Search

Time series anomaly detection is a task of significant importance and has been widely employed in realistic scenarios. Most of existing methods conduct time series anomaly detection in an unsupervised manner, ignoring the limited number of labeled anomalies that are commonly available in practical situations. However, how to take advantage of these limited but valuable labeled anomalies to benefit time series anomaly detection requires fully exploration. To this end, we propose a weakly supervised time series anomaly detection method with dual-association discrepancy to effectively identify anomalies. Specifically, the proposed method utilizes the limited number of anomalies to enlarge the distance between the association discrepancy of the anomaly and that of the normal one, enforcing significant differences between abnormal and normal samples in discriminant places. We also design a masking strategy to enrich the representations of anomalies in latent feature space. Experiments with public available datasets have demonstrated the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Time series anomaly detection is crucial in various industrial applications to identify unusual behaviors within the time series data. Due to the challenges associated with annotating anomaly events, time series reconstruction has become a prevalent approach for unsupervised anomaly detection. However, effectively learning representations and achieving accurate detection results remain challenging due to the intricate temporal patterns and dependencies in real-world time series. In this paper, we propose a cross-dimension attentive feature fusion network for time series anomaly detection, referred to as CAFFN. Specifically, a series and feature mixing block is introduced to learn representations in 1D space. Additionally, a fast Fourier transform is employed to convert the time series into 2D space, providing the capability for 2D feature extraction. Finally, a cross-dimension attentive feature fusion mechanism is designed that adaptively integrates features across different dimensions for anomaly detection. Experimental results on real-world time series datasets demonstrate that CAFFN performs better than other competing methods in time series anomaly detection. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Harbin Institute of Technology. Social Sciences Edition / Haerbin Gongye Daxue Xuebao. Shehui Kexue Ban is the property of Harbin Institute of Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

This paper presents an extensive empirical study on the integration of dimensionality reduction techniques with advanced unsupervised time series anomaly detection models, focusing on the MUTANT and Anomaly-Transformer models. The study involves a comprehensive evaluation across three different datasets: MSL, SMAP, and SWaT. Each dataset poses unique challenges, allowing for a robust assessment of the models’ capabilities in varied contexts. The dimensionality reduction techniques examined include PCA, UMAP, Random Projection, and t-SNE, each offering distinct advantages in simplifying high-dimensional data. Our findings reveal that dimensionality reduction not only aids in reducing computational complexity but also significantly enhances anomaly detection performance in certain scenarios. Moreover, a remarkable reduction in training times was observed, with reductions by approximately 300% and 650% when dimensionality was halved and minimized to the lowest dimensions, respectively. This efficiency gain underscores the dual benefit of dimensionality reduction in both performance enhancement and operational efficiency. The MUTANT model exhibits notable adaptability, especially with UMAP reduction, while the Anomaly-Transformer demonstrates versatility across various reduction techniques. These insights provide a deeper understanding of the synergistic effects of dimensionality reduction and anomaly detection, contributing valuable perspectives to the field of time series analysis. The study underscores the importance of selecting appropriate dimensionality reduction strategies based on specific model requirements and dataset characteristics, paving the way for more efficient, accurate, and scalable solutions in anomaly detection.

This paper introduces a novel anomaly detection approach tailored for time series data with exclusive reliance on normal events during training. Our key innovation lies in the application of kernel-density estimation (KDE) to scrutinize reconstruction errors, providing an empirically derived probability distribution for normal events post-reconstruction. This non-parametric density estimation technique offers a nuanced understanding of anomaly detection, differentiating it from prevalent threshold-based mechanisms in existing methodologies. In post-training, events are encoded, decoded, and evaluated against the estimated density, providing a comprehensive notion of normality. In addition, we propose a data augmentation strategy involving variational autoencoder-generated events and a smoothing step for enhanced model robustness. The significance of our autoencoder-based approach is evident in its capacity to learn normal representation without prior anomaly knowledge. Through the KDE step on reconstruction errors, our method addresses the versatility of anomalies, departing from assumptions tied to larger reconstruction errors for anomalous events. Our proposed likelihood measure then distinguishes normal from anomalous events, providing a concise yet comprehensive anomaly detection solution. The extensive experimental results support the feasibility of our proposed method, yielding significantly improved classification performance by nearly 10% on the UCR benchmark data.

Welding defect detection in the manufacturing of hot water tanks is still often performed by human visual inspection or with the help of classical non-destructive tests such as liquid penetrant. These quality inspection methods can be time-consuming and have an important error rate. As a potential alternative, machine learning algorithms can be employed to automate fault detection in this process. In this paper, we propose an approach for the detection of welding faults by identifying abnormal subsequences of the welding voltage signal. The novelty of our approach is in employing the one-class SVM with distance substitution kernels, which, unlike previous works employing the one-class SVM, allows us to work directly with the raw subsequences. This permits easier generalization of the detection across different types of signals and achieves higher detection accuracy. The results show that the approach is both accurate and fast compared to existing approaches, which makes it more suitable for real-time welding monitoring. We further propose an approach for the automatic diagnosis of welding defects using the random convolutional kernel transform. [ABSTRACT FROM AUTHOR]

Anomaly detection of liquid levels in molds is an important task in continuous casting. Data that consists of a series of liquid levels in mold during a continuous casting process can be viewed as a time series, on which Time Series Anomaly Detection (TSAD) methods can be applied. However, the abnormal and normal data in the liquid data in the mold sequence share similar features. And due to manual control limitations, the anomaly sequence in liquid level in mold data lasts longer. Therefore, using existing TSAD methods based on AutoEncoders (AEs) often results in high false positive rates. In this paper, a novel framework is proposed for anomaly detection of liquid level in mold by using unsupervised deep-learning-based TSAD. The framework decomposes a time series into normal and error sequences. A forecasting network reconstructs the normal sequence to solve the first issue, which allows the proposed method to consider the context. An error extraction network generates errors from the original sequence to solve the second issue. It removes anomalies from the original sequence during training to prevent anomaly pollution and allows the forecasting network's training to be free from anomaly pollution. A new dynamic threshold method is proposed to identify anomalies. The proposed method is evaluated on the actual casting dataset by comparing it with baseline methods. The experiment results indicate that the proposed framework outperforms some of the best anomaly detection methods in terms of accuracy, precision, and F1 score. [ABSTRACT FROM AUTHOR]

In this study, we present and implement the SAnDet (SDN anomaly detector) architecture, an anomaly-based intrusion detection system designed to take advantage of the capabilities offered by software-defined networking (SDN) architecture, as a controller application. The SAnDet system is composed of three modules: statistics collection, anomaly detection, and anomaly prevention. In particular, we utilize replicator neural networks (RNN), which is a specialized variant of the autoencoder, and the LSTM-based encoder–decoder (EncDecAD) method, which is a special type of long short-term memory (LSTM) network that has demonstrated a strong performance on data series particularly, to identify unknown attacks using flow features collected from OpenFlow switches. In our experiments, we utilize flow-based features extracted from network traffic data containing various types of attacks as input to our models in the form of time series. We evaluate the performance of our methods using the accuracy and area under the receiver operating characteristic curve (AUC) metrics. Our experimental results demonstrate that EncDecAD outperforms RNN and that our approach offers several benefits over previously conducted research. [ABSTRACT FROM AUTHOR]

Anomaly detection is a significant task in sensors' signal processing since interpreting an abnormal signal can lead to making a high-risk decision in terms of sensors' applications. Deep learning algorithms are effective tools for anomaly detection due to their capability to address imbalanced datasets. In this study, we took a semi-supervised learning approach, utilizing normal data for training the deep learning neural networks, in order to address the diverse and unknown features of anomalies. We developed autoencoder-based prediction models to automatically detect anomalous data recorded by three electrochemical aptasensors, with variations in the signals' lengths for particular concentrations, analytes, and bioreceptors. Prediction models employed autoencoder networks and the kernel density estimation (KDE) method for finding the threshold to detect anomalies. Moreover, the autoencoder networks were vanilla, unidirectional long short-term memory (ULSTM), and bidirectional LSTM (BLSTM) autoencoders for the training stage of the prediction models. However, the decision-making was based on the result of these three networks and the integration of vanilla and LSTM networks' results. The accuracy as a performance metric of anomaly prediction models showed that the performance of vanilla and integrated models were comparable, while the LSTM-based autoencoder models showed the least accuracy. Considering the integrated model of ULSTM and vanilla autoencoder, the accuracy for the dataset with the lengthier signals was approximately 80%, while it was 65% and 40% for the other datasets. The lowest accuracy belonged to the dataset with the least normal data in its dataset. These results demonstrate that the proposed vanilla and integrated models can automatically detect abnormal data when there is sufficient normal data for training the models. [ABSTRACT FROM AUTHOR]

Consider the online testing of a stream of hypotheses where a real-time decision must be made before the next data point arrives. The error rate is required to be controlled at all decision points. Conventional simultaneous testing rules are no longer applicable due to the more stringent error constraints and absence of future data. Moreover, the online decision-making process may come to a halt when the total error budget, or alpha-wealth, is exhausted. This work develops a new class of structure-adaptive sequential testing (SAST) rules for online false discovery rate (FDR) control. A key element in our proposal is a new alpha-investing algorithm that precisely characterizes the gains and losses in sequential decision making. SAST captures time varying structures of the data stream, learns the optimal threshold adaptively in an ongoing manner and optimizes the alpha-wealth allocation across different time periods. We present theory and numerical results to show that SAST is asymptotically valid for online FDR control and achieves substantial power gain over existing online testing rules. [ABSTRACT FROM AUTHOR]

Recently, Transformers and MLPs based models have dominated and made significant progress in time series analysis. However, these methods struggle to capture the complete periodic features to model the global dependencies of time series. To this end, we provide a new perspective to explore the potential of time series in the frequency domain. Specifically, we propose the DFT-based A mplitude- P hase D ecoupling Network (APDNet), which learns the significance changes and periodicity characteristics of time series by decoupling the real and imaginary parts in the complex frequency domain. First, we propose the Fourier Temporal/Variable Interaction Module to model time dependency and dependency multiple variables, respectively. Secondly, we innovatively propose the Dimensional Expanded Feature Embedding Module (DEFEM), which expands time series into a higher-dimensional representation space to preserve more independent features. Finally, through extensive experiments on multiple real-world datasets, our APDNet achieves state-of-the-art performance in various tasks such as long-term and short-term forecasting, anomaly detection, imputation, and classification. Code is available at: https://github.com/CodeJester196/APDNet. [ABSTRACT FROM AUTHOR]

This paper investigates the use of an unsupervised hybrid statistical–local outlier factor algorithm to detect anomalies in time-series flight data. Flight data analysis is an activity carried out by airlines primarily as a means of improving the safety and operation of their fleet. Traditionally, this is performed by checking exceedances in pre-set limits to the flight data parameters. However, this method highlights single events during a flight, making this analysis laborious. The process also fails to establish trends or reflect potential unknown hazards. This research took advantage of machine learning techniques to recognize patterns in large datasets by implementing the local outlier factor (LOF). In order to minimize human input, a statistical approach was adopted to establish the threshold value above which the flights are considered to be anomalous and interpret the scores. This paper shows that LOF quantifies the degree of outlier-ness of an outlier rather than binary categorizing a point into inlier or outlier, as in the case of clustering algorithms. Thus, with LOF, for the first time, we demonstrated that in the aviation industry, anomalous flights could not only be identified but also be given an anomaly score to compare two anomalous flights in an unsupervised manner. Furthermore, LOF helps to track anomalous behavior in time during the flight. This is insightful when a flight is abnormal, only for some seconds or short duration. For the first time, we attempted to detect flight parameters responsible for anomalous behavior or at least give direction to human experts looking for the cause of abnormal behavior. This was all analyzed with real-life flight data in an unsupervised manner in contrast to simulated data. [ABSTRACT FROM AUTHOR]

Anomaly detection of liquid levels in molds is an important task in continuous casting. Data that consists of a series of liquid levels in mold during a continuous casting process can be viewed as a time series, on which Time Series Anomaly Detection (TSAD) methods can be applied. However, the abnormal and normal data in the liquid data in the mold sequence share similar features. And due to manual control limitations, the anomaly sequence in liquid level in mold data lasts longer. Therefore, using existing TSAD methods based on AutoEncoders (AEs) often results in high false positive rates. In this paper, a novel framework is proposed for anomaly detection of liquid level in mold by using unsupervised deep-learning-based TSAD. The framework decomposes a time series into normal and error sequences. A forecasting network reconstructs the normal sequence to solve the first issue, which allows the proposed method to consider the context. An error extraction network generates errors from the original sequence to solve the second issue. It removes anomalies from the original sequence during training to prevent anomaly pollution and allows the forecasting network’s training to be free from anomaly pollution. A new dynamic threshold method is proposed to identify anomalies. The proposed method is evaluated on the actual casting dataset by comparing it with baseline methods. The experiment results indicate that the proposed framework outperforms some of the best anomaly detection methods in terms of accuracy, precision, and F1 score.

Context—Anomaly detection in a data center is a challenging task, having to consider different services on various resources. Current literature shows the application of artificial intelligence and machine learning techniques to either log files or monitoring data: the former created by services at run time, while the latter produced by specific sensors directly on the physical or virtual machine. Objectives—We propose a model that exploits information both in log files and monitoring data to identify patterns and detect anomalies over time both at the service level and at the machine level. Methods—The key idea is to construct a specific dictionary for each log file which helps to extract anomalous n-grams in the feature matrix. Several techniques of Natural Language Processing, such as wordclouds and Topic modeling, have been used to enrich such dictionary. A clustering algorithm was then applied to the feature matrix to identify and group the various types of anomalies. On the other side, time series anomaly detection technique has been applied to sensors data in order to combine problems found in the log files with problems stored in the monitoring data. Several services (i.e., log files) running on the same machine have been grouped together with the monitoring metrics. Results—We have tested our approach on a real data center equipped with log files and monitoring data that can characterize the behaviour of physical and virtual resources in production. The data have been provided by the National Institute for Nuclear Physics in Italy. We have observed a correspondence between anomalies in log files and monitoring data, e.g., a decrease in memory usage or an increase in machine load. The results are extremely promising. Conclusions—Important outcomes have emerged thanks to the integration between these two types of data. Our model requires to integrate site administrators' expertise in order to consider all critical scenarios in the data center and understand results properly. [ABSTRACT FROM AUTHOR]

• Novel forecasting-based time series anomaly detection framework with three components. • Two components are evolved from existing methods but more considerate and powerful while the rest one is completely novel . • Better performances than fifteen baselines on ten benchmarks and diverse settings Forecasting-based method is one of prevalent unsupervised time series anomaly detection approaches. Currently, large portions of existing forecasting-based methods are devoted to discussing the feature extraction of input sequences and targeting at accurate predictions. In essence, their frameworks and core ideas are identical to the pure forecasting models. However, the distinctiveness of anomalies is affected by not only forecasting accuracy, but also many other factors. This paper summarizes three other dominant factors: (1) Scale disparity ; (2) Discrete variate ; (3) Input anomaly. They are common and non-negligible in real-world anomaly detection. Moreover, we propose AFMF: a time series A nomaly detection F ramework with M odified F orecasting to solve them respectively by its three key components, i.e., Local Instance Normalization, Lopsided Forecasting and Progressive Adjacent Masking. The first two are refined descendants of existing mechanisms while the third component is completely novel. Extensive experiments on ten benchmarks verify that AFMF can be combined with any forecasting or forecasting-based anomaly detection method to achieve SOTA anomaly detection performances. The source code is available at https://github.com/OrigamiSL/AFMF. [ABSTRACT FROM AUTHOR]

Multivariate time series usually have entangled temporal patterns and various anomaly types. Meanwhile, they often contain both continuous and discrete features. Many existing methods directly model correlations in complex multivariate time series to conduct anomaly detection. Decomposing time series into different components, such as the overall trend and fluctuations, can contribute to better extracting semantic information and detecting anomalies. Existing decomposition-based anomaly detection methods still have several limitations. First, they directly decompose all features without considering that discrete features are unsuitable for decomposition because they do not have trends or fluctuations. Second, they adopt the same networks for different components with different characteristics, limiting their ability to extract semantic information. Moreover, due to the nature of Transformers, existing reconstruction-based methods using Transformers rarely form information bottlenecks, reducing the differentiation between the reconstruction errors of normal data and anomalies. This paper proposes a multivariate time series anomaly detection method with separation, decomposition, and dual Transformer-based autoencoder (SDDformer). Different from existing methods, SDDformer separates continuous and discrete features and only decomposes continuous features into trend and residual components. Considering the different characteristics of different components, SDDformer adopts Crossformer and the vanilla Transformer as the backbone of two different autoencoders to reconstruct the trend and residual components. Information bottlenecks are better formed using an extra token as the latent variable between the encoder and the decoder. SDDformer regards reconstructing a discrete feature as a classification task and calculates Cross-Entropy as its reconstruction error. Three different metrics are adopted in this paper to compare SDDformer with a variety of typical anomaly detection methods on public data sets, and the experimental results prove that SDDformer can achieve state-of-the-art performance. • A two-stream multivariate time series anomaly detection method is proposed. • Continuous and discrete features are separated before decomposition. • Only continuous features are decomposed into the trend and residual components. • Two different autoencoders are proposed to reconstruct different components. • The reconstruction of a discrete feature is regarded as a classification task. [ABSTRACT FROM AUTHOR]

Anomaly detection is a significant task in sensors’ signal processing since interpreting an abnormal signal can lead to making a high-risk decision in terms of sensors’ applications. Deep learning algorithms are effective tools for anomaly detection due to their capability to address imbalanced datasets. In this study, we took a semi-supervised learning approach, utilizing normal data for training the deep learning neural networks, in order to address the diverse and unknown features of anomalies. We developed autoencoder-based prediction models to automatically detect anomalous data recorded by three electrochemical aptasensors, with variations in the signals’ lengths for particular concentrations, analytes, and bioreceptors. Prediction models employed autoencoder networks and the kernel density estimation (KDE) method for finding the threshold to detect anomalies. Moreover, the autoencoder networks were vanilla, unidirectional long short-term memory (ULSTM), and bidirectional LSTM (BLSTM) autoencoders for the training stage of the prediction models. However, the decision-making was based on the result of these three networks and the integration of vanilla and LSTM networks’ results. The accuracy as a performance metric of anomaly prediction models showed that the performance of vanilla and integrated models were comparable, while the LSTM-based autoencoder models showed the least accuracy. Considering the integrated model of ULSTM and vanilla autoencoder, the accuracy for the dataset with the lengthier signals was approximately 80%, while it was 65% and 40% for the other datasets. The lowest accuracy belonged to the dataset with the least normal data in its dataset. These results demonstrate that the proposed vanilla and integrated models can automatically detect abnormal data when there is sufficient normal data for training the models.

A key performance indicator (KPI) consists of critical time series data that reflect the runtime states of network systems (e.g., response time and available bandwidth). Despite the importance of KPI, datasets for KPI anomaly detection available to the public are very limited, due to privacy concerns and the high overhead in manually labelling the data. The insufficiency of public KPI data poses a great barrier for network researchers and practitioners to evaluate and test what-if scenarios in the development of artificial intelligence for IT operations (AIOps) and anomaly detection algorithms. To tackle the difficulty, we develop a univariate time series generation tool called TSAGen, which can generate KPI data with anomalies and controllable characteristics for KPI anomaly detection. Experiment results show that the data generated by TSAGen can be used for comprehensive evaluation of anomaly detection algorithms with diverse user-defined what-if scenarios. [ABSTRACT FROM AUTHOR]

This paper investigates the use of an unsupervised hybrid statistical–local outlier factor algorithm to detect anomalies in time-series flight data. Flight data analysis is an activity carried out by airlines primarily as a means of improving the safety and operation of their fleet. Traditionally, this is performed by checking exceedances in pre-set limits to the flight data parameters. However, this method highlights single events during a flight, making this analysis laborious. The process also fails to establish trends or reflect potential unknown hazards. This research took advantage of machine learning techniques to recognize patterns in large datasets by implementing the local outlier factor (LOF). In order to minimize human input, a statistical approach was adopted to establish the threshold value above which the flights are considered to be anomalous and interpret the scores. This paper shows that LOF quantifies the degree of outlier-ness of an outlier rather than binary categorizing a point into inlier or outlier, as in the case of clustering algorithms. Thus, with LOF, for the first time, we demonstrated that in the aviation industry, anomalous flights could not only be identified but also be given an anomaly score to compare two anomalous flights in an unsupervised manner. Furthermore, LOF helps to track anomalous behavior in time during the flight. This is insightful when a flight is abnormal, only for some seconds or short duration. For the first time, we attempted to detect flight parameters responsible for anomalous behavior or at least give direction to human experts looking for the cause of abnormal behavior. This was all analyzed with real-life flight data in an unsupervised manner in contrast to simulated data.

• An anomaly detection framework was first adopted for wearable gait analysis. • A semi-supervised model was proposed for lameness detection. • Dairy cow gaits were reconstructed by using LSTM-Autoencoder. • The LSTM-Autoencoder has a high accuracy for lameness detection. Early lameness detection is crucial to ensure the welfare and productivity of dairy cows. However, current research on early lameness identification using wearable analysis relies on the limited robustness of indirect behavioral measures, which are susceptible to individual variations and imbalances in lameness samples. In this study, we propose a semi-supervised Long short-term memory (LSTM)-Autoencoder algorithm for early lameness detection in dairy cows through time series data reconstruction. We collected gait data from all four limbs of 30 dairy cows using four IMUs. A LSTM-Autoencoder with three LSTM hidden layers was trained to learn the time series features of healthy gaits. Each gait was reconstructed, and anomaly gaits exceeding a threshold were identified by comparing reconstructed gaits with actual gaits. The gait symmetry was measured by comparing the percentage of anomaly gait between opposite limbs as an indicator of lameness severity. With a high accuracy of 97.78% and a true negative rate of 98.33%, our integrated approach outperforms traditional methods in early lameness detection and lame limb identification, enabling real-time monitoring and timely identification of lameness. The study is the first attempt at using a time series anomaly detection framework with deep learning-based gait reconstruction for lameness detection. Wearable gait analysis offers portability and real-time capabilities, providing continuous, accurate, and comprehensive gait information unaffected by lighting and field-of-view limitations. This approach holds promise for enhancing animal welfare and optimizing management practices in the dairy industry through timely identification and continuous monitoring of lameness. [ABSTRACT FROM AUTHOR]

Time-series anomaly detection is a task of detecting data that do not follow normal data distribution among continuously collected data. It is used for system maintenance in various industries; hence, studies on time-series anomaly detection are being carried out actively. Most of the methodologies are based on Long Short-Term Memory (LSTM) and Convolution Neural Network (CNN) to model the temporal structure of time-series data. In this study, we propose an unsupervised prediction-based time-series anomaly detection methodology using Transformer, which shows superior performance to LSTM and CNN in learning dynamic patterns of sequential data through a self-attention mechanism. The prediction model consists of an encoder comprising multiple Transformer encoder layers and a decoder that includes a 1D convolution layer. The output representation of each Transformer layer is accumulated in the encoder to obtain a representation with multi-level, rich information. The decoder fuses this representation through a 1d convolution operation. Consequently, the model can perform predictions considering both the global trend and local variability of the input time-series. The anomaly score is defined as the difference between the predicted and the actual value at the corresponding timestamp, assuming that the trained model produces the predictions that follow the normal data distribution. Finally, the data with an anomaly score above the threshold is detected as an anomaly. Experiments on the benchmark datasets show that the proposed method has performance superior to those of the baselines. • A prediction-based unsupervised time-series anomaly detection method using Transformer is proposed. • The proposed method can detect anomalies, considering both the global trend and the local variability of the given time-series. • The proposed method achieves superior performance compared to the baselines. • The proposed method is more robust to the data characteristics. [ABSTRACT FROM AUTHOR]

Time-series discord is widely used in data mining applications to characterize anomalous subsequences in time series. Compared to some other discord search algorithms, the direct search algorithm based on the recurrence plot shows the advantage of being fast and parameter free. The direct search algorithm, however, relies on quasi-periodicity in input time series, an assumption that limits the algorithm's applicability. In this paper, we eliminate the periodicity assumption from the direct search algorithm by proposing a reference function for subsequences and a new sampling strategy based on the reference function. These measures result in a new algorithm with improved efficiency and robustness, as evidenced by our empirical evaluation. [ABSTRACT FROM AUTHOR]

Time series anomaly detection has become a crucial and challenging task driven by the rapid increase of streaming data with the arrival of the Internet of Things. Existing methods are either domain-specific or require strong assumptions that cannot be met in realistic datasets. Reinforcement learning (RL), as an incremental self-learning approach, could avoid the two issues well. However, the current investigation is far from comprehensive. In this paper, we propose a generic policy-based RL framework to address the time series anomaly detection problem. The policy-based time series anomaly detector (PTAD) is progressively learned from the interactions with time-series data in the absence of constraints. Experimental results show that it outperforms the value-based temporal anomaly detector and other state-of-the-art detection methods whether training and test datasets come from the same source or not. Furthermore, the tradeoff between precision and recall is well respected by the PTAD, which is beneficial to fulfill various industrial requirements. [ABSTRACT FROM AUTHOR]