MOEA/HD: A Multiobjective Evolutionary Algorithm Based on Hierarchical Decomposition.

Area of Science:

• Computational intelligence and optimization algorithms.
• Evolutionary computation for complex problem-solving.

Background:

• Multiobjective optimization problems (MOPs) are frequently addressed using multiobjective evolutionary algorithms (MOEAs).
• The widely adopted MOEA/D method decomposes MOPs using fixed weight vectors, which limits performance on ill-conditioned MOPs with complex Pareto fronts.

Purpose of the Study:

• To introduce a novel MOEA based on hierarchical decomposition (MOEA/HD).
• To address the limitations of fixed decomposition strategies in MOEA/D for challenging MOPs.

Main Methods:

• MOEA/HD employs a hierarchical decomposition strategy, layering subproblems.
• It adaptively adjusts search directions of lower-hierarchy subproblems based on higher-hierarchy results.
• The proposed algorithm was empirically compared against four state-of-the-art MOEAs using standard performance metrics.

Main Results:

• MOEA/HD demonstrated promising performance across all tested MOPs, including those with complex Pareto front characteristics.
• The adaptive search direction adjustment proved effective in handling difficulties faced by traditional MOEA/D on ill MOPs.

Conclusions:

• MOEA/HD offers a robust advancement over existing MOEA/D methods, particularly for MOPs with intricate solution spaces.
• The hierarchical decomposition and adaptive search strategy provide superior performance and broader applicability in multiobjective optimization.