Constraint multi-objective evolution method based on population image sharpening
By introducing population image sharpening technology and Sobel operator in the constrained multi-objective optimization algorithm, the existing algorithms have solved the problems of constraint ability, diversity and convergence balance and computational efficiency, and achieved efficient population reproduction and evolution information Mining improves algorithm performance and stability.
Patent Information
- Application Number
- CN202411075191.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the constraint multi-objective optimization problem, existing algorithms have challenges in dealing with constraint capabilities, diversity and convergence balance, computing efficiency and parameter sensitivity, and it is difficult to effectively use limited population data for evolutionary information learning.
The constrained multi-objective evolution method based on population image sharpening is adopted. By introducing the column-oriented Sobel operator and image sharpening PIS, it is embedded into the multi-objective evolution algorithm to reproduce populations and improve the generation efficiency of offspring populations.
It is realized that the evolution information of parent populations is efficiently mined in the constraint multi-objective optimization problem, and the high-quality offspring populations are generated, which significantly improves the performance and stability of the algorithm, especially in large-scale CMOPs.
Smart Images

Figure CN118940823A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of image processing, and in particular to a constrained multi-objective evolutionary method based on population image sharpening. Background Art
[0002] Constrained multi-objective optimization problems (CMOPs) refer to the optimization of multiple conflicting objectives, while the decision variables need to satisfy one or more equality / inequality constraints. In recent years, multi-objective evolutionary algorithms (CMOEAs) have become the mainstream approach to solve this problem. By continuously generating promising populations with various customized evolutionary strategies, a dynamic balance between the feasibility, convergence, and diversity of solutions is achieved in the objective space. However, learning evolutionary information from a limited amount of population data remains challenging.
[0003] These drawbacks mainly include the limitation of the ability to handle constraints, the difficulty of balancing diversity and convergence, computational efficiency, parameter sensitivity, and challenges in fitness evaluation. These drawbacks reflect the main obstacles that the algorithm may encounter in practical applications.
[0004] Therefore, it is necessary to provide a constrained multi-objective evolutionary method based on population image sharpening to solve the above problems. Summary of the invention
[0005] The purpose of the present invention is to provide a constrained multi-objective evolution method based on population image sharpening, which utilizes image sharpening technology to participate in population reproduction and efficiently generates offspring populations by mining detailed evolutionary information of parent populations.
[0006] To achieve the above object, the present invention provides a constrained multi-objective evolutionary method based on population image sharpening, using population image sharpening PIS for population propagation, introducing a column-oriented Sobel operator to participate in population propagation, and embedding population image sharpening PIS into CMOEAs to form an algorithm PIS-CMOEA. The execution steps of PIS-CMOEA are as follows:
[0007] S1: Input population size NP, problem dimension D, maximum population size T max and sharpening likelihood β;
[0008] S2: Assign the current evolutionary generation t = 1, P t is a two-dimensional image with NP × D elements;
[0009] S3: If the stop condition is not met, repeat S4 and S6 to reassemble P t Form V t ;
[0010] S4: Using simulated binary crossover to process population P t, generate the crossover population V t ;
[0011] S5: Execute operation U t =PIS(V t ,β),U t Mutation to form Q t , population R t =Q t ∪P t ;
[0012] S6: Use non-dominated sorting from R t The offspring population P is generated t+1 , update the current evolutionary generation by t=t+1;
[0013] S7: When the stopping condition is met, output the final population P Tmax , Tmax is the maximum evolutionary number. Preferably, in step S2, P t It is a two-dimensional image with NP×D elements. Each row represents a population individual and each column represents an independent variable. In order to sharpen the population image, the column-oriented Sobel operator is used:
[0014]
[0015] Where α∈[0,1] is the sharpening strength parameter, i∈[2,NP], j∈[1,D], NP and D are the population size and problem dimension respectively, α=rand, where rand generates a random floating point number between 0 and 1, and the population evolution should be restricted by a probability parameter β. The equation is executed only when rand<β.
[0016] Preferably, in step S3, the algorithm execution termination condition is set to that the error between the found optimal solution and the currently known optimal solution is less than 0.0001 or the number of function evaluations reaches 10,000 times.
[0017] Preferably, in step S4, the specific operation is, for P t The i-th individual P in the population i,t , randomly select two parent solutions x1 and x2, and use V i,t =0.5[(1+β)x1+(1-β)x2] and V i,t =0.5[(1-β)x1+(1+β)x2], where β is between 0.15 and 0.25.
[0018] Preferably, in step S5, the time complexity of PIS is related to NP, D, and β. Without Sobel operation, the time complexity of PIS is: O((NP-1)×D×β), where β is between 0.15 and 0.25.
[0019] Preferably, the specific steps in step S6 are:
[0020] S61: put all Rt individuals into the to-be-sorted set T, and find all non-dominated solutions in the current Rt;
[0021] S62: classify all non-dominated solutions in step S61 as the first-level frontier F1;
[0022] S63: Remove F1 from T, repeat the non-dominated sorting for the remaining individuals, and find the second-level frontier F2;
[0023] S64: Repeat steps S62-S63 until T is empty, and obtain the frontiers F1, F2, ..., Fn of all levels;
[0024] S65: Calculate the frontier crowding degree. For all individuals in the frontier surface, calculate the crowding distance. The crowding degree reflects the distribution density of individuals in the target space.
[0025] S66: Starting from the first frontier surface, individuals are selected. If the number of individuals on the frontier surface plus the number of selected individuals exceeds the population size, the individuals in the frontier surface are sorted according to the crowding degree, and the individuals with the largest crowding degree are selected until the preset population size is reached.
[0026] Therefore, the present invention adopts the above-mentioned constrained multi-objective evolution method based on population image sharpening, uses image sharpening technology to participate in population reproduction, and efficiently generates offspring populations by mining detailed evolutionary information of parent populations.
[0027] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is a flow chart of a constrained multi-objective evolutionary method based on population image sharpening of the present invention. DETAILED DESCRIPTION
[0029] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0030] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.
[0031] The words "include" or "comprises" and the like used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms "inside", "outside", "upper", "lower", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly specified and limited, the terms "attachment" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0032] like Figure 1 As shown, the present invention provides a constrained multi-objective evolutionary method based on population image sharpening, uses population image sharpening PIS for population propagation, introduces a column-oriented Sobel operator to participate in population propagation, embeds population image sharpening PIS into CMOEAs, and constructs the algorithm PIS-CMOEA. The execution steps of PIS-CMOEA are as follows:
[0033] S1: Input population size NP, problem dimension D, maximum population size T max and sharpening likelihood β;
[0034] S2: Assign time step t=1, P t is a two-dimensional image with NP×D elements; in step S2, P t It is a two-dimensional image with NP×D elements. Each row represents a population individual and each column represents an independent variable. In order to sharpen the population image, the column-oriented Sobel operator is used:
[0035]
[0036] Where α∈[0,1] is the sharpening strength parameter, i∈[2,NP], j∈[1,D], NP and D are the population size and problem dimension respectively, α=rand, where rand generates a random floating point number between 0 and 1, and the population evolution should be restricted by a probability parameter β. The equation is executed only when rand<β.
[0037] S3: If the stop condition is not met, repeat S4 and S6 to reassemble P t Form V t; In step S3, the algorithm execution termination condition is set to that the error between the found optimal solution and the currently known optimal solution is less than 0.0001 or the number of function evaluations reaches 10,000 times.
[0038] S4: Using simulated binary crossover to process population P t , generate the crossover population V t In step S4, the specific operation is, for P t The i-th individual P in the population i,t , randomly select two parent solutions x1 and x2, and use V i,t =0.5[(1+β)x1+(1-β)x2] and V i,t =0.5[(1-β)x1+(1+β)x2], where β is between 0.15 and 0.25.
[0039] S5: Execute operation U t =PIS(V t ,β),U t Mutation to form Q t , population R t =Q t ∪P t ; In step S5, the time complexity of PIS is related to NP, D, and β. Without Sobel operation, the time complexity of PIS is: O((NP-1)×D×β), where β is between 0.15 and 0.25.
[0040] S6: Use non-dominated sorting from R t The offspring population P is generated t+1 , update the current evolutionary generation by t=t+1;
[0041] The specific steps in step S6 are:
[0042] S61: put all Rt individuals into the to-be-sorted set T, and find all non-dominated solutions in the current Rt;
[0043] S62: classify all non-dominated solutions in step S61 as the first-level frontier F1;
[0044] S63: Remove F1 from T, repeat the non-dominated sorting for the remaining individuals, and find the second-level frontier F2;
[0045] S64: Repeat steps S62-S63 until T is empty, and obtain the frontiers F1, F2, ..., Fn of all levels;
[0046] S65: Calculate the frontier crowding degree. For all individuals in the frontier surface, calculate the crowding distance. The crowding degree reflects the distribution density of individuals in the target space.
[0047] S66: Starting from the first frontier surface, individuals are selected. If the number of individuals on the frontier surface plus the number of selected individuals exceeds the population size, the individuals in the frontier surface are sorted according to the crowding degree, and the individuals with the largest crowding degree are selected until the preset population size is reached.
[0048] S7: When the stopping condition is met, output the final population P Tmax , Tmax is the maximum evolutionary generation.
[0049] Comparative Example
[0050] Three experiments are conducted to evaluate the performance of this scheme;
[0051] Experiment 1: By comparing with five traditional constrained multi-objective optimization algorithms, the effectiveness of this scheme on the benchmark problem is verified; Experiment 2: By comparing with the most advanced algorithms, the superiority of this scheme on the benchmark problem is verified; Scheme 3: Study the impact of parameters on algorithm stability.
[0052] The test suites selected are MW and LIRCMOP. The test suites cover four common types of CMOPs:
[0053] 1) The unconstrained Pareto front (UPF) and CPF completely overlap;
[0054] 2) Partial UPF is feasible, and CPF is part of UPF;
[0055] 3) UPF is feasible in some areas, and CPF and UPF partially overlap;
[0056] 4) The UPF is located in an infeasible area and the CPF is completely separated from the UPF.
[0057] The results show that the selected test suite is representative and helpful for testing the proposed approach.
[0058] The number of optimization objectives and the dimension of decision variables are set as follows:
[0059] For LIR-CMOP1-LIR-CMOP12, M=2, D=15; for LIR-CMOP13 and LIR-CMOP14, M=3, D=15; for MW4, MW8, MW14, M=3, D=15; for other MW test instances, M=2, D=15.
[0060] In Experiment 2, about comparing algorithms and parameter settings;
[0061] The most advanced CMOEAs such as C3M[7], tDEACPBI[8], TSTI[9], AGEMOEAII
[10] are selected as the embedding objects of PIS. The variant of PIS is called "PIS-*". The effectiveness of the PIS algorithm is verified by comparing the performance of the "PIS-*" algorithm with the original algorithm. In order to illustrate the superiority of "PIS-*", this paper selects CMOCSO
[11] , CMOQLMT
[11] , CMOSMA
[12] , IMTCMO
[13] , cDPEA
[14] , MFOSPEA2
[15] , URCMO
[16] and other 7 state-of-the-art CMOEAs as comparison algorithms.
[0062] HV was selected as the performance indicator. In order to evaluate the significant differences in the performance indicators of the compared methods, the Wilcoxon Rank-sum test was used, and the significance level was set to 0.05. The symbols "+", "-", and "=" are used to indicate that the algorithm is significantly better, worse, or no different from the competitor.
[0063] The comparison of PIS-*” and embedded algorithms on HV indicators is shown in Table 1;
[0064] Table 1
[0065]
[0066] Regarding the effectiveness of the scheme, the comparison of “PIS-C3M” with the state-of-the-art CMOEAs on HV indicators is shown in Table 2. For all problems, each algorithm is run independently 30 times to obtain the HV performance indicators. Their means and standard deviations are listed in Table 1, where the best values are highlighted. The comparison shows that “PIS-*” significantly outperforms the embedded algorithms on up to 17 test problems and at least 14 test problems. Specifically, the advantages of PIS are mainly on MW4-6 and 11-14, while the performance on LIRCMOPS is different from that of the algorithms. On LIRCMOP9-10, PIS fails to improve the performance of C3M and tDEACPBI, but is effective for TSTI and AGEMOEAII. On LIRCMOP14, PIS does not successfully improve the performance indicators of TSTI and tDEACPBI, but is effective for C3M and AGEMOEAII. However, the effectiveness of PIS is limited by the performance of the embedded algorithms. On LIRCMOP5-6, PIS cannot find a feasible solution for the embedded algorithm, although it has significantly improved the performance of C3M on both problems. This is because the population is mainly generated by the crossover and mutation operators, while PIS is only a fine-tuning operator.
[0067] Table 2
[0068]
[0069] In Experiment III, the advancement of PIS is illustrated by comparison with state-of-the-art CMOEAs. C3M is used to embed PIS, and the new variant is called PIS-C3M. For all problems, each algorithm is run independently 30 times to obtain HV performance indicators. Their means and standard deviations are listed in Table 2, where the best values are highlighted. The results show that the PIS-C3M algorithm significantly outperforms the compared algorithms on a maximum of 24 and a minimum of 15 test problems. Overall, PIS-C3M has the best comprehensive performance. Specifically, the advantage of PIS is mainly reflected in MW3-7 and LIRCMOPS5-8, while the performance on other problems is different from the algorithm. On LIRCMOPS1-4, PIS-C3M has similar performance to the compared algorithms. On LIRCMOPS9-12, PIS-C3M shows disadvantages compared to URCMO.
[0070] The impact of β on the algorithm performance is shown in Table 3, specifically the performance of AGEMOEAII with different β values.
[0071] Considering the parameter β controlling the probability of performing overall image sharpening, AGEMOEAII variants β = 0.01, 0.05, 0.1, 0.15, 0.20, 0.25 and 0.3 are compared. For all problems, each algorithm is run 30 times independently to obtain the HV performance indicators. Their means and standard deviations are listed in Table III, where the best values are highlighted. Some interesting experimental phenomena can be summarized as follows. In the range of β = 0.01 to 0.05, the performance of AGEMOEAII has been greatly improved. In the range of β = 0.1 to 0.25, the performance of AGEMOEAII is relatively stable, and there is no significant difference between the AGEMOEAII variants. When β = 0.25 to 0.3, the performance of AGEMOEAII begins to decline. In summary, it is easy to use PIS, and the optimal interval of β is β∈[0.15,0.2].
[0072] Table 3
[0073]
[0074] Example
[0075] Application of this scheme in the optimal scheduling problem of integrated coal mine energy system (ICMES)
[0076] Fifteen new scenarios with different cold, heat, electricity load, light and wind intensity data were constructed, named ICMES1~15. The optimization objectives are to minimize the economic cost and the carbon emission limit. For all ICMES problems, D=384 and the maximum number of function evaluations is 50,000. Each problem was run independently 30 times to obtain the mean and standard deviation of HV. PIS is embedded in NSGA-II, called PIS-NSGA-II, and compared with CMOCSO, CMOQLMT, CMOSMA, IMTCMO, IMTCMO, MFOSPEA2 and URCMO. The experimental results are shown in Table 4. It is obvious that PIS-NSGA-II significantly outperforms the other algorithms on all 15 instances, while CMOCSO, CMOSMA, IMTCMO, MFOSPEA2 and URCMO cannot find a feasible solution on any instance. Although CMOQLMT can find feasible solutions for ICMES2~15, its HV index is significantly worse than that of IMTCMO, and the performance of IMTCMO is also far from that of PIS-NSGA-II.
[0077] The experimental results prove the effectiveness of PIS in improving the performance of the original CMOEAs, and the progress of PIS is verified by comparing with existing CMOEAs; PIS only introduces one parameter and is easy to control; when β∈[0.15,0.2], PIS has the best performance and is stable within this parameter range. The application of PIS in the optimization of integrated mine energy system scheduling shows the scalability of PIS in a wide range of CMOPs. Compared with other algorithms that cannot find feasible solutions, this method can not only find feasible solutions, but also has significant advantages in the quality of solutions.
[0078] Therefore, the present invention adopts the above-mentioned constrained multi-objective evolution method based on population image sharpening, uses image sharpening technology to participate in population reproduction, and efficiently generates offspring populations by mining detailed evolutionary information of parent populations.
[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A constrained multi-objective evolutionary method based on population image sharpening, characterized by: The population image sharpening PIS is used for population propagation. A column-oriented Sobel operator is introduced to participate in population propagation. The population image sharpening PIS is embedded into CMOEAs to form the algorithm PIS-CMOEA. The execution steps of PIS-CMOEA are as follows: S1: Input population size NP, problem dimension D, maximum population size T max and sharpening likelihood β; S2: Assign the current evolutionary generation t = 1, P t is a two-dimensional image with NP × D elements; S3: If the stop condition is not met, repeat S4 and S6 to reassemble P t Form V t ; S4: Using simulated binary crossover to process population P t , generate the crossover population V t ; S5: Execute operation U t =PIS(V t ,β), Ut mutates to form Q t , population R t =Q t ∪P t ; S6: Use non-dominated sorting from R t The offspring population P is generated t+1 , update the current evolutionary generation by t=t+1; S7: When the stopping condition is met, output the final population P Tmax , Tmax is the maximum evolutionary generation.
2. The constrained multi-objective evolutionary method based on population image sharpening according to claim 1, characterized in that: In step S2, P t It is a two-dimensional image with NP×D elements. Each row represents a population individual and each column represents an independent variable. In order to sharpen the population image, the column-oriented Sobel operator is used: Where α∈[0,1] is the sharpening strength parameter, i∈[2,NP], j∈[1,D], NP and D are the population size and problem dimension respectively, α=rand, where rand generates a random floating point number between 0 and 1, and the population evolution should be restricted by a probability parameter β. The equation is executed only when rand<β.
3. The constrained multi-objective evolutionary method based on population image sharpening according to claim 1, characterized in that: In step S3, the algorithm execution termination condition is set as follows: the error between the found optimal solution and the currently known optimal solution is less than 0.0001 or the number of function evaluations reaches 10,000 times.
4. The constrained multi-objective evolutionary method based on population image sharpening according to claim 1, characterized in that: In step S4, the specific operation is as follows: t The i-th individual P in the population i,t , randomly select two parent solutions x1 and x2, and use V i,t =0.5[(1+β)x1+(1-β)x2] and V i,t =0.5[(1-β)x1+(1+β)x2], where β is between 0.15 and 0.
25.
5. The constrained multi-objective evolutionary method based on population image sharpening according to claim 1, characterized in that: In step S5, the time complexity of PIS is related to NP, D, and β. Without Sobel operation, the time complexity of PIS is: O((NP-1)×D×β), where β is between 0.15 and 0.
25.
6. The constrained multi-objective evolutionary method based on population image sharpening according to claim 1, characterized in that: The specific steps in step S6 are: S61: put all Rt individuals into the to-be-sorted set T, and find all non-dominated solutions in the current Rt; S62: classify all non-dominated solutions in step S61 as the first-level frontier F1; S63: Remove F1 from T, repeat the non-dominated sorting for the remaining individuals, and find the second-level frontier F2; S64: Repeat steps S62-S63 until T is empty, and obtain the frontiers F1, F2, ..., Fn of all levels; S65: Calculate the frontier crowding degree. For all individuals in the frontier surface, calculate the crowding distance. The crowding degree reflects the distribution density of individuals in the target space. S66: Starting from the first frontier surface, individuals are selected. If the number of individuals on the frontier surface plus the number of selected individuals exceeds the population size, the individuals in the frontier surface are sorted according to the crowding degree, and the individuals with the largest crowding degree are selected until the preset population size is reached.
Citation Information
Patent Citations
Cascade hydropower station multi-objective optimization scheduling method based on improved NSGA-III
CN108805434A
Constraint multi-objective optimization method based on adaptive Epsilon constraint processing
CN115660183A
