Time Series Segmentation Research Articles

An adaptive time series segmentation algorithm based on visibility graph and particle swarm optimization

Time series segmentation is a crucial area of research in time series analysis as it can reveal meaningful patterns or segments hidden within time series data. In this paper, we present an accurate and efficient time series segmentation method that combines the visibility graph method, particle swarm optimization, and community detection algorithm. We start by applying visibility graph theory to process time series data, resulting in a corresponding complex network. Next, we introduce an adaptive particle swarm optimization algorithm with modularity Q as the objective function to optimize community detection. Finally, mapping the communities back to the nodes of the time series yields the segmented sequence. Our proposed method offers high segmentation accuracy and low time complexity (O(n2)). Experimental results demonstrate that our approach outperforms existing methods in terms of segmentation accuracy on two different synthetic datasets. Furthermore, when applied to the S&P500 index dataset, it accurately identifies financial cycles and key financial events.

An evolutionary vehicle scanning method for bridges based on time series segmentation and change point detection

In this paper, the problem of bridge health monitoring is formulated as an unsupervised semantic segmentation where the aim is to identify the onset of structural damage by detecting unexpected irregularities in the time series data. In particular, the main objective of this study is to approach this problem through an indirect sensing framework based on the concept of vehicle-mounted scanning. Thus, the sole source of information is the acceleration responses of the moving vehicle passing over the bridge. This is achieved through automatic feature learning by adopting the concept of matrix profile to conduct an all-pair-similarity search for a given time series, where the location of the contextual change is determined by the corresponding Corrected Arc Crossings (CAC) profile. The performance of the method is first evaluated through numerical investigations for different damage cases where the effect of operational uncertainties such as road roughness, and variability in vehicle properties are considered. Further, the method is experimentally evaluated using a lab-scale bridge model with different damage cases. Through these investigations, it is consistently observed that the location of the change point always coincides with the minimum point of the CAC profile. The proposed framework has three major advantages making it more appealing compared to the competing techniques. First, it has been fully formulated in the context of unsupervised learning where no access to labeled data is required. Second, it is capable of working with a limited amount of data yet providing promising results. Finally, it can be easily applied to online settings where data is streaming in real-time. These successful results demonstrate the potential of the proposed method as a cost-effective and rigorous bridge inspection tool and open up opportunities in the field of indirect bridge health monitoring.

Learning design thinking well: organizing (introductory) workshops for optimal participant experience

A better understanding of the implementation of design thinking (DT) as a means to increase the assimilation of its practices is an important concern for R&D and innovation managers because more frequent use of DT practices raises organizational innovative capabilities. Introducing DT practices in workshops is a common, but also challenging approach to learning DT. Participants often have little to no prior encounters with the tools, techniques and principles used to practice DT and often mistrust or misinterpret the DT concept after a bad experience during the first exposure. For that reason, a suboptimal experience in introductory workshops can prevent or reduce the assimilation of DT practices. This study investigates how participant experience during first exposure as a vital precursor to the assimilation of DT practices can be improved via the organization of introductory workshops. We examine flow as the core experience measure in a quantitative‐inductive, quasi‐experimental study of two 3‐day DT workshops and compare periods with step‐by‐step acquisition of DT practices under guidance, periods of flexible, independent application of DT practices and periods of recovery. Our research employs a segmented time series analysis on more than 1100 real‐time observations collected with the experience sampling method (ESM). This study contributes to the scarce empirical research on improving the implementation of DT practices by indicating that a positive initial experience depends on workshop length, the logic of guided periods, the combination of guided and unguided periods, timing and type of recovery, as well as on team formation. Our results lead to five organizing principles, which can be used as testable propositions for future research, as well as in the planning and execution of (introductory) DT workshops.

Semantic segmentation of anomalous diffusion using deep convolutional networks

Heterogeneous dynamics commonly emerges in anomalous diffusion with intermittent transitions of diffusion states but proves challenging to identify using conventional statistical methods. To effectively capture these transient changes of diffusion states, we propose a deep learning model (U-AnDi) for the semantic segmentation of anomalous diffusion trajectories. This model is developed with the dilated causal convolution (DCC), gated activation unit (GAU), and U-Net architecture. The study addresses two key subtasks related to trajectory segmentation and changepoint detection, concentrating on variations in diffusion exponents and dynamic models. Additionally, extended analyses are conducted on the segmentation of single-model trajectories, multistate biological trajectories, and anomalous diffusion with added correlation functions. By rationally designing comparative models and evaluating the performance of U-AnDi against these models, we discover that U-AnDi consistently outperforms other models across all segmentation tasks, thereby affirming its superiority in the field. This performance edge also sheds light on the interpretability of U-AnDi's core components: DCC, GAU, and U-Net. The clarity with which these components contribute to U-AnDi's success underscores their congruence with the intrinsic physics underlying anomalous diffusion. Furthermore, our model is examined using real-world anomalous diffusion data: the diffusion of transmembrane proteins on cell membrane surfaces, and the segmentation results are highly consistent with experimental observations. Our findings could offer a heuristic deep learning solution for the detection of heterogeneous dynamics in single-molecule/particle tracking experiments, and have the potential to be generalized as a universal scheme for time-series segmentation. Published by the American Physical Society 2024

Examining the Dynamics of Vegetation in South Korea: An Integrated Analysis Using Remote Sensing and In Situ Data

Vegetation constitutes the primary component of terrestrial ecosystems and plays a crucial role in examining global climate change and its impacts. Assessing vegetation dynamics over significant periods of time can provide critical information on changes in vegetation cover and regional climate. Satellite-based remote sensing products offer a unique opportunity to analyze these changes at a country scale. South Korea has a unique history of forest regeneration, and studying this relationship provides an opportunity to examine its impact. To achieve this, we employ the vegetation–precipitation relationship using time series segmentation on residual trends (TSS-RESTREND), a statistical method that calculates the correlation between the annual maximum vegetation index and cumulative rainfall. This method helps identify time-step changes related to structural changes in the ecosystem. Using the enhanced vegetation index (EVI) and rainfall, this study aims to uncover the underlying dynamics and driving forces behind observed ecosystem changes, facilitating a deeper comprehension of the intricate interactions between vegetation and precipitation through the application of TSS-RESTREND. The findings increase knowledge of the complex relationship between vegetation growth and rainfall, given the country’s commendable forest regeneration program, even in the face of rapid urbanization. Moreover, the integration of finer resolution land cover images derived from in situ data not only provides a nuanced understanding of vegetation dynamics across diverse regions, but also underscores the critical need for continued collection of in situ data to provide more accurate and comprehensive analyses. The findings underline the pivotal role played by land resource management practices and underscores the significance of implementing adaptive conservation strategies, thereby contributing to the preservation of a stable and sustainable vegetation cover. The findings serve as a call for the ecosystem monitoring and strategic adaptation required to examine the environmental and ecosystem structural changes in the region. This study sheds light on the critical relationship between land management practices, environmental sustainability, and the resilience of the region’s vegetation amidst an ever-changing landscape.

Larval presence and settlement of the American lobster (Homarus americanus) in a changing southern New England estuary

During the late 1980s and early 1990s, the American lobster (Homarus americanus) fishery boomed in southern New England, but starting in the early 2000s lobsters declined in the nearshore regions and the fishery retracted. While previous research has focused on the causes for the declining numbers of adult lobsters in the region as well as lobster settlement, lobster larvae were not sampled in Rhode Island waters. Additionally research was mixed on the origin of lobster larvae found inshore. To examine larval presence in coastal Rhode Island, weekly net tows were conducted at the mouth of Narragansett Bay during summer from 2019 to 2021, and light traps were deployed in 2020 and 2021. Time series segmentation and generalized linear models assessed trends and connections between lobsters caught in trawl surveys, commercial landings data, lobster settlement data, bottom water temperatures, and climate indices. Larval supply to Narragansett Bay is very low and mostly concentrated in the East Passage where ocean water flows into the bay, suggesting that a failure in larval supply from offshore sources is harming local population levels. Time series segmentation and models indicate that declines in lobster settlement followed the decline in adult lobsters in coastal Rhode Island, not Narragansett Bay. Altogether, the data suggest that larval lobster are transported inshore from spawning stock outside the bay. Present circulation patterns combined with the movement of lobsters into deeper waters has made the transport of larvae to Narragansett Bay unlikely.

Unsupervised Time Series Segmentation: A Survey on Recent Advances

Unsupervised Time Series Segmentation: A Survey on Recent Advances

On the Improvements of Metaheuristic Optimization-Based Strategies for Time Series Structural Break Detection

Structural break detection is an important time series analysis task. It can be treated as a multi-objective optimization problem, in which we ought to find a time series segmentation such that time series theoretical models constructed on each segment are well-fitted and the segments are long enough to bear meaningful information. Metaheuristic optimization can help us solve this problem. This paper introduces a suite of new cost functions for the structural break detection task. We demonstrate that the new cost functions allow for achieving quantitatively better precision than the cost functions employed in the literature of this domain. We show particular advantages of each new cost function. Furthermore, the paper promotes the use of Particle Swarm Optimization (PSO) in the domain of structural break detection, which so far has relied on the Genetic Algorithm (GA). Our experiments show that PSO outperforms GA for many analysed time series examples. Last but not least, we introduce a non-trivial generalization of the top-performing state-of-the-art approach to the structural break detection problem based on the Minimum Description Length (MDL) rule with autoregressive (AR) model to MDL ARIMA (autoregressive integrated moving average) model.

ОБНАРУЖЕНИЕ РАЗЛАДОК В НЕСТАЦИОНАРНЫХ РЯДАХ НАБЛЮДЕНИЙ: ОБЗОР ТЕХНОЛОГИЙ И КРИТИЧЕСКИЙ АНАЛИЗ

A critical analysis of modern mathematical technologies used to solve the problem of nonstationary time series change point, taking into account the chaotic nature of the observed processes, is considered. The known methods of the quickest change point detection in nonstationary time series described by traditional models of random processes are considered as initial assumptions. The discrepancy between the known time series observation models and the real processes occurring in unstable immersion environments described by the concept of stochastic chaos is shown. Time series segmentation technologies based on a posteriori analysis of changes in the properties of observation sites of chaotic processes are presented. Options for building online algorithms for the quickest change point detection based on methods of multiregressive data analysis and metric machine learning technologies are considered.

Detection of Obstructive Sleep Apnoea Using Features Extracted From Segmented Time-Series ECG Signals With a One Dimensional Convolutional Neural Network

Detection of Obstructive Sleep Apnoea Using Features Extracted From Segmented Time-Series ECG Signals With a One Dimensional Convolutional Neural Network

Research and Application of Deep Learning Selection Method for Magnetotelluric Power Spectrum Based on Rhoplus Correction

Magnetotelluric (MT) is an electromagnetic exploration method using natural field sources, which are unstable, irregular, and vulnerable to interference from electromagnetic signals. The total power spectrum of the magnetotelluric signal is superimposed by a sub-power spectrum of segmented time series. To reduce the influence of electromagnetic interference, the pre-processing of magnetotelluric data will delete the severely interfered sub-power spectrum. The selection of power spectrum often requires human-computer interaction, which is time-consuming and laborious, heavily dependent on personal experience, and has great randomness. This article innovation is put forward based on the classification of Rhoplus correction and deep learning automatic power spectrum method, first of all, the MT power spectrum impedance estimation results are Rhoplus one dimensional inversion and obtain fitting, using neural network learning fitting differential characteristics and classification, and then for the power spectrum of MT again after low-quality quantum power spectrum to eliminate overlay, Impedance estimation is performed, and the above process is repeated to find the optimal power spectrum combination. Through the comparison and verification of measured MT data, the automatic power spectrum selection method based on Rhoplus correction and deep learning classification achieve similar results to manual editing.

An air combat maneuver pattern extraction based on time series segmentation and clustering analysis

Target maneuver recognition is a prerequisite for air combat situation awareness, trajectory prediction, threat assessment and maneuver decision. To get rid of the dependence of the current target maneuver recognition method on empirical criteria and sample data, and automatically and adaptively complete the task of extracting the target maneuver pattern, in this paper, an air combat maneuver pattern extraction based on time series segmentation and clustering analysis is proposed by combining autoencoder, G-G clustering algorithm and the selective ensemble clustering analysis algorithm. Firstly, the autoencoder is used to extract key features of maneuvering trajectory to remove the impacts of redundant variables and reduce the data dimension; Then, taking the time information into account, the segmentation of Maneuver characteristic time series is realized with the improved FSTS-AEGG algorithm, and a large number of maneuver primitives are extracted; Finally, the maneuver primitives are grouped into some categories by using the selective ensemble multiple time series clustering algorithm, which can prove that each class represents a maneuver action. The maneuver pattern extraction method is applied to small scale air combat trajectory and can recognize and correctly partition at least 71.3% of maneuver actions，indicating that the method is effective and satisfies the requirements for engineering accuracy. In addition, this method can provide data support for various target maneuvering recognition methods proposed in the literature, greatly reduce the workload and improve the recognition accuracy.

109 Animal Behavior Analysis Using Unsupervised Machine Learning Techniques

Abstract With recent technological advances, Global Positioning Systems (GPS) and other tracking sensors can collect large amounts of data and transfer information to the cloud and then to producers to remotely monitor livestock health and well-being. Currently, supervised machine learning, such as random forest and linear or quadratic discriminant analysis techniques are used to develop algorithms to identify changes in animal behavior that may be associated with well-being concerns. Supervised machine learning requires behavior observation of monitored animals and may hinder normal expression. However, recording behavioral observations is time consuming and expensive. Our goal was to design a new unsupervised machine learning framework to identify animal behavior without the utilization of human observations. The framework contains two steps. The first step is to segment the tracking data of the animal using time series segmentation, and the second step is to group the segments into clusters where each cluster represents one type of behavior. To validate the applicability of our proposed framework, we utilize GPS tracking data collected at 2-minute intervals from eight cows from May 28 to June 22, 2018, in a 1096 ha rangeland pasture near Prescott, Arizona. After extensive experiments, our framework can partition the movement of the cow using the speed, direction, and distance of the cow from water. These segments were grouped into meaningful behavior clusters using clustering analysis. Speed was the most successful feature for clustering into behaviors. Results are similar to approaches based only on expert knowledge, that rely on speed for classification. Our study demonstrated that we can directly use unlabeled data to group animal behaviors from GPS tracking data. The proposed unsupervised two-step framework allows the analysis of cattle tracking data without direct human observation of behaviors. It is applicable for analyzing the immense amount of data that can be obtained from real-time tracking and sensor devices.

Early Warning of Loss and Kick for Drilling Process Based on Sparse Autoencoder With Multivariate Time Series

Complicated geological environments lead to a high risk of drilling incidents. Early warning of loss and kick for the drilling process is essential to ensure process safety. On account of the nonlinear and temporal correlation of drilling parameters, an early warning method for loss and kick based on sparse autoencoder with multivariate time series is proposed. The sparse autoencoder is utilized for multivariate time series abnormality detection of the drilling process. Abnormal drilling parameter isolation is performed through contribution analysis. Reconstruction analysis and time series segmentation approaches are integrated for abnormal time series trend evaluation. The characteristic of drilling parameters under normal operation learned by the sparse autoencoder and the property of the original time series are taken into account. The final early warning result can be obtained through expert rules based on the trend evaluation result. Case studies are presented based on the data from an actual drilling project. The experiment result shows the effectiveness of the proposed method.

Sequence signal reconstruction based multi-task deep learning for sleep staging on single-channel EEG

The temporal context information between sleep stage sequence contains sleep transition rules, which is important for improving sleep staging performance. Existing multi-task learning methods reconstruct EEG signals from one sleep stage, ignoring the importance of sequential temporal context in capturing long-term dependencies to enhance representation learning. To address these issues, we propose a multi-task deep learning model to jointly reconstruct sequence signal and segment time series. The model enhances the ability of time series segmentation task to capture sequential temporal context and improves the performance of single-channel EEG by optimizing the common encoder for sequence signal reconstruction task. In addition, we design a one-dimensional channel attention module to enhance the feature representation extracted for sleep sequence signal. The experimental results on four datasets show that the multi-task deep learning model can improve the generalization using sequence signal reconstruction. Compared with other state-of-the-art methods, the method proposed in this study obtained competitive performance in terms of metrics such as accuracy, which is 85.6% on the 2013 version of Sleep-EDF Database Expanded, 83.4% on the 2018 version of Sleep-EDF Database Expanded, 85.6% on Sleep Heart Health Study, and 77.4% on CAP Sleep Database.

A station-keeping maneuver detection method of non-cooperative geosynchronous satellites

Orbit maneuver detection is one of the major concerns for the work of Space Situation Awareness of GEO protected region. In this study, we propose a non-cooperative GEO satellites station-keeping maneuver detection approach based on the two-dimensional Convolutional Neural Network (CNN) and the public Two-Line Element (TLE). The proposed method focuses on identifying “station-keeping categories” using the time series segments of TLE data. The time series TLE of the selected objects are converted into several equal-sized segments using a window slicing method. These segments are then used to train a two-dimensional CNN classification model that can divide the input data into three categories of maneuver. The applicability of the proposed method on satellites in the same orbital region were further tested using other objects that had not been included during the modeling process. Experimental results demonstrated the feasibility and robustness of the proposed method, with a classification macro F1-score of 94.45%, macro precision of 94.48%, and macro recall of 94.46%. The results suggest that the proposed method provides a promising solution for detecting GEO station-keeping maneuvers.

Energy demand forecasting using adaptive ARFIMA based on a novel dynamic structural break detection framework

Forecasting energy demand has become increasingly important due to technological advances, especially new power systems and population growth. Accurate predictions of energy demand and supply are crucial for academics and policymakers. Also, energy consumption is non-stationary and dynamic in time, requiring an adaptive forecasting algorithm. This paper presents a new adaptive hybrid approach for energy time series forecasting using statistical methods. Some of the most popular energy demand time series applications are used to demonstrate the superiority of the proposed algorithm, which covers a wide range of critical energy demand forecasting situations at various scales. The paper's objectives are presented in three phases. In Phase I, ARFIMA is used to forecast the entire original time series as a statistical inspector and later for comparison. Although ARFIMA's fractionality feature helps to capture long and short-term memory behavior in time series, to be more adaptive and become close to a nonlinear algorithm that can handle non-stationarity, and Gaussian and non-Gaussian time series closely, a novel dynamic statistical structural break detection framework is developed, and used in phase II to identify the time series' change points and associated time indexes. Since time series are partitioned based on structural change detections, they approach one in terms of the Hurst exponent and exhibit pure long memory, which is ideal for ARFIMA modeling. Phase III captures non-stationary statistical properties and memory characteristics by applying adaptive ARFIMA modeling to segmented time series. Finally, the predicted partitions are concatenated. Various performance assessment metrics, e.g., Mean Absolute Percentage Error (MAPE), tested on the final achieved results show that the proposed adaptive algorithm outperforms the forecasting capabilities of the existing algorithms on the same energy demand time series used in the current paper. MAPE of the proposed approach for the four utilized real-life cases are 1.72%, 1.35%, 0.22%, and 4.4%, respectively. Therefore, the proposed model is a satisfactory method for energy demand forecasting due to its high accuracy.

Weighted Dynamic Time Warping for Time Series

Recurrence network is a typical time series analysis method. However, irregular sampling may overshadow the dynamic features characterized by traditional recurrence network method, which makes the method ineffective. This paper introduces dynamic time warping method to determine the distance between time series segments. The method can match the features of time series segments and reduce the distortion of distances caused by irregular sampling. For points in sparse sampling area, many characteristics may not be collected and thus the matching results are unreliable. To overcome this problem, we further propose weighted dynamic time warping method which assigns small weights to matching points with sparse local sampling so as to reduce their influence on distance. A recurrence network constructed by weighted dynamic time warping method can effectively capture underlying nonlinear and nonstationary system dynamics from irregular sampling time series. The effectiveness of the proposed method is illustrated by two case studies: the discrimination of different dynamic behaviors and detection of system parameter perturbation from time series of Rössler system.

Deep adaptive temporal network (DAT-Net): an effective deep learning model for parameter estimation of radar multipath interference signals

Accurate parameter estimation in radar systems is critically hindered by multipath interference, a challenge that is amplified in complex and dynamic environments. Traditional methods for parameter estimation, which concentrate on single parameters and rely on statistical assumptions, often struggle in such scenarios. To address this, the deep adaptive temporal network (DAT-Net), an innovative deep learning model designed to handle the inherent complexities and non-stationarity of time series data, is proposed. In more detail, DAT-Net integrates both the pruned exact linear time method for effective time series segmentation and the exponential scaling-based importance evaluation algorithm for dynamic learning of importance weights. These methods enable the model to adapt to shifts in data distribution and provide a robust solution for parameter estimation. In addition, DAT-Net demonstrates the capability to comprehend inherent nonlinearities in radar multipath interference signals, thereby facilitating the modeling of intricate patterns within the data. Extensive validation experiments conducted across parameter estimation tasks and demonstrates the robust applicability and efficiency of the proposed DAT-Net model. The architecture yield root mean squared error scores as low as 0.0051 for single-parameter estimation and 0.0152 for multiple-parameter estimation.

Time series analysis of daily reported number of new positive cases of COVID-19 in Japan from January 2020 to February 2023.

This study investigated temporal variations of the COVID-19 pandemic in Japan using a time series analysis incorporating maximum entropy method (MEM) spectral analysis, which produces power spectral densities (PSDs). This method was applied to daily data of COVID-19 cases in Japan from January 2020 to February 2023. The analyses confirmed that the PSDs for data in both the pre- and post-Tokyo Olympics periods show exponential characteristics, which are universally observed in PSDs for time series generated from nonlinear dynamical systems, including the so-called susceptible/exposed/infectious/recovered (SEIR) model, well-established as a mathematical model of temporal variations of infectious disease outbreaks. The magnitude of the gradient of exponential PSD for the pre-Olympics period was smaller than that of the post-Olympics period, because of the relatively high complex variations of the data in the pre-Olympics period caused by a deterministic, nonlinear dynamical system and/or undeterministic noise. A 3-dimensional spectral array obtained by segment time series analysis indicates that temporal changes in the periodic structures of the COVID-19 data are already observable before the commencement of the Tokyo Olympics and immediately after the introduction of mass and workplace vaccination programs. Additionally, the possibility of applying theoretical studies for measles control programs to COVID-19 is discussed.