演化策略用于高维故障样本集最优统计聚类分析

针对液体火箭发动机推进系统超高维故障样本数据的聚类问题 ,提出基于演化策略的最优统计聚类算法。为预防算法过早收敛 ,演化策略采用了父本适应值的动态调整值与共享函数 ,并针对超高维数据聚类提出了控制参数的适应性调整技术 ;为使算法能最终跳出局部最优死区 ,提出算法的局部调整策略。该算法用于液体火箭发动机典型故障仿真数据集分析 ,并取得了最优聚类结果。此外 ,还基于IRIS数据集比较了该算法与FKCN模糊自主聚类算法。仿真分析表明了算法在高维数据聚类分析中的优点。

Optimal statistical clustering for high dimensional fault sample using evolution strategies

A clustering algorithm based on Evolution Strategies was proposed to make analysis on high dimensional data of liquid rocket engine propulsion system In order to prevent the solution population from premature convergence, the dynamic fitness adaptation technique and all sharing function were introduced An adaptive regulation scheme for evolution control parameters was specially presented for clustering analysis of high dimensional data A local clustering deadlock can also be overcome by the deadlock check and cluster recombination & collapse strategies This algorithm was used in the optimal clustering analysis for the 560 data samples of 14 sorts of common faults, each of 68 dimensions, which were simulated for a liquid rocket engine In addition, comparison with fuzzy Kohonen clustering networks (FKCN) has also been made,based on IRIS data The simulation results show that the evolution strategies based on clustering algorithm is superior to other non evolutionary clustering algorithms, particularly when the data is of high dimensions