A Multi-level Classification Assisted Evolutionary Algorithm Based on Distribution Prediction

In decomposition-based multi-objective evolutionary algorithms (MOEAs), a set of uniformly distributed reference vectors is employed to decompose a multi-objective problem into different subproblems, and solutions are selected based on various subproblem evaluation functions. However, during the solution selection process, there is a tendency for individuals to be assigned to subproblems associated with reference vectors that are relatively distant. This can lead to the exclusion of individuals with better convergence properties if they are far from the reference vectors, resulting in wasted solutions and reduced convergence speed of the population.To address this issue and retain individuals with better convergence properties while using a classification surrogate to select higher-quality candidates for actual evaluation, this paper introduces an algorithm that incorporates a distribution prediction-based multi-layer classification surrogate model. This model effectively resolves the problem of improper individual solution allocation in traditional decomposition-based MOEAs. By optimizing classification and regression models, the proposed algorithm demonstrates improved convergence and solution quality in multi-objective optimization problems. Experimental results indicate that our proposed algorithm outperforms comparative algorithms on most of the test problems.