The two steps training method is proposed to replace the iterative way of local principal component analysis
algorithm in the regularity model-based multiobjective estimation of distribution algorithm (RM-MEDA). In the phase of
training model of improvement algorithm, the ??-means clustering method is used to partition the points in population into
primary ?? disjoint clusters at the first step, and then the clustering method based manifold is used to partition at the second
step. Simulation results of the six benchmark instances show that the improvement algorithm can maintain the convergence
and diversity performance, and decline the computation time greatly.