DDFS method optimizes large-scale problems with differentiated fuzzy search
New algorithm handles 1,000 variables by assigning tailored search granularities.
A new paper from Boxi Xiao and colleagues introduces DDFS, a Decision variable analysis-guided Differentiated Fuzzy Search method designed to tackle large-scale multi-objective optimization problems (LSMOPs). These problems involve hundreds or thousands of decision variables, making traditional evolutionary algorithms slow and prone to premature convergence. The key innovation is recognizing that not all variables contribute equally: some drive convergence toward the Pareto front, while others maintain population diversity. Existing fuzzy search methods apply a uniform granularity to all variables, wasting computational resources on less critical dimensions.
DDFS first analyzes each variable's role and search sensitivity, then assigns a differentiated fuzzy search granularity to each group. This means high-impact convergence variables get fine-grained searches, while diversity-related variables use coarser exploration. Additionally, DDFS incorporates a Dual-Indicator Stage Transition Mechanism that adjusts the fuzzy-updating intensity over time: initially compressing the search space for faster convergence, then refining precision in later stages. Experiments on LSMOP and UF benchmarks with up to 1,000 variables show DDFS consistently achieves competitive or superior performance compared to state-of-the-art large-scale multi-objective evolutionary algorithms.
- Maps decision variable roles (convergence vs. diversity) to differentiated fuzzy search granularities, avoiding uniform treatment.
- Dual-Indicator Stage Transition Mechanism dynamically adjusts fuzzy-updating intensity across evolutionary stages.
- Validated on up to 1,000 decision variables in LSMOP and UF benchmarks, outperforming existing large-scale MOEAs.
Why It Matters
Enables faster, more accurate optimization for high-dimensional real-world problems like engineering design and logistics.