Constrained multi-objective evolutionary optimization method with a deep-embedded fitness evaluation allocation strategy
Many real-world problems can be framed as constrained multi-objective optimization problems.While various existing methods address these issues,efficiently allocating fitness evaluation resources across the global search space while balancing solution feasibility,convergence,and diversity remains a challenge.To tackle this,we propose a novel constrained multi-objective evolutionary optimization method that incorporates a deep-embedded fitness evaluation allocation strategy.It identifies key regions in the search space and guides population evolution efficiently.The method employs a denoising autoencoder to create a dimensionality reduction model for the evolutionary population,revealing its low-dimensional manifold.By clustering the reduced-dimension population and analyzing constraint violation variances within each cluster,the algorithm can sense global and local search ranges for individual population members.The denoising autoencoder then maps these insights back to the original search space,enabling the precise allocation of fitness evaluation resources.This versatile method can be integrated into existing evolutionary algorithms,enhancing their performance to varying degrees.We validated our approach on 33 benchmark test problems and 15 dispatch optimization scenarios of the integrated mine energy system.The results demonstrate the effectiveness of our proposed method in solving constrained multi-objective optimization problems across diverse applications.
constrained multi-objective optimizationfitness evaluation allocationdeep embeddingevolutionary optimizationintegrated mine energy system
NETL
NSTL
万方数据
