TS2Vec: Towards Universal Representation of Time Series

Proceedings of the AAAI Conference on Artificial Intelligence, Год журнала: 2022, Номер 36(8), С. 8980 - 8987

Опубликована: Июнь 28, 2022

This paper presents TS2Vec, a universal framework for learning representations of time series in an arbitrary semantic level. Unlike existing methods, TS2Vec performs contrastive hierarchical way over augmented context views, which enables robust contextual representation each timestamp. Furthermore, to obtain the sub-sequence series, we can apply simple aggregation corresponding timestamps. We conduct extensive experiments on classification tasks evaluate quality representations. As result, achieves significant improvement SOTAs unsupervised 125 UCR datasets and 29 UEA datasets. The learned timestamp-level also achieve superior results forecasting anomaly detection tasks. A linear regression trained top outperforms previous forecasting. present detection, establishes SOTA literature. source code is publicly available at https://github.com/yuezhihan/ts2vec.

Язык: Английский

---

Temporal Convolutional Networks Applied to Energy-Related Time Series Forecasting

Applied Sciences, Год журнала: 2020, Номер 10(7), С. 2322 - 2322

Опубликована: Март 28, 2020

Modern energy systems collect high volumes of data that can provide valuable information about consumption. Electric companies now use historical to make informed decisions on production by forecasting the expected demand. Many deep learning models have been proposed deal with these types time series problems. Deep neural networks, such as recurrent or convolutional, automatically capture complex patterns in and accurate predictions. In particular, Temporal Convolutional Networks (TCN) are a specialised architecture has advantages over networks for tasks. TCNs able extract long-term using dilated causal convolutions residual blocks, also be more efficient terms computation time. this work, we propose TCN-based model improve predictive performance demand forecasting. Two energy-related from Spain studied: national electric power at charging stations vehicles. An extensive experimental study conducted, involving than 1900 different architectures parametrisations. The TCN proposal outperforms accuracy Long Short-Term Memory (LSTM) which considered state-of-the-art field.

Язык: Английский

---

A Hybrid Prediction Method for Realistic Network Traffic With Temporal Convolutional Network and LSTM

IEEE Transactions on Automation Science and Engineering, Год журнала: 2021, Номер 19(3), С. 1869 - 1879

Опубликована: Май 21, 2021

Accurate and real-time prediction of network traffic can not only help system operators allocate resources rationally according to their actual business needs but also them assess the performance a analyze its health status. In recent years, neural networks have been proved suitable predict time series data, represented by model long short-term memory (LSTM) temporal convolutional (TCN). This article proposes novel hybrid method named SG TCN-based LSTM (ST-LSTM) for such prediction, which synergistically combines power Savitzky–Golay (SG) filter, TCN, as well LSTM. ST-LSTM employs three-phase end-to-end methodology serving prediction. It first eliminates noise in raw data using then extracts features from sequences applying captures long-term dependence exploiting Experimental results over real-world datasets demonstrate that proposed outperforms state-of-the-art algorithms terms accuracy. Note Practitioners —This work considers high-accuracy traffic. is highly important capturing effectively extracting high- low-frequency information data. Yet, it big challenge achieve because there are unstable characteristics strong nonlinear due continuous expansion scale fast emergence new services. Current methods usually oversimplified theoretical assumptions, need significant memory, or suffer problems gradient disappearance early convergence. Thus, they fail capture large-scale sequences. (ST-LSTM), integrates merits (TCN), (LSTM), smoothing series, local features, dependence, respectively. based on real-life dataset achieves better accuracy than peers, including TCN be readily implemented deployed many industrial areas smart city, edge computing, cloud centers.

Язык: Английский

---

Temporal convolutional networks interval prediction model for wind speed forecasting

Electric Power Systems Research, Год журнала: 2020, Номер 191, С. 106865 - 106865

Опубликована: Окт. 3, 2020

Язык: Английский

---

Long Short-Term Memory Network-Based Metaheuristic for Effective Electric Energy Consumption Prediction

Applied Sciences, Год журнала: 2021, Номер 11(23), С. 11263 - 11263

Опубликована: Ноя. 27, 2021

The Electric Energy Consumption Prediction (EECP) is a complex and important process in an intelligent energy management system its importance has been increasing rapidly due to technological developments human population growth. A reliable accurate model for EECP considered key factor appropriate policy. In recent periods, many artificial intelligence-based models have developed perform different simulation functions, engineering techniques, optimal forecasting order predict future demands on the basis of historical data. this article, new metaheuristic based Long Short-Term Memory (LSTM) network proposed effective EECP. After collecting data sequences from Individual Household Power (IHEPC) dataset Appliances Load (AEP) dataset, refinement accomplished using min-max standard transformation methods. Then, LSTM with Butterfly Optimization Algorithm (BOA) BOA used select hyperparametric values which precisely describe EEC patterns discover time series dynamics domain. This extensive experiment conducted IHEPC AEP datasets shows that obtains minimum error rate relative existing models.

Язык: Английский

---

An Advanced CNN-LSTM Model for Cryptocurrency Forecasting

Electronics, Год журнала: 2021, Номер 10(3), С. 287 - 287

Опубликована: Янв. 26, 2021

Nowadays, cryptocurrencies are established and widely recognized as an alternative exchange currency method. They have infiltrated most financial transactions a result cryptocurrency trade is generally considered one of the popular promising types profitable investments. Nevertheless, this constantly increasing market characterized by significant volatility strong price fluctuations over short-time period therefore, development accurate reliable forecasting model essential for portfolio management optimization. In research, we propose multiple-input deep neural network prediction movement. The proposed utilizes inputs different data handles them independently in order to exploit useful information from each separately. An extensive empirical study was performed using three consecutive years with highest capitalization i.e., Bitcoin (BTC), Etherium (ETH), Ripple (XRP). detailed experimental analysis revealed that has ability efficiently mixed data, reduces overfitting decreases computational cost comparison traditional fully-connected networks.

Язык: Английский

---

A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anomaly Detection

IEEE Transactions on Pattern Analysis and Machine Intelligence, Год журнала: 2024, Номер 46(12), С. 10466 - 10485

Опубликована: Авг. 14, 2024

Time series are the primary data type used to record dynamic system measurements and generated in great volume by both physical sensors online processes (virtual sensors). analytics is therefore crucial unlocking wealth of information implicit available data. With recent advancements graph neural networks (GNNs), there has been a surge GNN-based approaches for time analysis. These can explicitly model inter-temporal inter-variable relationships, which traditional other deep network-based methods struggle do. In this survey, we provide comprehensive review analysis (GNN4TS), encompassing four fundamental dimensions: forecasting, classification, anomaly detection, imputation. Our aim guide designers practitioners understand, build applications, advance research GNN4TS. At first, task-oriented taxonomy Then, present discuss representative works introduce mainstream applications A discussion potential future directions completes survey. This first time, brings together vast array knowledge on research, highlighting foundations, practical opportunities

Язык: Английский

---

Temporal Convolutional Networks with RNN approach for chaotic time series prediction

Applied Soft Computing, Год журнала: 2022, Номер 133, С. 109945 - 109945

Опубликована: Дек. 17, 2022

Язык: Английский

---

A TCN-Based Hybrid Forecasting Framework for Hours-Ahead Utility-Scale PV Forecasting

IEEE Transactions on Smart Grid, Год журнала: 2023, Номер 14(5), С. 4073 - 4085

Опубликована: Янв. 16, 2023

This paper presents a Temporal Convolutional Network (TCN) based hybrid PV forecasting framework for enhancing hours-ahead utility-scale forecasting. The consists of two models: physics-based trend (TF) model and data-driven fluctuation (FF) model. Three TCNs are integrated in the for: i) blending inputs from different Numerical Weather Prediction sources TF to achieve superior performance on hourly profiles, ii) capturing spatial-temporal correlations between detector sites target site FF more accurate forecast intra-hour power drops, iii) reconciling results obtain coherent with both trends fluctuations well preserved. To automatically identify most contributive neighboring forming network, scenario-based correlation analysis method is developed, which significantly improves capability large caused by cloud movements. tested, validated using actual data collected 95 farms North Carolina. Simulation show that 6 hours ahead improved 20% - 30% compared state-of-the-art methods.

Язык: Английский

---

Deep Learning for Time Series Forecasting: Advances and Open Problems

Information, Год журнала: 2023, Номер 14(11), С. 598 - 598

Опубликована: Ноя. 4, 2023

A time series is a sequence of time-ordered data, and it generally used to describe how phenomenon evolves over time. Time forecasting, estimating future values series, allows the implementation decision-making strategies. Deep learning, currently leading field machine applied forecasting can cope with complex high-dimensional that cannot be usually handled by other learning techniques. The aim work provide review state-of-the-art deep architectures for underline recent advances open problems, also pay attention benchmark data sets. Moreover, presents clear distinction between are suitable short-term long-term forecasting. With respect existing literature, major advantage consists in describing most such as Graph Neural Networks, Gaussian Processes, Generative Adversarial Diffusion Models, Transformers.

Язык: Английский

---