Multi-stage Constrained Multi-objective Optimization Algorithm Improved By Knowledge-transfer Mechanism And Its Application

Constrained Multi-objective Optimization Problems(CMOP)are commonly encountered in social life.Constrained Multi-objective Evolutionary Algorithms(CMOEAs)have been widely applied to solve CMOPs and yield numerous successes.Multi-stage Constrained Multi-objective Algorithm(Multi-stage CMOEA)is one of the popular research directions in CMOEA,which divides the optimization process into multiple stages that differ from each other but work collaboratively.In a revolutionary approach known as MSCMO(Multi-stage Constrained Multi-objective Optimization),the optimization process is split into two stages: the unconstrained search stage and the constrained search stage.The constrained search stage is further divided into many sub-stages based on the priority of the constraints,which is decided in the unconstrained search stage based on the complexity of handling constraints.The algorithm then progressively applies constraint handling in these sub-stages.MSCMO has certain drawbacks.(1)Search direction of MSCMO is relatively single,and the population is prone to local optima.(2)When the population is reinitialized,the data gathered in MSCMO’s unconstrained search stage is lost.(3)At the final stages of optimization,the search engine used in MSCMO is unable to successfully balance diversity and convergence.(4)The stage transition determination strategy of MSCMO fails in ignorance of extreme solutions.This thesis first embeds Knowledge-transfer mechanism into MSCMO to obtain MSCMO/KT,analyzes the role of Knowledge-transfer mechanism in stage transition timing,calculates the contribution of each knowledge source to the stages,and thus verifies the effectiveness of Knowledge-transfer mechanism.Next,to address the problem of the diversity and convergence of Helper Task’s population affecting the quality of Main Task’s population,an improved Differential Evolution with a Mixed-selection mechanism is proposed as a search engine for Helper Task,and the effectiveness of MSCMO-DE/KT is compared through experiments.Lastly,a stage transition determination strategy based on crowding distance is suggested for C-MSCMO-DE/KT in order to solve the weakness of MSCMO in using the objective function change rate as the determination basis for stage transition.The strategy first performs non-dominated sorting on the population,and then for each layer of the population,includes the edge individuals and excludes individuals whose crowding distance less than the average crowding distance.In the experimental part,C-MSCMO-DE/KT is compared with MSCMO and various current mainstream high-performance algorithms on multiple test suites,and analyzed based on multiple indicators.The experimental findings show that C-MSCMO-DE/KT performs exceptionally well on the majority of test problems.Also,the parameter design of front rail of vehicle is optimized using the C-MSCMO-DE/KT.The suggested algorithm successfully increase the energy absorption capacity of the collision component and decrease the impact force of the vehicle collision on passengers,as shown by optimized design schemes by C-MSCMODE/KT.