论文部分内容阅读
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。
Aiming at the problem of multiple collinearity and excessive quantity of engine performance evaluation parameters, a clustering method based on inter-class variance and distance discrimination is proposed. Based on the support vector data description (Support Vector Data Description) algorithm, the hyper-sphere nuclear distance measure is introduced, and the multi-parameter transformation As a single parameter, to solve the problem of too many parameters of mutual contradictions. The distance from the hypersphere center in the feature space characterizes the degree of performance degradation of the engine and gives a threshold curve of performance degradation and performance degradation. Considering the different contributions of clustering parameters to engine performance evaluation, an improved particle swarm optimization algorithm is proposed to optimize the parameters of multi-scale kernel function and penalty factor C. The simulation results show that, considering the multi-scale parameters, the performance of the engine can be better than the single-scale parameters in line with the actual usage. After clustering, the multi-scale parameters are basically consistent with the original parameters.