提出两步训练法,改进了基于规则模型的多目标分布估计算法.在算法的模型训练环节,首先采用均值分簇法进行初步聚类;然后采用基于流形分簇法进行细致聚类,代替原算法中采用局部主元分析算法需要循环迭代的聚类分簇方法.通过6个Benchmark测试函数验证,改进算法保持了原算法的收敛性和多样性,并缩短了寻优的时间. This paper proposes a two-step training method to improve the multi-objective distribution estimation algorithm based on the rule model.At first, the mean clustering method is used to carry out the initial clustering in the model training part of the algorithm, and then the clustering based on the manifold clustering method is adopted, The original principal component analysis (PCA) algorithm needs iterative clustering and clustering method, which is verified by six Benchmark test functions. The improved algorithm preserves the convergence and diversity of the original algorithm and shortens the optimization time.