Efficient Nondomination Level Update Method for Steady-State Evolutionary Multiobjective Optimization.

Area of Science:

• Evolutionary Computation
• Multiobjective Optimization
• Algorithm Design

Background:

• Nondominated sorting (NDS) is crucial for evolutionary multiobjective optimization (EMO) algorithms, dividing populations into nondomination levels (NDLs).
• Traditional NDS methods are computationally expensive in steady-state evolution models, especially with large populations and many objectives.
• Existing NDS approaches often require recalculating from scratch, leading to inefficiency.

Purpose of the Study:

• To propose an efficient Nondominated Level (NDL) update method for steady-state evolutionary multiobjective optimization (EMO).
• To reduce the computational cost associated with maintaining the NDL structure in EMO algorithms.
• To offer an alternative NDS approach beneficial for both steady-state and generational evolution models.

Main Methods:

• Developed an incremental NDL update method that leverages existing NDL structure knowledge.
• The method updates NDLs for a limited number of solutions rather than recomputing the entire structure.
• The update procedure is applied twice per iteration: post-reproduction and post-environmental selection.

Main Results:

• The proposed method significantly reduces unnecessary comparisons compared to five state-of-the-art NDS methods.
• Experimental results demonstrate efficiency gains on both synthetic datasets and real-world optimization problems.
• The approach proved effective in reducing computational overhead in steady-state EMO.

Conclusions:

• The proposed efficient NDL update method offers a substantial improvement for steady-state EMO algorithms.
• This method effectively minimizes computational cost without compromising the integrity of nondomination levels.
• The developed technique is also applicable and beneficial to generational EMO models, enhancing its versatility.