Adaptive Feasible Region Estimation via Solution Space Partitioning for Constrained Optimization

Feasible regions in constrained optimization problems are often extremely small or fragmented, making feasible solutions difficult to locate and causing premature convergence within isolated feasible regions. To tackle these persistent challenges, this paper introducesadaptive feasible region estimation via solution space partitioning(AFRESP), which provides a new perspective by explicitly modeling the distribution of feasible regions. AFRESP first partitions the search space into subspaces and estimate the constraint violation of each subspace. These subspaces are then adaptively clustered into ε-feasible and ε-infeasible regions. An adaptive tree-search framework intelligently allocates computational resources to balance global exploration across the search space with targeted local exploitation. Comprehensive evaluations on IEEE CEC2006, CEC2010, CEC2017 benchmarks and a set of engineering applications demonstrate that AFRESP consistently delivers competitive or statistically superior results. Moreover, examining the constraint-violation landscapes of representative test problems provides deeper insight into the underlying sources of problem difficulty.