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...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。展开更多
The conventional digital core models are usually small in size and have difficulty in representing the complex structures of heterogeneous rocks;Therefore,the parameters of simulated rock physics are difficult to be r...The conventional digital core models are usually small in size and have difficulty in representing the complex structures of heterogeneous rocks;Therefore,the parameters of simulated rock physics are difficult to be referenced.In this study,we propose a feasible simulation method for obtaining multi-scale and multi-component digital cores based on three types of sandstone samples.In the proposed method,the plug and subplug samples are scanned via micro-computed tomography at different resolutions.Furthermore,the images are precisely registered using the proposed hybrid image registration method.In case of high-resolution images,the traditional segmentation method is used to segment the cores into pores and minerals.Subsequently,we established the relations between the gray values and the porosity/mineral content in case of the low-resolution images based on the registered domains and the relation curves were applied to the segmentation of the low-resolution images.The core images constitute the multi-scale and multi-component digital core models after segmentation.Further,the elastic properties of the three samples were simulated at both fine and coarse scales based on the multi-scale and multi-component digital core models,and four component models were considered.The results show that the multi-scale and multi-component digital core models can overcome the representative limits of the conventional digital core models and accurately characterize pores and minerals at different scales.The numerical results of the elastic modulus are more representative at large scales,and considerably reliable results can be obtained by appropriately considering the minerals.展开更多
文摘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。
基金supported by the National Natural Science Foundation of China Research(Nos.41574122 and 41374124)National Science and Technology major Project(No.2016ZX05006002-004)。
文摘The conventional digital core models are usually small in size and have difficulty in representing the complex structures of heterogeneous rocks;Therefore,the parameters of simulated rock physics are difficult to be referenced.In this study,we propose a feasible simulation method for obtaining multi-scale and multi-component digital cores based on three types of sandstone samples.In the proposed method,the plug and subplug samples are scanned via micro-computed tomography at different resolutions.Furthermore,the images are precisely registered using the proposed hybrid image registration method.In case of high-resolution images,the traditional segmentation method is used to segment the cores into pores and minerals.Subsequently,we established the relations between the gray values and the porosity/mineral content in case of the low-resolution images based on the registered domains and the relation curves were applied to the segmentation of the low-resolution images.The core images constitute the multi-scale and multi-component digital core models after segmentation.Further,the elastic properties of the three samples were simulated at both fine and coarse scales based on the multi-scale and multi-component digital core models,and four component models were considered.The results show that the multi-scale and multi-component digital core models can overcome the representative limits of the conventional digital core models and accurately characterize pores and minerals at different scales.The numerical results of the elastic modulus are more representative at large scales,and considerably reliable results can be obtained by appropriately considering the minerals.
