基于聚类和多尺度优化的超球体核距离评估的航空发动机性能衰退

Aeroengine Performance Deterioration Evaluation Using Clustering and Multi-Scaling Optimal Hyper Sphere Kernel Distance Assessment

关键词:  聚类  多尺度参数优化  超球体核距离  性能衰退  改进粒子群算法 

Key words:  Cluster  Multi-scaling parameter optimization  Hyper sphere kernel distance  Performance deterioration  Improved particle swarm optimization

摘要:

针对发动机性能评估参数存在多重共线性且数量过多的问题,提出一种依据类间方差和距离判别的聚类方法。将相似个体化为一类,并取类中均值作为分析对象,大大减少了参数维数；在支持向量数据描述(Support Vector Data Description)算法基础上,引入超球体核距离度量,将多参数转化为单参数,解决了参数过多相互矛盾的问题。特征空间上一点与超球体中心的距离表征发动机的性能衰退程度,并给出了性能开始衰退与性能明显恶化的阀值曲线。考虑聚类后类中参数对发动机性能评估的贡献不同,提出基于改进粒子群算法优化多尺度核函数参数和惩罚因子C。仿真结果表明:考虑了多尺度参数后,发动机性能状况较单尺度参数能更好的符合实际使用情况。聚类后多尺度参数与原参数的评估结果基本一致。 

Abstract:

Aiming at repeated linearity and excessiveness of engine performance evaluating parameter, a cluster method according to variation and distance judgement was presented. Similar individuals were grouped, mean of every group was adopted as analysis objective, which reduced dimension greatly. On the basis of Support Vector Data Description, hyper sphere kernel distance metric was introduced while multi-parameter was converted to single parameter, and contradiction leaded by parameter overfull was settled. Distance between a point in the character space and core of hyper sphere denotes performance deterioration. Valve curve of beginning deterioration and worsening was obtained. Considering contribution of parameter after clustering to performance evaluation, multi-scaling kernel parameter and punish coefficient C are optimized by improved Particle Swarm Optimization algorithm. Results indicate that engine performance can be consistent to factual condition after considering multi-scaling parameter. Evaluating results of multi-scaling parameter after clustering are consistent with original parameter. 