GPU加速的改进PAM聚类算法研究与应用

空间聚类是空间数据挖掘的重要方法,而K-Medoids是一种常用的空间聚类算法。K-Medoids聚类算法存在初始点选择问题,而且计算复杂。为了提高算法的有效性和时间效率,本文结合模拟退火算法思想,改进了传统的K-Medoids算法PAM,提出一种基于GPU计算的并行模拟退火PAM算法。类比矩阵乘法运算,定义了一种新的矩阵计算方法,可以有效减少数据在GPU全局内存和共享内存之间的传输,提高了算法在GPU中的执行效率。利用模拟退火算法搜索聚类中心点,保证了聚类结果的全局最优性。基于不同的数据集,将串行和并行模拟退火PAM算法以及已有的遗传PAM算法进行比较,结果表明并行模拟退火PAM算法聚类结果正确,且时间效率高。最后,应用本文改进算法对贵州省安监系统的安全监管隐患数据进行聚类分析,发现了隐患聚集中心,相关结果对政府的决策具有一定的实际应用价值。     

LAW AND THE INTERPRETIVE TURN IN THE SOCIAL SCIENCES

Legal cases are increasingly used in the social sciences as the raw material for social analysis. While this is entirely laudable as a research strategy, the analytical methods used in the social sciences to study cases are too often simplistic and reductionist. Positivism is too often the preferred analytical mode. I argue that the legal texture of cases ought to be retained by social scientists, just as we ought to be more aware of the problematic nature of interpretation, an issue embedded within legal discourse. In contrast to those who contend that the interpretive turn in the social sciences is nihilistic and subjective, I suggest reasons for understanding the interpretive turn as an organized social practice dependent upon social standards of right and wrong. These issues are illustrated by reference to a recent court case involving a corporation that secretly and illegally relocated production so as to avoid its pension obligations.     

贵州安顺市PM2.5气象输送条件及贡献源空间分布

利用贵州省安顺市2015—2019年大气污染物资料和气象资料,分析安顺市空气质量特征和主要大气污染物特征,通过TrajStat软件中HYSPLIT模型的后向轨迹模式,结合GDAS气象数据、PM_2.5浓度,分析不同季节输送途径及其污染轨迹,采用潜在源贡献作用和浓度权重轨迹分析方法,分析研究期内所有PM_2.5污染日（PM_2.5日浓度高于75 μg·m^-3）输送轨迹垂直与水平方向分布特征。结果表明: PM_2.5是安顺城区主要大气污染物,冬季输送污染轨迹占比较大,输送方向主要为贵州东北方向、偏南方向; 污染日PM_2.5输送路径以贵州东北方向近距离输送为主,该类轨迹基本分布在880—980 hPa高度; 潜在源高值区主要集中在贵阳整个地区、毕节织金县、黔西市、金沙县等,高贡献值区主要集中在安顺紫云县、镇宁县、毕节织金县、大方县等。     

Stratigraphic delineation by three-dimensional clustering of piezocone data

Piezocone soundings are a fast and economical approach for geotechnical site characterization, providing three separate and continuous channels of data with depth, including: tip resistance q _T, porewater pressure u ₂ and sleeve friction f _s. Literally hundreds to thousands of data points are collected by a single sounding. Since these readings are functions of both soil type and soil behaviour, they can be used for the delineation of soil stratigraphy.

One way to process large amounts of data involves clustering. Cluster analysis is an efficient statistical way to analyse the stratigraphic vertical profiling of geomaterials and means to detect the inherent similarity between data sets and group them together. Clustering in previous geotechnical research was based on only two channels of piezocone data (q _T and u₂). The method works well for soils that are under the groundwater table and was applied to soundings in clay deposits.

In the present paper, a new cluster analysis approach is developed based on all three channels of data, thus extending the method to soils above the water table and applicable to sands, silts, and clays. Example soil profiles derived by three-channel cluster analysis are presented herein and compared with conventional soil boring and sampling data.     

北京高科技产业集群衍生效应及其影响分析——基于中关村科技园区的案例研究

以北京高科技产业集群衍生效应的分析为核心,将衍生效应分为两大类:企业衍生效应和技术衍生效应.通过实证研究.分析出衍生效应的主要表现形式,并以此为基础设定分析标准,用以反映北京高科技产业集群的发展特点.研究发现:衍生效应能够提升竞争优势和智力资源的利用效率,对高科技产业集群的发展有积极意义;同时,衍生效应能够强化集群专业化分工,促进技术扩散;技术衍生效应能促进企业规模化和专业化,而企业衍生效应强化了集群内部的竞争,能促进技术创新和研发投入.     

联合DBSCAN聚类采样和SVM分类的滑坡易发性评价

针对基于机器学习的滑坡易发性评价中非滑坡样本选取不规范导致的分类精度较低问题,本文提出联合基于密度的噪声应用空间聚类(Density-Based Spatial Clustering of Applications with Noise,DBSCAN)采样策略和支持向量机(Support Vector Machine,...     

Hong kong direct investment in china’s mainland: A spatial study with special reference to the locational behaviour of investors

The study shows that the regional differentiation of Hong Kong direct investment (HKDI) in China was enlarged in the 1980s and has been reducing since 1991. The concentration of HKDI in the southern China and the coast has been replaced by the trend of moving northwards and inlandwards after 1989. But in terms of manufacturing sectors this trend has not clearly appeared until now. Those changing patterns are clearly related to the behaviour of Hong Kong industrial investors. The survey results suggest that “enjoying lower production cost” is the most important motive for their investment in China. In consequence, “close to Hong Kong”, “cheaper labour” and “favourable policy” are major factors for determining the investment locations. This behaviour, together with the lack of comparison study in location selection among most investors, explains that Guangdong absorbed most Hong Kong investment in the 1980s. As the production costs have been increasing in Guangdong, some companies have started to invest in non-Guangdong locations since the end of the 1980s. From the mental map of Hong Kong investors, however, only some of the non-Guangdong locations are practically attractive. Supported by National Natural Science Foundation of China and the University Graduates Association (Hong Kong). The author gratefully acknowledges the help from Professors Kam-hon Lee and Yue-man Yeung and Dr. David Chu at the Chinese University of Hong Kong, and Mr. Chai-ming Mak at Electricity Advisory Services Limited (HK) during the early stage of the study.     

产业转型背景下创业区位选择和集群空间演化

以广州软件和信息技术产业集群为案例,分析产业范式转变背景下地方创业区位选择对产业集群空间演化的影响。结果显示：初始资源富集和政策环境导向使特定空间成为创业的优势区位,早期第一代创业企业在其中集聚发展形成集群雏形,并在关系网络和区域认同作用下形成产业集群空间核心区。第二代创业企业在产业转型契机下选择开拓缝隙市场,新的区位诉求由此产生并作用在集群的空间演化上。总的来说,产业转型期同时也是集群空间转变的关键时期,创业企业通过区位选择可以改变集群的空间形态和层次。     

Fast numerical solution of the linearized Molodensky problem

 When standard boundary element methods (BEM) are used in order to solve the linearized vector Molodensky problem we are confronted with two problems: (1) the absence of O(|x|⁻²) terms in the decay condition is not taken into account, since the single-layer ansatz, which is commonly used as representation of the disturbing potential, is of the order O(|x|⁻¹) as x→∞. This implies that the standard theory of Galerkin BEM is not applicable since the injectivity of the integral operator fails; (2) the N×N stiffness matrix is dense, with N typically of the order 10⁵. Without fast algorithms, which provide suitable approximations to the stiffness matrix by a sparse one with O(N(logN) ^s ), s≥0, non-zero elements, high-resolution global gravity field recovery is not feasible. Solutions to both problems are proposed. (1) A proper variational formulation taking the decay condition into account is based on some closed subspace of co-dimension 3 of the space of square integrable functions on the boundary surface. Instead of imposing the constraints directly on the boundary element trial space, they are incorporated into a variational formulation by penalization with a Lagrange multiplier. The conforming discretization yields an augmented linear system of equations of dimension N+3×N+3. The penalty term guarantees the well-posedness of the problem, and gives precise information about the incompatibility of the data. (2) Since the upper left submatrix of dimension N×N of the augmented system is the stiffness matrix of the standard BEM, the approach allows all techniques to be used to generate sparse approximations to the stiffness matrix, such as wavelets, fast multipole methods, panel clustering etc., without any modification. A combination of panel clustering and fast multipole method is used in order to solve the augmented linear system of equations in O(N) operations. The method is based on an approximation of the kernel function of the integral operator by a degenerate kernel in the far field, which is provided by a multipole expansion of the kernel function. Numerical experiments show that the fast algorithm is superior to the standard BEM algorithm in terms of CPU time by about three orders of magnitude for N=65 538 unknowns. Similar holds for the storage requirements. About 30 iterations are necessary in order to solve the linear system of equations using the generalized minimum residual method (GMRES). The number of iterations is almost independent of the number of unknowns, which indicates good conditioning of the system matrix. Received: 16 October 1999 / Accepted: 28 February 2001