基于快速模拟退火的组合聚类算法

应用模拟退火算法解决组合聚类问题有两方面，一是有效利用基础聚类作为先验信息，以获得尽可能好的组合聚类结果；二是降低模拟退火过程的随机性，提高算法收敛速度。针对这2个问题，提出了基于投票的快速模拟退火（BV-RSA）模型。该模型利用基础聚类对样本划分的完全或部分一致性作为启发信息，构建超点集合和超点投票箱，由超点取代其代表的样本子集参与退火过程，超点运动方向在投票箱范围内随机选择，降低了超点运动随机性，加速了组合聚类过程。数据集实验表明，BV-RSA模型在聚类精度和鲁棒性方面表现良好。 

Ensemble clustering algorithm based on rapid simulated annealing

There are two key issues in applying simulated annealing algorithm to solve the problem of ensemble clustering. One is how to use basic partition information in annealing process to obtain better result, and the other is how to accelerate the algorithm convergence. In this paper, the rapid simulated annealing based on voting (BV-RSA) model is presented, in which the complete and partial consensuses of basic partitions are used to recognize super-objects and construct voting box for each super-object. In the process of simulated annealing, some data samples represented by a super-object are controlled to move in a group, and the motion direction of a super-object is selected randomly in the scope of its voting box, thus reducing moving randomness and speeding up the clustering of super-objects. Experiments on multiple data sets demonstrate that the BV-RSA model performs well in both clustering accuracy and robustness.