Dynamic multi-target service combination optimization method and system based on preference perception

The method addresses dynamic user preference changes by dividing populations based on preference levels and using a dominance relationship within a social learning algorithm, ensuring accurate and responsive service composition that meets user needs.

CN119090689APending Publication Date: 2024-12-06HENAN CULTURAL TOURISM INVESTMENT GROUP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411067229.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing systems fail to effectively perceive and respond to dynamic changes in user preferences, leading to inefficient and resource-intensive searches for service combination solutions that do not meet user needs, affecting user experience in service composition optimization.

Method used

A dynamic multi-objective service composition optimization method that divides populations based on user preference levels, incorporates environmental change detection, and uses a preference level dominance relationship within a social learning optimization algorithm to guide the population towards the user's area of interest, ensuring accurate and responsive service composition.

Benefits of technology

The method efficiently perceives and adapts to user preference changes, providing service combinations that meet user needs, enhancing user satisfaction and resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090689A_ABST
    Figure CN119090689A_ABST
Patent Text Reader

Abstract

The invention relates to the field of dynamic multi-target service combination optimization, in particular to a dynamic multi-target service combination optimization method and system based on preference perception, and the method comprises the steps: generating a region of interest of a user based on the minimum and maximum preference values of each target provided by the user; dividing the population into sub-populations of different grades according to the closeness degree of the individuals and the region of interest of the user, incorporating user preferences in the service combination optimization process, obtaining the dominating relationship between the individuals, guiding the population to evolve towards the region of interest of the user, and improving the service combination optimization efficiency. A dynamic preference perception strategy and a preference level dominating relation are combined with an improved social learning optimization algorithm, so that the change of user preferences can be effectively perceived and responded, a service combination scheme of the user is actively adjusted according to the dynamic change of the user preferences, a more accurate service combination scheme meeting the user preferences is provided, and the user experience is improved. And the service resource utilization rate and the user satisfaction are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic multi-objective service composition optimization, and in particular to a dynamic multi-objective service composition optimization method and system based on preference perception. Background Art

[0002] With the continuous development of cloud computing technology, more and more new functions and services are being developed and deployed on cloud platforms. These services cover multiple fields and provide users with strong technical support and convenient service experience. At the same time, user needs are becoming increasingly complex and personalized. Service platforms need to continue to innovate and optimize, and provide more flexible and customizable service combinations to meet the diverse needs of users.

[0003] Service composition is an important technology for combining multiple candidate services to build value-added services. It is an important research issue in the field of service computing. With the evolution of people's lifestyles and the development of information technology, the preference requirements of users in dynamic multi-objective service composition optimization problems are becoming increasingly complex. In practical applications, users are often only interested in a part of the service combination solution set, rather than all the optimal solutions. Blindly searching for all the optimal service combination solutions not only wastes resources and time, but also makes it difficult for users to choose from a large number of service combination solutions.

[0004] User preferences may change dynamically with changes in the environment. How to efficiently and accurately perceive the dynamic changes in user preferences is one of the key issues that need to be solved to achieve preference-aware dynamic multi-objective service composition optimization. Existing research has not considered the dynamic changes in user preferences and cannot effectively perceive and respond to changes in user preferences. Research on preference-aware dynamic multi-objective service composition is relatively rare. In addition, the dynamic nature of user preferences makes it impossible for the constructed service composition scheme to meet the user's new preference requirements, which seriously affects the user experience. Therefore, how to effectively respond to changes in user preferences in a dynamic environment and construct a service composition scheme that meets the user's new preferences is also a problem that needs to be considered. Summary of the invention

[0005] The present invention provides a preference-aware dynamic multi-objective service combination optimization method and system.

[0006] The technical solution of the present invention is as follows:

[0007] A dynamic multi-objective service composition optimization method based on preference perception includes the following steps:

[0008] Based on the proximity between individuals and the user's area of ​​interest, the initial population is divided into sub-populations with different preference levels, specifically:

[0009] S1.1 obtains the boundary points of the subpopulation with a preference level of 1 based on the minimum preference value and the maximum preference value of each target provided by the user. The area where the subpopulation with a preference level of 1 is located is the area of ​​interest of the user;

[0010] S1.2 Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas of the subpopulations with different preference levels other than preference level 1 are obtained;

[0011] S2, environmental change detection: detect changes in service resource quality attributes and user goals between the current moment and the previous moment.

[0012] If a change is detected, step S3 is executed.

[0013] If no change is detected, step S4 is executed;

[0014] S3, environmental change response: change the attribute values ​​and constraints of service resource quality;

[0015] S4, detect whether the user preference has changed,

[0016] If there is a change, execute step S5.

[0017] If no change occurs, execute step S6;

[0018] S5. Adjust preference level:

[0019] Based on the location of each individual, the individual is divided into different preference areas and the preference level is re-determined by comparing the location with the boundary values ​​of different preference areas on different dimensional targets;

[0020] S6. Based on the preference level, the preference level dominance relationship between individuals is obtained; the preference level dominance relationship is introduced into the social learning optimization algorithm, and the observational habitual weight and imitation learning weight are obtained based on the preference level dominance relationship, so as to guide the population to converge to the user's area of ​​interest and retain the individual with the highest fitness value that meets the user's preference;

[0021] S7. If the iteration number threshold has been reached, the loop ends, and the individual with the highest fitness value is output as the optimal individual to obtain a service combination solution that meets the user's preferences;

[0022] Otherwise, the individual with the highest fitness value is taken as the optimal individual and step S2 is executed.

[0023] Specifically, in S1, based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained, and the area where the sub-population with a preference level of 1 is located is the user's area of ​​interest, as follows:

[0024]

[0025] Among them, min i is the minimum value of the i-th dimension target, d is the value range of the i-th dimension target, W i,min is the user’s minimum preference value for the i-th dimension target, W i,max is the user’s maximum preference value for the i-th dimension target, is the minimum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, is the maximum boundary value of the target of the i-th dimension in the region where the subpopulation with preference level 1 is located.

[0026] Furthermore, in S1, based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas where the subpopulations with different preference levels other than the preference level of 1 are located are obtained, as follows:

[0027]

[0028] in, is the minimum boundary value of the target of the i-th dimension in the region where the subpopulation with preference level other than 1 is located, is the maximum boundary value of the target of the i-th dimension in the region where the subpopulation other than the preference level 1 is located, is the minimum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, is the maximum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, R j,i Based on the user's area of ​​interest, the area where the subpopulation with preference level j is located is wider than the user's area of ​​interest on the i-th dimension target.

[0029] Specifically, in S5, based on the location of each individual, the individual is divided into different preference areas by comparing the location with the boundary values ​​of different preference areas on different dimensional targets, specifically:

[0030]

[0031] Among them, PR is the user's preferred region, is the total dimension, m is the target dimension; x i is the objective function value of individual x on the i-th dimension target, l i is the next value of the preferred region, u i is the previous value of the preferred area.

[0032] Furthermore, the preference level dominance relationship is determined in S5, specifically:

[0033] If individual x Pareto dominates individual y, or:

[0034] Individual x and individual y do not dominate each other, and individual g x <g y ,

[0035] The level of the subpopulation where x belongs is less than the level of the subpopulation where individual y belongs.

[0036] It is assumed that the preference level of individual x dominates that of individual y, where g x is the preference level of individual x, g y is the preference level of individual y.

[0037] Specifically, in S6, the preference level dominance relationship is introduced into the social learning optimization algorithm to guide the population to converge to the user's area of ​​interest, and retain the individual with the highest fitness value that meets the user's preference, specifically:

[0038] In the microspace, based on the set crossover rate and mutation rate, the individuals in the population are sequentially subjected to selection, crossover and mutation operations;

[0039] In the learning space, the learning habit weights and disturbance learning factors based on the preference level dominance relationship are introduced to perform observation learning operations; the imitation learning weights based on the preference level dominance relationship are introduced to perform imitation learning operations;

[0040] In the belief space, knowledge updating operations and cultural influence operations are performed to retain individuals with the highest fitness values ​​that meet user preferences.

[0041] Furthermore, in the learning space, the learning habit weight and the disturbance learning factor based on the preference level dominance relationship are introduced to perform observation learning operation, specifically:

[0042]

[0043] in, is the service number used to perform the kth subtask in the i-th service combination scheme after the observation learning operation; x ik is the service number used to execute the kth subtask in the i-th service combination scheme; R (x,y) is the learning habit weight based on the preference level dominance relationship, To learn the perturbation factor, is the service number of the optimal individual j for executing the kth subtask; g x is the preference level of individual x, g y is the preference level of individual y, g max is the maximum preference level, g min is the minimum preference level.

[0044] Furthermore, the imitation learning weights based on the preference level dominance relationship are introduced to perform the imitation learning operation, specifically:

[0045]

[0046] Among them, r1 and r2 are two different individuals selected randomly, r1,r2∈{1,...,N}, N is the population size; S∈[0,1] is the scaling factor, is the imitation learning weight based on the preference level dominance relationship, [] is the rounding operation, is the service number used to perform the kth subtask in the i-th service combination scheme obtained after the imitation learning operation is completed, x ik is the service number used to execute the kth subtask in the i-th service combination scheme, is the service number of the optimal individual j for executing the kth subtask, represents the service number used to execute the kth subtask in any individual r1, Represents the service number used to execute the kth subtask in any individual r2.

[0047] The present invention also provides a preference-aware dynamic multi-objective service composition optimization system, comprising:

[0048] Initial population division module: Based on the proximity between individuals and the user's area of ​​interest, the initial population is divided into sub-populations with different preference levels, specifically:

[0049] Based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained. The area where the sub-population with a preference level of 1 is located is the user's area of ​​interest.

[0050] Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas where the subpopulations with different preference levels other than preference level 1 are located are obtained;

[0051] Environment change detection module: performs change detection on the service resource quality attributes and user goals at the current moment and the previous moment. If a change is detected, the environment change response module is executed. If no change is detected, the user preference change detection module is executed.

[0052] Environmental change response module: changes the attribute values ​​and constraints of service resource quality;

[0053] User preference change detection module: detects whether the user preference has changed. If so, the preference level adjustment module is executed. If not, the social learning optimization algorithm optimization module is executed.

[0054] Preference level adjustment module: Based on the location of each individual, it is compared with the boundary values ​​of different preference areas on different dimensional targets, and the individual is divided into different preference areas to redefine the preference level;

[0055] Social learning optimization algorithm optimization module: Based on the preference level, obtain the preference level dominance relationship between individuals; introduce the preference level dominance relationship into the social learning optimization algorithm, obtain the observational habitual weight and imitation learning weight based on the preference level dominance relationship, guide the population to converge to the user's area of ​​interest, and retain the individual with the highest fitness value that meets the user's preference;

[0056] Output module: If the iteration number threshold is reached, the loop ends and the individual with the highest fitness value is output as the optimal individual to obtain a service combination solution that meets the user's preferences;

[0057] Otherwise, the individual with the highest fitness value is taken as the optimal individual and the environmental change detection module is executed.

[0058] The beneficial effects of the present invention are:

[0059] 1. The present invention proposes a dynamic preference perception strategy based on preference regions. Based on the minimum and maximum preference values ​​of each target provided by the user, the user's region of interest is generated, and the population is divided into sub-populations of different levels according to the degree of proximity between the individual and the user's region of interest. Incorporating user preferences into the service combination optimization process can effectively perceive the dynamic changes of user preferences, ensure the diversity of the service combination solution set, and avoid stagnation in the local optimum.

[0060] 2. The present invention proposes a dominance relationship based on preference levels. Based on the distance between an individual and the user's area of ​​interest, the user's preference is considered on the basis of Pareto dominance, and the individual is divided into different preference areas. The preference level is determined to obtain the dominance relationship between individuals. This dominance relationship can guide the population to evolve in the direction of the user's area of ​​interest based on the preference information given by the user, and more accurately search for service combination solutions that meet the user's preferences.

[0061] 3. The present invention proposes a preference-aware dynamic multi-objective service combination optimization method. Based on the preference level dominance relationship, the social learning optimization algorithm is improved, and the dynamic preference perception strategy and the preference level dominance relationship are combined with the improved social learning optimization algorithm. It can effectively perceive and respond to changes in user preferences, actively adjust the user's service combination plan according to the dynamic changes in user preferences, provide a service combination plan that is more accurate to user preferences, and improve service resource utilization and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In the attached picture:

[0063] Figure 1 Schematic diagram of a dynamic multi-objective service composition optimization method based on preference perception in an embodiment;

[0064] Figure 2 It is a schematic diagram of the running time of the dynamic multi-objective service composition optimization method based on preference perception under different numbers of candidate services in the embodiment;

[0065] Figure 3 It is a schematic diagram of the running time of the dynamic multi-objective service composition optimization method based on preference perception under different numbers of subtasks in the embodiment;

[0066] Figure 4 Schematic diagram of service quality attribute values ​​of a solution set of a dynamic multi-objective service composition optimization method based on preference perception under different problem scales in the embodiment;

[0067] Figure 5 A schematic diagram of service quality attribute values ​​of a solution set of a dynamic multi-objective service composition optimization method based on preference perception under different preference information in an embodiment;

[0068] Figure 6 This is a schematic diagram of algorithm performance comparison in the embodiments;

[0069] Figure 7 A schematic diagram of the effectiveness verification of the preference perception strategy in the embodiment;

[0070] Figure 8 Schematic diagram of the effectiveness verification of the preference level dominance strategy in the embodiment. DETAILED DESCRIPTION

[0071] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings.

[0072] Example

[0073] This embodiment provides a dynamic multi-objective service composition optimization method based on preference perception, see Figure 1 , including the following steps:

[0074] Step 1: Based on the proximity between individuals and the user's area of ​​interest, the initial population is divided into sub-populations with different preference levels, specifically:

[0075] 1.1 Based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained. The area where the sub-population with a preference level of 1 is located is the area of ​​interest to the user;

[0076] 1.2 Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas of subpopulations with different preference levels other than preference level 1 are obtained.

[0077] It also includes taking the three service quality attributes of cost, response time and reliability as optimization goals and setting global constraints, specifically:

[0078]

[0079] Among them, minF(x, t) is the minimum value of the three-dimensional objective function, is the cost attribute value of the i-th component service at time t, is the time attribute value of the i-th component service at time t, is the reliability attribute value of the ith component service at time t, x = {x1, x2, ..., x n} is the n-dimensional decision vector in the decision space Ω, t is the time variable, Cost(x, t) is the cost objective function, Time(x, t) is the response time objective function, Rel(x, t) is the reliability objective function, Cost max is the maximum cost attribute value, Time max is the maximum time attribute value, Rel max is the maximum reliability attribute value.

[0080] Based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained. The area where the sub-population with a preference level of 1 is located is the user's area of ​​interest, as follows:

[0081]

[0082] Among them, min i is the minimum value of the i-th dimension target, d is the value range of the i-th dimension target, W i,min is the user’s minimum preference value for the i-th dimension target, W i,max is the user’s maximum preference value for the i-th dimension target, is the minimum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, is the maximum boundary value of the target of the i-th dimension in the region where the subpopulation with preference level 1 is located.

[0083] Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas of subpopulations with different preference levels other than preference level 1 are obtained, as follows:

[0084]

[0085] in, is the minimum boundary value of the target of the i-th dimension in the region where the subpopulation with preference level other than 1 is located, is the maximum boundary value of the target of the i-th dimension in the region where the subpopulation other than the preference level 1 is located, is the minimum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, is the maximum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, R j,i Based on the user's area of ​​interest, the area where the subpopulation with preference level j is located is wider than the user's area of ​​interest on the i-th dimension target.

[0086] Step 2: Environmental change detection:

[0087] Perform change detection on the service resource quality attributes and user goals at the current moment and the previous moment.

[0088] If a change is detected, step S3 is executed.

[0089] If no change is detected, step S4 is executed.

[0090] Specifically, service resource quality attributes include cost, response time, and reliability.

[0091] Step 3: Response to environmental changes: Change the attribute values ​​and constraints of service resource quality.

[0092] Step 4: Detect whether user preferences have changed.

[0093] If there is a change, execute step S5.

[0094] If there is no change, execute step S6.

[0095] Step 5: Adjust preference level:

[0096] Based on the location of each individual, the individual is divided into different preference areas and the preference level is re-determined by comparing the location with the boundary values ​​of different preference areas on different dimensional targets;

[0097] Specifically, based on the location of each individual, it is divided into different preference areas by comparing the boundary values ​​of different preference areas on different dimensional targets, specifically:

[0098]

[0099] Among them, PR is the user's preferred region, is the total dimension, m is the target dimension; x i is the objective function value of individual x on the i-th dimension target, l i is the next value of the preferred region, u i is the previous value of the preferred area.

[0100] Determine the preference level dominance relationship, specifically:

[0101] If individual x Pareto dominates individual y, or:

[0102] Individual x and individual y do not dominate each other, and individual g x <g y ,

[0103] The level of the subpopulation where x belongs is less than the level of the subpopulation where individual y belongs.

[0104] It is assumed that the preference level of individual x dominates that of individual y, where g x is the preference level of individual x, g y is the preference level of individual y.

[0105] Preference level dominance uses the distance between individuals and the user's area of ​​interest to determine the dominance relationship between individuals, and iteratively updates the population through this dominance relationship to obtain the optimal service combination solution that meets the user's preference requirements. The preference level dominance relationship can not only guide the entire population to converge to the Pareto frontier through the Pareto dominance relationship, but also increase the chances of individuals closer to the preference area being selected and retained.

[0106] Step 6: Based on the preference level, obtain the preference level dominance relationship between individuals; introduce the preference level dominance relationship into the social learning optimization algorithm, and obtain the observational habitual weight and imitation learning weight based on the preference level dominance relationship, guide the population to converge to the user's area of ​​interest, and retain the individual with the highest fitness value that meets the user's preference.

[0107] The preference hierarchy dominance relationship is introduced into the social learning optimization algorithm to guide the population to converge to the user's area of ​​interest and retain the individuals with the highest fitness value that meets the user's preference. Specifically:

[0108] In the microspace, based on the set crossover rate and mutation rate, the individuals in the population are sequentially subjected to selection, crossover and mutation operations.

[0109] The present invention adopts the roulette selection method as the selection operation, and the crossover operation is expressed as follows:

[0110]

[0111] Among them, X i =(x i1 ,...,x ik ,...,x in ) and X j =(x j1 ,...,x jk ,...,x jn ) are any two different individuals, pc represents the crossover rate, q is a random integer in the interval [1,n], n represents the number of subtasks, and are the two individuals after the crossover operation, x iq is the service number for executing the qth subtask in the i-th service combination scheme; is x jq is the service sequence number used to execute the qth subtask in the jth service combination solution.

[0112] After the crossover operation, in individual X i and Between, X j and Greedy operations are performed between them, and better individuals are retained.

[0113] Mutation operation: select any individual X in the population i =(x i1 ,...,x iw ,...,x in ), randomly select a mutation gene point x iw , using a randomly generated new gene x il Replace X i The variant gene point x iw , and obtain the individual X after the mutation operation i ', after executing the mutation operation, the greedy operation is used to retain the better individuals. The mutation operation is expressed as follows:

[0114]

[0115] Further, in the learning space, learning habit weights and perturbation learning factors based on preference level dominance are introduced to perform observation learning operations; imitation learning weights based on preference level dominance are introduced to perform imitation learning operations;

[0116] In the learning space, the learning habit weight and disturbance learning factor based on the preference level dominance relationship are introduced to perform observation learning operations, specifically:

[0117]

[0118] in, is the service number used to perform the kth subtask in the i-th service combination scheme after the observation learning operation; x ik is the service number used to execute the kth subtask in the i-th service combination scheme; R (x,y) is the learning habit weight based on the preference level dominance relationship, To learn the perturbation factor, is the service number of the optimal individual j for executing the kth subtask; g x is the preference level of individual x, g y is the preference level of individual y, g max is the maximum preference level, g min is the minimum preference level.

[0119] The imitation learning weight based on the preference level dominance relationship is introduced to perform the imitation learning operation, specifically:

[0120]

[0121] Among them, r1 and r2 are two different individuals selected randomly, r1,r2∈{1,...,N}, N is the population size; S∈[0,1] is the scaling factor, is the imitation learning weight based on the preference level dominance relationship, [] is the rounding operation, is the service number used to perform the kth subtask in the i-th service combination scheme obtained after the imitation learning operation is completed, x ik is the service number used to execute the kth subtask in the i-th service combination scheme, is the service number of the optimal individual j for executing the kth subtask, represents the service number used to execute the kth subtask in any individual r1, Represents the service number used to execute the kth subtask in any individual r2.

[0122] In the belief space, knowledge updating operations and cultural influence operations are performed to retain individuals with the highest fitness values ​​that meet user preferences. The knowledge updating operations and cultural influence operations are existing technologies and will not be described in detail in the present invention.

[0123] Step 7: If the iteration number threshold has been reached, the loop ends, and the individual with the highest fitness value is output as the optimal individual to obtain a service combination solution that meets the user's preferences;

[0124] Otherwise, the individual with the highest fitness value is taken as the optimal individual and step 2 is executed.

[0125] In this embodiment, the fitness value is calculated using the Pareto dominance concept, and the specific calculation method is:

[0126]

[0127] Among them, fit is the fitness value, dom is the number of dominated solutions, and |PF| is the number of all solutions.

[0128] The present invention also provides a preference-aware dynamic multi-objective service composition optimization system, comprising:

[0129] Initial population division module: Based on the proximity between individuals and the user's area of ​​interest, the initial population is divided into sub-populations with different preference levels, specifically:

[0130] Based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained. The area where the sub-population with a preference level of 1 is located is the user's area of ​​interest.

[0131] Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas where the subpopulations with different preference levels other than preference level 1 are located are obtained;

[0132] Environment change detection module: performs change detection on the service resource quality attributes and user goals at the current moment and the previous moment. If a change is detected, the environment change response module is executed. If no change is detected, the user preference change detection module is executed.

[0133] Environmental change response module: changes the attribute values ​​and constraints of service resource quality;

[0134] User preference change detection module: detects whether the user preference has changed. If so, the preference level adjustment module is executed. If not, the social learning optimization algorithm optimization module is executed.

[0135] Preference level adjustment module: Based on the location of each individual, it is compared with the boundary values ​​of different preference areas on different dimensional targets, and the individual is divided into different preference areas to redefine the preference level;

[0136] Social learning optimization algorithm optimization module: Based on the preference level, obtain the preference level dominance relationship between individuals; introduce the preference level dominance relationship into the social learning optimization algorithm, obtain the observational habitual weight and imitation learning weight based on the preference level dominance relationship, guide the population to converge to the user's area of ​​interest, and retain the individual with the highest fitness value that meets the user's preference;

[0137] Output module: If the iteration number threshold is reached, the loop ends and the individual with the highest fitness value is output as the optimal individual to obtain a service combination solution that meets the user's preferences;

[0138] Otherwise, the individual with the highest fitness value is taken as the optimal individual and the environmental change detection module is executed.

[0139] Experimental content

[0140] The number of candidate services for each component service is set to be the same, and each candidate service has three service quality attributes, namely response time, cost, and reliability. The quality attribute values ​​of the candidate services are randomly generated within a certain range, as shown in Table 1. Considering the randomness of the data set, in order to eliminate the random influence of the final experimental results, each group of experiments is run independently 20 times, and the average of these 20 test results is taken as the final experimental result.

[0141] Table 1 Range of service quality attribute values ​​for candidate services

[0142]

[0143] Set the expansion standard R = {0%, 15%, 30%}, where the first value 0% represents the user's area of ​​interest, 15% and 30% represent the expansion degree of the corresponding area in each dimension based on the user's area of ​​interest. In addition, according to the distance between the service combination solution set and the user's area of ​​interest, the service combination solution set is divided into four levels, namely, the first level user area of ​​interest solution set, the second level short distance solution set, the third level medium distance solution set, and the fourth level long distance solution set.

[0144] The experimental part of the present invention implements a preference-aware dynamic multi-objective service composition optimization method using Java programming language, where the crossover probability is set to 0.85, the mutation probability is set to 0.05, the knowledge update selection ratio is set to 10%, and the cultural influence update interval is set to 5. In the imitation learning, the number of individuals in the learning group is 3.

[0145] The present invention proposes a preference-aware dynamic multi-objective service composition optimization method, hereinafter referred to as PA-DMOSCO.

[0146] 1. PA-DMOSCO feasibility verification

[0147] (1) Performance verification of PA-DMOSCO at different iteration numbers

[0148] In order to verify the feasibility of the PA-DMOSCO algorithm in solving the preference-aware dynamic multi-objective service composition problem under different numbers of iterations, two environments with different numbers of candidate services and different numbers of subtasks are set. For different numbers of candidate services, the number of subtasks is fixed to 10, and the number of candidate services corresponding to the subtasks is set to 10, 20, 50, 100, and 200 respectively. The preference weight vector of the user's area of ​​interest is set to W2 = [0.3, 0.7], and the running time of the algorithm under different numbers of iterations is recorded. The experimental results are shown in Figure 2 As shown, m represents the number of candidate services corresponding to each subtask.

[0149] Through experimental results Figure 7 From the data in , we can observe that the running time of the PA-DMOSCO algorithm for solving the preference-aware dynamic multi-objective service composition problem is linearly related to the size of the candidate services. In other words, as the number of candidate services increases, the running time of the algorithm does not increase exponentially. Based on this observation, we can conclude that the PA-DMOSCO method is feasible in solving the preference-aware dynamic multi-objective service composition problem at different sizes of candidate services.

[0150] For different numbers of subtasks, the number of candidate services corresponding to each subtask is fixed to 50, and the number of subtasks is set to 5, 10, 15, 20, and 25 respectively. The preference weight vector of the user's area of ​​interest is set to W2 = [0.3, 0.7]. The PA-DMOSCO algorithm is executed, and the running time of the algorithm at different iterations is recorded. The experimental results are shown in Figure 3 As shown, where n represents the number of subtasks.

[0151] By observation Figure 8 It can be found that when the number of candidate services corresponding to the component service is 50, the running time of the PA-DMOSCO algorithm increases with the number of iterations. The running time when the number of subtasks is 5 shows a certain linear relationship with the running time when the number of subtasks is 10. In addition, when the number of subtasks is 15, 20, and 25, the running time of the PA-DMOSCO algorithm does not show an obvious correlation. In short, the running time of the PA-DMOSCO algorithm is linearly correlated with the number of different subtasks. In addition, as the number of subtasks increases, the running time of the algorithm does not increase exponentially.

[0152] Based on the experimental results under different numbers of candidate services and different numbers of subtasks, it can be concluded that the PA-DMOSCO algorithm is feasible in solving preference-aware dynamic multi-objective service composition optimization problems.

[0153] (2) Performance verification of PA-DMOSCO under different problem scales

[0154] In order to verify the adaptability of the PA-DMOSCO algorithm for solving preference-aware dynamic multi-objective service composition optimization problems under different problem scales, different problem scales are set by setting the number of different subtasks and the number of candidate services corresponding to different subtasks. The experiment sets five problem scales, and the number of subtasks and candidate services are 5*10, 10*20, 20*50, 30*100, and 50*200, respectively. The preference weight vector of the user's area of ​​interest is set to W2=[0.3, 0.7], the number of iterations is set to 1000, and the service quality attribute values ​​(abbreviated as QOS in the figure) of the service combination solution set searched by the algorithm under different problem scales are recorded. The experimental results are shown in the figure. Figure 4 As shown in the figure, the horizontal axis represents the problem scale, the vertical axis represents the service quality attribute value of the service combination solution set, and Cost, Time, and Reliability represent the cost, response time, and reliability of the service combination solution set respectively.

[0155] from Figure 4 It can be seen that the service quality attribute values ​​of the service combination solution set searched by PA-DMOSCO fluctuate within the user preference range. When the problem scale increases from 30*100 to 50*200, the cost and time attributes of the solution set decrease. This is because when the problem scale increases, the number of service combination solutions also increases, which can provide more and better choices for the algorithm. Therefore, the PA-DMOSCO algorithm has a certain adaptability to solving the preference-aware dynamic multi-objective service combination problem.

[0156] (3) Performance verification of PA-DMOSCO under different preference information

[0157] In order to verify the adaptability of the proposed method to different user preferences, this paper sets three different user interest area scenarios to represent different user preferences. The preference weight vector of the first user interest area is set to W1 = [0, 1, 0, 4], the preference weight vector of the second user interest area is set to W2 = [0, 3, 0, 7], and the preference weight vector of the third user interest area is set to W3 = [0, 6, 1, 0]. In addition, the number of iterations is set to 1000, the number of subtasks is set to 10, the number of candidate services for subtasks is set to 50, and the service quality attribute values ​​of the service combination solution set searched by the algorithm under different preference information are recorded. The experimental results are shown in the figure. Figure 5As shown in the figure, the horizontal axis represents different user interest regions (abbreviated as ROI in the figure), the vertical axis represents the service quality attribute value of the service combination solution (abbreviated as QOS in the figure), the service quality attribute value of User's Preference represents the user's preferred service quality attribute value, and the service quality attribute value of PA-DMOSCO represents the service quality attribute value of the service combination solution set.

[0158] Figure 5 The results show that PA-DMOSCO is adaptable to different preference information. In three different user interest areas, the service quality attribute values ​​of the service combination solutions solved by PA-DMOSCO are close to the service quality attribute values ​​of user preferences. The experimental comparison results show that PA-DMOSCO is reasonable and effective in solving the preference-aware dynamic multi-objective service combination problem, which is conducive to guiding the population to evolve towards the user's interest area and providing users with more accurate service combination solutions.

[0159] 2. Performance comparison experiment

[0160] In order to verify the performance of the PA-DMOSCO algorithm, DACP, DAR, SLO and PA-DMOSCO algorithms are used to solve the same preference-aware dynamic multi-objective service composition optimization problem. The initialization method, objective function and constraints of the original SLO are modified to make it suitable for the preference-aware dynamic multi-objective service composition optimization problem. The number of subtasks of the four algorithms is set to 10, the number of candidate services corresponding to the subtasks is set to 50, the initial population size is set to 100, and the preference weight vector of the user's area of ​​interest is set to W2 = [0.3, 0.7]. The fitness values ​​of the service combination solution sets searched by the four algorithms at different iteration times are recorded. The experimental results are shown in Figure 2. Figure 6 shown.

[0161] from Figure 6 It can be seen that under the same environment, as the number of iterations increases, the convergence speed of the PA-DMOSCO algorithm is faster than that of the other algorithms, and some of the solutions searched by the PA-DMOSCO algorithm are better than those of the other two algorithms. In addition, the solution obtained by the PA-DMOSCO algorithm is far better than the SLO, and the convergence speed of the PA-DMOSCO algorithm is faster than that of the SLO algorithm, which proves that the improved SLO based on preference level domination is effective. Based on the above experimental results, it can be concluded that the PA-DMOSCO algorithm can effectively solve the preference-aware dynamic multi-objective service composition optimization problem, and is highly competitive compared to other algorithms.

[0162] 3. Experimental verification of dynamic preference perception strategy

[0163] In order to verify the effectiveness of the dynamic preference-aware strategy, based on the PA-DMOSCO method, the dynamic preference-aware strategy is removed to obtain a variant of PA-DMOSCO, which is called PG-SLO. PG-SLO and PA-DMOSCO are used to solve the same preference-aware dynamic multi-objective service composition optimization problem, and the results are compared. In the experimental setting, the number of subtasks of the two algorithms is set to 10, the number of candidate services corresponding to the subtasks is set to 50, and the initial population size is set to 100. The algorithm is executed, and the service quality attribute values ​​of the service combination solution set searched by the two algorithms under different preference information are recorded. The experimental results are shown in the figure. Figure 7 As shown, the horizontal axis represents different user interest areas, and the vertical axis represents the service quality attribute value of the solution set.

[0164] from Figure 7 It can be seen that in the same preference-aware service combination environment, the service combination solutions solved by the PA-DMOSCO algorithm are all within the preference region, while the PG-SLO algorithm only pursues the service combination solution with the optimal service quality attribute value. Therefore, the experimental comparison results verify the effectiveness of the dynamic preference-aware strategy proposed in this paper.

[0165] 4. Experimental verification of the dominance relationship of preference levels

[0166] In order to verify the effectiveness of the proposed preference level dominance relationship, the Pareto dominance relationship is used to replace the preference level dominance relationship in PA-DMOSCO. This variant is called PA-DMOSCO-PG, and PA-DMOSCO is compared with PA-DMOSCO-PG. In the experimental setting, the number of subtasks of the two algorithms is set to 10, the number of candidate services corresponding to the subtasks is set to 50, the initial population size is set to 100, and the preference weight vector of the user's area of ​​interest is set to W2 = [0.3, 0.7]. The fitness values ​​of the service combination solution sets searched by the two algorithms under different degrees of environmental change are recorded. The experimental results are shown in Figure 2. Figure 8 shown.

[0167] from Figure 8 It can be seen that PA-DMOSCO has a faster convergence speed than PA-DMOSCO-PG, and the solution obtained is also better than the PA-DMOSCO-PG algorithm. The experimental comparison results show that the preference level dominance relationship is conducive to accelerating the convergence of the population and obtaining a better service combination solution set. Based on the above experimental results, the effectiveness of the preference level dominance relationship can be verified.

Claims

1. A dynamic multi-objective service composition optimization method based on preference perception, characterized in that: The following steps are involved: S1. Based on the proximity between individuals and the user's area of ​​interest, the initial population is divided into sub-populations with different preference levels, specifically: S1.1 obtains the boundary points of the subpopulation with a preference level of 1 based on the minimum preference value and the maximum preference value of each target provided by the user. The area where the subpopulation with a preference level of 1 is located is the area of ​​interest of the user; S1.2 Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas of the subpopulations with different preference levels other than preference level 1 are obtained; S2, environmental change detection: detect changes in service resource quality attributes and user goals between the current moment and the previous moment. If a change is detected, step S3 is executed. If no change is detected, step S4 is executed; S3, environmental change response: change the attribute values ​​and constraints of service resource quality; S4, detect whether the user preference has changed, If there is a change, execute step S5. If no change occurs, execute step S6; S5. Adjust preference level: Based on the location of each individual, the individual is divided into different preference areas and the preference level is re-determined by comparing the location with the boundary values ​​of different preference areas on different dimensional targets; S6. Based on the preference levels, obtain the preference level dominance relationship between individuals; The preference level dominance relationship is introduced into the social learning optimization algorithm. Based on the preference level dominance relationship, the observational habitual weight and imitation learning weight are obtained to guide the population to converge to the user's area of ​​interest and retain the individuals with the highest fitness value that meets the user's preference. S7. If the iteration number threshold has been reached, the loop ends, and the individual with the highest fitness value is output as the optimal individual to obtain a service combination solution that meets the user's preferences; Otherwise, the individual with the highest fitness value is taken as the optimal individual and step S2 is executed.

2. The preference-aware dynamic multi-objective service composition optimization method according to claim 1 is characterized in that: In S1, based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained, and the area where the sub-population with a preference level of 1 is located is the user's area of ​​interest, as follows: Among them, min i is the minimum value of the i-th dimension target, d is the value range of the i-th dimension target, W i,min is the user’s minimum preference value for the i-th dimension target, W i,max is the user’s maximum preference value for the i-th dimension target, is the minimum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, is the maximum boundary value of the target of the i-th dimension in the region where the subpopulation with preference level 1 is located.

3. The preference-aware dynamic multi-objective service composition optimization method according to claim 1 is characterized in that: In S1, based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas where the subpopulations with different preference levels other than the preference level of 1 are located are obtained, as follows: in, is the minimum boundary value of the target of the i-th dimension in the region where the subpopulation with preference level other than 1 is located, is the maximum boundary value of the target of the i-th dimension in the region where the subpopulation other than the preference level 1 is located, is the minimum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, is the maximum boundary value of the target of the ith dimension in the region where the subpopulation with preference level 1 is located, R j,i Based on the user's area of ​​interest, the area where the subpopulation with preference level j is located is wider than the user's area of ​​interest on the i-th dimension target.

4. The preference-aware dynamic multi-objective service composition optimization method according to claim 1 is characterized in that: In S5, based on the location of each individual, the individual is divided into different preference areas by comparing the location with the boundary values ​​of different preference areas on different dimensional targets, specifically: Among them, PR is the user's preferred region, is the total dimension, m is the target dimension; x i is the objective function value of individual x on the i-th dimension target, l i is the next value of the preferred region, u i is the previous value of the preferred area.

5. The preference-aware dynamic multi-objective service composition optimization method according to claim 1 is characterized in that: The preference level dominance relationship is determined in S5, specifically: If individual x Pareto dominates individual y, or: Individual x and individual y do not dominate each other, and individual g x <g y , The level of the subpopulation where x belongs is less than the level of the subpopulation where individual y belongs. It is assumed that the preference level of individual x dominates that of individual y, where g x is the preference level of individual x, g y is the preference level of individual y.

6. The preference-aware dynamic multi-objective service composition optimization method according to claim 1 is characterized in that: In S6, the preference level dominance relationship is introduced into the social learning optimization algorithm to guide the population to converge to the user's area of ​​interest and retain the individual with the highest fitness value that meets the user's preference, specifically: In the microspace, based on the set crossover rate and mutation rate, the individuals in the population are sequentially subjected to selection, crossover and mutation operations; In the learning space, the learning habit weights and disturbance learning factors based on the preference level dominance relationship are introduced to perform observation learning operations; the imitation learning weights based on the preference level dominance relationship are introduced to perform imitation learning operations; In the belief space, knowledge updating operations and cultural influence operations are performed to retain individuals with the highest fitness values ​​that meet user preferences.

7. The preference-aware dynamic multi-objective service composition optimization method according to claim 6 is characterized in that: In the learning space, the learning habit weight and the disturbance learning factor based on the preference level dominance relationship are introduced to perform observation learning operations, specifically: in, is the service number used to perform the kth subtask in the i-th service combination scheme after the observation learning operation; x ik is the service number used to execute the kth subtask in the i-th service combination scheme; R (x,y) is the learning habit weight based on the preference level dominance relationship, To learn the perturbation factor, is the service number of the optimal individual j for executing the kth subtask; g x is the preference level of individual x, g y is the preference level of individual y, g max is the maximum preference level, g min is the minimum preference level.

8. The preference-aware dynamic multi-objective service composition optimization method according to claim 6 is characterized in that: The imitation learning weights based on the preference level dominance relationship are introduced to perform the imitation learning operation, specifically: Among them, r1 and r2 are two different individuals selected randomly, r1,r2∈{1,...,N}, N is the population size; S∈[0,1] is the scaling factor, is the imitation learning weight based on the preference level dominance relationship, [] is the rounding operation, is the service number used to perform the kth subtask in the i-th service combination scheme obtained after the imitation learning operation is completed, x ik is the service number used to execute the kth subtask in the i-th service combination scheme, is the service number of the optimal individual j for executing the kth subtask, represents the service number used to execute the kth subtask in any individual r1, Represents the service number used to execute the kth subtask in any individual r2.

9. A dynamic multi-objective service composition optimization system based on preference perception, characterized in that: include: Initial population division module: Based on the proximity between individuals and the user's area of ​​interest, the initial population is divided into sub-populations with different preference levels, specifically: Based on the minimum and maximum preference values ​​of each target provided by the user, the boundary points of the sub-population with a preference level of 1 are obtained. The area where the sub-population with a preference level of 1 is located is the user's area of ​​interest. Based on the boundary points of the subpopulation with a preference level of 1 and the extended criteria provided by the user, the boundary points of the preference areas where the subpopulations with different preference levels other than preference level 1 are located are obtained; Environment change detection module: performs change detection on the service resource quality attributes and user goals at the current moment and the previous moment. If a change is detected, the environment change response module is executed. If no change is detected, the user preference change detection module is executed. Environmental change response module: changes the attribute values ​​and constraints of service resource quality; User preference change detection module: detects whether the user preference has changed. If so, the preference level adjustment module is executed. If not, the social learning optimization algorithm optimization module is executed. Preference level adjustment module: Based on the location of each individual, it is compared with the boundary values ​​of different preference areas on different dimensional targets, and the individual is divided into different preference areas to redefine the preference level; Social learning optimization algorithm optimization module: Based on the preference level, obtain the preference level dominance relationship between individuals; introduce the preference level dominance relationship into the social learning optimization algorithm, obtain the observational habitual weight and imitation learning weight based on the preference level dominance relationship, guide the population to converge to the user's area of ​​interest, and retain the individual with the highest fitness value that meets the user's preference; Output module: If the iteration number threshold is reached, the loop ends and the individual with the highest fitness value is output as the optimal individual to obtain a service combination solution that meets the user's preferences; Otherwise, the individual with the highest fitness value is taken as the optimal individual and the environmental change detection module is executed.