Representation Of Time Series Data Research Articles

BIMO: Bootstrap Inter–Intra Modality at Once Unsupervised Learning for Multivariate Time Series

It is difficult to learn meaningful representations of time-series data since they are sparsely labeled and unpredictable. Hence, we propose bootstrap inter–intra modality at once (BIMO), an unsupervised representation learning method based on time series. Unlike previous works, the proposed BIMO method learns both inter-sample and intra-temporal modality representations simultaneously without negative pairs. BIMO comprises a main network and two auxiliary networks, namely inter-auxiliary and intra-auxiliary networks. The main network is trained to learn inter–intra modality representations sequentially by regulating the use of each auxiliary network dynamically. Thus, BIMO thoroughly learns inter–intra modality representations simultaneously. The experimental results demonstrate that the proposed BIMO method outperforms the state-of-the-art unsupervised methods and achieves comparable performance to existing supervised methods.

Multi-task self-supervised time-series representation learning

Time-series representation learning is crucial for extracting meaningful representations from time-series data with temporal dynamics and sparse labels. Contrastive learning, a powerful technique for exploiting the inherent data patterns, has been applied to explore the diverse consistencies in time-series data, achieved through careful selection of contrastive pairs and design of appropriate loss. Encouraging such consistency is essential for acquiring comprehensive representations of time-series data. In this paper, we propose a new framework for time-series representation learning that combines the advantages of contextual, temporal, and transformation consistencies. This framework enables the network to learn general representations suitable for different tasks and domains. First, positive and negative pairs are generated to establish a multi-task learning setup. Then, contrastive losses are formulated to explore contextual, temporal, and transformation consistencies, which are jointly optimized to learn general time-series representations. In addition, we investigate an uncertainty weighting approach to enhance the effectiveness of multi-task learning. To evaluate the performance of our framework, we conduct experiments on three downstream tasks: time-series classification, forecasting, and anomaly detection. The experimental results demonstrate the superior performance of our framework compared to benchmark models across different tasks. Furthermore, our framework shows efficiency in cross-domain transfer learning scenarios.

Spatial–Temporal Fusion Gated Transformer Network (STFGTN) for Traffic Flow Prediction

Traffic flow prediction is essential for smart city management and planning, aiding in optimizing traffic scheduling and improving overall traffic conditions. However, due to the correlation and heterogeneity of traffic data, effectively integrating the captured temporal and spatial features remains a significant challenge. This paper proposes a model spatial–temporal fusion gated transformer network (STFGTN), which is based on an attention mechanism that integrates temporal and spatial features. This paper proposes an attention mechanism-based model to address these issues and model complex spatial–temporal dependencies in road networks. The self-attention mechanism enables the model to achieve long-term dependency modeling and global representation of time series data. Regarding temporal features, we incorporate a time embedding layer and a time transformer to learn temporal dependencies. This capability contributes to a more comprehensive and accurate understanding of spatial–temporal dynamic patterns throughout the entire time series. As for spatial features, we utilize DGCN and spatial transformers to capture both global and local spatial dependencies, respectively. Additionally, we propose two fusion gate mechanisms to effectively accommodate to the complex correlation and heterogeneity of spatial–temporal information, resulting in a more accurate reflection of the actual traffic flow. Our experiments on three real-world datasets illustrate the superior performance of our approach.

Learning reservoir dynamics with temporal self-modulation

Reservoir computing (RC) can efficiently process time-series data by mapping the input signal into a high-dimensional space via randomly connected recurrent neural networks (RNNs), which are referred to as a reservoir. The high-dimensional representation of time-series data in the reservoir simplifies subsequent learning tasks. Although this simple architecture allows fast learning and facile physical implementation, the learning performance is inferior to that of other state-of-the-art RNN models. In this study, to improve the learning ability of RC, we propose self-modulated RC (SM-RC) that extends RC by adding a self-modulation mechanism. SM-RC can perform attention tasks where input information is retained or discarded depending on the input signal. We find that a chaotic state can emerge as a result of learning in SM-RC. Furthermore, we demonstrate that SM-RC outperforms RC in NARMA and Lorenz model tasks. Because the SM-RC architecture only requires two additional gates, it is physically implementable as RC, thereby providing a direction for realizing edge artificial intelligence.

Development of a CNN-LSTM Approach with Images as Time-Series Data Representation for Predicting Gold Prices

Predicting gold prices is not easy due to its non-linear, unpredictable, volatile, and uncontrollable price movements. In this research, a combination of convolutional neural network (CNN) and long short-term memory (LSTM) is used to predict gold prices. By combining these two methods, the prediction model can leverage the strengths of CNN and LSTM to improve accuracy and learning performance. In addition, this CNN-LSTM model is enriched with input in the form of images that represent the timeseries data, where the gramian angular field (GAF) technique is used in timeseries data to images transformation. Experimental results showed that the proposed approach performs significantly better compared to the benchmark model.

Advancing chipboard milling process monitoring through spectrogram-based time series analysis with Convolutional Neural Network using pretrained networks

This paper presents a novel approach to enhance chipboard milling process monitoring in the furniture manufacturing sector using Convolutional Neural Networks (CNNs) with pretrained architectures like VGG16, VGG19, and RESNET34. The study leverages spectrogram representations of time-series data obtained during the milling process, providing a unique perspective on tool condition monitoring. The efficiency of the CNN models in accurately classifying tool conditions into distinct states (`Green', `Yellow', and `Red') based on wear levels is thoroughly evaluated. Experimental results demonstrate that VGG16 and VGG19 achieve high accuracy, however with longer training times, while RESNET34 offers faster training at the cost of reduced precision. This research not only highlights the potential of pretrained CNNs in industrial applications but also opens new avenues for predictive maintenance and quality control in manufacturing, underscoring the broader applicability of AI in industrial automation and monitoring systems.

GLOBE: A novel pruning-based sparse modeling with application to time series data representation

In the past years, Sparse Modeling (SM) has been used in a variety of applications, such as data representation. In this type of data modeling, the data is divided into small segments, and each one is processed independently using SM techniques to find the best few coefficients that capture the most important features of the segment. This paper proposes a backward stage for SM to process all segments at the same time. The main idea is to select significant coefficients that capture the global features of the data, not just the features of the small segments. To evaluate the effectiveness of the proposed algorithm, extensive experiments were conducted on different time series data. The results showed that the proposed algorithm works better with time series data that has localized features, such as speech data. The study also compared the proposed algorithm to other recent backward techniques and found that it outperforms the others significantly, especially when using learned bases.

Study and Analyze the Time Series of Daily Prices Oil World by Using ARIMA(p,d,q) Models

The research aims to use the ARIMA models to study and analyze the time series of Daily Prices Oil World 2010-2013 to find the best model for prediction. The results of the application showed that the appropriate and efficient model for the representation of time series data is the ARIMA (2.1.1). According to the estimation results of this model, the prediction results time series of Daily Prices Oil World for the period 2010-2013 were consistent with those of the original time series

A Complete Air Pollution Monitoring and Prediction Framework

Forecasting air pollution is crucial for understanding the phenomenological and contextual variety of mechanisms underlying pollution in a particular area or region. Analyzing high-dimensional data with spatial and temporal dependencies is pointed out as a major challenge for traditional machine learning approaches for air pollution forecasting. The unprecedented advances in deep learning employed on massive quantities of IoT sensor data raise high hopes for the future of air pollution forecasting. Drawing on past and current experiences and solutions advanced in a growing body of research on the topic, we proposed and evaluated four encoder-decoder architectures with attention for forecasting particulate matter (PM) levels, with a general applicability that is both location- and season-independent. The attention mechanism was employed to better learn representation of time series data. By selectively focusing the attention on elements that contribute the most, superior performance gains have been obtained by the proposed models. Sensor measurements obtained in the past 7 or 14 days as well a set of external factors known to affect the pollutant levels, such as weather conditions, location and time frame (e.g., season, time of day) were taken into account. In this research, we carried out a case study on PM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2.5</sub> forecasting for the city of Skopje, allowing us to discuss the relevance of the results obtained by the proposed solutions. The ability to draw valid inferences from data has critical importance when forecasting air pollution. In this respect, any air pollution forecasting model based on deep learning would require a component for generation of realistic data samples to augment the time series dataset. We address the challenges in missing data by proposing and evaluating two adversarial networks for data augmentation. We have conducted a number of experiments to investigate the performance of the predictive models with and without augmenting the training datasets and obtain superior performance gains by the proposed adversarial models for data augmentation. The deep neural architectures advanced in this research are general enough to be used for predictive and generative tasks for other pollutants; moreover, they could be adopted for related tasks handling time series data in other domains.

An algorithm to improve the performance of convolutional neural networks for TSC tasks

AbstractThe convolutional neural network (CNN) is an important network model for processing time series classification tasks. However, there can be a lot of noise in the data itself. The clarity of spatial feature expression of time‐series samples and the freeness and balance of data samples both have a big effect on how well CNNs can handle time series classification (TSC) tasks. This paper proposes a multi‐pose marking learning algorithm (MPML), which optimizes the spatial representation of data samples through multiple graphical representations of time series data and then solves the problems of freeness and heterogeneity of data samples through labeling and preferential learning of poorly trained sample data. The algorithm improves the model's robustness to noise by improving the spatial expression sentiment of the time series and the labeling learning of the feature data, thereby improving the convolutional neural network's performance to perform the TSC task. We show how well the algorithm works by testing it on 46 datasets from the UCR Time Series Classification Archive and comparing the results.

Image-based time series forecasting: A deep convolutional neural network approach

Inspired by the successful use of deep learning in computer vision, in this paper we introduce ForCNN, a novel deep learning method for univariate time series forecasting that mixes convolutional and dense layers in a single neural network. Instead of using conventional, numeric representations of time series data as input to the network, the proposed method considers visual representations of it in the form of images to directly produce point forecasts. Three variants of deep convolutional neural networks are examined to process the images, the first based on VGG-19, the second on ResNet-50, while the third on a self-designed architecture. The performance of the proposed approach is evaluated using time series of the M3 and M4 forecasting competitions. Our results suggest that image-based time series forecasting methods can outperform both standard and state-of-the-art forecasting models.

Fault signal reconstruction for multi-sensors in gas turbine control systems based on prior knowledge from time series representation

Improving efficiency through intelligence is the current development trend of industrial gas turbines. Among the fault statistics of gas turbines, the number of sensor fault is the highest during use. The fault signal diagnosis and reconstruction are of great significance to the efficient and safe operation of gas turbines. In order to eliminate the sensor fault signal and transmit the normal signal to the control system, a multivariate fault signal reconstruction method based on the prior knowledge of the time-series representation was proposed in this work. The proposed multivariate signal reconstruction method can reconstruct almost all fault cases with high accuracy by training only one model. Firstly, the prior knowledge is applied to improve the conventional time series data representation. Secondly, three steps are employed to utilize spatial or temporal information and obtain three intermediate data. The masks combine the third intermediate data and the original time series to obtain the final reconstruction results. Then, reconstruction data sets are built based on exhaust gas temperatures (EGTs) from real-world power plant to verify the effectiveness of the proposed method. Different evaluation metrics and visualization reveal the high accuracy results. Compared results with different reconstruction algorithms reflect both the robustness and high speed of this reconstruction method. Finally, three typical fault signal reconstruction cases reveal the generalizability of this model.

Attention-CNN 을 이용한 유도탄 비행데이터의 이상성 검출 기법 연구

This study proposes a new deep learning-based anomaly detection algorithm in a supervised framework for the Monte-Carlo (MC) simulation data of a missile control system. First, in a similar fashion to the famous BERT(Bidirectional Encoder Representations from Transformers) algorithm in NLP(Natural Language Programming), a pre-trained transformer-based encoder is utilized to provide an alternative representation of time-series data, replacing conventional data-preprocessing. For a more direct perception of differences in normal and abnormal data, a multi-head attention score is also considered as an input of a two-dimensional convolutional neural network classifier along with the output of transformer encoder. The training of the proposed framework is comprised of two steps: unsupervised pre-training of the transformer encoder with all generated MC simulation data and supervised training of the classifier with labelled data. To generate the MC simulation data, the high-fidelity missile 6DOF simulation program is developed with random variables that represent the uncertainties of realistic scenarios. The resulting anomaly detection performance from the generated MC simulation dataset indicates that the proposed framework provides significantly better detection performance than other supervised machine/deep learning-based algorithms. Furthermore, using a multi-head attention score from the transformer encoder as an input of the classifier not only improves the detection performance but also contributes to reducing the necessity of labelled train data. Hence, the proposed method has the potential to greatly reduce the human effort in detecting anomalies in MC simulation data.

Writer-independent online signature verification based on 2D representation of time series data using triplet supervised network

Online signature verification (OSV) as a critical link of the paperless office still has tremendous challenges. In this paper, we propose a novel Writer-Independent (WI) OSV framework that includes three parts. (1) The two-dimensional (2D) representation method is designed to transform the original time-series signature data into stroke images with dynamic-static hybrid information. (2) The Channel-wise Weight Learning (CWL) mechanism is integrated into the feature extractor to mine the potential relationship between three dynamic attributes (altitude, azimuth, and pressure). (3) A new Triplet Supervised Network (TSN) that contains three weight-shared convolution neural network (CNN) streams was provided to measure the distance of [Anchor, Positive, Test]. The experimental results show that our model has at least 1.29% accuracy improvement when confronted with skilled forgery signatures than that of other classical CNNs and 11.43% than lightweight models. Moreover, the TSN model is superior to the previous OSV algorithms in the WI pattern.

Graft: A graph based time series data mining framework

Rapid technology integration causes a high dimensional time series data accumulation in multiple domains and applying the classical data mining tools and techniques becomes a challenging task. Hence, the time series data representation have gained popularity over the years, which ease the task of mining, analysis and visualization. Graph based representation is one such emerging tool in which the time series data is represented as nodes and edges of graph. The current graph based representation is designed either to mine the motif or discords from a single time series or cluster the time series where each node represents a time series sample. Such representation technique causes information loss and also no further analysis could be performed other than clustering. To address these challenges, we propose a unique graph representation for time series dataset that works on multiple domains. Novelty of the graph representation is that it is unique for multiple time series and it acts as a framework for whole time series clustering, temporal pattern extraction from each cluster and temporally dependent rare event discovery. A new research direction for the proposed graph based framework is shown. Comparative analysis reveal the superiority of the proposed framework particularly as a clustering technique. The key contributions of the paper are: (i) transformation strategy of time series database from time domain to graph structure in topological domain (ii) time series clustering using path level analysis (iii) identification of temporally dependent co-occurring patterns (iv) rare event detection using component level analysis

Multiresolution dendritic cell algorithm for network anomaly detection.

Anomaly detection in computer networks is a complex task that requires the distinction of normality and anomaly. Network attack detection in information systems is a constant challenge in computer security research, as information systems provide essential services for enterprises and individuals. The consequences of these attacks could be the access, disclosure, or modification of information, as well as denial of computer services and resources. Intrusion Detection Systems (IDS) are developed as solutions to detect anomalous behavior, such as denial of service, and backdoors. The proposed model was inspired by the behavior of dendritic cells and their interactions with the human immune system, known as Dendritic Cell Algorithm (DCA), and combines the use of Multiresolution Analysis (MRA) Maximal Overlap Discrete Wavelet Transform (MODWT), as well as the segmented deterministic DCA approach (S-dDCA). The proposed approach is a binary classifier that aims to analyze a time-frequency representation of time-series data obtained from high-level network features, in order to classify data as normal or anomalous. The MODWT was used to extract the approximations of two input signal categories at different levels of decomposition, and are used as processing elements for the multi resolution DCA. The model was evaluated using the NSL-KDD, UNSW-NB15, CIC-IDS2017 and CSE-CIC-IDS2018 datasets, containing contemporary network traffic and attacks. The proposed MRA S-dDCA model achieved an accuracy of 97.37%, 99.97%, 99.56%, and 99.75% for the tested datasets, respectively. Comparisons with the DCA and state-of-the-art approaches for network anomaly detection are presented. The proposed approach was able to surpass state-of-the-art approaches with UNSW-NB15 and CSECIC-IDS2018 datasets, whereas the results obtained with the NSL-KDD and CIC-IDS2017 datasets are competitive with machine learning approaches.

WaDGAN-AD를 이용한 전력 소비 패턴의 비지도 학습 기반 이상 탐지

Anomaly detection in time series is essential because it can detect outlying patterns such as a breakdown in machines and fraudulent customers. Among many anomaly detection domains, detecting abnormal patterns in energy consumption is used to detect technical breakdown in factories, general buildings, or energy theft in households. To overcome the limitations of previous studies, this paper suggests WaDGAN-AD, which combines generative adversarial network (GAN) and Long Short-Term Memory (LSTM) and applies two structural improvements. WaDGAN-AD has stacked discriminator LSTM layers to more precisely learn feature representations of time series data. Also, it has different numbers of hidden units in each hidden layer of LSTM to consider multiple cycles appearing in a single time-series data. Experimental results based on synthetic datasets and real datasets show that WaDGAN-AD can better detect abnormal energy consumption than benchmark methods.

Gold price prediction using Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM)

Gold has an important role in worldwide economics. Gold is not only used in jewelry but also can be a good deal for investment however several factors can affect the fluctuation in gold which can make the risk of investing in gold is bigger for many people. Therefore, is very important to predict the gold price for people who invest in gold in order to help reduce the investment risk. This study will implement a hybrid method from Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). CNN can extract useful knowledge and learn the internal representation of time-series data, and LSTM networks will identify short-term and long-term dependencies effectively. This research will use daily time frame data and weekly time frame data. This research also tried some experiments to find the best hyperparameters of batch size and epochs in ratio data 60:40 and 80:20. The best result obtained in the daily time of ratio data 60:40 with RMSE 13.67953 and MAE 9,40998, while in ratio data 80:20 has RMSE 15,53199 and MAE 12,78120. In weekly time has obtained the RMSE 38,01949 and MAE 28,32035 for ratio data 60:40 while in ratio data 80:20 the result was RMSE 32,61283 and MAE 22,74638. Those results shows that CNN-LSTM model can predict the trend of daily time frame gold price.

Effect of Sparse Representation of Time Series Data on Learning Rate of Time-Delay Neural Networks

In this paper, we examine how sparsifying input to a time-delay neural network (TDNN) can significantly improve the learning time and accuracy of the TDNN for time series data. The sparsifying of input is done through a sparse transform input layer. Many applications that involve prediction or forecasting of the state of a dynamic system can be formulated as a time series forecasting problem. Here, the task is to forecast some state variable, which is represented as a time series in applications such as weather forecasting, energy consumption prediction or predicting future state of a moving object. While there are many tools for time-delay forecasting, TDNNs have recently received more attention. We show that through applying a sparsifying input transform layer to the TDNN, we can considerably improve the learning time and accuracy. Through analyzing the learning process, we demonstrate the mathematical reasons for this improvement. Experiments with several datasets are used to show the improvement and the reason behind it. We use data from national weather forecast datasets, vehicle speed time series and synthetic data. Several different sparse representations are evaluated including principal component analysis (PCA), discrete cosine transform (DCT) and a mixture of DCT and Haar transforms. It is observed that the higher sparsity leads to better performance. The relative simplicity of TDNNs, compared with deep networks, and the use of sparse transforms for quicker learning open up possibilities for online learning in small embedded devices that do not have powerful computing capabilities.

Recognizing Activities of Daily Living using 1D Convolutional Neural Networks for Efficient Smart Homes

Human activity recognition is considered a challenging task in sensor-based monitoring systems. In ambient intelligent environments, such as smart homes, collecting data from ambient sensors is useful for recognizing activities of daily living, which can then be used to provide assistance to inhabitants. Activities of daily living are composed of complex multivariable time series data that has high dimensionality, is huge in size, and is updated constantly. Thus, developing methods for analyzing time series data to extract meaningful features and specific characteristics would help solve the problem of human activity recognition. Based on the noticeable success of deep learning in the field of time series classification, we developed a model called a deep one-dimensional convolutional neural network (Deep 1d-CNN) for recognizing activities of daily living in smart homes. Our model contains several one-dimensional convolution layers coupled with max-pooling technique to learn the internal representation of time series data and automatically generate very deep features for recognizing different activity types. For the performance evaluation, we tested our deep model on the new real-life dataset, ContextAct@A4H, and the results showed that our model achieved a high F1 score (0.90). We also extended our study to show the potential energy saving in smart homes through recognizing activities of daily living. We built a recommendation system based on the activities recognized by our deep model to detect the devices that are wasting energy, and recommend the user to execute energy optimization actions. The experiment indicated that recognizing activities of daily living can result in energy savings of around 50%.