簇中心初始选择策略与更新异权机制相耦合的MDBA算法

MDBA algorithm coupled with the initial selection strategy of the cluster center and the updated weight mechanism

在聚类任务中,初始簇中心的选取和更新方式影响聚类结果的准确性.针对现有DBA算法初始簇中心选择的不确定性、簇中心更新序列的差异性以及算法复杂度高、收敛性差等问题,提出了一种融合簇中心初始选择策略与更新异权机制的MDBA算法. MDBA算法针对DBA算法中初始簇中心选取的不确定性问题,通过选取数据集中惯性最小的时间序列作为初始簇中心以消除其随机性;同时,利用更新异权机制更新簇中心以改善DBA算法中簇中心更新时数据集中序列存在差异性问题.数值实验结果表明,相比于原算法,簇中心初始选择策略迭代的最终惯性值接近多次随机的惯性均值;簇中心更新异权机制能够有效提高算法惯性收敛性,减少算法迭代次数,降低算法复杂度;MDBA算法降低原算法复杂度的同时提高簇中心的质量.

In clustering tasks, the selection and the updating of the initial cluster center affect the accuracy of clustering results. In view of the uncertainty of the selection of the initial cluster center of the existing DTW barycenter averaging(DBA) algorithm, the difference between the cluster center update sequence and the high complexity and poor convergence of the algorithm, a merging DTW barycenter averaging(MDBA) algorithm is proposed to fuse the initial cluster center selection strategy and the cluster center update weight mechanism. Aiming at the uncertainty of the initial cluster center selection in the DBA algorithm, the MDBA algorithm selects the time series with the least inertia as the initial cluster center to eliminate its randomness. At the same time, a new weight mechanism is used to update the cluster center to improve the differences in the sequence of the data set in the DBA algorithm when the cluster center is updated. The numerical results show that compared with DBA algorithm, the final inertial value of the initial cluster center selection strategy iteration is close to the random mean of the initial cluster center. Cluster center weight mechanism can improve algorithm convergence, reduce algorithm iteration times, and thus reduce algorithm complexity. The MDBA algorithm reduces algorithm complexity and improves cluster center quality.