A dynamic multi-task multi-objective optimization method based on adaptive migration prediction

Abstract

This invention provides a dynamic multi-task multi-objective optimization method based on adaptive transfer prediction, belonging to the field of dynamic multi-task multi-objective optimization technology. The invention obtains an initial population for the current task and uses this initial population as the current population; it then performs environment detection on the current population; if the environment changes, it generates an initial population for the new environment of the current task by designing a diversity knowledge preservation strategy, an adaptive multi-region prediction strategy, and a transfer strategy based on Easy transfer learning; otherwise, it optimizes the current population of the current task to obtain the optimization result. The calculation ends when the preset number of optimization attempts is reached. Through the synergy of three major strategies—diversity knowledge preservation, adaptive multi-region prediction, and knowledge transfer—it achieves accurate perception, rapid response, and stable optimization of the dynamic environment, significantly improving the optimization stability and efficiency in dynamic multi-task scenarios.

Description

Technical Field

[0001] This invention relates to the field of dynamic multi-task multi-objective optimization technology, specifically a dynamic multi-task multi-objective optimization method based on adaptive migration prediction. Background Technology

[0002] With the rapid development of information technology, flexible workshop scheduling systems are increasingly exhibiting characteristics such as multi-task concurrency, multi-objective conflict, and dynamic time-varying. Given limited resources and a constantly changing environment, how to efficiently coordinate multiple tasks and achieve dynamic balance and optimization among multiple mutually constraining objectives has become crucial for improving the overall system efficiency and adaptability.

[0003] Most current research considers shop floor scheduling problems in static environments or dynamic scheduling problems under single tasks. However, in practical applications, shop floors typically need to adapt to multi-task production demands while simultaneously handling dynamic events. Dynamic multi-task multi-objective optimization techniques can process multiple related or independent tasks in parallel within the same framework and adjust optimization strategies in real time according to environmental changes, thereby achieving an adaptive balance among multiple objectives. Therefore, dynamic multi-task multi-objective optimization techniques can be used to solve dynamic scheduling problems with multi-task requirements.

[0004] The core requirement of dynamic multi-task multi-objective optimization is to simultaneously handle multiple interrelated optimization tasks, where the objective functions and constraints of these tasks dynamically change with environmental parameters. This dynamic multi-task multi-objective optimization problem poses a challenge to traditional optimization algorithms: it must handle the trade-offs between multiple competing objectives, coordinate knowledge transfer between different optimization tasks, and maintain a rapid response capability to dynamic changes in environmental parameters.

[0005] In existing technologies, traditional dynamic optimization methods focus only on the dynamic response of a single task, lacking the collaborative processing of relationships between multiple tasks; traditional multi-task optimization methods are difficult to adapt to dynamic environmental changes, and fixed optimization frameworks and prediction models cannot quickly respond to real-time fluctuations in environmental parameters. Furthermore, traditional dynamic optimization methods and multi-task optimization methods cannot effectively solve the dynamic multi-task multi-objective optimization problem existing in flexible workshop scheduling systems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to propose a dynamic multi-task multi-objective optimization method based on adaptive migration prediction. This method is applicable to complex optimization scenarios involving dynamic environmental changes, simultaneous optimization of multiple tasks, and conflicting objectives. Through the synergy of three major strategies—diversity knowledge retention, adaptive multi-region prediction, and knowledge transfer—it achieves accurate perception, rapid response, and stable optimization of the dynamic environment.

[0007] This invention provides a dynamic multi-task, multi-objective optimization method based on adaptive migration prediction, comprising:

[0008] Step 1: Generate an initial population Pop based on the workshop production task; the workshop production task includes multiple workpieces to be processed in the current workshop;

[0009] Step 2: Use the initial population Pop as the current production population. Based on the current production population, perform change detection on the current production environment in the workshop. If a change in the current production environment is detected, proceed to Step 3. If no change in the current production environment is detected, directly use the current production population as the current production population of the new production environment and proceed to Step 4.

[0010] Step 3: Process the current production population to generate a new production population adapted to the new production environment;

[0011] Step 4: Optimize the current production task population using a static optimization algorithm to obtain the target production population. Count the number of iterations from Step 2 to Step 4. If the number of iterations has not reached the preset number of iterations, return to Step 2. If the number of iterations has reached the preset number of iterations, use the target production population as the final production population, and the optimization process ends.

[0012] Optionally, step 1 specifically includes:

[0013] Obtain the workshop production task, obtain the processing equipment number corresponding to each workpiece in the current workshop production environment, and use the equipment number and workpiece processing sequence as production task variables.

[0014] Set the upper and lower boundary vectors of the production task variables. Within the range of the upper and lower boundary vectors, generate N scheduling schemes. The N scheduling schemes form the initial population Pop, where N is the preset population size parameter.

[0015] Optionally, step 2 specifically includes:

[0016] Step 2.1: Select a preset number of scheduling schemes from the current production population to form a detector, denoted as Pop. d ;

[0017] Step 2.2: Set the objective function for the production tasks of M workshops, and calculate the difference value S based on the objective function. This is achieved through the following formula:

[0018] ;

[0019] Where A represents Pop d The number of scheduling schemes, x a Pop dThe a-th scheduling scheme in the diagram, where t0 represents the production environment at the time of the last environmental check, t r Indicates the current production environment, f m (•) represents the m-th objective function;

[0020] Step 2.3: When the difference value S is 0, it indicates that the current production environment has not changed. The current production population is directly used as the current production population of the new production environment, and step 4 is executed. When the difference value S is not 0, it indicates that the current production environment has changed, and step 3 is executed.

[0021] Optionally, step 3 specifically includes:

[0022] Step 3.1: Determine multiple optimal solutions in the current production population based on the objective function, form an optimal solution set, and store the optimal solution set of the current production environment in the storage space;

[0023] Step 3.2: When the number of production environments stored in the storage space exceeds a preset threshold P, calculate the congestion distance for each of the first P-1 production environments. :

[0024] ;

[0025] Where D(•) is the Euclidean distance; p i For the i-th production environment among the first P-1 production environments, p j Let j be the j-th production environment among the first P-1 production environments, and p j With p i different, For p i The center point of the optimal solution set, For p j The center point of the optimal solution set;

[0026] Among the congestion distances of the first P-1 production environments, the production environment with the smallest congestion distance is removed to obtain the final storage space;

[0027] Step 3.3: Calculate the severity of changes in the current production environment, and then calculate the number of subspaces K;

[0028] Step 3.4: Set the maximum and minimum values ​​for each objective function. The interval between the minimum and maximum values ​​forms the target space. Each production environment in the final storage space is used as the target environment.

[0029] Divide the target space of the target environment into K subspaces, and calculate the size, upper limit, and lower limit of each subspace;

[0030] Step 3.5: Determine the inflection points of each subspace;

[0031] Step 3.6: Based on the inflection point of the subspace, generate the objective solution of the subspace using the following formula. The objective solutions of all subspaces constitute the first solution set.

[0032] ;

[0033] ;

[0034] in, This represents the inflection point of the l-th subspace under the t-th target environment. This represents the inflection point of the l-th subspace under the (t-1)-th target environment. Let represent the evolution direction of the l-th subspace under the t-th target environment. This represents the objective solution in the l-th subspace under the t-th objective environment. This represents the objective solution in the l-th subspace under the (t+1)-th objective environment;

[0035] The migration similarity between the current production environment and the final storage space compared to historical production environments (excluding the current production environment) is calculated using the following formula:

[0036] ;

[0037] in, Indicates the current production environment and historical production environment The migration similarity, where n is the number of scheduling schemes in the optimal solution of the historical production environment in the storage space. Indicates historical production environment The objective function value of the w-th optimal solution. Indicates the current production environment The objective function value of the wth optimal solution;

[0038] The optimal solution set of the historical production environment with the highest migration similarity is obtained as the second solution set;

[0039] Step 3.7: When K is less than or equal to a preset threshold, obtain a preset first percentage of solutions from the first solution set as predicted solutions, and obtain a preset second percentage of solutions from the second solution set as predicted solutions, to form a predicted solution set; when K is greater than a preset threshold, obtain a preset second percentage of solutions from the first solution set as predicted solutions, and obtain a preset first percentage of solutions from the second solution set as predicted solutions, to form a predicted solution set.

[0040] Step 3.8: Perform non-dominated sorting on the predicted solution set, determine multiple non-dominated solutions to form an approximate population, and expand the approximate population by interpolation using the following formula until the number of solutions in the approximate population reaches a threshold. Use the expanded approximate population as the target domain.

[0041] ;

[0042] in, For the first in an approximate population One predicted solution, This is the p-th predicted solution in the approximate population. The new predicted solution generated by the interpolation method;

[0043] Step 3.9: Calculate the distribution similarity between the target domain and the optimal solution set of each historical production environment, specifically using the following formula:

[0044] ;

[0045] in, Indicates the target domain. This represents the optimal solution set for the historical production environment. This represents the similarity in distribution between the optimal solution sets of the target domain and the historical production environment. This indicates the number of predicted solutions in the target domain. This represents the z-th predicted solution in the target domain. For mapping functions;

[0046] Obtain the optimal solution set of the historical production environment corresponding to the maximum value of distributed similarity as the source domain;

[0047] Step 3.10: Take the non-dominated solutions in the source domain as excellent solutions and the dominated solutions as non-excellent solutions, construct the non-parametric transfer classifier of Easy transfer learning, classify the target domain based on the source domain and the non-parametric transfer classifier, and obtain the excellent solutions as the current production population of the new production environment.

[0048] Optionally, step 3.3 calculates the severity of changes in the current production environment, and then calculates the number of subspaces K, specifically using the following formula:

[0049] ;

[0050] ;

[0051] ;

[0052] ;

[0053] in, Indicates the current production environment tr The severity of the changes, This represents the q-th scheduling scheme in the current production population. and For intermediate parameters, Indicates the current production environment t r The m-th objective function value for the q-th scheduling scheme is given below. This represents the current production environment t in the final storage space. r The objective function value of the m-th scheduling scheme in the previous production environment. Indicates the current production environment t r The maximum value of the m-th objective function value. Indicates the current production environment t r Let K1 and K2 represent the lower and upper bounds of the m-th objective function value.

[0054] Optionally, in step 3.4, the size, upper limit, and lower limit of each subspace are calculated using the following formulas:

[0055] ;

[0056] in, This represents the size of the l-th subspace of the m-th objective function under the t-th objective environment. This represents the maximum value of the m-th objective function under the t-th objective environment. Let m represent the minimum value of the m-th objective function under the t-th objective environment;

[0057] ;

[0058] ;

[0059] in, Let be the upper bound of the l-th subspace. This is the lower bound of the l-th subspace.

[0060] Optionally, step 3.5 includes:

[0061] Define the extreme value line L of the child control as follows:

[0062] ;

[0063] Where a, b, c are determined by two extreme points in the non-dominated solution set corresponding to the subspace, and f1, f2 are two objective function values ​​of a solution in the non-dominated solution set;

[0064] Calculate the solutions in the non-dominated solution set. Distance to the extreme line L Specifically, this is achieved through the following formula:

[0065] ;

[0066] Among all solutions in the non-dominated solution set, the solution with the largest distance to the extreme line L is selected and used as the inflection point of the subspace.

[0067] The beneficial effects of adopting the above technical solution are as follows:

[0068] This invention employs diverse knowledge retention strategies to preserve more types of environmental information within limited storage space, laying a data foundation for cross-scenario knowledge reuse. Through an adaptive multi-region prediction strategy, it senses the intensity of environmental changes in real time and dynamically adjusts the number of regions and the prediction scheme to achieve precise adaptation to dynamic scenarios. Furthermore, by using a transfer strategy based on Easy transfer learning, it transfers effective historical knowledge to the new environment based on the similarity matching results between the current environment and historical knowledge, avoiding negative transfer problems caused by invalid information interference and improving the optimization stability and efficiency of the algorithm in dynamic multi-task scenarios. Attached Figure Description

[0069] Figure 1 This is a dynamic multi-task multi-objective optimization method based on adaptive migration prediction in an embodiment of the present invention;

[0070] Figure 2 This is a block diagram of the diversity knowledge retention process in an embodiment of the present invention;

[0071] Figure 3 This is a block diagram of the adaptive multi-region prediction process in an embodiment of the present invention;

[0072] Figure 4 This is a block diagram of the knowledge transfer process based on EasyTL in an embodiment of the present invention. Detailed Implementation

[0073] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0074] Addressing the shortcomings of existing technologies, this approach is suitable for complex optimization scenarios involving dynamic environmental changes, simultaneous optimization of multiple tasks, and conflicting objectives.

[0075] This invention provides a dynamic multi-task, multi-objective optimization method based on adaptive migration prediction, combined with... Figure 1 This may include the following steps:

[0076] Step 1: Generate an initial population Pop based on the workshop production tasks;

[0077] Obtain the workshop production task, which includes multiple workpieces to be processed in the current workshop. Obtain the processing equipment number corresponding to each workpiece in the current workshop production environment, and use the equipment number and workpiece processing order as production task variables.

[0078] Set the upper and lower boundary vectors of the production task variables. Within the range of the upper and lower boundary vectors, generate N scheduling schemes. The N scheduling schemes form the initial population Pop, where N is the preset population size parameter.

[0079] Specifically, based on the production workshop's needs, determine the number of tasks T to be optimized, the number of objectives M to be optimized for each task, and identify the production task variables affecting the optimization problem, as well as the number and value limits of these variables (i.e., the search region of the problem): upper boundary vector. and lower boundary vector Initialize the current function evaluation by FE=0 and set the population size to N=100. In this problem, MOEA / D is used as the optimization algorithm for a static environment.

[0080] Step 2: Use the initial population Pop as the current production population. Based on the current production population, perform change detection on the current production environment in the workshop. If a change in the current production environment is detected, proceed to Step 3. If no change in the current production environment is detected, directly use the current production population as the current production population of the new production environment and proceed to Step 4.

[0081] Combination Figure 2 A flowchart of the process of preserving diverse knowledge is provided, and step 3 is described in detail through the following steps;

[0082] Step 2.1: Select a preset number (5%N) of scheduling schemes from the current production population to form a detector, denoted as Pop. d ;

[0083] Step 2.2: Set the objective function for the production tasks of M workshops, denoted as . The difference value S is calculated based on the objective function, specifically through the following formula:

[0084] ;

[0085] Where A represents Pop d The number of scheduling schemes, x a Pop d The a-th scheduling scheme in the diagram, where t0 represents the production environment at the time of the last environmental check, t r Indicates the current production environment, f m (•) represents the m-th objective function;

[0086] Step 2.3: When the difference value S is 0, it indicates that the current production environment has not changed. The current production population is directly used as the current production population of the new production environment, and step 4 is executed. When the difference value S is not 0, it indicates that the current production environment has changed, and step 3 is executed.

[0087] Step 3: Process the current production population to generate a new production population adapted to the new production environment; specifically, generate the initial population under the new environment of the current production task through diversity knowledge preservation, adaptive multi-region prediction and knowledge transfer.

[0088] Step 3.1: Determine multiple optimal solutions in the current production population based on the objective function, form an optimal solution set, and store the optimal solution set of the current production environment in the storage space;

[0089] Step 3.2: When the number of production environments stored in the storage space exceeds a preset threshold P, calculate the congestion distance for each of the first P-1 production environments. :

[0090] ;

[0091] Where D(•) is the Euclidean distance; p i For the i-th production environment among the first P-1 production environments, p j Let j be the j-th production environment among the first P-1 production environments, and p j With p i different, For p i The center point of the optimal solution set, For p j The center point of the optimal solution set;

[0092] Among the congestion distances of the first P-1 production environments, the production environment with the smallest congestion distance is removed to obtain the final storage space;

[0093] Step 3.3: Combining Figure 3 The flowchart of the adaptive multi-region prediction process calculates the severity of changes in the current production environment, and then calculates the number of subspaces K, which is specifically implemented through the following formula:

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] in, Indicates the current production environment t r The severity of the changes, This represents the q-th scheduling scheme in the current production population. and For intermediate parameters, Indicates the current production environment t r The m-th objective function value for the q-th scheduling scheme is given below. This represents the current production environment t in the final storage space. r The objective function value of the m-th scheduling scheme in the previous production environment. Indicates the current production environment t r The maximum value of the m-th objective function value. Indicates the current production environment t r The minimum value of the m-th objective function, where K1 and K2 represent the lower and upper bounds of K;

[0099] Step 3.4: Set the maximum and minimum values ​​for each objective function. The interval between the minimum and maximum values ​​forms the target space. Each production environment in the final storage space is used as the target environment.

[0100] The target space of the target environment is divided into K subspaces, and the size, upper limit, and lower limit of each subspace are calculated using the following formula:

[0101] ;

[0102] in, This represents the size of the l-th subspace of the m-th objective function under the t-th objective environment. This represents the maximum value of the m-th objective function under the t-th objective environment. Let m represent the minimum value of the m-th objective function under the t-th objective environment;

[0103] ;

[0104] ;

[0105] in, Let be the upper bound of the l-th subspace. Let l be the lower bound of the l-th subspace;

[0106] Step 3.5: Determine whether a point is an inflection point by calculating the distance from the point to the extreme line or hyperplane, and identify the inflection points of each subspace;

[0107] Specifically, the extreme value line L of the child control is defined as follows:

[0108] ;

[0109] Where a, b, c are determined by two extreme points in the non-dominated solution set corresponding to the subspace. The non-dominated solution set is the set of all solutions that cannot further improve other objectives without sacrificing a certain objective. f1 and f2 are two objective function values ​​of a solution in the non-dominated solution set.

[0110] Calculate the solutions in the non-dominated solution set. Distance to the extreme line L Specifically, this is achieved through the following formula:

[0111] ;

[0112] Among all the solutions in the non-dominated solution set, the solution with the largest distance to the extreme line L is selected and used as the inflection point of the subspace.

[0113] Step 3.6: The prediction model is constructed using historical environmental information from the last two generations. Specifically, based on the inflection point of the subspace, the objective solution of the subspace is generated using the following formula, and the objective solutions of all subspaces form the first solution set.

[0114] ;

[0115] ;

[0116] in, This represents the inflection point of the l-th subspace under the t-th target environment. This represents the inflection point of the l-th subspace under the (t-1)-th target environment. Let represent the evolution direction of the l-th subspace under the t-th target environment. This represents the objective solution in the l-th subspace under the t-th objective environment. This represents the objective solution in the l-th subspace under the (t+1)-th objective environment;

[0117] The migration similarity between the current production environment and the final storage space compared to historical production environments (excluding the current production environment) is calculated using the following formula:

[0118] ;

[0119] in, Indicates the current production environment and historical production environment The migration similarity, where n is the number of scheduling schemes in the optimal solution of the historical production environment in the storage space. Indicates historical production environment The objective function value of the w-th optimal solution. Indicates the current production environment The objective function value of the wth optimal solution;

[0120] The optimal solution set of the historical production environment with the highest migration similarity is obtained as the second solution set, which is generated by the migration model;

[0121] Step 3.7: When K is less than or equal to a preset threshold, obtain a preset first percentage (70%) of solutions from the first solution set as predicted solutions, and obtain a preset second percentage (30%) of solutions from the second solution set as predicted solutions, to form a predicted solution set; when K is greater than a preset threshold, obtain a preset second percentage of solutions from the first solution set as predicted solutions, and obtain a preset first percentage of solutions from the second solution set as predicted solutions, to form a predicted solution set.

[0122] Step 3.8: Combining Figure 4 Based on the flowchart of the knowledge transfer process of EasyTL, the non-dominated sorting method in the NSGA-II algorithm is used to sort the predicted solution set, and multiple non-dominated solutions are determined to form an approximate population. The approximate population is expanded by interpolation using the following formula until the number of solutions in the approximate population reaches a threshold, so as to construct a population that approximates the position and distribution of the true Pareto solution set. The expanded approximate population is used as the target domain.

[0123] ;

[0124] in, For the first in an approximate population One predicted solution, This is the p-th predicted solution in the approximate population. The new predicted solution generated by the interpolation method;

[0125] Step 3.9: Calculate the distribution similarity between the target domain and the optimal solution set of each historical production environment, specifically using the following formula:

[0126] ;

[0127] in, Indicates the target domain. This represents the optimal solution set for the historical production environment. This represents the similarity in distribution between the optimal solution sets of the target domain and the historical production environment. This indicates the number of predicted solutions in the target domain. This represents the z-th predicted solution in the target domain. For mapping functions;

[0128] Obtain the optimal solution set of the historical production environment corresponding to the maximum value of distributed similarity as the source domain;

[0129] Step 3.10: Take the non-dominated solutions in the source domain as excellent solutions and the dominated solutions as non-excellent solutions, construct the non-parametric transfer classifier of Easy transfer learning, classify the target domain based on the source domain and the non-parametric transfer classifier, and obtain the excellent solutions as the current production population of the new production environment;

[0130] First, construct a nonparametric transfer classifier for Easy transfer learning, and let... Let C represent the class label, and C represent the number of class labels. Define the probability annotation matrix. and its elements , Describes the solution x in the source domain u Belongs to class The probability is given by U, where U represents the number of solutions in the source domain.

[0131] Easy transfer learning focuses on learning the probability annotation matrix Pr. Therefore, the classification learning cost function of Easy transfer learning can be formalized as:

[0132] ;

[0133] in, x represents u The Euclidean distance between the source domain and the c-th class center. The value is a measure of x u The confidence probability belonging to class c has the following constraints:

[0134] ;

[0135] ;

[0136] The target domain is classified by predicting its class label based on the learning probability matrix of the source domain. The best classification results are then used as the initial population for the new environment.

[0137] Step 4: Optimize the current production task population using a static optimization algorithm to obtain the target production population. Count the number of iterations from Step 2 to Step 4. If the number of iterations has not reached the preset number of iterations, return to Step 2. If the number of iterations has reached the preset number of iterations, use the target production population as the final production population, and the optimization process ends.

[0138] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A dynamic multi-task, multi-objective optimization method based on adaptive migration prediction, characterized in that, include: Step 1: Generate an initial population Pop based on the workshop production task; the workshop production task includes multiple workpieces to be processed in the current workshop; Step 2: Use the initial population Pop as the current production population. Based on the current production population, perform change detection on the current production environment in the workshop. If a change in the current production environment is detected, proceed to Step 3. If no change in the current production environment is detected, directly use the current production population as the current production population of the new production environment and proceed to Step 4. Step 3: Process the current production population to generate a new production population adapted to the new production environment; Step 4: Optimize the current production task population using a static optimization algorithm to obtain the target production population. Count the number of iterations from Step 2 to Step 4. If the number of iterations has not reached the preset number of iterations, return to Step 2. If the number of iterations has reached the preset number of iterations, use the target production population as the final production population, and the optimization process ends.

2. The dynamic multi-task multi-objective optimization method based on adaptive migration prediction according to claim 1, characterized in that, Step 1 specifically includes: Obtain the workshop production task, obtain the processing equipment number corresponding to each workpiece in the current workshop production environment, and use the equipment number and workpiece processing sequence as production task variables. Set the upper and lower boundary vectors of the production task variables. Within the range of the upper and lower boundary vectors, generate N scheduling schemes. The N scheduling schemes form the initial population Pop, where N is the preset population size parameter.

3. The dynamic multi-task multi-objective optimization method based on adaptive migration prediction according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Select a preset number of scheduling schemes from the current production population to form a detector, denoted as Pop. d ; Step 2.2: Set the objective function for the production tasks of M workshops, and calculate the difference value S based on the objective function. This is achieved through the following formula: ； Where A represents Pop d The number of scheduling schemes, x a Pop d The a-th scheduling scheme in the diagram, where t0 represents the production environment at the time of the last environmental check, t r Indicates the current production environment, f m (•) represents the m-th objective function; Step 2.3: When the difference value S is 0, it indicates that the current production environment has not changed. The current production population is directly used as the current production population of the new production environment, and step 4 is executed. When the difference value S is not 0, it indicates that the current production environment has changed, and step 3 is executed.

4. The dynamic multi-task multi-objective optimization method based on adaptive migration prediction according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Determine multiple optimal solutions in the current production population based on the objective function, form an optimal solution set, and store the optimal solution set of the current production environment in the storage space; Step 3.2: When the number of production environments stored in the storage space exceeds a preset threshold P, calculate the congestion distance for each of the first P-1 production environments. : ； Where D(•) is the Euclidean distance; p i For the i-th production environment among the first P-1 production environments, p j Let j be the j-th production environment among the first P-1 production environments, and p j With p i different, For p i The center point of the optimal solution set, For p j The center point of the optimal solution set; Among the congestion distances of the first P-1 production environments, the production environment with the smallest congestion distance is removed to obtain the final storage space; Step 3.3: Calculate the severity of changes in the current production environment, and then calculate the number of subspaces K; Step 3.4: Set the maximum and minimum values ​​for each objective function. The interval between the minimum and maximum values ​​forms the target space. Each production environment in the final storage space is used as the target environment. Divide the target space of the target environment into K subspaces, and calculate the size, upper limit, and lower limit of each subspace; Step 3.5: Determine the inflection points of each subspace; Step 3.6: Based on the inflection point of the subspace, generate the objective solution of the subspace using the following formula. The objective solutions of all subspaces constitute the first solution set. ； ； in, This represents the inflection point of the l-th subspace under the t-th target environment. This represents the inflection point of the l-th subspace under the (t-1)-th target environment. Let represent the evolution direction of the l-th subspace under the t-th target environment. This represents the objective solution in the l-th subspace under the t-th objective environment. This represents the objective solution in the l-th subspace under the (t+1)-th objective environment; The migration similarity between the current production environment and the final storage space compared to historical production environments (excluding the current production environment) is calculated using the following formula: ； in, Indicates the current production environment and historical production environment The migration similarity, where n is the number of scheduling schemes in the optimal solution of the historical production environment in the storage space. Indicates historical production environment The objective function value of the w-th optimal solution. Indicates the current production environment The objective function value of the wth optimal solution; The optimal solution set of the historical production environment with the highest migration similarity is obtained as the second solution set; Step 3.7: When K is less than or equal to a preset threshold, obtain a preset first percentage of solutions from the first solution set as predicted solutions, and obtain a preset second percentage of solutions from the second solution set as predicted solutions, to form a predicted solution set; when K is greater than a preset threshold, obtain a preset second percentage of solutions from the first solution set as predicted solutions, and obtain a preset first percentage of solutions from the second solution set as predicted solutions, to form a predicted solution set. Step 3.8: Perform non-dominated sorting on the predicted solution set, determine multiple non-dominated solutions to form an approximate population, and expand the approximate population by interpolation using the following formula until the number of solutions in the approximate population reaches a threshold. Use the expanded approximate population as the target domain. ； in, For the first in an approximate population One predicted solution, This is the p-th predicted solution in the approximate population. The new predicted solution generated by the interpolation method; Step 3.9: Calculate the distribution similarity between the target domain and the optimal solution set of each historical production environment, specifically using the following formula: ； in, Indicates the target domain. This represents the optimal solution set for the historical production environment. This represents the similarity in distribution between the optimal solution sets of the target domain and the historical production environment. This indicates the number of predicted solutions in the target domain. This represents the z-th predicted solution in the target domain. For mapping functions; Obtain the optimal solution set of the historical production environment corresponding to the maximum value of distributed similarity as the source domain; Step 3.10: Take the non-dominated solutions in the source domain as excellent solutions and the dominated solutions as non-excellent solutions, construct the non-parametric transfer classifier of Easy transfer learning, classify the target domain based on the source domain and the non-parametric transfer classifier, and obtain the excellent solutions as the current production population of the new production environment.

5. The dynamic multi-task multi-objective optimization method based on adaptive migration prediction according to claim 4, characterized in that, Step 3.3 calculates the severity of changes in the current production environment, and then calculates the number of subspaces K, specifically using the following formula: ； ； ； ; in, Indicates the current production environment t r The severity of the changes, This represents the q-th scheduling scheme in the current production population. and For intermediate parameters, Indicates the current production environment t r The m-th objective function value for the q-th scheduling scheme is given below. This represents the current production environment t in the final storage space. r The objective function value of the m-th scheduling scheme in the previous production environment. Indicates the current production environment t r The maximum value of the m-th objective function value. Indicates the current production environment t r Let K1 and K2 represent the lower and upper bounds of the m-th objective function value.

6. The dynamic multi-task multi-objective optimization method based on adaptive migration prediction according to claim 4, characterized in that, Step 3.4 calculates the size, upper limit, and lower limit of each subspace using the following formula: ； in, This represents the size of the l-th subspace of the m-th objective function under the t-th objective environment. This represents the maximum value of the m-th objective function under the t-th objective environment. Let m represent the minimum value of the m-th objective function under the t-th objective environment; ； ； in, Let be the upper bound of the l-th subspace. This is the lower bound of the l-th subspace.

7. The dynamic multi-task multi-objective optimization method based on adaptive migration prediction according to claim 4, characterized in that, Step 3.5 includes: Define the extreme value line L of the child control as follows: ； Where a, b, c are determined by two extreme points in the non-dominated solution set corresponding to the subspace, and f1, f2 are two objective function values ​​of a solution in the non-dominated solution set; Calculate the solutions in the non-dominated solution set. Distance to the extreme line L Specifically, this is achieved through the following formula: ； Among all solutions in the non-dominated solution set, the solution with the largest distance to the extreme line L is selected and used as the inflection point of the subspace.

Images