融合优化可调Q因子小波变换的改进密度峰值聚类算法

为提升时间序列的聚类精度，提出一种融合优化可调Q因子小波变换的改进密度峰值聚类（improved density peaks clustering based on optimal tunable Q-factor wavelet transform，OTQWT-IDPC）算法，该算法利用可调Q因子小波变换的能量优化选择策略及改进粒子群优化算法确定的最佳Q因子分解时序信号，通过最优特征子带的能量、均值、标准差和模糊熵构建特征子空间，并采用主成分分析降低特征维度，以减少特征冗余。同时，考虑到距离较远而周围密集程度较大的K近邻样本对局部密度的贡献率，引入权重系数及K近邻重新定义DPC的局部密度，并利用共享最近邻描述样本间的相似性。在BONN癫痫脑电信号和CWRU滚动轴承数据集上进行对比实验，结果表明，该算法的聚类精度分别为95%、94%，且Jacarrd、FMI和F1值指标均优于其他对比算法，证明了OTQWT-IDPC算法的有效性。

Improved density peaks clustering algorithm combining optimal tunable Q-factor wavelet transform

In order to improve the accuracy of time series clustering, this paper proposed an improved density peak clustering algorithm based on optimal tunable Q-factor wavelet transform. The algorithm used energy optimization strategy of adjustable Q-factor wavelet transform and the optimal Q-factor determined by improved particle swarm optimization algorithm to decompose the time series signal. Through the energy, mean, standard deviation and fuzzy entropy of the optimal characteristic sub-bands, the algorithm constructed the characteristic subspace. And it used principal component analysis to reduce feature dimensions to debase feature redundancy. At the same time, considering the contribution rates of K-nearest neighbor samples that are far away and have a higher degree of surrounding density to local density, it introduced weight coefficients and K-nearest neighbors to redefine the local density of DPC, and used shared nearest neighbors to describe the similarity between samples. OTQWT-IDPC algorithm was compared with its comparison algorithms using BONN epileptic EEG and CWRU rolling bearings datasets. The experimental results show that the accuracy of OTQWT-IDPC algorithm on BONN and CWRU are 94% and 92%. Its evaluation indicators such as Jacarrd, FMI and F1 are superior to other comparative algorithms, which proves the effectiveness.