CN-117273055-B - Multi-objective optimization device based on dynamic elite learning and dynamic environment selection

CN117273055BCN 117273055 BCN117273055 BCN 117273055BCN-117273055-B

Abstract

The invention provides a multi-objective optimization device based on dynamic elite learning and dynamic environment selection, which comprises an input module, an initializing module, a population storage module, a population iteration module and an output module, wherein the input module is used for inputting m objective functions and N decision variable limit range functions constructed according to a designated multi-objective optimization task by a user, the initializing module is used for constructing and initializing a population P comprising N individuals, the iteration number G is set to be 1, the population storage module is used for storing the population P, the count storage module is used for storing the iteration number G and a preset maximum iteration number, the population iteration module is used for carrying out iteration update on the population P according to the iteration number G, the maximum iteration number, the decision variable limit range functions and the objective functions to obtain the population P F , and the output module is used for displaying the population P F to the user as a preferred scheme of the multi-objective optimization task. In summary, the method can more quickly obtain a more accurate and specific preferred scheme of the multi-target task.

Inventors

DAI YIRU
CHEN ZIHAO
ZHOU SHENGQING

Assignees

同济大学

Dates

Publication Date: 20260505
Application Date: 20231007

Claims (9)

1. A multi-objective optimization device based on dynamic elite learning and dynamic environment selection for providing an optimal planning scheme for an urban rainwater drainage system, comprising: an input module for user input of a task constructed according to the specified multi-objective optimization task Target function of A decision variable boundary range function, The objective functions are constructed by collecting rainfall-related data, drainage network data, on-site retention storage facility data, process plant data and receiving direct current data of the body of water, Each of the objective functions includes a rank Water network costs, storage facility costs, processing facility costs, expected flood loss costs and expected economic losses due to flood, The decision variable limit range functions include local retention storage capacity, maximum processing rate and maximum allowable overflow rate; an initialization module for constructing and initializing a device comprising Population of individuals And set the iteration number 1 Is shown in the specification; a population storage module for storing the population ; A count storage module for storing the iteration times And a preset maximum iteration number ; A population iteration module for carrying out the iteration times Said maximum number of iterations The decision variable limit range function and the objective function are applied to the population Iterative updating is carried out to obtain population ; An output module for displaying the population to the user As a preferred solution to the multi-objective optimization task, Wherein the population iteration module comprises an iteration judging sub-module, a population updating sub-module, a dynamic elite learning sub-module and a dynamic environment selecting sub-module, The iteration judging submodule is used for judging the iteration times in the counting storage module Whether or not to be greater than the maximum iteration number If yes, the population in the population storage module is selected As said population And input the output module, if not, the population in the population storage module is stored The population update sub-module is entered into, The population updating submodule is used for inputting the population Pairing, crossing, mutation and merging are sequentially carried out to obtain a population , The dynamic elite learning submodule is used for carrying out population selection according to the objective function and the decision variable limit range function Dynamic elite learning is carried out to obtain population , The dynamic environment selection submodule is used for selecting the group according to the objective function and the population For the population in the population storage module Updating and counting the number of iterations in the memory module 1 Is added in the process, so that the temperature of the product is increased, The dynamic elite learning submodule comprises a first convergence index calculation unit, a first screening unit, a random number generation unit, an acceptance probability generation unit, a second screening unit, an elite learning unit, an evaluation unit and a population The generation unit is configured to generate a first output signal, The first convergence index calculation unit is used for calculating the population according to the objective function Convergence index corresponding to each individual in the system , The first screening unit is used for all the convergence indexes Ordering from small to large, and sorting the first 10% of the convergence index Corresponding individuals as elite collections The latter 25% of the convergence index Corresponding individuals as a set of extreme differences , The random number generation unit is used for integrating the range points Generates a corresponding random number for each individual in the database, The acceptance probability generation unit stores a preset acceptance probability formula for the current iteration times Said maximum number of iterations And the acceptance probability formula calculates to obtain acceptance probability , The second screening unit is used for making all the random numbers smaller than the acceptance probability As the set of elite studies, The elite learning unit stores a preset elite learning formula for each individual in the elite learning set From the elite collection Randomly selecting an individual According to the individual The individual Calculating with the elite learning formula to obtain the individual Corresponding individual , The evaluation unit is used for evaluating each individual according to the decision variable limit range function The decision variables in the model are adjusted to obtain corresponding individuals , The population is The generation unit is used for grouping the population Each of the individuals in (a) Is replaced by the corresponding individual Obtaining the population , The expression of each objective function is: , , In the middle of As a total number of the objective functions, Is the first The number of functions of the object is the number of functions, In order to make a decision vector, For the total number of decision variables, Is the first And decision variables.
2. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 1, wherein: wherein the population updating submodule comprises a pairing unit, a crossing unit, a mutation unit and a merging unit, The pairing unit is used for preparing the group Randomly selecting said individual constructs A set of parents, each set of said parents comprising two of said individuals, The crossing unit is used for pairing Respectively performing cross operation on the parent groups to obtain The sub-generation of individuals is taken as the main generation, The mutation unit is used for pairing The sub-individuals are subjected to mutation operation respectively to obtain The individual child individuals are updated with the updated information, The merging unit is used for according to the Individual updated offspring individuals and said population In (a) and (b) Individual construction of the population 。
3. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 1, wherein: Wherein the convergence index The calculation formula of (2) is as follows: , In the middle of For individuals Corresponding convergence index , For individuals According to the first The target values calculated by the individual objective functions, The expression of the acceptance probability formula is: , the expression of the elite learning formula is: , In the middle of For individuals Is the first of (2) The number of decision variables is a function of the number of decision variables, For individuals Is the first of (2) The number of decision variables is a function of the number of decision variables, In order for the rate of learning to be high, Is a random number between 0 and 1, For individuals Is the first of (2) The number of decision variables is a function of the number of decision variables, Is the total number of individual decision variables.
4. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 1, wherein: Wherein the decision variable limit range function comprises an upper threshold and a lower threshold of the corresponding decision variable, The specific process of the evaluation unit for adjusting the decision variable is as follows: Judging whether the value of the decision variable is larger than the upper threshold value of the corresponding decision variable limit range function, if so, taking the upper threshold value as the value of the decision variable, And judging whether the value of the decision variable is smaller than the lower threshold value of the corresponding decision variable limit range function, and if so, taking the lower threshold value as the value of the decision variable.
5. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 1, wherein: Wherein the dynamic environment selection submodule comprises an initialization unit, a second convergence index calculation unit, a non-dominant layer generation unit and a population Generating unit, first judging unit and population Generating unit, population Storage unit and population A storage unit, a dual layer selection policy unit and a first updating unit, The initialization unit is used for setting an identifier 1, And constructing an empty population Population of Sum population , The second convergence index calculating unit is used for calculating the population according to the objective function Convergence index corresponding to each individual in the system , The non-dominant layer generating unit is used for the population Non-dominated ordering is achieved as a plurality of non-dominated layers, The population is A generation unit for adding individuals in each non-dominant layer into the population in turn Until the total number of individuals in the next non-dominant layer and the population The sum of the individual total numbers in (a) is greater than The non-dominant layer is taken as a key layer And combining the populations The first judgment unit is input to the first judgment unit, The first judging unit is used for judging the population Whether the total number of individuals in a pool is equal to If so, the population is then As said population Inputting the first updating unit, if not, inputting the key layer All individuals in (a) are added to the population And inputting the population The generation unit is configured to generate a first output signal, The population is The generation unit stores a preset elite maintenance strategy for maintaining the performance of the system according to the elite maintenance strategy and the convergence index From the population Extraction of Individual elite individuals are added to the population And stored to the population A storage unit for extracting the population As said population Stored to the population The memory unit is provided with a memory unit, The population is The storage unit is used for storing the population , The population is The storage unit is used for storing the population , The double-layer selection strategy unit is used for selecting the population according to the population Population of memory cells Said sign And the population For the population Updating, and updating the population The first updating block is entered and, The first updating unit is used for updating the population in the population storage module All individuals replaced by the entered population And store the number of iterations in the count storage module 1 Is added.
6. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 5, wherein: Wherein the convergence index The calculation formula of (2) is as follows: , In the middle of For individuals Corresponding convergence index , For individuals According to the first Target values calculated by the objective functions.
7. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 5, wherein: Wherein, the Individual elite individual is Individual boundary individuals 、 Individual with optimal convergence And Individual corner , The said Individual boundary individuals For the population Respectively with The unit vectors on the coordinate axes have the smallest vector angle The calculation formula of the vector angle is as follows: , , In the middle of For individuals And (3) with The smallest vector angle among all vector angles corresponding to the unit vectors on the coordinate axes, Is that A set of unit vectors on each axis, For individuals And individuals Is used for the vector angle of (a), For individuals The target vector after the normalization is performed, For individuals Target vector and individual of (a) Is used to determine the inner product of the target vectors, For individuals First, the The target value after the normalization is used for the comparison, The said Individual with optimal convergence For the population Corresponding minimum in (3) Individual convergence index Is a group of the individuals of the group (a), The said Individual corner For the population Respectively with each of said objective functions having a minimum target value Individual, the first Individual corresponding to the minimum target value of the individual objective function The expression of (2) is: , , In the middle of For individuals And the first Target values corresponding to the target functions.
8. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 5, wherein: Wherein the double-layer selection strategy unit comprises a calculation control subunit, a vector angle calculation subunit and a population Generating subunits, individuals A generating subunit, a second updating subunit and a second judging subunit, The computation control subunit is used for judging the identifier If the vector angle is 1, controlling the vector angle calculation subunit to calculate if the vector angle is 1, and marking the mark Setting to 0, if not, controlling the population The generation subunit performs the calculation(s), The vector angle calculation subunit stores a preset vector angle calculation formula for calculating the population The vector angle between any two individuals is used for obtaining the minimum vector angle corresponding to each individual as the minimum vector angle of the individual, The population is The generation subunit stores a preset number of A calculation formula for the population Calculating the number Selecting from all minimum vector angles Individuals corresponding to the maximum value are used as the population , The individual Generating subunits for generating a population from said population Is selected to have the smallest convergence index As individuals of (a) , The second updating subunit is used for updating the individual Adding said population And separating it from the population The deletion of the group(s), The second judging subunit is used for judging the population Each individual of (a) and the individual Whether the vector angle of the (B) is smaller than the minimum vector angle corresponding to the individual, if so, the individual and the individual are combined As the minimum vector angle of the individual, and separately storing the population And the population To the population Storage unit and population A storage unit for storing the population respectively if not And the population To the population Storage unit and population And a memory cell.
9. The dynamic elite learning and dynamic environment selection based multi-objective optimization device according to claim 8, wherein: Wherein the number is The expression of the calculation formula is: , , In the middle of In order to round up the function, Is a population Is a total number of individuals in the group.

Description

Multi-objective optimization device based on dynamic elite learning and dynamic environment selection Technical Field The invention relates to the technical field of high-dimensional multi-objective optimization, in particular to a multi-objective optimization device based on dynamic elite learning and dynamic environment selection. Background The multi-objective optimization problem (MOPs) refers to an optimization problem with multiple conflicting objectives, and since a solution that enables all objectives to reach an optimal value at the same time cannot be found, the final objective of multi-objective optimization is usually to obtain a well-distributed and well-converged set of solutions to approach Pareto Front (PF), i.e. a true optimal solution set. The basic idea of the evolutionary algorithm is to start from a randomly generated initial population, generate new individuals by simulating genetic, crossover, mutation and other operations, and then evaluate the fitness of these individuals according to a certain evaluation method. Individuals with better fitness have a greater chance to be selected as parents of the next generation, thus allowing excellent individuals to remain in the population. The multi-objective evolutionary algorithm (MOEAs) can effectively solve two-objective or three-objective MOPs by virtue of its strong meta-heuristic search capability and population-based framework. However, as the number of targets increases, MOEAs faces many challenges when used to solve the high-dimensional multi-target problem (MaOPs) with target dimensions greater than 3. On the one hand, the proportion of non-dominant solutions in a population increases dramatically due to the weakening of the ability of the dominant relationships to distinguish individuals with increasing target numbers, and on the other hand, MOEAs is more difficult to maintain in terms of distribution due to the sparsity of the solutions in high-dimensional spatial distribution. In recent years, three major classes of high-dimensional multi-objective evolutionary algorithms have been proposed to address these challenges in the high-dimensional multi-objective optimization problem, which are dominant-based evolutionary algorithms, index-based evolutionary algorithms, and decomposition-based evolutionary algorithms, respectively. These three classes of algorithms solve the above-mentioned challenges to some extent, but all suffer from the problem that the convergence and diversity of the population cannot be well balanced, which has a great impact on the performance of the evolutionary algorithm. Disclosure of Invention The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a multi-objective optimization apparatus based on dynamic elite learning and dynamic environment selection. The invention provides a multi-target optimization device based on dynamic elite learning and dynamic environment selection, which is characterized by comprising an input module, an initialization module, a population storage module, a population iteration module, an output module and a population model, wherein the input module is used for inputting m target functions and N decision variable limit range functions constructed according to a designated multi-target optimization task by a user, the initialization module is used for constructing and initializing a population P comprising N individuals, the iteration number G is set to be 1, the population storage module is used for storing the population P, the count storage module is used for storing the iteration number G and a preset maximum iteration number G max, the population iteration module is used for carrying out iterative update on the population P according to the iteration number G, the maximum iteration number G max, the decision variable limit range functions and the target functions to obtain a population P F, the output module is used for displaying the population P F to the user as a preferred scheme of the multi-target optimization task, the population iteration judgment module comprises an iteration judgment sub-module, a population updating sub-module, a dynamic elite learning sub-module and a dynamic environment selection sub-module, the iteration judgment sub-module is used for judging whether the number G in the count storage sub-module is larger than the maximum iteration number G mdx, if the population P in the count storage sub-module is larger than the maximum iteration number G mdx, the population P in the iteration number storage module is used for storing the iteration number P, the decision sub-population P is used as a decision variable limit function, the optimal population P is used for carrying out iterative update on the population P, if the population P is used as a decision variable threshold function, and the optimal population P is used for the decision sub-target optimization task is obtained by t