and at the same time, the water level should not be higher than the maximum limit to prevent floods. B k,min and B k,max represent the minimum gate opening height and the maximum gate opening height of the kth sluice dam respectively; B k (t) is the operating state of the kth sluice dam at time t; δ is the change in the operation of the sluice dam. Usually, this value should not be too small, and its opening can meet the regulation requirements.

[0144] The specific construction steps of the multi-objective particle swarm optimization algorithm in Step 3 are as follows:

[0145] Step 3.1, construction of the multi-objective particle swarm algorithm. Based on the standard particle swarm algorithm (PSO), this algorithm develops MOPSO by introducing Pareto solutions. This algorithm is still an evolutionary algorithm that imitates the foraging behavior of bird flocks. Different from the standard PSO, MOPSO uses an external particle archive to save the non-dominated solutions obtained during the evolutionary process, and then other particles will use them to guide their flight. In the l-th iteration, the velocity and position of the i-th particle in the d-th dimensional search space are updated:

[0146]

[0147]

[0148] Among them, ω is the inertia weight; c1 and c 2 are the individual learning and social learning factors of the particle, respectively; r 1 and r 2 are random numbers between 0 and 1; pbest i d is the individual optimal solution of the i-th particle; gbest d is the global optimal solution of all particles in the d-dimensional space; x and V are the position and velocity of the particle, respectively;

[0149] Step 3.2, in order to ensure that the algorithm has better global search ability in the early stage and has a certain ability to jump out of the local optimal trap in the later stage, an inertial mutation weight ω is introduced with the idea of mutation:

[0150]

[0151] where ω min and ω max are the minimum and maximum values of the weight; when the change of the velocity update in the later stage is very small, that is, V i d (l + 1) - V i d (l) ≤ 0.01V i d (l), the algorithm can jump out of the local optimum through a random mutation value, and its mutation probability is p, and σ is the mutation factor, taking a random number between (0, 1).

[0152] In step 4, the specific steps of the coupling of the river channel sluice dam and the surface water dynamic model are as follows:

[0153] Step 4.1, sluice dam grid marking processing. Mark the positions of the sluice dams to be processed in the topographic file of the hydrodynamic model. For example, mark the sluice dam topographic grid cells as 10, as Figure 5 shown.

[0154] Step 4.2, calculation of the flow through the sluice. Regard the water passing gate ( Figure 5 at the grid boundary of the gate) as a closed boundary, and stop the calculation of the flux by the HLLC Riemann solver at the side where the grid is located; take the water depth h of the grid in front of the gate numbered 10 as the head input parameter of the sluice orifice outflow formula to calculate the flow rate of each upstream grid passing through the gate. Taking the flat-bottomed flat gate with free outflow as an example, the sluice orifice outflow formula is as follows:

[0155]

[0156] where Q o is the flow rate passing through a single grid, m 3 / s; B is the grid width, m; e is the gate opening; H oThe water head on each grid, m; g is the acceleration due to gravity, m / s 2 ; μ o is the comprehensive discharge coefficient, specifically:

[0157] μ 0 = 0.6 - 0.18×(e / H o )(10)

[0158] Step 4.3, exchange of water volume information. Calculate the change in water depth Δh within the same time step for each grid cell at the gate through the discharge Q o at the gate, as shown in formula (11). The water level of each grid upstream of the gate minus the change in water depth Δh is used as the water depth value at the next moment, while for the grid downstream of the gate, the change in water depth Δh is added as the water depth value at the next moment, specifically:

[0159]

[0160] where c l is the length of the grid cell, c k is the width of the grid cell, and dt is the time step of the model calculation.

[0161] In step 7, the specific steps of coupling the machine learning model and the particle swarm algorithm through online data-driven are as follows:

[0162] Step 7.1, according to the machine learning model established in step 6, select a particle individual in the offspring population as the evolutionary guide through the filling criterion. Different from the single-objective optimization problem, the solution of the multi-objective optimization problem is a set of non-dominated solution sets, and the particles selected by the filling criterion may all be non-dominated solutions of the offspring. To avoid the larger value of the EI criterion falling into the local optimum, first improve this criterion by introducing a weighting term. The improved EI criterion comprehensively considers the predicted mean and standard deviation of the Kriging model and can balance the global optimum and the local optimum. The expression is as follows:

[0163]

[0164] where y min is the optimal response value of the current sample point; the predicted value of any unknown point x follows a normal distribution, that is Φ and φ represent the cumulative distribution function and probability density function of the standard normal distribution respectively; the predicted mean of the Kriging model is smaller, the first term of the WEI criterion is larger, and the predicted standard deviation of the model is larger, the second term of the WEI criterion is larger; w e is used as the weighting coefficient of this criterion to balance the global optimum and the local optimum, and is determined by the following formula:

[0165]

[0166] Wherein, the weighting coefficient w e Changes non-linearly between (0, 1). This weight coefficient has a larger value in the early stage of the algorithm, with better global search ability, and its value decays rapidly in the later stage, with stronger local search ability.

[0167] The predicted mean of the Kriging model The smaller it is, the larger the first term of the WEI criterion, and the predicted standard deviation of the model The larger it is, the larger the second term of the WEI criterion. This sampling criterion tends to add new sample points where the predicted value of the Kriging model is small and the prediction uncertainty is large, taking into account both local search ability and global search ability, which not only improves the optimization solution accuracy but also ensures that the sample points are avoided from falling into local optima.

[0168] Step 7.2, to solve the problem of sampling in multi-objective optimization, with the idea of matrix, the approximate Pareto solution set of the multi-objective optimization problem is regarded as the expansion of the single-objective optimal solution in two directions. In terms of dimension, the multi-objective expands from 1 dimension to multiple dimensions; in terms of the number of solutions, the multi-objective expands from 1 optimal solution to multiple Pareto solutions, so it will finally expand into a two-dimensional matrix. The improved filling criterion in Step 7.1 is also applicable. By integrating the research points on each objective and the single-objective expected improvement for each approximate point beyond the Pareto front, a two-dimensional WEI matrix can be obtained:

[0169]

[0170] Wherein, X is an n-dimensional point, X = [x 1 , x 2 , …, x n ; M is the number of objective functions; J is the number of points on the Pareto front; each element in the WEIM matrix is a one-dimensional expected improvement function.

[0171] Step 7.3, the WEIM matrix cleverly describes the expected improvement of the prediction points on each Pareto front point in each objective direction. However, the filling criterion for selecting new samples must be a scalar rather than a matrix, so it cannot be directly applied to the algorithm optimization process. By integrating the elements in the WEIM matrix in Step 7.2 to form a scalar to measure the expected improvement of the prediction points on the overall Pareto front. Considering that the objective function is in a multi-dimensional space, the elements of WEIM are combined into a scalar function through the Euler distance improvement function:

[0172]

[0173] Step 7.4: Based on the evolutionary guiding particles selected in Step 7.3, use the coupled sluice-dam hydrodynamic model established in Step 4 for calculation, and calculate the true fitness function value through Step 5. This true value is used on the one hand for the evolutionary process of the optimization algorithm in Step 3, and on the other hand for updating the training database of the machine learning model in Step 6.

[0174] The present invention shows good effects in the application of the actual cascade sluice-dam flood control system. Figure 6 The Pareto scheduling rules obtained by the present invention are shown, taking the selected representative solutions as references, as Figure 7 shown. This optimization rule prevents the superposition of flood peaks in the downstream area, effectively alleviates the flood control pressure in the downstream river section, and at the same time, the highest water levels of each sluice-dam section in the upstream do not exceed the upper limit, ensuring the safety of the upstream area. The optimized scheduling scheme obtained in the case of the present invention is obtained through hydrodynamic model calculation, thus ensuring the calculation accuracy. At the same time, through the method proposed by the present invention, the original 3500 iterative calculations are reduced to 130 times, providing sufficient lead time for emergency response for sluice-dam scheduling management personnel. The 19 Pareto scheduling rules obtained also provide more decision-making references for schedulers. It is an accurate, efficient and practical flood control scheduling method.

[0175] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A multi-objective optimization scheduling method for dam flood control system based on online data-driven evolutionary optimization, characterized in that: include: Collect information on the dam dispatching system, express the required dispatching objectives and constraints in the form of functions, form objective functions and constraints, and build a multi-objective flood control optimization dispatching model for dams and gates; Construct a multi-objective particle swarm optimization algorithm based on mutation factor improvement; Based on the topographic data, hydrological data, number and location of dams and gates, and operation data of dams and gates dispatching system, a two-dimensional hydrodynamic model of coupled dams and gates is constructed; Construct a machine learning model with the optimization decision variables as input data and the hydraulic parameters of the two-dimensional hydrodynamic model obtained under different flood conditions as output data, and establish a mapping relationship between input and output; The machine learning model is coupled into the multi-objective particle swarm optimization algorithm process in an online data-driven manner. Based on the prediction results of the machine learning model, the expected improvement matrix filling criterion suitable for multiple objectives is introduced to select potential non-dominated candidate solutions in the offspring population of the evolutionary process as evolutionary guides. Based on the selected evolutionary guided individuals, a two-dimensional hydrodynamic model of coupled dams is used for real calculations to solve the multi-objective flood control optimization scheduling model of dams and gates, and to achieve the iteration of population evolution and the update of machine learning models. After reaching the limit filling number, a set of Pareto solutions for the multi-objective flood control optimization scheduling of dams and gates are obtained. The information of the gate and dam dispatching system is collected, and the required dispatching target and the restricted conditions are expressed in the form of functions to form the objective function and the constraint conditions, which are specifically: The total number of control variables of the dam is Among them, K is the number of different dams on the river, the decision variable is the dam operating height or gate opening, T k It indicates the number of decision-making dispatching intervals that each dam's operating time in a flood event can be divided into, and each dam has its own operating rules within each dispatching interval; The competition objectives for optimization are: the lowest water level in front of the upstream dam and the smallest flood peak flow in the downstream. The following formula is used to describe the competition objectives: F2=min{maxQ down (t) / Q ori } (2) Among them, F1 is the upstream flood control target, which minimizes the sum of the highest water levels of the upstream dams; F2 is the downstream flood control target, which minimizes the peak flow at the downstream control point; H k (t) is the upstream water level of the Kth dam at time t; D k,max is the maximum allowable water level of the kth dam; Q down (t) is the flow rate at the downstream control point at time t; Q ori It is the flood discharge when no dam or sluice gate regulation is performed; The multi-objective flood control optimization scheduling model for dams is constructed as follows: The multi-objective flood control constraints of dams and gates include water level restrictions, flow restrictions, and the operation of dams and gates: D k,min <H k (t)<D k,max (3) Q k,min <Q k (t)<Q k,max (4) B k,min <B k (t)+δ<B k,max δ∈S (5) Among them, D k,min , and D k,max are the maximum and minimum permissible water depths of dam k respectively; Q k,min and Q k,max They represent the minimum flow and maximum flow of the dam k respectively; each section of the river needs to ensure the minimum ecological flow requirement, and the water level should not be higher than the maximum limit to prevent floods; B k,min and B k,max They represent the minimum and maximum heights of the gate opening of the kth dam; B k (t) is the operating state of the kth dam at time t; δ is the change in the operation of the dam, and its opening can meet the control requirements; The hydraulic parameters of the two-dimensional hydrodynamic model are as follows: the position of the dam to be processed is marked in the hydrodynamic model terrain file, the gate grid boundary is regarded as a closed boundary, and the calculation of the flux by the HLLC Riemann solver is stopped at the edge of the grid; the grid water depth h in front of the gate is used as the head input parameter of the gate hole outflow formula to calculate the flow rate of each grid upstream through the gate; for a flat-bottomed flat gate with free outflow, the gate hole outflow formula is as follows: Among them, Q o is the flow rate through a single grid, m 3 / s; B is the grid width, m; e is the gate opening; H o is the water head on each grid, m; g is the gravitational acceleration, m / s 2 ;μ o is the comprehensive flow coefficient, specifically: μ0=0.6-0.18×(e / H o ) (10) Flow rate through the gate Q o Calculate the change in water depth Δh in the same time step of each grid unit at the gate. Subtract the change in water depth Δh from the water level of each grid upstream of the gate as the water depth value at the next moment, and add the change Δh to the grid downstream of the gate as the water depth value at the next moment. Specifically: In the formula, c l is the grid unit length, c k is the grid cell width, and dt is the model calculation time step.

2. The multi-objective optimization scheduling method for dam flood control system based on online data-driven evolutionary optimization according to claim 1 is characterized in that: The multi-objective particle swarm optimization algorithm based on mutation factor improvement is constructed as follows: based on the standard particle swarm algorithm, MOPSO is developed by introducing the Pareto solution; in the lth iteration, the speed and position of the i-th particle in the d-dimensional search space will be updated: Among them, ω is the inertia weight; c1 and c2 are the individual learning and social learning factors of the particles respectively; r1 and r2 are random numbers between 0 and 1; pbest i d is the individual optimal solution of the i-th particle; gbest d is the global optimal solution for all particles in the d-th dimension space; x and V are the position and velocity of the particle respectively; In order to ensure that the algorithm has better global search capabilities in the early stage and the ability to jump out of the local optimal trap in the later stage, the inertial mutation weight ω is introduced with the idea of ​​mutation: Among them, ω min and ω max is the minimum and maximum value of the weight; when the later speed update changes to V i d (l+1)-V i d (l)≤0.01V i d (l), the algorithm can jump out of the local optimum through random mutation value, the mutation probability is p, σ is the mutation factor, and it takes a random number between (0,1).

3. The multi-objective optimization scheduling method for dam flood control system based on online data-driven evolutionary optimization according to claim 1 is characterized in that: Based on the prediction results of the machine learning model, the expected improvement matrix filling criterion suitable for multiple objectives is introduced to select potential non-dominated candidate solutions as evolution guides in the offspring population of the evolution process. Specifically, in the machine learning model, a particle individual is selected as the evolution guide in the offspring population through the filling criterion; in order to avoid the value of the EI criterion falling into the local optimum, the criterion is improved by introducing a weighted term. The improved EI criterion comprehensively considers the predicted mean and standard deviation of the Kriging model, and balances the global optimal value and the local optimal value. The expression is: Among them, y min is the optimal response value of the current sample point; the predicted value of any unknown point x follows the normal distribution, that is, Φ and φ represent the cumulative distribution function and probability density function of the standard normal distribution, respectively; the predicted mean of the Kriging model The smaller it is, the larger the first term of the WEI criterion is, and the predicted standard deviation of the model is The larger the value is, the larger the second term of the WEI criterion is; e The weighted coefficient of this criterion balances the global optimum and the local optimum and is determined by the following formula: In the formula, the weighting coefficient W e The weighted coefficient changes in a nonlinear relationship between (0,1). The global search capability is better in the early stage of the algorithm, while the weighted coefficient decays faster in the later stage, and the local search capability is stronger.

4. The multi-objective optimization scheduling method for dam flood control system based on online data-driven evolutionary optimization according to claim 3 is characterized in that: The introduction of the expected improvement matrix filling criterion applicable to multiple objectives selects potential non-dominated candidate solutions in the offspring population of the evolutionary process as an evolution guide, and also includes: with the help of the matrix idea, the approximate Pareto solution set of the multi-objective optimization problem is regarded as the expansion of the single-objective optimal solution in two directions; in terms of dimension, the multi-objective is expanded from one dimension to multiple dimensions; in terms of the number of solutions, the multi-objective is expanded from one optimal solution to multiple Pareto solutions, and finally expanded to a two-dimensional matrix; by integrating the research point on each objective, for each single-objective expected improvement beyond the Pareto frontier approximate point, a two-dimensional WEI matrix is ​​obtained: Where X is an n-dimensional point, X=[x1,x2,…,x n ]; M is the number of objective functions; J is the number of points on the Pareto front; each element in the WEIM matrix is ​​a one-dimensional expected improvement function.

5. The multi-objective optimization scheduling method for dam flood control system based on online data-driven evolutionary optimization according to claim 4 is characterized in that: It also includes processing the WEIM matrix, specifically: integrating the elements in the WEIM matrix to form a scalar measure of the expected improvement of the predicted point on the overall Pareto frontier; considering that the objective function is a multidimensional space, the elements of WEIM are combined into a scalar function through the Euler distance improvement function: According to the evolutionary guided particles, the coupled sluice-dam hydrodynamic model is used for calculation and the true fitness function value is calculated; the fitness function value is used to optimize the evolutionary process of the algorithm and update the machine learning model training database.

6. A multi-objective optimization scheduling system for dam flood control systems based on online data-driven evolutionary optimization, characterized in that: include: A scheduling model module, wherein the scheduling model module collects information of the dam and gate scheduling system, expresses the required scheduling objectives and the constraints in the form of functions, forms objective functions and constraints, and constructs a multi-objective flood control optimization scheduling model for dam and gate; An optimization algorithm module, wherein the optimization algorithm module constructs a multi-objective particle swarm optimization algorithm based on mutation factor improvement; A hydrodynamic calculation module, which constructs a two-dimensional hydrodynamic model of coupled dams based on topographic data, hydrological data, the number and location of dams, and operation data of dams and sluices of the dam dispatching system; A machine learning module, wherein the machine learning module constructs a machine learning model, takes the optimization decision variables as input data, takes the hydraulic parameters of the two-dimensional hydrodynamic model obtained under different flood conditions as output data, and establishes a mapping relationship between the input and the output; An online driving module, wherein the online driving module couples the machine learning model into the multi-objective particle swarm optimization algorithm process in an online data-driven manner, and introduces an expected improvement matrix filling criterion applicable to multiple objectives to select potential non-dominated candidate solutions in the offspring population of the evolution process as an evolution guide based on the prediction results of the machine learning model; The iterative update module uses a two-dimensional hydrodynamic model of coupled dams to perform real calculations based on the selected evolutionary guided individuals to solve the multi-objective flood control optimization scheduling model of dams and gates, and realize the iteration of population evolution and the update of the machine learning model. After reaching the limit filling number, a set of Pareto solutions for the multi-objective flood control optimization scheduling of dams and gates are obtained. The information of the gate and dam dispatching system is collected, and the required dispatching target and the restricted conditions are expressed in the form of functions to form the objective function and the constraint conditions, which are specifically: The total number of control variables of the dam is Among them, K is the number of different dams on the river, the decision variable is the dam operating height or gate opening, T k It indicates the number of decision-making dispatching intervals that each dam's operating time in a flood event can be divided into, and each dam has its own operating rules within each dispatching interval; The competition objectives for optimization are: the lowest water level in front of the upstream dam and the smallest flood peak flow in the downstream. The following formula is used to describe the competition objectives: F2=min{maxQ down (t) / Q ori } (2) Among them, F1 is the upstream flood control target, which minimizes the sum of the highest water levels of the upstream dams; F2 is the downstream flood control target, which minimizes the peak flow at the downstream control point; H k (t) is the upstream water level of the Kth dam at time t; D k,max is the maximum allowable water level of the kth dam; Q down (t) is the flow rate at the downstream control point at time t; Q ori It is the flood discharge when no dam or sluice gate regulation is performed; The multi-objective flood control optimization scheduling model for dams is constructed as follows: The multi-objective flood control constraints of dams and gates include water level restrictions, flow restrictions, and the operation of dams and gates: D k,min <H k (t)<D k,max (3) Q k,min <Q k (t)<Q k,max (4) B k,min <B k (t)+δ<B k,max δ∈S (5) Among them, D k,min , and D k,max are the maximum and minimum permissible water depths of dam k respectively; Q k,min and Q k,max They represent the minimum flow and maximum flow of the dam k respectively; each section of the river needs to ensure the minimum ecological flow requirement, and the water level should not be higher than the maximum limit to prevent floods; B k,min and B k,max They represent the minimum and maximum heights of the gate opening of the kth dam; B k (t) is the operating state of the kth dam at time t; δ is the change in the operation of the dam, and its opening can meet the control requirements; The hydraulic parameters of the two-dimensional hydrodynamic model are as follows: the position of the dam to be processed is marked in the hydrodynamic model terrain file, the gate grid boundary is regarded as a closed boundary, and the calculation of the flux by the HLLC Riemann solver is stopped at the edge of the grid; the grid water depth h in front of the gate is used as the head input parameter of the gate hole outflow formula to calculate the flow rate of each grid upstream through the gate; for a flat-bottomed flat gate with free outflow, the gate hole outflow formula is as follows: Among them, Q o is the flow rate through a single grid, m 3 / s; B is the grid width, m; e is the gate opening; H o is the water head on each grid, m; g is the gravitational acceleration, m / s 2 ;μ o is the comprehensive flow coefficient, specifically: μ0=0.6-0.18×(e / H o ) (10) Flow rate through the gate Q o Calculate the change in water depth Δh in the same time step of each grid unit at the gate. Subtract the change in water depth Δh from the water level of each grid upstream of the gate as the water depth value at the next moment, and add the change Δh to the grid downstream of the gate as the water depth value at the next moment. Specifically: In the formula, c l is the grid unit length, c k is the grid cell width, and dt is the model calculation time step.

Images