Density Clustering Algorithm Based on KD-Tree and Voting Rules

Traditional clustering algorithms often struggle to produce satisfactory results when dealing with datasets withuneven density. Additionally, they incur substantial computational costs when applied to high-dimensional datadue to calculating similarity matrices. To alleviate these issues, we employ the KD-Tree to partition the dataset andcompute the K-nearest neighbors (KNN) density for each point, thereby avoiding the computation of similaritymatrices. Moreover, we apply the rules of voting elections, treating each data point as a voter and casting a votefor the point with the highest density among its KNN. By utilizing the vote counts of each point, we develop thestrategy for classifying noise points and potential cluster centers, allowing the algorithm to identify clusters withuneven density and complex shapes. Additionally, we define the concept of “adhesive points” between two clustersto merge adjacent clusters that have similar densities. This process helps us identify the optimal number of clustersautomatically. Experimental results indicate that our algorithm not only improves the efficiency of clustering butalso increases its accuracy.

The Heterogeneity and Its Influencing Factors of Soil Nutrients in Peak-Cluster Depression Areas of Karst Region

By selecting a typical peak-cluster depression area of karst region in Southwest China, we evaluated the effect of land use types and topographic factors on soil nutrients. Grid and line sampling methods were used to sample soil in depression and slope lands respectively, and classical statistical tools were applied to analyze the spatial variability character of soil organic carbon （SOC）, total nitrogen （TN）, total phosphorus （TP）, total potassium （TK）, available nitrogen （AN）, available phosphorus （AP）, available potassium （AK）, pH, and C/N. It was found that land use type was the dominant factor that effected the spatial heterogeneity of SOC, TN, TP, TK, AN, and AP. The content of SOC, TN, and AN decreased with the increase of land use intensity. Due to high fertilizer input, TP and AP in tillage fields were higher than those in the other land use types. TK had no obvious change trend among various land use types. Topographic factors had a significant effect on SOC, TN, TP, AN, AP, AK, and pH. Habitat factor was the dominant factor that effected AK. Altitude factor was the dominant factor for pH. However, all of these factors had no significant effect on C/N. Tillage practice had important effect on soil nutrients loss and soil degradation in the fragile karst ecosystem, and the input of organic manure should be increased in this region.

The features of soil erosion and soil leakage in karst peak-cluster areas of Southwest China

Through utilizing water flow monitoring, rock scratching, soil wood piles and radionuclide ^(137)Cs tracing in the Longhe karst ecological experimental site(hereinafter referred to"Longhe site"), Pingguo County, Guangxi Province, the features and values of soil erosion and soil leakage in different geomorphologic locations and land uses in the karst peak-cluster depressions are showed clearly. There are four kinds of geomorphologic locations in the karst peak-cluster depression, namely peaks, strip, slopes and depression. The soil leakage modulus in the peaks and strips respectively occupy 92.43% and 96.24% of the total mean soil erosion modulus at experimental sites. On the slope, soil leakage accounted for about 75%. At the bottom of depression, surface water was the main factor of soil erosion, and at last most soil leaked into underground rivers from sinkholes. The total soil erosion modulus and the contribution rates of relative surface soil erosion in regard of peaks, slopes and depressions gradually increased. There are also five major types of land use in the karst peak-cluster depressions, farmland, Kudingcha tea plantations, young Lignum Sappan fields, shrub-grassland and pastures. The soil erosion modulus of slope farmland has the highest value with an increasing trend year by year. But soil erosion modulus of other four land use types decreased by year, which shows the "grain for green" will result in better soil protection. By handling with rocky desertification and ecological rehabilitation in Longhe site, the mean soil erosion modulus of the karst peak-cluster depression has decreased about 80% from 2003 to 2015.

Fast density peak-based clustering algorithm for multiple extended target tracking

The key challenge of the extended target probability hypothesis density (ET-PHD) filter is to reduce the computational complexity by using a subset to approximate the full set of partitions. In this paper, the influence for the tracking results of different partitions is analyzed, and the form of the most informative partition is obtained. Then, a fast density peak-based clustering (FDPC) partitioning algorithm is applied to the measurement set partitioning. Since only one partition of the measurement set is used, the ET-PHD filter based on FDPC partitioning has lower computational complexity than the other ET-PHD filters. As FDPC partitioning is able to remove the spatially close clutter-generated measurements, the ET-PHD filter based on FDPC partitioning has good tracking performance in the scenario with more clutter-generated measurements. The simulation results show that the proposed algorithm can get the most informative partition and obviously reduce computational burden without losing tracking performance. As the number of clutter-generated measurements increased, the ET-PHD filter based on FDPC partitioning has better tracking performance than other ET-PHD filters. The FDPC algorithm will play an important role in the engineering realization of the multiple extended target tracking filter.

基于聚类和深度学习的车联网轨迹隐私保护机制

针对车联网轨迹发布中用户面临的隐私泄露问题,提出基于聚类和深度学习的轨迹隐私保护机制(PPCDL).考虑轨迹中的时间因素,通过时间戳将轨迹空间划分为多个区域,获取区域中的轨迹分布点.对每个区域进行改进稳定隶属度多峰值聚类,根据区域轨迹密度进行隐私预算矩阵的预分配.利用时间图卷积网络模型提取轨迹数据的时空特征,对隐私预算预分配矩阵进行训练和预测.根据预测结果添加相应的拉普拉斯噪声,在轨迹数据发布前进行扰动.理论分析和实验结果表明,PPCDL相较于对比机制,时间开销更少,能够更精确地预测隐私预算.利用PPCDL可以合理地在轨迹数据中添加拉普拉斯噪声,有效地提高了轨迹数据的可用性.

面向密度分布不均数据的加权逆近邻密度峰值聚类算法

针对密度分布不均数据,密度峰值聚类算法易忽略类簇间样本的疏密差异,导致误选类簇中心;分配策略易将稀疏区域的样本误分到密集区域,导致聚类效果不佳的问题,本文提出一种面向密度分布不均数据的加权逆近邻密度峰值聚类算法。该算法首先在局部密度公式中引入基于sigmoid函数的权重系数,增加稀疏区域样本的权重,结合逆近邻思想,重新定义了样本的局部密度,有效提升类簇中心的识别率;其次,引入改进的样本相似度策略,利用样本间的逆近邻及共享逆近邻信息,使得同一类簇样本间具有较高的相似度,可有效改善稀疏区域样本分配错误的问题。在密度分布不均、复杂形态和UCI数据集上的对比实验表明,本文算法的聚类效果优于IDPC-FA、FNDPC、FKNN-DPC、DPC和DPCSA算法。

高光谱影像逆近邻密度峰值聚类的波段选择算法

密度峰值聚类波段选择算法利用局部密度描述波段的密度信息,然而现有的局部密度容易忽略波段分布的全局信息,不能有效描述波段的分布特征,导致波段子集分类精度有限。为解决上述问题,本文提出一种基于逆近邻的密度峰值聚类波段选择算法。首先,利用波段与其K近邻构建K近邻有向图,获取波段的逆近邻,以及波段之间的共享近邻和共享逆近邻;然后,利用共享近邻和共享逆近邻并集的个数作为波段之间的相似度,利用波段与其逆近邻的平均欧氏距离和相似度构造增强型局部密度;最后,将增强型局部密度、距离因子、信息熵三者的乘积作为权重值,根据权重值挑选波段子集。为提高试验效率和实用性,本文算法还提出一种自动获得K值的自适应K值方法。在3个高光谱标准数据集上的试验结果表明,本文算法得到的波段子集比其他先进算法挑选的波段有更好的分类性能,尤其是在波段数较少的情况下,而且计算效率较高。

结合聚类边界采样的主动学习

主动学习是一种机器学习方法,需要选择最有价值的样本进行标注。目前,主动学习在应用时面临着一些挑战,其依赖分类器的先验假设,这容易导致分类器性能意外下降,同时需要一定规模的样本作为启动条件。聚类可以降低问题规模,是主动学习的一种有效手段。为此,结合密度聚类边界采样,开展主动学习方法的研究。针对容易产生分类错误的聚类边界区域,通过计算样本密度,提出一种密度峰值聚类边界点采样方法;在此基础上,给出密度熵的定义,并利用密度熵对聚类边界区域进行启发式搜索,提出一种基于聚类边界采样的主动学习方法。试验结果表明,与文献中的5种主动学习算法相比,该算法能够以更少标记量获得同等甚至更高的分类性能,是一种有效的主动学习算法;在标记不足,无标签样本总量20%的情况下,算法在Accuracy、F-score等指标上取得较好的结果。

局部标准差优化的密度峰值聚类算法

密度峰值聚类(clustering by fast search and find of density peaks,DPC)算法是一种基于密度的聚类算法,它可以发现任意形状和维度的类簇,是具有里程碑意义的聚类算法。然而,DPC算法的样本局部密度定义不适用于同时发现数据集的稠密簇和稀疏簇;此外,DPC算法的一步分配策略使得一旦有一个样本分配错误,将导致更多样本的错误分配,产生“多米诺骨牌效应”。针对这些问题,提出一种新的样本局部密度定义,采用局部标准差指数定义样本局部密度,克服DPC的密度定义缺陷;采用两步分配策略代替DPC的一步分配策略,克服DPC的“多米诺骨牌效应”,得到ESDTS-DPC算法。与DPC及其改进算法KNN-DPC、FKNN-DPC、DPC-CE和经典密度聚类算法DBSCAN的实验比较显示,提出的ESDTS-DPC算法具有更好的聚类准确性。

基于网格近邻优化的密度峰值聚类算法

密度峰值聚类(DPC)将数据样本点的局部密度和相对距离进行结合,能对任意形状数据集进行聚类处理,但密度峰值聚类算法存在主观选择截断距离、简单分配策略和较高时间复杂度等问题。为此,提出了一种基于网格近邻优化的密度峰值聚类算法(KG-DPC算法)。首先对数据空间进行网格化,减少了样本数据点之间距离的计算量;在计算局部密度时不仅考虑了网格自身的密度值,而且考虑了周围k个近邻的网格密度值,降低了主观选择截断距离对聚类结果的影响,提高了聚类准确率,设定网格密度阈值,保证了聚类结果的稳定性。通过实验结果表明,KG-DPC算法比DBSCAN、DPC和SDPC算法在聚类准确率上有很大提升,在聚类平均消耗时间上DPC、SNN-DPC和DPC-NN算法分别降低38%、44%和44%。在保证基本聚类准确率的基础上,KG-DPC算法在聚类效率上有特定优势。

基于密度峰值的标签传播算法

随着智能技术应用的推广,高质量社区的检测已成为社会网络研究的热点之一。由于具有线性时间复杂度,且无需预定义目标函数和社团数,标签传播算法(LPA)已得到广泛关注。然而,在标签传播过程中,LPA具有不确定性和随机性,进而影响检测社区结果的准确性和稳定性。为此,提出一种基于密度峰值的标签传播社区检测方法(DPC-RWL)。首先,采用密度峰值聚类算法查找出社区的核心节点集合,计算节点与核心节点集之间的权重,选取最大值为该节点赋予权值。最后,使用基于标签传播算法的归属度函数进行传播。真实网络和LFR人工基准网络的对比实验表明,所提算法能准确高效地识别出社区结构。

密度峰值聚类在塔机损伤诊断中的应用研究

建立塔机有限元模型,获取塔机完好状态和各损伤工况的各采集点的动态位移。提出了两种模型建立方法,基于悬臂梁的双输入单输出模型和基于时域数据的动态双输入单输出模型,对基于时域数据的双输入单输出模型首先利用最小二乘法计算参数初值,进一步利用粒子群优化方法进行参数优化,提高了模型精度。以完好工况的塔机数据为基础,建立基于悬臂梁的双输入单输出模型和基于时域数据的双输入单输出模型,计算参数,建立损伤识别模型,用待检状态的位移数值拟合模型,用两种模型计算出的残差方差做损伤因子,利用密度峰值聚类方法对损伤因子进行分析,实现了对塔机的损伤判定和损伤位置的确定。这种基于密度峰值聚类的诊断方法可对塔机微小损伤进行智能诊断和位置确定,该方法只需要塔机完好状态的数据和待检状态的数据即可自动诊断,解决了塔机损伤识别中损伤数据难以获取,因而无法实现智能训练和诊断的问题。

A Health State Prediction Model Based on Belief Rule Base and LSTM for Complex Systems

In industrial production and engineering operations,the health state of complex systems is critical,and predicting it can ensure normal operation.Complex systems have many monitoring indicators,complex coupling structures,non-linear and time-varying characteristics,so it is a challenge to establish a reliable prediction model.The belief rule base(BRB)can fuse observed data and expert knowledge to establish a nonlinear relationship between input and output and has well modeling capabilities.Since each indicator of the complex system can reflect the health state to some extent,the BRB is built based on the causal relationship between system indicators and the health state to achieve the prediction.A health state prediction model based on BRB and long short term memory for complex systems is proposed in this paper.Firstly,the LSTMis introduced to predict the trend of the indicators in the system.Secondly,the Density Peak Clustering(DPC)algorithmis used todetermine referential values of indicators for BRB,which effectively offset the lack of expert knowledge.Then,the predicted values and expert knowledge are fused to construct BRB to predict the health state of the systems by inference.Finally,the effectiveness of the model is verified by a case study of a certain vehicle hydraulic pump.

The Deconvolution Method for Obtaining Correspondence in Data-Independent Acquisition Mass Spectrometry Data

Although data-independent acquisition (DIA) shows powerful potential in achieving comprehensive peptide information acquisition, the difficulty in determining the precursor m/z and distinguishing fragment ions has posed challenges in DIA data analysis. To address this challenge, a common approach is to recover the correspondence between precursor ions and fragment ions, followed by peptide identification using traditional data-dependent acquisition (DDA) database searching. In this study, we propose a cosine similarity-based deconvolution method that rapidly establishes the correspondence between chromatographic profiles of precursor ions and fragment ions through matrix calculations. Experimental results demonstrate that our method, referred to as CosDIA, yields a peptide identification count close to that of DIA-umpire. However, compared to DIA-umpire, we can establish the correspondence between original MS/MS spectra and pseudo-MS/MS spectra. Furthermore, compared to the CorrDIA method, our approach achieves higher efficiency in terms of time, reducing the time cost of the analysis process. These results highlight the potential advantages of the CosDIA method in DIA data analysis, providing a powerful tool and method for large-scale proteomics research.

改进DPC聚类算法的离群点检测与解释方法

为解决全局离群点检测方法无法对局部离群点进行检测,以及局部异常因子在面对大量局部离群点时性能下降的问题,利用k近邻(KNN)和核密度估计方法(KDE)提出一种基于改进快速搜索和发现密度峰值聚类算法(KDPC)的离群点检测与解释方法,该方法能够同时对数据点的全局和局部进行分析。首先,利用k近邻和核密度估计方法计算数据点的局部密度,代替传统DPC算法中根据截断距离计算的局部密度。其次,将数据点的k近邻距离之和作为全局异常值,并通过KDPC聚类算法计算簇密度以及数据点的局部异常值。最后,将数据点的全局与局部异常值进行乘积作为最终异常得分,选取异常得分最高的Top-n作为离群点,通过构建全局-局部异常值决策图对全局和局部离群点进行解释。利用人工数据集和UCI数据集进行实验并与10种常用离群点检测方法进行比较。结果表明,该方法对全局和局部离群点都有着较高的检测精度和检测性能,并且AUC方面受k值影响较小。同时,利用该方法对NBA球员数据进行分析讨论,进一步证明了该方法的实用性和有效性。

贝壳形屋盖风压系数密度峰值聚类分区研究

鉴于贝壳形屋盖表面风压分布的特殊性,提出基于快速搜索技术的密度峰值聚类风压系数分区方法。以某贝壳形屋盖表面风压分布试验结果为基础,进行密度峰值聚类风压系数分区,采用SD有效性指标验证分区数的合理性,并与改进k-means聚类分区结果进行对比。结果表明:密度峰值聚类风压系数分区以风压系数相对距离、局部密度和综合指数为特征参数,能较好反映屋面风压分布特性,有效保证类聚合性和类类分离性;相较于改进k-means分区法,不同风向角下密度峰值聚类得到的分区数与SD最优聚类数接近;密度峰值聚类分区结果能更准确反映贝壳形屋盖表面风压分布特性,充分体现测点风压系数局部密度和相对距离特征值较大的基本原则,对于贝壳形屋盖的风压系数分区具有更好的适用性;贝壳形屋盖密度峰值聚类分区最大负风压系数在-0.59~-1.74之间。

时序加权密度峰值聚类算法及用电负荷特性分类模型

针对现有的密度峰值快速搜索算法没有考虑数据的时序性、无法处理动态时间序列数据的问题,在密度峰值快速搜索算法基础之上,增加时序加权因子对数据点间的拓扑关系进行改进,提出了时序加权密度峰值聚类算法,使密度峰值快速搜索算法具有处理动态时序数据的能力.使用基于时序加权密度峰值聚类算法的用户负荷分类模型对OpenEI公布的用户电力负荷数据集进行处理,其聚类效果对比基于密度峰值聚类算法的用户负荷分类模型结果更准确,且统计学评价指标均有所提升.

基于密度峰值聚类的Tri-training算法

Tri-training利用无标签数据进行分类可有效提高分类器的泛化能力,但其易将无标签数据误标,从而形成训练噪声。提出一种基于密度峰值聚类的Tri-training(Tri-training with density peaks clustering,DPC-TT)算法。密度峰值聚类通过类簇中心和局部密度可选出数据空间结构表现较好的样本。DPC-TT算法采用密度峰值聚类算法获取训练数据的类簇中心和样本的局部密度,对类簇中心的截断距离范围内的样本认定为空间结构表现较好,标记为核心数据,使用核心数据更新分类器,可降低迭代过程中的训练噪声,进而提高分类器的性能。实验结果表明:相比于标准Tritraining算法及其改进算法,DPC-TT算法具有更好的分类性能。

喀斯特峰丛洼地3个建群树种“植物-凋落物-土壤”系统氮同位素特征

以喀斯特峰丛洼地3种常见次生林及其建群树种为研究对象,测定3种林分建群种的不同器官及凋落物、土壤的氮含量及氮同位素丰度值(δ^(15)N),探讨植物-凋落物-土壤氮含量及稳定氮同位素组成变化及内在联系。结果显示:3种次生林建群种植物中叶片氮含量最高,茎的氮含量最低,3种次生林建群树种叶片氮含量均显著高于掉落物和土壤氮含量,0~30 cm土壤氮含量显著高于30~90 cm土壤氮含量,土壤养分主要集中于表层土壤;白栎、栓皮栎和光皮桦器官中δ^(15)N变化范围分别为-2.82‰~14.94‰、-1.37‰~9.35‰和-4.39‰~26.06‰,3种建群种器官间氮同位素组成均表现出显著差异(P<0.05);3种建群种群落表层土壤δ^(15)N差异显著,3种林分中0~30 cm土壤δ^(15)N均显著低于30~60 cm和60~90 cm的土壤,δ^(15)N可能是这些群落土壤氮循环过程中的敏感指标,光皮桦叶片氮含量略大于白栎和栓皮栎,光皮桦林土壤表层δ^(15)N大于白栎林和栓皮栎林,光皮桦可能比白栎和栓皮栎更为适应喀斯特生境。