Time Series Segmentation Research Articles

Unsupervised Retrieval Based Multivariate Time Series Anomaly Detection and Diagnosis with Deep Binary Coding Models

Retrieval based multivariate time series anomaly detection and diagnosis refer to identifying abnormal status in certain time steps and pinpointing the root cause input variables, i.e., sensors, by comparing a current time series segment and its relevant ones that are retrieved from huge amount of historical data. Binary coding with a deep neural network can be applied to reduce the computational cost of the retrieval tasks. However, it is hard to pinpoint the root cause sensors that are responsible for the anomaly, once multivariate time series segments are transformed into binary codes. In this paper, we present an unsupervised retrieval based multivariate time series anomaly detection and diagnosis method with deep binary coding model, to secure both efficiency and explainability. Specifically, we first transform input multivariate time series segments into low dimensional features with a temporal encoder. Subsequently, two hash functions predict two binary codes with different lengths from each feature. The binary codes with two different lengths can contribute to accelerate both anomaly detection and anomaly diagnosis. Experiments performed on datasets from various domains including real optical network, demonstrate the effectiveness and efficiency of the proposed method.

Anomaly prediction of CT equipment based on IoMT data

BackgroundLarge-scale medical equipment, which is extensively implemented in medical services, is of vital importance for diagnosis but vulnerable to various anomalies and failures. Most hospitals that conduct regular maintenance have been suffering from medical equipment-related incidents for years. Currently, the Internet of Medical Things (IoMT) has emerged as a crucial tool in monitoring the real-time status of the medical equipment. In this paper, we develop an IoMT system of Computed Tomography (CT) equipment in the West China Hospital, Sichuan University and collected the system status time-series data. Novel multivariate time-series classification models and frameworks are proposed to predict the anomalies of CT equipment. The important features that are closely related to the equipment anomalies are identified with the model.MethodsWe extracted the real-time CT equipment status time-series data of 11 equipment between May 19, 2020 and May 19, 2021 from the IoMT, which includes the equipment oil temperature, anode voltage, etc. The arcs are identified as labels of anomalies due to their relationship with decreased imaging quality and CT equipment failures. To improve prediction accuracy, the statistics and transformations of the raw historical time-series data segment in the sliding time window are used to construct new features. Due to the particularity of time-series data, two frameworks are proposed for splitting the training and test sets. Then the Decision Tree, Support Vector Machine, Logistic Regression, Naive Bayesian, and K-Nearest Neighbor classification models are used to classify the system status. We also compare our model to state-of-the-art models.ResultsThe results show that the anomaly prediction accuracy and recall of our method are 79% and 77%, respectively. The oil temperature and anode voltage are identified as the decisive features that may lead to anomalies. The proposed model outperforms the others when predicting the anomalies of the CT equipment based on our dataset.ConclusionsThe proposed method could predict the state of CT equipment and be used as a reference for practical maintenance, where unexpected anomalies of medical equipment could be reduced. It also brings new insights into how to handle non-uniform and imbalanced time series data in practical cases.

Arousal detection by using ultra-short-term heart rate variability (HRV) analysis

Introduction: The study of arousal is crucial as it helps to understand the role of increased physiological and psychological activation in emotions, motivation, cognitive performance, stress responses, sleep-wake cycles, and clinical application. Recent studies have shown that arousal is commonly associated with physiological changes including heart rate variability (HRV) as indicated by RR intervals. In some applications, the analysis requires analyzing short segments of RR time series, shorter than the usual 5 min HRV. The objective of this study is to check the performance of ultra-short-term HRV indices to track changes in arousal.Method: In this study, to follow arousal changes, 31 healthy subjects were examined in both non-arousal (relaxed) and aroused states. Two states of 5 minutes each are used to measure the relaxed and arousal states. After data collection, RR time series segments were obtained randomly for each subject in arousal and relaxed states in the 30s, 60s, 120s, and 240s time windows. Next, 17 ultra-short-term HRV indices were computed for each time window for RR intervals in relaxed and aroused states.Results and Discussion: Due to the findings, novel indices such as ACI and fnQ may aid in the recognition of arousal from relaxed status. The odds of ACI being higher for the same subject during a randomly selected arousal interval than during a randomly selected relax interval are 78%, 79%, 84%, and 89% for the 30s, 60s, 120s, and 240s time windows respectively. Similarly, the odds of fnQ being higher during arousal than during a relaxed state are 79%, 81%, 84%, and 85% for the 30s, 60s, 120s, and 240s time windows respectively. Therefore, ACI and fnQ provided the best performances in intra-individual arousal detection by using ultra-short-term HRV analysis among all of the obtained indices. Nevertheless, when pooling the indices for all the subjects, the inter-subject variability causes a moderate classification performance for all indices. In this case, the best performing index is fnQ with an area under the receiver operator curve (AUC) of 75%, 77%, 79%, and 80% for the 30s, 60s, 120s, and 240s time windows respectively.

A statistical examination of utilization trends in decentralized applications

Decentralized applications (dApp) have proliferated in recent years, but their long-term viability is a topic of debate. However, for dApps to be sustainable, and suitable for integration into a larger service networks, they need to attract users and promise reliable availability. Therefore, assessing their longevity is crucial. Analyzing the utilization trajectory of a service is, however, challenging due to several factors, such as demand spikes, noise, autocorrelation, and non-stationarity. In this study, we employ robust statistical techniques to identify trends in currently popular dApps. Our findings demonstrate that a significant proportion of dApps, across a range of categories, exhibit statistically significant positive overall trends, indicating that success in decentralized computing can be sustainable and transcends specific fields. However, there is also a substantial number of dApps showing negative trends, with a disproportionately high number from the decentralized finance (DeFi) category. Furthermore, a more detailed inspection of time series segments shows a clearly diminishing proportion of positive trends from mid-2021 to the present. In summary, we conclude that the dApp economy might have lost some momentum, and that there is a strong element of uncertainty regarding its future significance.

National increase in the community supply of take‐home naloxone associated with a mass media campaign in Scotland: a segmented time series analysis

BackgroundTake-home naloxone (THN) programmes have been associated with reductions in opioid-related mortality. In response to high rates of drug-related deaths in Scotland, the Scottish Government commissioned the ‘How to save a life’ (HTSAL) mass media campaign to: (1) increase awareness of drug-related deaths and how to respond to an overdose, and (2) increase the supply of THN. The aim of this study was to assess the effect of the campaign on the supply of THN. MethodsWe used an interrupted time series design to assess the effect of the HTSAL mass media campaign on the national community supply of THN. The study time period was August 2020–December 2021. We modelled two key dates: the start of the campaign (week beginning (w/b) 30th of August 2021) and after the end of the main campaign (w/b 25th of October 2021). ResultsThe total number of THN kits distributed in the community in Scotland during the study period was 27,064. The mean number of THN kits distributed per week (relative to the pre-campaign period), increased by 126% during the campaign and 57% post-campaign. In segmented regression analyses, the pre-campaign trend in the number of THN kits supplied was increasing by an average of 1% each week (RR=1.01, 95% CI 1.01 to 1.01, p<0.001). Once the campaign started, a significant change in level was observed, and the number of kits increased by 75% (RR=1.75, 95% CI 1.29 to 2.40, p<0.001). The trend during the campaign was stable (i.e. not increasing or decreasing) but a significant change in level was observed when the campaign ended, and the number of THN kits supplied decreased by 32% (RR=0.68, 95% CI 0.46 to 0.98, p = 0.042). The trend during the post-campaign period was stable. ConclusionsThe HTSAL campaign had a short term, but large and significant impact, on the community supply of THN in Scotland. Mass media campaigns could be combined with other interventions and strategies to maintain the increased uptake of THN outside of campaign periods.

An intelligent method for TBM surrounding rock classification based on time series segmentation of rock-machine interaction data

Accurate, continuous real-time perception of surrounding rock conditions on the tunnel boring machine (TBM) tunnel face provides particularly important prior information to ensure safe, economic, and efficient tunneling. This study presents an intelligent method for assessing the TBM surrounding rock conditions using a combination of time series segmentation and a deep clustering model (SSAE-FCM). In data preprocessing, fifteen operational parameters were selected as input features. And an adaptive piecewise constant approximation method was employed for the segmentation and compression of stable phase data. Then, a total of 69,214 data samples that can characterize the stable rock-machine interaction were obtained to construct a dataset. Thereafter, the SSAE-FCM model was established and trained to classify the surrounding rock conditions into four clusters. The distribution result of the main operational parameters and composite indexes (TPI, FPI, and SE) within these clusters demonstrated that the classification results are compatible with rock-machine interaction in engineering practice. Additionally, the rock mass assessment performance of the classification results was compared to BQ classification in terms of various rock mass parameters (UCS, Jv), fault sections, and collapse sections. These results indicate that the proposed classification method accurately reflects the stability and boreability of rock mass and is more sensitive to unfavorable geology than the BQ classification. From the three clustering evaluation indexes (SC, CHI, and DBI) perspective, SSAE-FCM was superior to the other clustering models (BIRCH + k-means++, FCM, and SSAE + FCM). In conclusion, the proposed intelligent method for TBM surrounding rock classification is reasonable and applicable for the evaluation of rock mass quality, as well as accurate and sensitive in real-time perception.

Similarity Measurement and Classification of Temporal Data Based on Double Mean Representation

Time series data typically exhibit high dimensionality and complexity, necessitating the use of specific approximation methods to perform computations on the data. The currently employed compression methods suffer from varying degrees of feature loss, leading to potential distortions in similarity measurement results. Considering the aforementioned challenges and concerns, this paper proposes a double mean representation method, SAX-DM (Symbolic Aggregate Approximation Based on Double Mean Representation), for time series data, along with a similarity measurement approach based on SAX-DM. Addressing the trade-off between compression ratio and accuracy in the improved SAX representation, SAX-DM utilizes the segment mean and the segment trend distance to represent corresponding segments of time series data. This method reduces the dimensionality of the original sequences while preserving the original features and trend information of the time series data, resulting in a unified representation of time series segments. Experimental results demonstrate that, under the same compression ratio, SAX-DM combined with its similarity measurement method achieves higher expression accuracy, balanced compression rate, and accuracy, compared to SAX-TD and SAX-BD, in over 80% of the UCR Time Series dataset. This approach improves the efficiency and precision of similarity calculation.

A novel feature engineering approach for high-frequency financial data

Feature engineering for high-frequency financial data based on constructing dynamic data subsets, defined by time intervals in which high-frequency trends occur, is proposed. These intervals are obtained through time series segmentation. This methodology allows us to extract and analyze variables by intraday trends as well as to feed artificial intelligence models to forecast response variables in future trends. Furthermore, to show how to use this feature engineering, this methodology is applied to estimate high-frequency volatility, duration and direction linked to future intraday trends, developing multiclass classification models based on the machine learning method extreme gradient boosting. Experimentation was conducted using high-frequency financial data from the Brazil Stock Exchange, corresponding to 206 trading days related to 20 listed assets from this financial market.

Considering the self-adaptive segmentation of time series in interval prediction of remaining useful life for lithium-ion battery

In order to solve the limitations of numerical prediction in RUL prediction of lithium-ion battery, this paper studies the essence of interval prediction, and proposes a feasible strategy to achieve the RUL interval prediction of lithium-ion battery. In order to make the proposed interval prediction strategy applicable to degradation data sets with different change states, a self-adaptive time series segmentation algorithm is proposed. The algorithm can identify and distinguish smooth degradation stages and obvious fluctuation stages of the degradation series, so as to retain the fluctuation information of the original degradation data as much as possible in the subsequent processing, and is also adaptive for different battery degradation data sets. Firstly, the proposed self-adaptive segmentation algorithm is used to segment the original capacity degradation data, and then the segmentation results are processed by fuzzy information granulation into granule sequences with upper and lower limits. Finally, the least squares support vector machine is used to model the granule data to realize interval prediction. Four groups of battery data sets are used to verify the feasibility and effectiveness of the proposed self-adaptive segmentation algorithm, and the determination of an important parameter in this algorithm is also discussed.

ℓ0 Trend Filtering

The [Formula: see text] trend filtering ([Formula: see text]-TF) is a new effective tool for nonparametric regression with the power of automatic knot detection in function values or derivatives. It overcomes the drawback of [Formula: see text]-TF that is known to have bias issues. To solve the [Formula: see text]-TF problem, we propose an alternating minimization induced active set (AMIAS) search method based on the necessary optimality conditions derived from an augmented Lagrangian framework. The proposed method takes full advantage of the primal and dual variables with complementary supports, and decouples the high-dimensional problem into two subsystems on the active and inactive sets, respectively. A sequential AMIAS algorithm with warm start initialization is developed for efficient determination of the cardinality parameter, along with the output of solution paths. Theoretically, the oracle estimator of [Formula: see text]-TF is justified to behave like regression splines under the continuous time setting with mild conditions. Our numerical experiments include simulation studies for comparing [Formula: see text]-TF to [Formula: see text]-TF and free-knot splines on several synthetic examples, and a real data application of time series segmentation on Hong Kong PM2.5 indexes. History: Accepted by Antonio Frangioni, Area Editor for Design &amp; Analysis of Algorithms – Continuous. Funding: This work was supported in part by Hong Kong General Research Fund [No. 17306519]. C. Wen’s research is partially supported by National Science Foundation of China [12171449] and Fundamental Research Funds for the Central Universities [WK3470000027, YD2040002019]. X. Wang’s research is partially supported by National Natural Science Foundation of China [Grants 72171216, 12231017, 71921001, and 71991474], and the National Key R&amp;D Program of China [No. 2022YFA1003803]. Supplemental Material: The e-companion is available at https://doi.org/10.1287/ijoc.2021.0313 . The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2021.0313 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2021.0313 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

Ultra-short-term multi-step probability interval prediction of photovoltaic power: A framework with time-series-segment feature analysis

Power prediction can effectively mitigate the uncertainty in photovoltaic power generation, enabling better operation and scheduling of power grids. Therefore, in this study, a multi-step interval prediction method for ultra-short-term photovoltaic power from time-series-segment (TSS) feature analysis is proposed. First, three TSS indicators are defined to determine the fluctuation characteristics of historical data and combined with fuzzy C-means clustering to address the time mismatch problem. Subsequently, a deterministic multi-step prediction method is proposed based on the optimal membership search using deep recurrent neural networks, improving the prediction stability. Finally, based on the difference in the TSS types, an improved interval prediction method is proposed in combination with Gaussian process regression, narrowing the average interval width. Experiments are conducted to compare the performances of the conventional and proposed methods using measured data from Australia. Compared with the baseline scheme, the proposed scheme enhances the accuracy of multi-step prediction by 19.7%, and the average error of each step does not exceed 5%.The average interval width is reduced by 45.6% while ensuring more than 95% interval coverage in the probability interval prediction. The experimental findings demonstrate that the TSS feature analysis can effectively reveal the potential patterns of PV power output under various weather conditions. This enables the algorithm to learn clearer sample features and thus enhances the performance of multi-step probability interval prediction.

Automatic IHR-based sleep stage detection using features of residual neural network

Untreated sleep disorders can harm bodily functions, and a sleep study and monitoring of sleep stages are the first steps in diagnosing these disorders. Using Polysomnography (PSG), signal scoring for sleep stage determination has become a familiar investigation in recent years. Despite its effectiveness, the procedure is time-consuming and costly. This study presents a cost-effective method for sleep classification based on Electrocardiogram (ECG) input signals. We proposed a multi-ethnic study of the Atherosclerosis dataset, including 1700 PSG, to develop a Residual Neural Network (RNN) classifier to stage sleep from Instantaneous Heart Rate (IHR) extracted from the ECG signals. The proposed system follows the following steps: ECG collection, signal preprocessing (including ECG normalization and segmentation, instant heart rate calculation and normalization, resampling, and filtering), and classification using an RNN. A Convolutional Neural Network (CNN) is used to detect sleep stages using preprocessed segments of the IHR time series of 240 samples centered on 30-s epochs as inputs. The proposed algorithm in the five-fold cross-validation achieved an accuracy of 85.32%, a kappa of 77.11%, a Sensitivity of 81.14%, a Specificity of 82.68%, and an F-1 score of 81.87%. The results show that ECG data provide valuable information about sleep stages for a large population.

A Total Variation Based Method for Multivariate Time Series Segmentation

A Total Variation Based Method for Multivariate Time Series Segmentation

A Multivariate Time-Series Segmentation Framework for Flight Condition Recognition

Helicopters usage monitoring has gained significant attention in recent years, due to the safety and cost management implications. At its core there is the Flight Condition Recognition algorithm, which enables to detect the maneuvers carried out by the aircraft through on-board sensors measurements. In this work we propose a multi-variate time-series segmentation framework, which uses supervised learning algorithms, sliding windows and stacking ensembles to produce reliable estimates of the flown flight regimes. We validate the proposed approach on a large dataset of 460 labeled load flights from two distinct helicopter models, demonstrating its efficacy in predicting a range of 49 different maneuver types.

Multivariate count time series segmentation with “sums and shares” and Poisson lognormal mixture models: a comparative study using pedestrian flows within a multimodal transport hub

Multivariate count time series segmentation with “sums and shares” and Poisson lognormal mixture models: a comparative study using pedestrian flows within a multimodal transport hub

Ordinal Poincaré sections: Reconstructing the first return map from an ordinal segmentation of time series.

We propose a robust algorithm for constructing first return maps of dynamical systems from time series without the need for embedding. A first return map is typically constructed using a convenient heuristic (maxima or zero-crossings of the time series, for example) or a computationally nuanced geometric approach (explicitly constructing a Poincaré section from a hyper-surface normal to the flow and then interpolating to determine intersections with trajectories). Our method is based on ordinal partitions of the time series, and the first return map is constructed from successive intersections with specific ordinal sequences. We can obtain distinct first return maps for each ordinal sequence in general. We define entropy-based measures to guide our selection of the ordinal sequence for a "good" first return map and show that this method can robustly be applied to time series from classical chaotic systems to extract the underlying first return map dynamics. The results are shown for several well-known dynamical systems (Lorenz, Rössler, and Mackey-Glass in chaotic regimes).

Latent space unsupervised semantic segmentation.

The development of compact and energy-efficient wearable sensors has led to an increase in the availability of biosignals. To effectively and efficiently analyze continuously recorded and multidimensional time series at scale, the ability to perform meaningful unsupervised data segmentation is an auspicious target. A common way to achieve this is to identify change-points within the time series as the segmentation basis. However, traditional change-point detection algorithms often come with drawbacks, limiting their real-world applicability. Notably, they generally rely on the complete time series to be available and thus cannot be used for real-time applications. Another common limitation is that they poorly (or cannot) handle the segmentation of multidimensional time series. Consequently, the main contribution of this work is to propose a novel unsupervised segmentation algorithm for multidimensional time series named Latent Space Unsupervised Semantic Segmentation (LS-USS), which was designed to easily work with both online and batch data. Latent Space Unsupervised Semantic Segmentation addresses the challenge of multivariate change-point detection by utilizing an autoencoder to learn a 1-dimensional latent space on which change-point detection is then performed. To address the challenge of real-time time series segmentation, this work introduces the Local Threshold Extraction Algorithm (LTEA) and a "batch collapse" algorithm. The "batch collapse" algorithm enables Latent Space Unsupervised Semantic Segmentation to process streaming data by dividing it into manageable batches, while Local Threshold Extraction Algorithm is employed to detect change-points in the time series whenever the computed metric by Latent Space Unsupervised Semantic Segmentation exceeds a predefined threshold. By using these algorithms in combination, our approach is able to accurately segment time series data in real-time, making it well-suited for applications where timely detection of changes is critical. When evaluating Latent Space Unsupervised Semantic Segmentation on a variety of real-world datasets the Latent Space Unsupervised Semantic Segmentation systematically achieves equal or better performance than other state-of-the-art change-point detection algorithms it is compared to in both offline and real-time settings.

Evaluation of duration of antibiotic therapy across hospitals in Scotland including the impact of COVID-19 pandemic: a segmented interrupted time series analysis

Background Little is known about the duration of antibiotic use in hospital settings. We evaluated the duration of hospital antibiotic therapy (as a quality indicator proxy) for four commonly prescribed antibiotics (amoxicillin, co-amoxiclav, doxycycline and flucloxacillin) including the assessment of COVID-19 impact. Methods A repeated, cross-sectional study using the Hospital Electronic Prescribing and Medicines Administration system (January/2019-March/2022). Monthly median duration of therapy/duration categories were calculated, stratified by routes of administration, age and sex. Impact of COVID-19 was assessed using segmented time-series analysis. Results There were significant variations in the median duration of therapy across routes of administration (P<0.05), with the highest value among those antibiotic courses comprised of both oral and IV antibiotics (“Both” group). Significantly higher proportions of prescriptions within the “Both” group had a duration of > 7 days compared to oral or IV. Duration of therapy overall differed significantly by age. Some small statistically significant changes in the level/trends of duration of therapy were observed in the post- COVID-19 period. Conclusions No evidence for prolonged duration of therapy were observed, even during COVID-19 pandemic. Duration of IV therapy was relatively short suggesting timely clinical review and consideration of IV to oral switch. Longer duration of therapy was observed among older patients.

Industrial Transfer Learning for Multivariate Time Series Segmentation: A Case Study on Hydraulic Pump Testing Cycles.

Industrial data scarcity is one of the largest factors holding back the widespread use of machine learning in manufacturing. To overcome this problem, the concept of transfer learning was developed and has received much attention in recent industrial research. This paper focuses on the problem of time series segmentation and presents the first in-depth research on transfer learning for deep learning-based time series segmentation on the industrial use case of end-of-line pump testing. In particular, we investigate whether the performance of deep learning models can be increased by pretraining the network with data from other domains. Three different scenarios are analyzed: source and target data being closely related, source and target data being distantly related, and source and target data being non-related. The results demonstrate that transfer learning can enhance the performance of time series segmentation models with respect to accuracy and training speed. The benefit can be most clearly seen in scenarios where source and training data are closely related and the number of target training data samples is lowest. However, in the scenario of non-related datasets, cases of negative transfer learning were observed as well. Thus, the research emphasizes the potential, but also the challenges, of industrial transfer learning.

PrecTime: A deep learning architecture for precise time series segmentation in industrial manufacturing operations

The fourth industrial revolution creates ubiquitous sensor data in production plants. To generate maximum value out of these data, reliable and precise time series-based machine learning methods like temporal neural networks are needed. This paper proposes a novel sequence-to-sequence deep learning architecture for time series segmentation called PrecTime which tries to combine the concepts and advantages of sliding window and dense labeling approaches. The general-purpose architecture is evaluated on a real-world industry dataset containing the End-of-Line testing sensor data of hydraulic pumps. We are able to show that PrecTime outperforms five implemented state-of-the-art baseline networks based on multiple metrics. The achieved segmentation accuracy of around 96% shows that PrecTime can achieve results close to human intelligence in operational state segmentation within a testing cycle.