Accelerating the Decomposition-Based Multiobjective Evolutionary Algorithm via Evolutionary Direction Vector

In the process of multi-objective optimization, a large amount of data is generated, which contains the problem-related information. Learning knowledge from this data and applying it to the evolutionary process can improve the quality of the offspring solutions. Most existing knowledge learning methods suffer from high algorithm complexity, and there are still few methods designed specifically for the decomposition-based MOEAs. This paper proposes an evolutionary direction vector (EDV)-guided multiobjective evolutionary algorithm based on decomposition (MOEA/D-EDV). It collects the high-quality evolutionary process data based on the solution's performance in the objective space and their movement directions in the decision space. It then extracts the problem-related knowledge using an efficient learning method to accelerate the evolutionary process. Experimental results on 30 benchmark problems demonstrate that the EDV-guided evolution can significantly accelerate the evolutionary process, and the overall performance of the proposed MOEA/D-EDV is significantly better than the comparative algorithms in the literature.