A flood control scheduling method based on the improved grey wolf algorithm with fusion of multiple strategies

Abstract

The invention discloses a flood control dispatching method based on a fusion multi-strategy improved gray wolf algorithm, comprising the following steps: obtaining sample data of floods and reservoirs; establishing a flood control dispatching model; improving the gray wolf algorithm, using an improved Tent chaotic map to initialize the population, mixing Lévy walk and tangent walk and adding an elite retention strategy to determine the position of the α wolf, parabolic spiral approach to determine the position of the β wolf, determining the position of the γ wolf according to the positions of the α wolf and the β wolf, combining Cauchy mutation and dynamic weight coefficient to determine the position of the ω wolf, adding a cosine convergence factor to obtain a fusion multi-strategy improved gray wolf algorithm: using the fusion multi-strategy improved gray wolf algorithm to simulate flood control dispatching. The invention realizes the combination of local search and global search, avoids the algorithm from falling into premature, improves the algorithm optimization accuracy, and meets the requirements of effectively reducing the flood peak in flood control dispatching.

Description

Technical Field

[0001] The invention relates to the field of reservoir flood control dispatching, and in particular to a flood control dispatching method based on an improved grey wolf algorithm integrating multiple strategies. Background Art

[0002] Swarm intelligence algorithms are generally used in the field of flood control scheduling for the following reasons: 1. Maximizing the utilization of flood resources: Optimization algorithms can help managers make the best decisions under limited water resources, such as minimizing reservoir discharge to ensure the safety of downstream flood control objects; 2. Reducing flood control risks: By simulating different design floods, optimization algorithms can help predict and reduce the risks caused by floods, including property losses and casualties; 3. Improving the efficiency of flood control scheduling: Optimization algorithms can speed up the decision-making process and quickly provide the best decision-making plan by automating the analysis of large amounts of data; 4. Enhancing the adaptability and flexibility of flood control scheduling: With climate change and changes in environmental conditions, optimization algorithms can adjust parameters to adapt to new flood conditions and provide flexible flood control strategies.

[0003] In general, the application of swarm intelligence algorithms in flood control scheduling provides an efficient, flexible and reliable way to deal with complex flood control problems and ensure that reservoirs can protect the lives and property of people downstream. At present, most flood control scheduling studies use: Whale Algorithm (WOA): This algorithm has a high convergence speed and convergence accuracy in continuous optimization problems, Grey Wolf Algorithm (GWO): This algorithm has a high convergence speed and global search ability when solving optimization problems, Particle Swarm Algorithm (PSO): This algorithm can handle various types of optimization problems, whether continuous or discrete. However, reservoir flood control scheduling problems have the characteristics of many decision variables and many constraints, especially large floods with long cycles. Most swarm intelligence algorithms are prone to "dimensional disaster" and the optimization effect is reduced. Summary of the invention

[0004] Purpose of the invention: The present invention aims to provide a flood control scheduling method based on the improved grey wolf algorithm integrating multiple strategies, which can escape the dimensionality disaster and enhance the ability to handle high-dimensional optimization problems.

[0005] Technical solution: The flood control dispatching method based on the improved grey wolf algorithm integrating multiple strategies described in the present invention comprises the following steps:

[0006] (1) Obtain sample data on floods and reservoirs;

[0007] (2) Based on the objective function minf(x) of minimizing the maximum downstream discharge during flood control scheduling, a flood control scheduling model is established by taking into account the water balance constraint, the outflow flow range constraint, the outflow flow fluctuation constraint and the upper limit constraint of reservoir flood storage;

[0008] (3) Improved Gray Wolf Algorithm: Use improved Tent chaotic mapping to initialize the population, update the position of wolf α by integrating Lévy walk and tangent walk, and finally determine the position of wolf α according to the elite retention strategy, use parabolic spiral approach to determine the position of wolf β, determine the position of wolf γ according to the positions of wolf α and wolf β, determine the position of wolf ω by combining Cauchy mutation and dynamic weight coefficient, and use cosine convergence factor in the calculation of exploration control factor to obtain the improved Gray Wolf Algorithm with fusion of multiple strategies;

[0009] (4) An improved Grey Wolf Algorithm with fusion of multiple strategies is used to simulate flood control.

[0010] Furthermore, in step (3), the Lévy walk and the tangent walk are integrated, and the rule for updating the position of the α wolf is as follows:

[0011]

[0012] Where, X a,new is the updated wolf position, X a To update the position of the wolf before α, Levy_step is the step length of the Levy walk, tan_step is the step length of the tangent walk, T is the number of iterations, and max_T is the maximum number of iterations.

[0013] Furthermore, in step (3), the elite retention strategy is as follows:

[0014]

[0015] Where X' a To finally determine the position of wolf α, f(X a,new ) is the fitness value of wolf α after update, f(X a ) is the fitness value of wolf α before updating.

[0016] Furthermore, in step (3), the step length of the Levy walk Levy_step is

[0017]

[0018] In the formula, x and y are variables that obey the normal distribution, and c is a constant with a value of 1.5.

[0019] The step length of the tangent walk tan_step is

[0020] tan_step=0.1·tan(teta)·sign(rand-0.5) / log(1+T)

[0021] Where teta is the angle control parameter, rand is a random number in (0,1), if the value of rand is greater than 0.5, sign() returns 1, if the value of rand is less than 0.5, sign() returns -1, if the value of rand is equal to 0.5, sign() returns 0.

[0022] The angle control parameter teta is

[0023] teta=rand(0,1)·π / 2.5

[0024] Where rand(0,1) is a random number between 0 and 1.

[0025] Furthermore, in step (3), the position of the β wolf is determined by using a parabolic spiral approach, as follows:

[0026]

[0027] Where, X β,new is the updated position of β wolf, X β To update the position of the previous β wolf, s is a parabolic control factor, rand(0,1) is a random number between 0 and 1, rand is a random number in (0,1), if the value of rand is greater than 0.5, sign() returns 1, if the value of rand is less than 0.5, sign() returns -1, if the value of rand is equal to 0.5, sign() returns 0.

[0028] Furthermore, in step (3), the position of the γ wolf is determined according to the positions of the α wolf and the β wolf, as follows:

[0029] X γ , new =(p-1)·X γ +(1-p)·X β,new +(1-p)·X'α

[0030] Where, X γ,new The updated position of wolf γ is determined according to the positions of wolf α and wolf β, X γ is the position of γ wolf before updating, and p is a random number between (0,1).

[0031] Furthermore, in step (3), the position X of the ω wolf is determined by combining the Cauchy mutation and the dynamic weight coefficient. ω :

[0032] X ω =X1·M1+X2·M2+X3·M3+0.01·tan(π·(rand(0,1)-0.5))

[0033]

[0034] Where X1, X2, and X3 represent the updated positions of the three common wolves ω under the leadership of wolves α, β, and δ, and M i represents the inertia weight coefficient adopted by ordinary wolf packs, i=1,2,3; m1, m2, m3 represent the inertia weight coefficients of wolves α, β, δ respectively, and rand is a random number in (0,1).

[0035] Furthermore, in step (3), the cosine convergence factor a is used in the calculation of the exploration control factor:

[0036]

[0037] Furthermore, in step (2), the flood control dispatch model is as follows:

[0038] minf(x)=max(QO(t))

[0039]

[0040] Where V(t) and V(t-1) are the reservoir capacities at the end of period t and period t-1, respectively, in units of 100 million m 3 ; QI(t) and QO(t) are the inflow and outflow of the reservoir at the end of period t, respectively, in m 3 / s; Δt is the dispatching period interval, 1 day for large floods and 1 hour for small floods; QO max (t) is the upper limit of the outbound flow rate in period t, in m 3 / s; ΔQO max is the maximum fluctuation constraint allowed by the reservoir, in m 3 / s; V max The maximum flood storage capacity allowed by the reservoir, in billion m 3 ; Z(1) represents the starting water level, Z0 represents the flood limit water level, and regulation starts from the flood limit water level after entering the flood control stage.

[0041] Furthermore, in step (3), the improved Tent chaotic map is

[0042]

[0043] In the formula, x i is a randomly generated population individual, x i+1 The population individuals are reinitialized using the improved Tent chaotic map, rand(0,1) is a random number between 0 and 1, and μ is a control parameter, which is 0.015 in the present invention.

[0044] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. The present invention adopts a leader-following strategy. First, the first leader α wolf is updated by combining Lévy walk with tangent walk and adding an elite retention strategy. Then, the β wolf is updated according to the position of the first leader using a parabolic spiral approach. Finally, the third leader γ wolf is updated according to the position of the α wolf and the β wolf. This strategy realizes the combination of local search and global search, avoids premature convergence of IGWO and improves the optimization accuracy, and meets the requirements of effectively reducing flood peaks in flood control scheduling; 2. The present invention integrates tangent flight with Lévy flight. Using tangent flight in the early and late stages of iteration helps the algorithm Development and convergence, using Levy flight in the middle of iteration to jump out of the local optimum as much as possible; 3. The present invention uses Cauchy mutation for ω wolf and adds dynamic weight update coefficient, which improves the ability of IGWO to jump out of the local optimal solution; 4. The present invention uses improved Tent chaos mapping for initialization, improves the distribution quality of the population in the search space, strengthens the global search ability of the algorithm, and enables IGWO to find better solutions than other algorithms in the early stage of iteration; 5. The present invention solves the problem that the existing algorithms used in flood control scheduling are prone to dimensional disasters. Compared with the existing methods, the present invention can effectively reduce the maximum downstream flow by about 50m during the scheduling process. 3 / s-100m 3 / s, which means that the method provided by the present invention is trustworthy in flood control scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the present invention;

[0046] Figure 2 This is a comparison chart between Levy Walk and Tangent Walk;

[0047] Figure 3 This is a graph showing the parabolic control factor S changing with the number of iterations. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with the accompanying drawings.

[0049] The flood control scheduling method based on the improved grey wolf algorithm with fusion of multiple strategies described in the present invention mainly includes the following seven steps:

[0050] Step 1: Obtain sample data of floods and reservoirs, including: flood process lines, reservoir water level and storage capacity relationship curves, characteristic water levels of reservoirs such as flood limit water level, design flood level, verification flood level, and controlled discharge of reservoirs under different flood control standards.

[0051] Step 2: Establish an objective function that minimizes the maximum downstream flow rate during flood control scheduling, and consider the flood control scheduling mathematical model that takes into account the water balance constraint, the outflow flow range constraint, the outflow flow fluctuation constraint, and the reservoir flood storage upper limit constraint: The present invention provides a more general objective function and constraint conditions in flood control scheduling, and the objective function is shown in formula (1). The constraint conditions are shown in formula (2).

[0052] minf(x)=max(QO(t)) (1)

[0053]

[0054] In formulas (1) and (2), QO(t) is the outbound flow in period t, m 3 / s; V(t) and V(t-1) are the reservoir capacities at the end of period t and period t-1, respectively, billion m 3 ; QI(t) and QO(t) are the inflow and outflow of the reservoir at the end of period t, respectively, m 3 / s; Δt is the dispatching period interval, 1 day for large floods and 1 hour for small floods; QO max (t) is the upper limit of the outbound flow rate in period t, m 3 / s; ΔQO max is the maximum fluctuation constraint allowed by the reservoir, m 3 / s; V max is the maximum flood storage capacity allowed by the reservoir, billion m 3 ; Z(1) represents the starting water level, Z0 represents the flood limit water level, and regulation starts from the flood limit water level after entering the flood control stage.

[0055] Step 3: Improving and executing the Gray Wolf Algorithm mainly includes the following steps:

[0056] (1) The improved Tent chaotic map is used to initialize the population. The position of the α wolf is determined by using a mixed Lévy walk and tangent walk with an elite retention strategy. The position of the β wolf is determined by using a parabolic spiral approach. The position of the γ wolf is determined based on the positions of the α wolf and the β wolf. The position of the ω wolf is determined by combining the approximate Cauchy mutation with the dynamic weight coefficient, and a cosine convergence factor is added.

[0057] (2) Run the improved Grey Wolf Algorithm: The previous content of this section has explained the basic data, constraints, and objective functions required for flood control scheduling problems. On this basis, the improved Grey Wolf Algorithm that integrates multiple strategies is used to simulate flood control scheduling.

[0058] The Gray Wolf Algorithm (GWO) is a meta-heuristic optimization algorithm proposed by Professor Mirjalili and others from Griffith University, Australia in 2014. It has attracted much attention due to its high efficiency and reliability and has been widely used in engineering practice. It finds the optimal solution by simulating the social structure and predation behavior of gray wolf groups. In the social structure of gray wolves, the first layer is the highest leader among the gray wolves, called α, who mainly leads hunting, roosting, and sharing; the second layer is the secondary leader among the gray wolves, called β, who assists α in managing the group and obeys α wolf; the third layer is γ, which is mainly responsible for reconnaissance, assists α and β wolves and obeys α and β; the fourth layer is the lowest-level gray wolf in the population, called ω, which needs to obey α, β, and γ. The hunting process of the wolf pack can be divided into searching for prey, surrounding prey, and attacking prey. First, the α wolf leads the wolf pack to search for prey. After finding the prey, it commands the β and γ wolves closest to it to attack. If the prey escapes, the ω wolf pack will besiege it. In the gray wolf algorithm, the exploration control factor A determines whether to surround the prey or continue to search for other prey. Its definition is shown in formula (3):

[0059]

[0060] In the formula, max_T is the maximum number of iterations, T is the current number of iterations, and r1 is a random number between (0, 1). When the absolute value of A is greater than 1, the wolf pack will continue to search for prey. When the absolute value of A is less than 1, the wolf pack will shrink the encirclement and besiege the prey.

[0061] Based on the exploration control factor A, the mathematical model of gray wolves searching for and surrounding prey can be obtained as shown in formula (4):

[0062]

[0063] Where C is a random number between (0,2), X(T) is the position of the T-th generation gray wolf, and X p (T) is the position of the Tth generation prey. In the gray wolf algorithm, it is considered that α, β, and γ wolves are closest to the prey. Therefore, their positions can be used to calculate the distance that the gray wolf moves to the prey. The mathematical model is shown in formula (5):

[0064]

[0065] Where C1, C2, and C3 are random numbers between (0, 2).

[0066] Based on formula (5), the position update formula of the ω wolf pack can be obtained, as shown in formula (6):

[0067]

[0068] On this basis, GWO is improved. The present invention proposes an improved gray wolf algorithm (IGWO) that integrates multiple strategies. Since there is no prior information, the gray wolf individuals in GWO are generated by random initialization. Random initialization can easily cause individuals to be unevenly distributed in the search space, lose population diversity, and cause individuals to be far away from the global optimal solution, resulting in poor optimization results of the algorithm. Chaos has the characteristics of ergodicity, randomness, and regularity. Some scholars have introduced Tent chaotic mapping into GWO to slightly improve the optimization effect, but Tent chaotic mapping has unstable points, which will affect the optimization performance and robustness of the algorithm. Therefore, the present invention proposes and introduces an improved Tent chaotic mapping. A random factor is added to the original Tent chaotic mapping, which helps the Tent chaotic mapping avoid unstable points, thereby improving the distribution quality of the population in the search space, strengthening the global search capability, and improving the optimization effect of the algorithm. The improved Tent chaotic mapping formula is shown in formula (7):

[0069]

[0070] In the formula, x i is a randomly generated population individual, x i+1 The population individuals are reinitialized using the improved Tent chaotic map, rand(0,1) is a random number between 0 and 1, and μ is a control parameter, which is 0.015 in the present invention.

[0071] The above content of this section has described the social system and hunting behavior of the real gray wolf group. Although the hierarchy of the gray wolf group is imitated in GWO, the leadership of α wolf in the gray wolf group is absolute. In GWO, the three gray wolf individuals with the best fitness are selected as α wolf, β wolf, and γ wolf, which means that the positions of β wolf and γ wolf have little to do with α wolf, which is inconsistent with the social system of the real gray wolf group. The present invention proposes a leader following strategy to let β wolf update its position according to the position of α wolf, and γ wolf updates its position according to the positions of α wolf and β wolf. In summary, the position of α wolf has a great influence on the optimization result, so the present invention introduces two wandering strategies to update α wolf: Lévy flight is widely used in many optimization algorithms such as cuckoo algorithm (CS) and marine predator algorithm (MPA) due to its unique search mode. It combines the advantages of local search and global search. Through occasional long-distance jumps, it can quickly explore unexplored areas in the search space to avoid the algorithm from falling into the local optimum. Through a shorter step length, it can also carefully search the adjacent area to improve the accuracy of the solution. Its mathematical model is shown in formula (8) and formula (9):

[0072]

[0073] Where Levy_step is the step length of Levy walk, x and y are two normally distributed variables, and c is a parameter whose value is generally 1.5.

[0074] The normal distribution of x and y is as follows:

[0075]

[0076] However, the walk steps generated by Levy flight are highly random. In most cases, the step length interval is narrow and cannot cover all areas of the search space. This may cause the algorithm to have a search distance that is too small in the early iteration and a search distance that is too large in the late iteration. Compared with Levy flight, tangent walk helps to explore the search space more effectively. It can also realize local search and global search through the tangent function, which can avoid the shortcomings of Levy flight. Its mathematical model is shown in equations (10) and (11):

[0077] tan_step=0.1·tan(teta)·sign(rand-0.5) / log(1+T) (10)

[0078] Where tan_step is the step size of the tangent walk, teta is the angle control parameter, rand is a random number in (0,1), if the value of rand is greater than 0.5, sign() returns 1, if the value of rand is less than 0.5, sign() returns -1, if the value of rand is equal to 0.5, sign() returns 0; T is the number of iterations.

[0079] The formula for the angle control parameter teta is as follows:

[0080] teta=rand(0,1)·π / 2.5 (11)

[0081] Where rand(0,1) is a random number between 0 and 1.

[0082] However, the long-distance jump distance of tangent walk is not as good as that of Levy flight, so the present invention combines tangent flight with Levy flight. Using tangent flight in the early and late stages of iteration is helpful for algorithm development and convergence, and using Levy flight in the middle stage of iteration is used to jump out of the local optimum as much as possible. Its mathematical model is shown in formula (12):

[0083]

[0084] Where T is the number of iterations and max_T is the maximum number of iterations.

[0085] According to the theory of natural selection and survival of the fittest, in order to avoid the fitness value of wolf α after wandering being inferior to the fitness value before wandering, the present invention adds a greedy strategy, and the final position update formula of wolf α is shown in formula (13):

[0086]

[0087] The leader-following strategy enables β wolf to update its position according to the position of α wolf. When searching for prey, β wolf sometimes explores in front of α wolf, and when attacking prey, it will follow α wolf and obey α wolf's command. The present invention proposes a parabolic spiral walk to imitate the behavior of β wolf. The parabolic control factor s combined with the spiral function makes β wolf have a 50% probability to continue to develop and explore the area in front of α wolf, and a 50% probability to follow α wolf to attack prey. Its mathematical model is shown in equation (14) and equation (15):

[0088]

[0089] Where s is the parabolic control factor, rand(0,1) is a random number between 0 and 1, rand is a random number in (0,1), if the value of rand is greater than 0.5, sign() returns 1, if the value of rand is less than 0.5, sign() returns -1, if the value of rand is equal to 0.5, sign() returns 0; T is the number of iterations.

[0090] The formula of parabolic control factor s is as follows:

[0091]

[0092] Where rand is a random number in (0,1). If the value of rand is greater than 0.5, sign() returns 1. If the value of rand is less than 0.5, sign() returns -1. If the value of rand is equal to 0.5, sign() returns 0. T is the number of iterations, and max_T is the maximum number of iterations.

[0093] Similarly, the leader-following strategy enables wolf γ to update its position according to the positions of wolf α and wolf β. The update method is shown in formula (16):

[0094] X γ , new =(p-1)·X γ +(1-p)·X β,new +(1-p)·Xα (16)

[0095] Where p is a random number between (0,1).

[0096] For the wolf ω, the lowest level in the wolf pack social hierarchy, it must obey the gray wolves of all other classes. The α wolf occupies an absolute leadership position. The leadership positions of the β wolf and the γ wolf in the wolf pack are sometimes exchanged. The present invention imitates this behavior and proposes a dynamic weight update strategy. At the same time, the Cauchy mutation is added to the ω wolf group to improve the probability of the population jumping out of the local optimum. Their mathematical formulas are shown in formulas (17) and (18):

[0097] X ω =X1·M1+X2·M2+X3·M3+0.01·tan(π·(rand(0,1)-0.5)) (17)

[0098] The dynamic weight coefficients used to update ω wolf are as follows:

[0099]

[0100] In formula (17), rand is a random number in (0,1).

[0101] The present invention introduces an exploration factor A, and the value of A is directly affected by the control factor a. From formula (3), it can be seen that the control factor a changes linearly with the number of iterations. However, the hunting process of wolves in reality is complex, and linear changes may not be able to effectively characterize their exploration and attack hunting process. The present invention proposes a cosine convergence factor. In the early stage of iteration, its value is wider than the original control factor, which makes the algorithm explore a wider range. In the later stage, it takes a smaller value than the original convergence factor, which helps to speed up the convergence speed of the algorithm. Its mathematical formula is shown in formula (19):

[0102]

[0103] Where T is the number of iterations and max_T is the maximum number of iterations.

[0104] Now, six benchmark test functions in the CEC2005 function test set are used to verify the performance of IGWO, and GWO, WOA, PSO, and GA are selected as comparison algorithms. The five algorithms are run independently 30 times (the results are shown in Table 1), and the population size is 100 and the number of iterations is 1000. The detailed information of the basic test function is shown in Table 2. From Table 1, it can be seen that IGWO is better than the comparison algorithm in all indicators on the test function.

[0105] Table 1 Results of each algorithm on the test function

[0106]

[0107]

[0108] Table 2 Test function details

[0109]

[0110] Next, the effectiveness and rationality of the invented method are illustrated by taking the flood control dispatch of the Longyangxia Reservoir in Lanzhou City, Gansu Province, which is a once-in-a-millennium flood. Longyangxia is located in Lanzhou City, Gansu Province, and is famous for its outstanding flood storage and disaster reduction capabilities. As the leading reservoir of the Yellow River, if the flood storage capacity of Longyangxia can be fully utilized, reducing the discharge flow of Longyangxia during the flood control dispatch will greatly reduce the flood control pressure of the downstream cascade reservoirs and protection objects of Longyangxia. In the context of a once-in-a-millennium flood, considering the safety of the downstream cascade reservoirs and protection objects, the maximum allowable discharge flow of Longyangxia is 4510m 3 / s. Considering the safety of Longyangxia itself, the maximum flood storage capacity allowed is 25.56 billion m 3 .

[0111] The present invention takes the minimum maximum discharge flow of Longyangxia as the objective function and adopts the IGWO algorithm for optimization scheduling. In order to prove the excellent optimization ability of IGWO, GWO, WOA, PSO and GA are selected as comparison algorithms. The five algorithms are independently run 30 times (the results are shown in Table 1), and the number of populations is 100 and the number of iterations is 1000. The IGWO flood control scheduling process is shown in Table 2. It can be seen from Table 2 and Table 3 that under the scheduling of IGWO, the discharge interval of Longyangxia is [3325, 3447], the discharge process is stable, and the mean and standard deviation of the maximum discharge flow and the optimal solution are better than the comparison algorithm.

[0112] Table 3 Flood control dispatch results of each algorithm

[0113]

[0114] Table 4 IGWO flood control dispatch process

[0115]

[0116]

Claims

1. A flood control dispatching method based on the improved grey wolf algorithm with fusion of multiple strategies, characterized in that: The following steps are involved: (1) Obtain sample data on floods and reservoirs; (2) Based on the objective function minf(x) of minimizing the maximum downstream discharge during flood control scheduling, a flood control scheduling model is established by taking into account the water balance constraint, the outflow flow range constraint, the outflow flow fluctuation constraint and the upper limit constraint of reservoir flood storage; (3) Improved Gray Wolf Algorithm: Use improved Tent chaotic mapping to initialize the population, update the position of wolf α by integrating Lévy walk and tangent walk, and finally determine the position of wolf α according to the elite retention strategy, use parabolic spiral approach to determine the position of wolf β, determine the position of wolf γ according to the positions of wolf α and wolf β, determine the position of wolf ω by combining Cauchy mutation and dynamic weight coefficient, and use cosine convergence factor in the calculation of exploration control factor to obtain the improved Gray Wolf Algorithm with fusion of multiple strategies; (4) Adopting the improved Grey Wolf Algorithm with fusion of multiple strategies to simulate flood control operations; In step (3), the rule for updating the position of α wolf by integrating Lévy walk and tangent walk is as follows: Where, X a,new is the updated wolf position, X a To update the position of the wolf α before, Levy_step is the step length of Levy walk, tan_step is the step length of tangent walk, T is the number of iterations, and max_T is the maximum number of iterations; In step (3), the step length of the Levy walk Levy_step is In the formula, x and y are variables that obey the normal distribution, c is a constant with a value of 1.5; The step length of the tangent walk tan_step is tan_step=0.1·tan(teta)·sign(rand-0.5) / log(1+T) Where teta is the angle control parameter, rand is a random number in (0,1), if the value of rand is greater than 0.5, sign() returns 1, if the value of rand is less than 0.5, sign() returns -1, if the value of rand is equal to 0.5, sign() returns 0; The angle control parameter teta is teta=rand(0,1)·π / 2.5 In the formula, rand(0,1) is a random number between 0 and 1; In step (3), the position of the β wolf is determined by using a parabolic spiral approach, as follows: Where, X β,new is the updated position of β wolf, X β is the position of the β wolf before updating, s is the parabolic control factor, rand(0,1) is a random number between 0 and 1, rand is a random number in (0,1), if the value of rand is greater than 0.5, sign() returns 1, if the value of rand is less than 0.5, sign() returns -1, if the value of rand is equal to 0.5, sign() returns 0; In step (3), the position X of the ω wolf is determined by combining the Cauchy mutation and the dynamic weight coefficient. ω : X ω =X1·M1+X2·M2+X3·M3+0.01·tan(π·(rand(0,1)-0.5)) Where X1, X2, and X3 represent the updated positions of the three common wolves ω under the leadership of wolves α, β, and δ, and M i represents the inertia weight coefficient adopted by ordinary wolves, i = 1, 2, 3; m1, m2, m3 represent the inertia weight coefficients of wolves α, β, δ respectively, and rand is a random number in (0, 1); In step (3), the improved Tent chaotic map is In the formula, x i is a randomly generated population individual, x i+1 To use the improved Tent chaotic map to reinitialize the population individuals, rand(0,1) is a random number between 0 and 1, and μ is a control parameter with a value of 0.

015.

2. The flood control dispatching method based on the improved grey wolf algorithm with fusion of multiple strategies according to claim 1 is characterized in that: In step (3), the elite retention strategy is as follows: Where X' a To finally determine the position of wolf α, f(X a,new ) is the fitness value of wolf α after update, f(X a ) is the fitness value of wolf α before updating.

3. The flood control dispatching method based on the improved grey wolf algorithm with fusion of multiple strategies according to claim 1 is characterized in that: In step (3), the position of wolf γ is determined according to the positions of wolf α and wolf β, as follows: X γ , new =(p-1)·X γ +(1-p)·X β,new +(1-p)·X'α Where, X γ,new The updated position of wolf γ is determined according to the positions of wolf α and wolf β, X γ is the position of γ wolf before updating, and p is a random number between (0,1).

4. The flood control dispatching method based on the improved grey wolf algorithm integrating multiple strategies according to claim 1 is characterized in that: In step (3), the cosine convergence factor a is used in the calculation of the exploration control factor:

5. The flood control dispatching method based on the improved grey wolf algorithm integrating multiple strategies according to claim 1 is characterized in that: In step (2), the flood control dispatch model is as follows: minf(x)=max(QO(t)) Where V(t) and V(t-1) are the reservoir capacities at the end of period t and period t-1, respectively, in units of 100 million m 3 ; QI(t) and QO(t) are the inflow and outflow of the reservoir at the end of period t, respectively, in m 3 / s; Δt is the dispatching period interval, 1 day for large floods and 1 hour for small floods; QO max (t) is the upper limit of the outbound flow rate in period t, in m 3 / s; ΔQO max is the maximum fluctuation constraint allowed by the reservoir, in m 3 / s; V max The maximum flood storage capacity allowed by the reservoir, in billion m 3 ; Z(1) represents the starting water level, Z0 represents the flood limit water level, and the water level will be adjusted starting from the flood limit water level after entering the flood prevention stage.

Images