Unsupervised Time Series Segmentation: A Survey on Recent Advances

Time series segmentation has attracted more interests in recent years,which aims to segment time series into different segments,each reflects a state of the monitored objects.Although there have been many surveys on time series segmentation,most of them focus more on change point detection(CPD)methods and overlook the advances in boundary detection(BD)and state detection(SD)methods.In this paper,we categorize time series segmentation methods into CPD,BD,and SD methods,with a specific focus on recent advances in BD and SD methods.Within the scope of BD and SD,we subdivide the methods based on their underlying models/techniques and focus on the milestones that have shaped the development trajectory of each category.As a conclusion,we found that:(1)Existing methods failed to provide sufficient support for online working,with only a few methods supporting online deployment;(2)Most existing methods require the specification of parameters,which hinders their ability to work adaptively;(3)Existing SD methods do not attach importance to accurate detection of boundary points in evaluation,which may lead to limitations in boundary point detection.We highlight the ability to working online and adaptively as important attributes of segmentation methods,the boundary detection accuracy as a neglected metrics for SD methods.

Analysis of Change Point in Surface Temperature Time Series Using Cumulative Sum Chart and Bootstrapping for Asansol Weather Observation Station, West Bengal, India

This paper aims to detect the short-term as well as long-term change point in the surface air temperature time series for Asansol weather observation station, West Bengal, India. Temperature data for the period from 1941 to 2010 of the said weather observatory have been collected from Indian Meteorological Department, Kolkata. Variations and trends of annual mean temperature, annual mean maximum temperature and annual minimum temperature time series were examined. The cumulative sum charts (CUSUM) and bootstrapping were used for the detection of abrupt changes in the time series data set. Statistically significant abrupt changes and trends have been detected. The major change point in the annual mean temperatures occurred around 1986 (0.57°C) at the period of 25 years in the long-term regional scale. On the other side, the annual mean maximum and annual mean minimum temperatures have distinct change points at level 1. There are abrupt changes in the year 1961 (Confidence interval 1961, 1963) for the annual mean maximum and 1994 (Confidence interval 1993, 1996) for the annual mean minimum temperatures at a confidence level of 100% and 98%, respectively. Before the change, the annual mean maximum and annual mean minimum temperatures were 30.90°C and 23.99°C, respectively, while after the change, the temperatures became 33.93°C and 24.84°C, respectively. Over the entire period of consideration (1941-2010), 11 forward and backward changes were found in total. Out of 11, there are 3 changes (1961, 1986 and 2001) in annual mean temperatures, 4 changes (1957, 1961, 1980 and 1994) in annual mean maximum temperatures, and rest 4 changes (1968, 1981, 1994 and 2001) are associated with annual mean minimum temperature data set.

On-line outlier and change point detection for time series

The detection of outliers and change points from time series has become research focus in the area of time series data mining since it can be used for fraud detection, rare event discovery, event/trend change detection, etc. In most previous works, outlier detection and change point detection have not been related explicitly and the change point detections did not consider the influence of outliers, in this work, a unified detection framework was presented to deal with both of them. The framework is based on ALARCON-AQUINO and BARRIA's change points detection method and adopts two-stage detection to divide the outliers and change points. The advantages of it lie in that: firstly, unified structure for change detection and outlier detection further reduces the computational complexity and make the detective procedure simple; Secondly, the detection strategy of outlier detection before change point detection avoids the influence of outliers to the change point detection, and thus improves the accuracy of the change point detection. The simulation experiments of the proposed method for both model data and actual application data have been made and gotten 100% detection accuracy. The comparisons between traditional detection method and the proposed method further demonstrate that the unified detection structure is more accurate when the time series are contaminated by outliers.

Change Point Detection and Trend Analysis for Time Series

Trend analysis and change point detection in a time series are frequent analysis tools.Change point detection is the identification of abrupt variation in the process behaviour due to natural or artificial changes,whereas trend can be defined as estimation of gradual departure from past norms.We analyze the time series data in the presence of trend,using Cox-Stuart methods together with the change point algorithms.We applied the methods to the nearsurface wind speed time series for Australia as an example.The trends in near-surface wind speeds for Australia have been investigated based upon our newly developed wind speed datasets,which were constructed by blending observational data collected at various heights using local surface roughness information.The trend in wind speed at 10 m is generally increasing while at 2 m it tends to be decreasing.Significance testing,change point analysis and manual inspection of records indicate several factors may be contributing to the discrepancy,such as systematic biases accompanying instrument changes,random data errors(e.g.accumulation day error)and data sampling issues.Homogenization technique and multiple-period trend analysis based upon change point detections have thus been employed to clarify the source of the inconsistencies in wind speed trends.

Comparative Analysis of Climatic Change Trend and Change-Point Analysis for Long-Term Daily Rainfall Annual Maximum Time Series Data in Four Gauging Stations in Niger Delta

The aim of this study is to establish the prevailing conditions of changing climatic trends and change point dates in four selected meteorological stations of Uyo, Benin, Port Harcourt, and Warri in the Niger Delta region of Nigeria. Using daily or 24-hourly annual maximum series (AMS) data with the Indian Meteorological Department (IMD) and the modified Chowdury Indian Meteorological Department (MCIMD) models were adopted to downscale the time series data. Mann-Kendall (MK) trend and Sen’s Slope Estimator (SSE) test showed a statistically significant trend for Uyo and Benin, while Port Harcourt and Warri showed mild trends. The Sen’s Slope magnitude and variation rate were 21.6, 10.8, 6.00 and 4.4 mm/decade, respectively. The trend change-point analysis showed the initial rainfall change-point dates as 2002, 2005, 1988, and 2000 for Uyo, Benin, Port Harcourt, and Warri, respectively. These prove positive changing climatic conditions for rainfall in the study area. Erosion and flood control facilities analysis and design in the Niger Delta will require the application of Non-stationary IDF modelling.

Crop Yield Forecasted Model Based on Time Series Techniques

Traditional studies on potential yield mainly referred to attainable yield： the maximum yield which could be reached by a crop in a given environment. The new concept of crop yield under average climate conditions was defined in this paper, which was affected by advancement of science and technology. Based on the new concept of crop yield, the time series techniques relying on past yield data was employed to set up a forecasting model. The model was tested by using average grain yields of Liaoning Province in China from 1949 to 2005. The testing combined dynamic n-choosing and micro tendency rectification, and an average forecasting error was 1.24%. In the trend line of yield change, and then a yield turning point might occur, in which case the inflexion model was used to solve the problem of yield turn point.

Clustering-Inverse: A Generalized Model for Pattern-Based Time Series Segmentation

Patterned-based time series segmentation (PTSS) is an important task for many time series data mining applications. In this paper, according to the characteristics of PTSS, a generalized model is proposed for PTSS. First, a new inter-pretation for PTSS is given by comparing this problem with the prototype-based clustering (PC). Then, a novel model, called clustering-inverse model (CI-model), is presented. Finally, two algorithms are presented to implement this model. Our experimental results on artificial and real-world time series demonstrate that the proposed algorithms are quite effective.

Time Sequence Change-Point Model of Electrostatic State Parameters of Aircraft Engine

Electrostatic monitoring technology of particle charging information can facilitate online monitoring of aero-engine,which effectively enhances engine fault diagnosis and health managements.Unlike traditional engine state monitoring technologies,aircraft engine monitoring by gas path electrostatic monitoring not only covers the predicted information source itself,but also detects the information that can provide an early warnings for initial fault states through gas path charging levels.This paper establishes a non-stationary time sequence change-point model for anomaly recognition of electrostatic signals based on change-point theory combined with difference method of non-stationary time series.Finally,electrostatic induction data were utilized by the engine life test for a particular aircraft to validate the proposed algorithm.The results indicate that the activity level and the event rate were0.5—0.8(nc)and 50%,respectively,which were far greater than 4—12(pc)and 0—4% under normal working conditions of the engine.

高拱坝时序多属性施工方案随机智能优化方法

为了解决当前施工方案优化方法大多忽略各关键节点施工进度指标的随机性对施工方案优化的影响,同时属性权重确定方法难以实现对高维度多层次的多属性决策问题进行整体优化的问题,提出了基于前景随机理论和麻雀搜索算法的高拱坝时序多属性施工方案随机智能优化方法。针对高拱坝工程建设特点,基于前景随机理论建立时序多属性随机智能优化模型,提出了基于阶段发展特征的动态参考点设置方法;基于麻雀搜索算法建立最优属性权重及时间权重搜索模型,并以差异最大化思想构造适应度函数,实现模型的求解。工程实例验证结果表明该优化方法具有合理性,优化结果与前景随机占优-CRITIC方法、随机占优方法优化结果一致,且具有更好的方案区分度。

基于注意力机制CNN-LSTM的毫米波雷达点云特征数据预测生成

在智能驾驶的环境感知领域,毫米波雷达是一种关键的传感器技术。然而,因数据量有限,其特征数据的采集具有一定的挑战性,这限制了环境感知分类模型的训练效果。针对这一难题,提出了一种融合自注意力机制的卷积长短期记忆网络模型,旨在预测并生成毫米波雷达点云的特征数据,以此来扩展雷达特征数据集。首先采集道路目标的运动状态数据,对数据进行二维快速傅里叶变换、恒虚警率检测,并利用多输入多输出(Multiple-Input Multiple-Output,MIMO)技术提升方位分辨率;接着执行点云聚类及特征提取;最后采用含注意力机制的卷积长短期记忆网络对特征数据进行进一步处理与预测。在真实采集的3类道路目标数据集上,与其他模型相比,该方法在不同道路目标运动特征的预测R^(2)上提高了1%~7%。

基于测点聚类的POD-BPNN风压重构方法

文章提出本征正交分解(POD)与聚类分析结合的结构表面风压测点分类与关键测点布置方法,基于少量测点的风压数据,通过POD与误差反向传播神经网络(BPNN)方法实现方柱结构表面风压时程的重构。机器学习数据集为多风向角均匀来流下单方柱测压风洞试验得到的测点风压时程。将44个测点的风压时程数据POD降维,并采用K-means++聚类分析得到方柱周向轮廓系数分布,并基于轮廓系数的多风向角平均值,得到12、16、20和24个关键测点的轴对称布置方案。以关键测点的风压时程数据为训练集,采用POD-BPNN方法重构方柱表面其余测点所在位置的风压时程,并将风压时程及其统计值同试验结果对比。从12~20测点方案,风压重构精度逐步提升;20测点和24测点方案的重构风压差异较小,二者都能较好地重构方柱表面风压分布,仅在0°风向角方柱脉动风压误差偏大。

DRG点数法付费实施效果实证研究——基于间断时间序列模型

[目的]评价DRG点数法付费实施效果,为医院管理和政策完善提供借鉴。[方法]以泰州市某三甲医院2017-2023年医保住院患者为研究对象,通过描述性分析、卡方检验及间断时间序列分析等方法分析DRG点数法实施对运营效率、医疗费用、就医负担和医保基金支出的影响。[结果]DRG实施后,平均住院日、次均住院费用、次均药品费用和次均自付费用呈下降趋势(P值均小于0.05),次均检验检查和次均医疗服务费用呈上升趋势(P值均小于0.05),月出院人次、次均耗材费用和次均统筹基金支出趋势变化差异无统计学意义(P值均大于0.05)。[结论]DRG实施对提高运营效率、降低次均住院费用、优化费用结构、减轻就医负担起到积极作用,但对医院服务量和医保基金支出影响较小。

考虑监测数据时序特征和空间分布的堆石坝参数反演研究

随着安全监测技术的发展,柔性智能位移计、管道机器人等新型监测技术逐步被用于堆石坝的安全监测。高堆石坝在其生命期内中积累了海量的监测数据,充分利用这些数据,开展参数反演分析,可以提高堆石坝数值模拟的准确性,有助于合理评估堆石坝安全性态。论文基于时间序列聚类从海量监测数据选择有代表性、多样性的测点组合,提取时序特征构造目标函数,反映堆石坝变形的时空演化特性,采用多目标优化算法进行堆石坝各分区的材料参数反演。与现有参数反演方法相比,本文方法能合理利用堆石坝大量监测数据,充分反映其在填筑、蓄水和运行过程中的变形发展和空间分布特性,基于反演分析的材料参数其计算沉降值与实测值吻合良好,能够显著提升参数反演的精度。

一种基于多阈值模板的快速分类在线检测方法

传统离线数据分析方法对于处理即时性高和流量大的数据存在缺陷,而在线检测模型可以满足数据流分析的实时性要求。文中提出了一种基于多阈值模板的在线检测方法。该方法结合多路搜索树突变点检测(Ternary Search Tree and Kolmogorov-Smirnov,TSTKS)算法进行在线检测,基于突变点密度更新窗口长度从而提高了突变点检测精度。采用等量分级策略实现对时序数据的自学习、匹配和分类,进而对大规模病变数据进行状态检测和预测。仿真实验和病变数据的实验结果表明,所提方法具有效果高、分类准确等优点,为大规模时序数据进行快速分类研究提供了新方法。

基于多突变点与模板匹配的用电设备在线状态监测方法

针对现阶段用电设备状态监测技术存在的处理速度较慢、准确率较低等问题,文中基于多突变点检测和模板匹配策略提出了一种用电设备在线状态监测方法。该方法在缓冲区模型和滑动窗口模型的基础上,利用多路搜索树突变点检测(Ternary Search Tree and Kolmogorov-Smirnov,TSTKS)算法形成窗口维度和缓冲区维度的特征向量,通过两种维度的模板匹配实现用电设备的运行状态匹配和状态切换时刻定位。基于家用电冰箱的仿真实验结果表明,所提方法具有检测速度快、准确率高等优点,可为用电设备状态监测领域提供参考。

一种基于深度学习的时序病变数据段分类方法

针对在大规模时序医疗数据的分析中现有检测方法检测精度低、检测速度慢等问题,文中提出了一种基于深度学习的时序病变数据段分类方法。该方法在TSTKS(Ternary Search Trees and modified Kolmogorov-Smirnov)算法和滑动窗口理论的基础上,利用深度学习技术实现了对病变数据段的快速准确分类。文中以利用该方法对病变数据段进行分类的结果作为依据,实现了滑动窗口大小的动态调整。通过对真实癫痫脑电信号(Electroencephalogram,EEG)进行分析,证明了所提病变数据段分类方法和基于该分类方法的滑动窗口动态调整机制具有检测速度快、精度较高等优点,可以为大规模时序数据的快速分析研究提供一种新选择。

船舶柴油机运行参数异常检测及分析

为保障船舶安全高效运行,提出4种船舶柴油机运行参数异常检测方法。该技术在空间向量统计的异常检测基础上,进一步提出新的方法监测运行参数时间序列的趋势异常。通过提取运行参数数据演化过程的趋势和规律,可以更早、更准确地识别异常趋势,为设备管理提供决策支持。实验结果表明,所提出的异常检测技术能够有效提高船舶柴油机异常检测的效率和准确性,及早发现安全隐患。

基于时序InSAR的沿江地面形变监测与精度分析

针对长江两岸部分城市区域出现的地面沉降问题,提取2018年1月至2023年4月覆盖长江沿线南京-马鞍山段部分区域的60余景Sentinel-1A SAR影像,结合高精度PS特征点和SBAS-InSAR技术,监测该区域的地表形变情况。融入PS点的SBAS-InSAR技术与常规SBAS-InSAR技术结果均与水准测量结果高度一致,前者一致性和精度更高,相关决定系数高达0.91。实验发现该区域有3个典型沉降漏斗,分布在南京浦口区长江沿岸、建邺区东北部与秦淮区西北部交界处以及江宁区与安徽马鞍山市交界处,该区域沉降的主要原因地层岩性及新构造运动、工程活动、地下水位变化等。文中研究结果表明,融入PS点的SBAS-InSAR技术既可以获取高精度的地面形变结果,又可以为相关部门进行灾害风险管理和城市规划提供一定的技术支持。

基于PS-InSAR的渤海新区地表沉降监测

利用2017年4月—2018年12月的47景COSMO-SkyMed雷达影像,采用永久散射体合成孔径雷达干涉测量技术获得了渤海新区地表沉降信息,并基于GIS空间分析方法进一步揭示了渤海新区地表沉降的空间分布特征。根据InSAR遥感解译结果,结合渤海新区城市规划建设背景,分析了渤海新区地表沉降的分布演化态势及部分区域形变因素。结果显示,超过95.4%的PS点形变速率在区间[-9,6]mm/a内,约95%的PS点累计形变量在[-15,10]mm之间,处在沉降状态的PS点数量偏多。表明渤海新区港城区整体上基本处于稳定状态,形变较严重的地区主要集中在沿海地区和项目区西北部,填海造陆及新建建筑物地表压实等是造成区域形变的原因。

基于关键点与迁移学习的用户用电能耗动态预测算法

用户用电能耗数据具有时变性特点,导致用户用电能耗动态预测准确性较低。因此,设计一个基于关键点与迁移学习的用户用电能耗动态预测算法。应用迁移学习模型挖掘数据,采用关键点方法计算两个数据的余弦相似度,计算完成后融合数据,进行数据降噪与无量纲化处理,建立变分模型,分解采集的信号,将其分解为若干个波动规律的简单信号分量。在此基础上,结合支持向量回归方法与PSO算法组,实现用户用电能耗动态预测。实验结果表明,所提出的预测方法在电锅炉电能消耗、空调电能消耗、工作日电能消耗以及非工作日的电能消耗预测上,都具有较高的准确性。