A multi-objective intelligent optimization scheduling method for reservoir groups based on multi-constraint coupling
Through the multi-objective intelligent optimization scheduling method of reservoir group based on multi-constraint coupling, combined with evolutionary algorithms and parallel processing technology, the complexity of the multi-objective optimization problem in reservoir scheduling is solved, and efficient scheduling scheme generation and constraint processing of cascade reservoir group are realized. Improve the scientificity and efficiency of water resource management.
Patent Information
- Application Number
- CN202411270014.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-11
AI Technical Summary
In reservoir scheduling, the multi-objective optimization problem has discrete and reduced solution space due to the existence of multi-dimensional constraints, making it difficult to achieve effective balance and optimal decision-making, and it is difficult for the existing technology to obtain accurate scheduling solutions efficiently and reasonably.
Adopt multi-objective intelligent optimization scheduling method based on multi-constraint coupling, combined with evolutionary algorithms and parallel processing technology to build a cascade reservoir group with large-scale multi-objective criterion for maximum flood control capacity, maximum peak cutting criterion for flood control sections and maximum power generation criterion. The scheduling model solves the optimal scheduling scheme of the reservoir through multi-core adaptive parallel allocation strategy, dual-population collaborative optimization strategy and dynamic joint constraint processing strategy.
It realizes efficient and rapid generation of cascade reservoir groups with good convergence and wide distribution, improves the accuracy and reliability of optimization decisions, avoids local optimal difficulties, and significantly improves model operation efficiency and constraints. Processing applicability.
Smart Images

Figure CN118798588B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of reservoir optimization and dispatching, and in particular relates to a multi-objective intelligent optimization and dispatching method for a reservoir group based on multi-constraint coupling. Background Art
[0002] As the core link of water resources management, reservoir operation faces multiple complex and conflicting objectives, and must also strictly abide by various constraints. This problem can be regarded as a typical constrained multi-objective optimization problem. In this optimization problem, it is necessary not only to find a balance between different objectives, but also to ensure that all constraints are met to achieve the best comprehensive benefits.
[0003] First, the multi-objective and complexity of reservoir operation determine the difficulty of its optimization problem. The main objectives of reservoir operation include maximizing power generation benefits, ensuring flood control safety, water supply reliability, and protecting the ecological environment. These objectives are often conflicting. For example, in order to maximize power generation benefits, reservoirs need to maintain a higher water level, which may increase the risk of downstream floods during the flood season; on the contrary, in order to ensure flood control safety, reservoirs need to lower water levels during the flood season, which may affect the benefits of power generation and water supply. In addition, ecological and environmental protection requires maintaining a healthy flow of river ecosystems, which also has potential conflicts with power generation and water supply goals. Therefore, in reservoir operation, a balance must be found between these conflicting objectives, which requires the use of multi-objective optimization methods.
[0004] In addition, in the multi-objective optimization problem of reservoir operation, the existence of constraints further increases the complexity of the problem. These constraints include but are not limited to water balance constraints, water level constraints, outflow constraints and time constraints. The water balance constraint requires that the inflow and outflow of the reservoir be balanced within a certain period of time to avoid reservoir overflow or drying up. The water level constraint ensures that the water level of the reservoir is within a safe range, neither exceeding the flood warning water level nor falling below the ecological protection water level. The outflow constraint stipulates that the outflow of the reservoir must meet the flow demand of the downstream river to ensure the realization of flood control and ecological requirements. The time constraint limits the implementation cycle and time period of the scheduling strategy to ensure that the scheduling measures are implemented within a suitable time range. Due to the addition of the above multi-dimensional constraints, the problem solution space undergoes discretization and reduction. At the same time, the complexity of the objective function and the diversity of constraints make it difficult to achieve effective balance and optimal decision-making in the overall problem solving.
[0005] Therefore, since the evolutionary algorithm based on the diversified group search strategy has the advantages of excellent optimization mechanism, high search efficiency and high-quality decision-making solutions, the present invention takes into account the requirements of reservoir scheduling engineering applications for computing efficiency, solving ability and solution accuracy, and combines the excellent search mechanism, parallel processing technology and constraint processing mechanism of the evolutionary algorithm to form a parallel constrained multi-objective optimization evolutionary algorithm to meet the needs of solving large-scale complex engineering problems in reservoir scheduling, thereby better responding to the multi-objective optimization challenges in reservoir scheduling and achieving more scientific, reasonable and efficient water resources management. Summary of the invention
[0006] The purpose of the present invention is to provide a multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling, so as to solve the problem that when solving the joint optimization scheduling problem of a cascade reservoir group, the addition of multi-dimensional constraints causes the problem solution space to be discrete, reduced, etc., and at the same time, the complexity of the objective function and the diversity of the constraints make it difficult to achieve effective balance and optimal decision-making in problem solving as a whole, thereby making it difficult to efficiently and reasonably obtain an accurate scheduling plan.
[0007] To achieve the above object, the present invention provides a multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling, comprising the following steps:
[0008] Step 1: Based on the basic information data of the reservoirs, a large-scale multi-objective scheduling model for cascade reservoirs is constructed with the maximum flood control storage capacity safety guarantee criteria, the maximum peak reduction criteria for flood control sections, and the maximum power generation criteria; at the same time, constraints are added to the large-scale multi-objective scheduling model for cascade reservoirs, the maximum number of iterations, population size, and elite population size of the optimization scheduling model are preset, and the outflow flow during the reservoir scheduling period is set as the decision variable;
[0009] Step 2, executing a multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling to solve the optimal scheduling plan for the reservoir;
[0010] Step 3: Output the optimal scheduling solution set for the reservoir group.
[0011] Preferably, the maximum flood control storage capacity safety guarantee criterion is to ensure that no flood exceeding the flood control standard occurs downstream by mobilizing less flood control storage capacity, so as to achieve the flood security dispatching goal. The objective function is expressed as:
[0012] (1)
[0013] in, is the size of the flood control storage capacity occupied; and are the number of reservoirs and the number of decision variables, respectively; and are the kth reservoir The water level storage capacity in each time period and the storage capacity corresponding to the limit water level.
[0014] Preferably, the maximum peak reduction criterion of the flood control section is to reduce the peak flow of the control section to the maximum extent under the premise of ensuring the safety of the reservoir itself and the safety of the downstream flood control target, so as to achieve the flood pressure reduction scheduling target. The objective function is expressed as:
[0015] (2)
[0016] in, The flow and volume of all flood control sections; N is the number of flood control points; For the n Flood control points are located in j Time period traffic.
[0017] Preferably, the maximum power generation principle is to maximize the total power generation of all considered reservoirs, thereby fully improving the economic benefits. The objective function is expressed as:
[0018] (3)
[0019] in, is the total power generation of all cascade reservoirs; and represent the number of reservoirs and the number of decision variables respectively; It is Total number of hours per cycle; , and Respectively represent Reservoir Output, power generation flow and water head for each period; For the The output coefficient of a reservoir.
[0020] Preferably, the constraints added in step 1 include:
[0021] Output constraints:
[0022] (4)
[0023] in, , Respectively Period The minimum and maximum output of each reservoir;
[0024] Storage capacity constraints:
[0025] (5)
[0026] in, , Respectively Period The minimum and maximum storage capacity of each reservoir;
[0027] Water balance constraints:
[0028] (6)
[0029] in, , Respectively Period The interval inflow and outflow of each reservoir; Remove lost water from outflows from upstream reservoirs; , Respectively and Period The storage capacity of a reservoir;
[0030] Outbound traffic constraints:
[0031] (7)
[0032] in, , Respectively Period The minimum and maximum outflow of each reservoir;
[0033] Power generation flow constraints:
[0034] (8)
[0035] in, , Respectively Period The minimum and maximum power generation flow of each reservoir;
[0036] Initial and final storage capacity constraints:
[0037] (9)
[0038] in, , Respectively k The initial and final storage capacity of a reservoir.
[0039] Preferably, the specific process of executing the multi-objective intelligent optimization scheduling method of a reservoir group based on multi-constraint coupling to solve the optimal scheduling plan of the reservoir in step 2 is as follows:
[0040] S201, Initialization: Setting , and represents the number of elite storage The default parameters are used; the initial population is generated according to the preset range of the decision variables; the fitness of individuals in the initial population is calculated; the multi-core adaptive parallel allocation strategy is implemented, and the individuals at the first level of the Pareto frontier are selected for the initial elite population;
[0041] S202, executing a dual-population collaborative optimization strategy: dividing the optimization population into a convergence population and a diversity population, executing a multi-parameter convergence optimization strategy and a diversity random perturbation strategy to balance the efficiency of solution set convergence and the diversity of exploration particles;
[0042] S203: Execute the search characteristic optimization strategy, dynamically select according to the search characteristics of convergent individuals and diverse individuals, and obtain the optimal search population X ;
[0043] S204. Execute dynamic joint constraint processing strategy: Combined with correction based on full population information The constraint strategy, the adaptive penalty constraint strategy based on target value conversion and the CDP strategy based on the strict constraint dominance principle are three constraint processing technologies. After the optimal population is constrained, the corrected optimal population is merged with the population and the offspring population is obtained according to the dominance relationship.
[0044] S205, perform iterative update: if the preset maximum number of iterations is reached, the optimal reservoir scheduling plan is obtained, and the fitness value of the elite population is output to obtain the reservoir group scheduling plan corresponding to the fitness value, otherwise return to step S202.
[0045] Preferably, the specific process of S201 is as follows:
[0046] S201-1. Generate an optimal population according to the preset range of decision variables and calculate individual fitness; the decision variables are initialized as follows:
[0047] (10)
[0048] in, For the i Individual The value of the decision variable; Indicates that in the interval Functions for generating uniformly distributed random numbers; is the number of individuals in the population;
[0049] S201-2, execute the multi-core adaptive parallel allocation strategy, first through The function determines the number of processors that can be enabled on the computer, and then sets the adaptive sub-population parallel allocation of different core processors based on the number of core processor startups, the main-sub core processor frequency, and the parallel model on and off; finally, the information interaction between populations is realized through a certain number of iterations. The specific expression is as follows:
[0050] (11)
[0051] (12)
[0052] (13)
[0053] (14)
[0054] in, is the number of parallel core processors that can be enabled, is the final number of parallel cores, The function is to determine the number of core processors in the computer. It is the maximum number of core processors available on the computer. If there is no manual setting, the maximum number of core processors on the computer will be used by default for parallel computing. is the parallel model startup function, It is a single-core non-parallel model. It is a multi-core parallel model; is the corresponding subpopulation number, is the total population size, is the number of reference vectors set, , are the number of targets and the number of decision variables, respectively.
[0055] Preferably, the specific process of S202 is as follows:
[0056] S202-1. Execute multi-parameter convergence optimization strategy: In order to fully consider and measure the high-quality information in the population optimization process, introduce extreme value distribution learning parameters in the convergent population , mean distribution learning parameters , variation perturbation learning parameters Three types of learning parameters are used to dynamically optimize the convergent population based on the information of the relative best / worst individuals in the offspring population and the elite population. The updated convergent population is obtained by dynamically adjusting the influence proportion of each parameter in different iteration stages. ; Use GA genetic algorithm to fully communicate within the population, and improve the selection probability of GA in the later iteration to improve the local optimization ability. The specific process is as follows:
[0057] (15)
[0058] In the formula, It is in The first The individual The value of the decision variable, and are the current number of iterations and the maximum number of iterations respectively, , is the learning factor;
[0059] in, Using the optimal solution , stay away from the worst solution The extreme value distribution learning strategy is updated, and the parameters The influence of gradually weakens as the search process progresses, so that the population can approach the dominant solution set area at a faster speed in the early stage of the search; Update based on the mean distribution learning strategy of the current best / worst population to guide the population out of the local optimal pseudo-Pareto optimal solution; The mutation disturbance strategy is adopted for updating, and the local exploration capability is increased through mutation disturbance. The specific expression is as follows:
[0060] (16)
[0061] in, Indicated in The extreme value distribution information obtained during the iteration process, , Respectively expressed in The environmental non-dominated sorting Selecting elite populations and the current optimal population The individuals with the best and worst Pareto frontier grades, is a random function that creates a normal distribution. is a random serial number;
[0062] (17)
[0063] in, Indicated in The mean distribution information obtained during the iteration process is Generate a random value in the interval [0,1], , Respectively expressed in The environmental non-dominated sorting Select the elite population and the current optimal population The best and worst Pareto frontier levels The individual decision variables, To find the best population Individual The value of the decision variable;
[0064] (18)
[0065] in, Indicated in The variation disturbance information obtained during the iteration process, and They are two randomly selected individuals in the optimal population;
[0066] S202-2, implement the diversity random perturbation strategy: Considering that the optimization process may fall into the local optimal dilemma due to the single convergence of the population, in order to avoid further affecting the reliability and effectiveness of the algorithm, a random perturbation strategy is introduced into the diverse population. Diverse individuals are selected through random generation, historical archive sets, and unconstrained environment non-dominated sorting, expanding the population exploration range to achieve an organic combination of convergence efficiency and population diversity. The specific process is as follows:
[0067] (19)
[0068] in, is the first individual; Represents non-dominated sorting in an unconstrained environment; Represents random generation individual; Indicates a random selection from the historical elite archives collection Individual.
[0069] Preferably, in S203, a search characteristic optimization strategy is executed to dynamically select the optimal search population according to the search characteristics of convergent individuals and diverse individuals. X , the specific expression is as follows:
[0070] (20)
[0071] (twenty one)
[0072] in, is the floor rounding function; Indicates from Random selection Function of integers; is the optimal population individual, It is in The first The individual The value of the decision variable, is the first Individuals, and are the current number of iterations and the maximum number of iterations respectively.
[0073] Preferably, the specific process of S204 is as follows:
[0074] S204-1. Adoption of amendments In order to make full use of the information of the objective function, the constraint strategy appropriately scales the allowed violation degree of the constraint, so that as the number of iterations increases, it gradually approaches the feasible solution. The specific expression is as follows:
[0075] (twenty two)
[0076] (twenty three)
[0077] (twenty four)
[0078] (25)
[0079] (26)
[0080] in, Indicates the number of iterations to stop scaling the allowed violation level; is the scaling parameter that stops the allowed violation from scaling up; is the ceiling function; It is in t The degree to which constraint violations are allowed in the generation; is the constraint violation value of the optimization population; and There are two constraint violation degree calculation models respectively; is the sorting objective function value of the individual violating the constraint for non-dominated sorting, and bre is the sequence number of the individual violating the constraint; is the objective function value corresponding to the population; is the constraint violation degree value corresponding to the individual violating the constraint;
[0081] S204-2, adopting the adaptive penalty constraint strategy, in order to make full use of the information of the objective function, by matching the objective function magnitude information, the population is approached to the feasible solution, so that the objective function position information can be fully utilized in the early stage of iteration. In addition, as the number of iterations increases, the constraint penalty strength is adaptively increased, so that the population approaches the feasible solution, and the parameters are adaptively adjusted to achieve a balanced search between feasible and infeasible areas. The specific expression is as follows:
[0082] (27)
[0083] (28)
[0084] in, is the individual serial number that violates the constraint; is the penalty value corresponding to the individual violating the constraint. If the constraint is not violated, the value is 0; is the sorting objective function value of the individuals violating the constraint for non-dominated sorting; is the objective function value corresponding to the individual violating the constraint, Represents the optimal population X The maximum fitness value on each target, Represents the optimal population X The constraint violation degree values of all individuals in ;
[0085] S204-3, joint constraint processing strategy, dynamic joint constraint processing strategy introduces correction through random perturbation and specified iteration stage dual restriction requirements The three constraint processing methods of constraint strategy, adaptive penalty constraint strategy and CDP strategy are used to classify and optimize the population, which not only improves the applicability of different complex constraint processing, but also further accelerates the constraint convergence speed. The specific expression is as follows:
[0086] (29)
[0087] (30)
[0088] (31)
[0089] (32)
[0090] in, , , They represent the individuals selected by the CDP strategy based on the strict constraint dominance principle, the correction based on the full information of the population, The individuals selected by the constraint strategy and the individuals selected by the adaptive penalty constraint strategy based on target value conversion, , , Represents the number of individuals selected by each constraint, and They represent the optimal population in the tth iteration process, For the population size.
[0091] Therefore, the present invention adopts the above-mentioned multi-objective intelligent optimization scheduling method for reservoir groups based on multi-constraint coupling, which has the following beneficial effects:
[0092] (1) The present invention has the advantages of simple function implementation, excellent optimization efficiency, strong global search capability, not prone to local optimality, and is applicable to a wide range of application scenarios. For the complex optimization scheduling problem of cascade reservoirs, it can efficiently and quickly generate a set of scheduling schemes with good convergence and wide distribution, thereby providing a more effective technical method for solving the multi-objective engineering optimization problem of cascade reservoirs;
[0093] (2) The present invention can significantly improve the model operation efficiency by implementing a multi-core adaptive parallel allocation strategy, and to a certain extent reduce the algorithm from falling into the local optimal dilemma, thereby improving the accuracy and reliability of optimization decision-making;
[0094] (3) The present invention can balance the efficiency of solution set convergence and the diversity of explored particles by implementing a dual-population collaborative optimization strategy, thereby preventing the algorithm from falling into the dilemma of prematurely falling into a local optimum or low optimization efficiency;
[0095] (4) The dynamic joint constraint processing strategy can make full use of the position information of the objective function of the individuals that violate the constraint, guide the population to adaptively approach the feasible solution, and effectively improve the applicability of different complex constraint processing.
[0096] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0097] Figure 1 It is an overall flow chart of a multi-objective intelligent optimization dispatching method for a reservoir group based on multi-constraint coupling of the present invention;
[0098] Figure 2 The Pareto front diagrams of the present invention and the comparative method under two flood control objectives;
[0099] Figure 3 It is the Pareto front graph of different algorithms in the scheduling engineering problem of the present invention;
[0100] Figure 41 and 1 are water level change diagrams of different reservoirs obtained by the present invention, wherein (a) is the water level change diagram of reservoir 1 - 50 solution sets are selected based on target sorting, (b) is the water level change diagram of reservoir 2 - 50 solution sets are selected based on target sorting, (c) is the water level change of reservoir 3 - 50 solution sets are selected based on target sorting, and (d) is the water level change of reservoir 4 - 50 solution sets are selected based on target sorting. DETAILED DESCRIPTION
[0101] 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 invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0102] See also Figure 1-4 , a multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling, comprising the following steps:
[0103] Step 1: Based on the basic information data of the reservoirs, a large-scale multi-objective scheduling model for cascade reservoirs is constructed with the maximum flood control storage capacity safety guarantee criteria, the maximum peak reduction criteria for flood control sections, and the maximum power generation criteria; at the same time, constraints are added to the large-scale multi-objective scheduling model for cascade reservoirs, the maximum number of iterations, population size, and elite population size of the optimization scheduling model are preset, and the outflow flow during the reservoir scheduling period is set as the decision variable;
[0104] Among them, the maximum flood control storage capacity safety guarantee criterion is to use less flood control storage capacity to ensure that no flood exceeding the flood control standard occurs downstream, so as to achieve the flood security dispatching goal. The objective function is expressed as:
[0105] (1)
[0106] in, is the size of the flood control storage capacity occupied; and are the number of reservoirs and the number of decision variables, respectively; and are the kth reservoir The water level storage capacity in each time period and the storage capacity corresponding to the limit water level.
[0107] The maximum peak reduction criterion of the flood control section is to reduce the peak flow of the control section to the maximum extent and achieve the flood pressure reduction scheduling target under the premise of ensuring the safety of the reservoir itself and the safety of the downstream flood control target. The objective function is expressed as:
[0108] (2)
[0109] in, The flow and volume of all flood control sections; N is the number of flood control points; For the n Flood control points are located in j Time period traffic.
[0110] The maximum power generation principle is to maximize the total power generation of all considered reservoirs, thereby fully improving the economic benefits. The objective function is expressed as:
[0111] (3)
[0112] in, is the total power generation of all cascade reservoirs; and represent the number of reservoirs and the number of decision variables respectively; It is Total number of hours per cycle; , and Respectively represent Reservoir Output, power generation flow and water head for each period; For the The output coefficient of a reservoir.
[0113] The constraints added in step 1 are:
[0114] Output constraints:
[0115] (4)
[0116] in, , Respectively Period The minimum and maximum output of each reservoir;
[0117] Storage capacity constraints:
[0118] (5)
[0119] in, , Respectively Period The minimum and maximum storage capacity of each reservoir;
[0120] Water balance constraints:
[0121] (6)
[0122] in, , Respectively Period The interval inflow and outflow of each reservoir; Remove lost water from outflows from upstream reservoirs; , Respectively and Period The storage capacity of a reservoir;
[0123] Outbound traffic constraints:
[0124] (7)
[0125] in, , Respectively Period The minimum and maximum outflow of each reservoir;
[0126] Power generation flow constraints:
[0127] (8)
[0128] in, , Respectively Period The minimum and maximum power generation flow of each reservoir;
[0129] Initial and final storage capacity constraints:
[0130] (9)
[0131] in, , Respectively k The initial and final storage capacity of a reservoir.
[0132] Step 2: Execute the multi-objective intelligent optimization scheduling method of reservoir groups based on multi-constraint coupling to solve the optimal scheduling plan of reservoirs; the specific process is as follows:
[0133] S201, Initialization: Setting , and represents the number of elite storage The default parameters are used; the initial population is generated according to the preset range of the decision variables; the fitness of individuals in the initial population is calculated; the multi-core adaptive parallel allocation strategy is implemented, and the individuals at the first level of the Pareto frontier are selected for the initial population execution environment as the elite population for initialization; the specific process is as follows:
[0134] S201-1. Generate an optimal population according to the preset range of decision variables and calculate individual fitness; the decision variables are initialized as follows:
[0135] (10)
[0136] in, For the i Individual The value of the decision variable; Indicates that in the interval Functions for generating uniformly distributed random numbers; is the number of individuals in the population;
[0137] S201-2, execute the multi-core adaptive parallel allocation strategy, first through The function determines the number of processors that can be enabled on the computer, and then sets the adaptive sub-population parallel allocation of different core processors based on the number of core processor startups, the main-sub core processor frequency, and the parallel model on and off; finally, the information interaction between populations is realized through a certain number of iterations. The specific expression is as follows:
[0138] (11)
[0139] (12)
[0140] (13)
[0141] (14)
[0142] in, is the number of parallel core processors that can be enabled, is the final number of parallel cores, The function is to determine the number of core processors in the computer. It is the maximum number of core processors available on the computer. If there is no manual setting, the maximum number of core processors on the computer will be used by default for parallel computing. is the parallel model startup function, It is a single-core non-parallel model. It is a multi-core parallel model; is the corresponding subpopulation number, is the total population size, is the number of reference vectors set, , are the number of targets and the number of decision variables, respectively.
[0143] S202, execute the dual population collaborative optimization strategy: divide the optimization population into a convergence population and a diversity population, execute the multi-parameter convergence optimization strategy and the diversity random perturbation strategy to balance the efficiency of solution set convergence and the diversity of exploration particles; the specific process is as follows:
[0144] S202-1. Execute multi-parameter convergence optimization strategy: In order to fully consider and measure the high-quality information in the population optimization process, introduce extreme value distribution learning parameters in the convergent population , mean distribution learning parameters , variation perturbation learning parameters Three types of learning parameters are used to dynamically optimize the convergent population based on the information of the relative best / worst individuals in the offspring population and the elite population. The updated convergent population is obtained by dynamically adjusting the influence proportion of each parameter in different iteration stages. ; Use GA genetic algorithm to fully communicate within the population, and improve the selection probability of GA in the later iteration to improve the local optimization ability. The specific process is as follows:
[0145] (15)
[0146] In the formula, It is in The first The individual The value of the decision variable, and are the current number of iterations and the maximum number of iterations respectively, , is the learning factor;
[0147] in, Using the optimal solution , stay away from the worst solution The extreme value distribution learning strategy is updated, and the parameters The influence of gradually weakens as the search process progresses, so that the population can approach the dominant solution set area at a faster speed in the early stage of the search; Update based on the mean distribution learning strategy of the current best / worst population to guide the population out of the local optimal pseudo-Pareto optimal solution; The mutation disturbance strategy is adopted for updating, and the local exploration capability is increased through mutation disturbance. The specific expression is as follows:
[0148] (16)
[0149] in, Indicated in The extreme value distribution information obtained during the iteration process, , Respectively expressed in The environmental non-dominated sorting Selecting elite populations and the current optimal population The individuals with the best and worst Pareto frontier grades, is a random function that creates a normal distribution. is a random serial number;
[0150] (17)
[0151] in, Indicated in The mean distribution information obtained during the iteration process is Generate a random value in the interval [0,1], , Respectively expressed in The environmental non-dominated sorting Select the elite population and the current optimal population The best and worst Pareto frontier levels The individual decision variables, To find the best population Individual The value of the decision variable;
[0152] (18)
[0153] in, Indicated in The variation disturbance information obtained during the iteration process, and They are two randomly selected individuals in the optimal population;
[0154] S202-2, implement the diversity random perturbation strategy: Considering that the optimization process may fall into the local optimal dilemma due to the single convergence of the population, in order to avoid further affecting the reliability and effectiveness of the algorithm, a random perturbation strategy is introduced into the diverse population. Diverse individuals are selected through random generation, historical archive sets, and unconstrained environment non-dominated sorting, expanding the population exploration range to achieve an organic combination of convergence efficiency and population diversity. The specific process is as follows:
[0155] (19)
[0156] in, is the first individual; Represents non-dominated sorting in an unconstrained environment; Represents random generation individual; Indicates a random selection from the historical elite archives collection Individual.
[0157] S203: Execute the search characteristic optimization strategy, dynamically select according to the search characteristics of convergent individuals and diverse individuals, and obtain the optimal search population X ; The specific expression is as follows:
[0158] (20)
[0159] (twenty one)
[0160] in, is the floor rounding function; Indicates from Random selection Function of integers; is the optimal population individual, It is in The first The individual The value of the decision variable, is the first Individuals, and are the current number of iterations and the maximum number of iterations respectively.
[0161] S204. Execute dynamic joint constraint processing strategy: Combined with correction based on full population information The constraint strategy, the adaptive penalty constraint strategy based on target value conversion, and the CDP strategy based on the strict constraint dominance principle are three constraint processing technologies. After the optimal population is constrained, the corrected optimal population is merged with the population and the offspring population is obtained according to the dominance relationship. The specific process is as follows:
[0162] S204-1. Adoption of amendments In order to make full use of the information of the objective function, the constraint strategy appropriately scales the allowed violation degree of the constraint, so that as the number of iterations increases, it gradually approaches the feasible solution. The specific expression is as follows:
[0163] (twenty two)
[0164] (twenty three)
[0165] (twenty four)
[0166] (25)
[0167] (26)
[0168] in, Indicates the number of iterations to stop scaling the allowed violation level; is the scaling parameter that stops the allowed violation from scaling up; is the ceiling function; It is in t The degree to which constraint violations are allowed in a generation; is the constraint violation value of the optimization population; and There are two constraint violation degree calculation models respectively; is the sorting objective function value of the individual violating the constraint for non-dominated sorting, and bre is the sequence number of the individual violating the constraint; is the objective function value corresponding to the population; is the constraint violation degree value corresponding to the individual violating the constraint;
[0169] S204-2, adopting the adaptive penalty constraint strategy, in order to make full use of the information of the objective function, by matching the objective function magnitude information, the population is approached to the feasible solution, so that the objective function position information can be fully utilized in the early stage of iteration. In addition, as the number of iterations increases, the constraint penalty strength is adaptively increased, so that the population approaches the feasible solution, and the parameters are adaptively adjusted to achieve a balanced search between feasible and infeasible areas. The specific expression is as follows:
[0170] (27)
[0171] (28)
[0172] in, is the individual serial number that violates the constraint; is the penalty value corresponding to the individual violating the constraint. If the constraint is not violated, the value is 0; is the sorting objective function value of the individuals violating the constraint for non-dominated sorting; is the objective function value corresponding to the individual violating the constraint, Represents the optimal population X The maximum fitness value on each target, Represents the optimal population X The constraint violation degree values of all individuals in ;
[0173] S204-3, joint constraint processing strategy, dynamic joint constraint processing strategy introduces correction through random perturbation and specified iteration stage dual restriction requirements The three constraint processing methods of constraint strategy, adaptive penalty constraint strategy and CDP (constraint dominance principle) strategy are used to classify and optimize the population, among which the correction The constraint strategy and adaptive penalty constraint strategy are introduced in S204-1 and S204-2. The CDP strategy is a conventional constraint strategy and is not described here. The specific expression is as follows:
[0174] (29)
[0175] (30)
[0176] (31)
[0177] (32)
[0178] in, , , They represent the individuals selected by the CDP strategy based on the strict constraint dominance principle, the correction based on the full information of the population, The individuals selected by the constraint strategy and the individuals selected by the adaptive penalty constraint strategy based on target value conversion, , , Represents the number of individuals selected by each constraint, and They represent the optimal population in the tth iteration process, For the population size.
[0179] S205, perform iterative update: if the preset maximum number of iterations is reached, the optimal reservoir scheduling plan is obtained, and the fitness value of the elite population is output to obtain the reservoir group scheduling plan corresponding to the fitness value, otherwise return to step S202.
[0180] Step 3: Output the optimal scheduling solution set for the reservoir group.
[0181] Example
[0182] The three-objective cascade reservoir group optimization scheduling model with the maximum flood control storage capacity safety guarantee criterion, the maximum peak shaving criterion of the flood control section, and the maximum power generation criterion was constructed. The present invention and CCMO were used to schedule the four cascade reservoirs in a certain basin for a typical flood scenario with a return period of once in 20 years. There were 80 scheduling periods, each with an interval of 6 hours. The constraint variables included water level inequality, final water level equation, total flood control storage capacity and other constraints. Therefore, the model had a total of 320 variables and 1049 constraint variables.
[0183] In order to compare and analyze the solution stability, solution efficiency and solution optimization breadth of the present invention and CCMO in the actual multi-objective scheduling problem of reservoirs, the multi-objective scheduling problem was calculated 10 times independently, and the Pareto fronts obtained by the three algorithms were used to sort the non-dominated solutions to obtain the ideal Pareto front.
[0184] Since the calculation of the inverse generation distance (IGD) indicator in actual scheduling requires knowing the true Pareto front (PF), and the true PF of the actual problem is almost impossible to obtain. Therefore, in this case, the IGD indicator cannot be calculated to evaluate the performance of the method. The accurate HV indicator requires the extreme points of the true PF on each target dimension to normalize the solution set. However, strictly speaking, the HV indicator is also immeasurable in actual cases. However, the exact value of the HV indicator is not what we care about. We are concerned about the relative ranking of the HV indicator obtained by each algorithm. Because no matter which algorithm is used as the standard, their relative order will not change. Therefore, the ideal extreme point obtained on each target dimension is set as the standard to normalize the solution set obtained by the method. The HV indicator calculated by this method can evaluate the relative performance of different actual situations. Therefore, the hypervolume (HV) indicator is calculated after the Pareto front obtained by each method and the ideal Pareto front are normalized. HV can measure the convergence and distribution of the solution set, especially for problems with unknown Pareto fronts. When the HV is larger, it can be said that the distribution of the scheduling solution set obtained by this method is better, so as to compare their differences in actual applications. Finally, the feasible solution ratio (FR) indicator is used to evaluate the optimization performance of the algorithm and the model running time is used to evaluate the solution efficiency.
[0185] Result analysis:
[0186] First, by calculating and analyzing the average HV index of the three methods and the best HV in 10 independent calculations, as shown in Table 1, it can be seen that the average HV index and the best HV of the present invention in the actual scheduling scenario are better than the comparison algorithm. Therefore, the stability and optimization breadth of the present invention have significant advantages over CCMO in multi-objective optimization scheduling problems. In addition, the average running time and the shortest running time of the present invention are both optimal, indicating that the present invention has efficient search efficiency and reliable stability. Therefore, the present invention has stronger exploration ability, search efficiency and solution efficiency for solving the multi-objective optimization problem of cascade reservoirs.
[0187] Table 1 Evaluation index values obtained by different algorithms
[0188] .
[0189] Finally, 3D graphics visualization is used to evaluate the performance of different algorithms from three aspects: Pareto front dominance relationship, solution set convergence, and solution set distribution. Convergence and distribution are measured based on the range of target values and population distribution on the coordinate axis, and Pareto front dominance relationship is measured based on the positional relationship between solution sets. A wider and more uniform solution set and a better Pareto front dominance relationship indicate better convergence performance. Therefore, the best HV value corresponding to the 10 independent calculations of each method is taken to draw the Pareto front diagram, see details. Figure 2 and Figure 3 It can be seen that: in this actual scheduling case study, compared with the other two algorithms, the dominance relationship, solution set convergence and solution set distribution of the present invention on different objectives proves that it has better global convergence ability. At the same time, the point distribution in the three-dimensional space is more uniform, indicating that its solution is more diverse and more representative. Therefore, the present invention performs well in convergence and distribution.
[0190] In addition, in the scheduling problem, before the flood season, water should be released as much as possible to empty the reservoir capacity in order to ensure flood control safety. After the flood peak, the water level should be returned to the starting level as much as possible under the constraint of equality to ensure the full utilization of the benefits of the reservoir after the flood season. It can be seen that the optimization of the multi-objective constraint joint scheduling of cascade reservoirs is difficult. Therefore, in order to further verify the rationality and effectiveness of the present invention in the scheduling problem, the water level change process lines of the four reservoirs under 50 scheduling scenarios are selected according to the first target value and equal distance (such as Figure 4In order to coordinate the conflict of flood control benefits between goal 1 and goal 2 to the greatest extent, the three upstream reservoirs were pre-discharged before the flood season, and there was a flood peak lag between the cascade reservoirs, which effectively avoided the problem of a surge in flood peak flow in the control section. Therefore, the present invention can play a good role in joint dispatching when encountering excessive floods, can pre-discharge in time, use the reservoir capacity to effectively reduce the flood peak, and provide a variety of effective alternatives for dispatching decisions to formulate the best decision plan.
[0191] Therefore, the present invention adopts the above-mentioned multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling. Taking into account the requirements of reservoir scheduling engineering applications for computing efficiency, solving ability and solution accuracy, the excellent search mechanism, parallel processing technology and constraint processing mechanism of the evolutionary algorithm are combined to form a parallel constrained multi-objective optimization evolutionary algorithm to meet the needs of solving large-scale and complex engineering problems in reservoir scheduling, thereby better responding to the multi-objective optimization challenges in reservoir scheduling and achieving more scientific, reasonable and efficient water resources management.
[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling, characterized in that: The following steps are involved: Step 1: Based on the basic information data of the reservoirs, a large-scale multi-objective scheduling model for cascade reservoirs is constructed with the maximum flood control storage capacity safety guarantee criteria, the maximum peak reduction criteria for flood control sections, and the maximum power generation criteria; at the same time, constraints are added to the large-scale multi-objective scheduling model for cascade reservoirs, the maximum number of iterations, population size, and elite population size of the optimization scheduling model are preset, and the outflow flow during the reservoir scheduling period is set as the decision variable; Step 2, executing a multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling to solve the optimal scheduling plan for the reservoir; Step 3: Output the optimal dispatching plan set of the reservoir group; The specific process of executing the multi-objective intelligent optimization scheduling method of reservoir groups based on multi-constraint coupling to solve the optimal scheduling plan of reservoirs in step 2 is as follows: S201, Initialization: Setting , and represents the number of elite storage is the default parameter; Generate an initial population according to the preset range of decision variables; calculate the fitness of individuals in the initial population; execute a multi-core adaptive parallel allocation strategy, and select the individuals at the first level of the Pareto frontier in the execution environment of the initial population as the initialized elite population; S202, executing a dual-population collaborative optimization strategy: dividing the optimization population into a convergence population and a diversity population, and executing a multi-parameter convergence optimization strategy and a diversity random perturbation strategy; S203: Execute the search characteristic optimization strategy, dynamically select according to the search characteristics of convergent individuals and diverse individuals, and obtain the optimal search population X ; S204. Execute dynamic joint constraint processing strategy: Combined with correction based on full population information The constraint strategy, the adaptive penalty constraint strategy based on target value conversion and the CDP strategy based on the strict constraint dominance principle are three constraint processing technologies. After the optimal population is constrained, the corrected optimal population is merged with the population and the offspring population is obtained according to the dominance relationship. S205, perform iterative update: if the preset maximum number of iterations is reached, the optimal reservoir dispatching plan is obtained, and the fitness value of the elite population is output to obtain the reservoir group dispatching plan corresponding to the fitness value, otherwise return to step S202; The specific process of S201 is as follows: S201-1. Generate an optimal population according to the preset range of decision variables and calculate individual fitness; the decision variables are initialized as follows: (10) in, For the i Individual The value of the decision variable; Indicates that in the interval Functions for generating uniformly distributed random numbers; is the number of individuals in the population; S201-2, execute the multi-core adaptive parallel allocation strategy, first through The function determines the number of processors that can be enabled on the computer, and then sets up adaptive sub-population parallel allocation for different core processors based on the number of core processor startups, the main-sub core processor frequency, and the parallel model on and off; finally, information interaction between populations is achieved through iteration. The specific expression is as follows: (11) (12) (13) (14) in, is the number of parallel core processors that can be enabled, is the final number of parallel cores, The function is to determine the number of core processors in the computer. It is the maximum number of core processors available on the computer. If there is no manual setting, the maximum number of core processors on the computer will be used by default for parallel computing. is the parallel model startup function, It is a single-core non-parallel model. It is a multi-core parallel model; is the corresponding subpopulation number, is the total population size, is the number of reference vectors set, , are the number of targets and the number of decision variables, respectively; The specific process of S202 is as follows: S202-1. Execute multi-parameter convergence optimization strategy: introduce extreme value distribution learning parameters into the convergence population , mean distribution learning parameters , variation perturbation learning parameters Three types of learning parameters are used to dynamically optimize the convergent population based on the relatively best / worst individual information in the offspring population and the elite population. The updated convergent population is obtained by dynamically adjusting the influence proportion of each parameter in different iteration stages. ; Use GA genetic algorithm to fully communicate within the population, and improve the selection probability of GA in the later iteration to improve the local optimization ability. The specific process is as follows: (15) In the formula, It is in The first The individual The value of the decision variable, and are the current number of iterations and the maximum number of iterations respectively, , is the learning factor; in, Using the optimal solution , stay away from the worst solution The extreme value distribution learning strategy is updated, and the parameters The influence of gradually weakens as the search process progresses, so that the population can approach the dominant solution set area at a faster speed in the early stage of the search; Update based on the mean distribution learning strategy of the current best / worst population to guide the population out of the local optimal pseudo-Pareto optimal solution; The mutation disturbance strategy is adopted for updating, and the local exploration capability is increased through mutation disturbance. The specific expression is as follows: (16) in, Indicated in The extreme value distribution information obtained during the iteration process, , Respectively expressed in The environmental non-dominated sorting Selecting elite populations and the current optimal population The individuals with the best and worst Pareto frontier grades, is a random function that creates a normal distribution. is a random serial number; (17) in, Indicated in The mean distribution information obtained during the iteration process is Generate a random value in the interval [0,1], , Respectively expressed in The environmental non-dominated sorting Select the elite population and the current optimal population The best and worst Pareto frontier levels The individual decision variables, To find the best population Individual The value of the decision variable; (18) in, Indicated in The variation disturbance information obtained during the iteration process, and They are two randomly selected individuals in the optimal population; S202-2. Execute the diversity random perturbation strategy: introduce the random perturbation strategy into the diversity population, select the diversity individuals through random generation, historical archive set, and unconstrained environment non-dominated sorting, and expand the population exploration range. The specific process is as follows: (19) in, is the first individual; Represents non-dominated sorting in an unconstrained environment; Represents random generation individual; Indicates a random selection from the historical elite archives collection individual; In S203, a search characteristic optimization strategy is executed to dynamically select the optimal search population based on the search characteristics of convergent individuals and diverse individuals. X , the specific expression is as follows: (20) (21) in, is the floor rounding function; Indicates from Random selection Function of integers; is the optimal population individual, It is in The first The individual The value of the decision variable, is the first Individuals, and are the current number of iterations and the maximum number of iterations respectively; The specific process of S204 is as follows: S204-1. Adoption of amendments Constraint strategy, the specific expression is as follows: (22) (23) (24) (25) (26) in, Indicates the number of iterations to stop scaling the allowed violation level; is the scaling parameter that stops the allowed violation from scaling up; is the ceiling function; It is in t The degree to which constraint violations are allowed in the generation; is the constraint violation value of the optimization population; and There are two constraint violation degree calculation models respectively; is the sorting objective function value of the individual violating the constraint for non-dominated sorting, and bre is the sequence number of the individual violating the constraint; is the objective function value corresponding to the population; is the constraint violation degree value corresponding to the individual violating the constraint; S204-2, using an adaptive penalty constraint strategy, the specific expression is as follows: (27) (28) in, is the individual serial number that violates the constraint; is the penalty value corresponding to the individual violating the constraint. If the constraint is not violated, the value is 0; is the sorting objective function value of the individuals violating the constraint for non-dominated sorting; is the objective function value corresponding to the individual violating the constraint, Represents the optimal population X The maximum fitness value on each target, Represents the optimal population X The constraint violation degree values of all individuals in ; S204-3, joint constraint processing strategy, dynamic joint constraint processing strategy introduces correction through random perturbation and specified iteration stage dual restriction requirements The three constraint processing methods of constraint strategy, adaptive penalty constraint strategy and CDP strategy are used to classify and optimize the population. The specific expressions are as follows: (29) (30) (31) (32) in, , , They represent the individuals selected by the CDP strategy based on the strict constraint dominance principle, the correction based on the full information of the population, The individuals selected by the constraint strategy and the individuals selected by the adaptive penalty constraint strategy based on target value conversion, , , Represents the number of individuals selected by each constraint, and They represent the optimal population in the t-th and t+1-th iterations, respectively. For the population size.
2. According to claim 1, a multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling is characterized by: The maximum flood control reservoir capacity safety guarantee criterion is to use the flood control reservoir capacity to ensure that no flood exceeding the flood control standard occurs downstream, so as to achieve the flood security dispatching goal. The objective function is expressed as: (1) in, is the size of the flood control storage capacity occupied; and are the number of reservoirs and the number of decision variables, respectively; and are the kth reservoir The water level storage capacity in each time period and the storage capacity corresponding to the limit water level.
3. The multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling according to claim 2 is characterized by: The maximum peak reduction criterion of the flood control section is to reduce the peak flow of the control section to the maximum extent and achieve the flood pressure reduction scheduling target under the premise of ensuring the safety of the reservoir itself and the safety of the downstream flood control target. The objective function is expressed as: (2) in, The flow and volume of all flood control sections; N is the number of flood control points; For the n Flood control points are located in j Time period traffic.
4. The multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling according to claim 3 is characterized by: The maximum power generation principle is to maximize the total power generation of all considered reservoirs. The objective function is expressed as: (3) in, is the total power generation of all cascade reservoirs; and represent the number of reservoirs and the number of decision variables respectively; It is Total number of hours per cycle; , and Respectively represent Reservoir Output, power generation flow and water head for each period; For the The output coefficient of a reservoir.
5. The multi-objective intelligent optimization scheduling method for a reservoir group based on multi-constraint coupling according to claim 4 is characterized in that: The constraints added in step 1 are: Output constraints: (4) in, , Respectively Period The minimum and maximum output of each reservoir; Storage capacity constraints: (5) in, , Respectively Period The minimum and maximum storage capacity of each reservoir; Water balance constraints: (6) in, , Respectively Period The interval inflow and outflow of each reservoir; Remove lost water from outflows from upstream reservoirs; , Respectively and Period The storage capacity of a reservoir; Outbound traffic constraints: (7) in, , Respectively Period The minimum and maximum outflow of each reservoir; Power generation flow constraints: (8) in, , Respectively Period The minimum and maximum power generation flow of each reservoir; Initial and final storage capacity constraints: (9) in, , Respectively k The initial and final storage capacity of a reservoir.
Citation Information
Patent Citations
Dynamic multi-objective optimization control system and method for flood season water level of giant reservoir
CN118114945A
