Article
Computer Science, Artificial Intelligence
Artemios-Anargyros Semenoglou, Evangelos Spiliotis, Vassilios Assimakopoulos
Summary: Data augmentation techniques can improve forecasting accuracy in univariate time series prediction, especially when deep neural networks are used. However, these improvements become less significant as the initial size of the training set increases.
PATTERN RECOGNITION
(2022)
Article
Multidisciplinary Sciences
Vitor Hugo Serravalle Reis Rodrigues, Paulo Roberto de Melo Barros Junior, Euler Bentes dos Santos Marinho, Jose Luis Lima de Jesus Silva
Summary: Developing accurate models for groundwater control is crucial for managing and planning water resources from aquifer reservoirs. The proposed Wavelet Gated Multiformer combines the strengths of a vanilla Transformer and a Wavelet Crossformer to improve the model's predictive capabilities by computing the relationships between time-series points and finding trending periodic patterns. This model outperforms other transformer-like models in terms of Mean Absolute Error reduction.
SCIENTIFIC REPORTS
(2023)
Article
Computer Science, Artificial Intelligence
Arun M. George, Sounak Dey, Dighanchal Banerjee, Arijit Mukherjee, Manan Suri
Summary: IoT-based automated systems require efficient online time-series analysis and forecasting, which is challenging to achieve on low-cost constrained edge devices. This study proposes a novel spiking reservoir based network that relies on temporal spike encoding and feedback-driven online learning mechanism for online time series forecasting. The network outperforms conventional methods like SARIMA, Online ARIMA, Stacked LSTM, achieving up to 8% higher R2 score while using negligible buffer memory.
Article
Computer Science, Artificial Intelligence
Rakshitha Godahewa, Kasun Bandara, Geoffrey Webb, Slawek Smyl, Christoph Bergmeir
Summary: Ensembling techniques are used to improve the performance of Global Forecasting Models (GFM) and univariate models in heterogeneous datasets. A new clustered ensembles methodology is proposed to train multiple GFMs on different clusters of series, achieving higher accuracy than baseline models.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Multidisciplinary Sciences
Pantelis Linardatos, Vasilis Papastefanopoulos, Theodor Panagiotakopoulos, Sotiris Kotsiantis
Summary: This article discusses the relationship between carbon emissions and climate change, as well as the efforts of countries and organizations to reduce CO2 emissions. A hybrid machine learning system was developed to forecast CO2 concentration in a smart city environment using IoT technologies, and it was compared to other methods. The results demonstrate the superior performance and interpretability of the proposed system, with deep learning approaches outperforming traditional methods in handling multivariate datasets and longer forecasting horizons.
SCIENTIFIC REPORTS
(2023)
Article
Engineering, Civil
Julien Monteil, Anton Dekusar, Claudio Gambella, Yassine Lassoued, Martin Mevissen
Summary: This work investigates the use of deep learning models for long-term large-scale traffic prediction tasks, focusing on scalability. By analyzing 14 weeks of speed observations from over 1000 segments in downtown Los Angeles, different machine learning and deep learning predictors were studied, along with their scalability to larger areas. The study shows that modeling temporal and spatial features into deep learning predictors can be beneficial for long-term predictions, while simpler predictors achieve satisfactory performance for link-based and short-term forecasting, with a trade-off in prediction accuracy, horizon, training time, and model sizing discussed.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Theory & Methods
Konstantinos Benidis, Syama Sundar Rangapuram, Valentin Flunkert, Yuyang Wang, Danielle Maddix, Caner Turkmen, Jan Gasthaus, Michael Bohlke-Schneider, David Salinas, Lorenzo Stella, Francois-Xavier Aubet, Laurent Callot, Tim Januschowski
Summary: Deep learning based forecasting methods have achieved remarkable success in time series prediction and have become widely used in industrial applications and forecasting competitions. This article provides an introduction to deep forecasting, discussing important building blocks and summarizing recent literature.
ACM COMPUTING SURVEYS
(2023)
Article
Computer Science, Information Systems
Ola Surakhi, Martha A. Zaidan, Pak Lun Fung, Naser Hossein Motlagh, Sami Serhan, Mohammad AlKhanafseh, Rania M. Ghoniem, Tareq Hussein
Summary: This paper investigates the impact of selecting an appropriate time-lag value on forecasting accuracy in time-series forecasting. The results show that the proposed LSTM model with heuristic algorithm is the best method for determining the optimal time-lag value.
Article
Computer Science, Artificial Intelligence
Ming Jin, Yu Zheng, Yuan-Fang Li, Siheng Chen, Bin Yang, Shirui Pan
Summary: This paper proposes a continuous model, MTGODE, to forecast multivariate time series by overcoming the limitations of discrete neural architectures, high complexity, and reliance on graph priors. MTGODE utilizes dynamic graph neural ordinary differential equations to unify spatial and temporal message passing, resulting in superior forecasting performance on benchmark datasets.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Engineering, Mechanical
Haitao Liu, Changjun Liu, Xiaomo Jiang, Xudong Chen, Shuhua Yang, Xiaofang Wang
Summary: This study proposes a data-driven deep probabilistic sequence model by combining deep generative models and state space models. The model utilizes recurrent neural networks (RNNs) to create a variational sequence model in an augmented recurrent input space, inducing rich stochastic sequence dependency. Extensive numerical experiments demonstrate the superior performance of the model in system identification and prediction tasks.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Andreas Kanavos, Fotios Kounelis, Lazaros Iliadis, Christos Makris
Summary: This paper focuses on the analysis and modeling of passenger demand dynamics in the aviation industry, proposing a method using time series and deep learning techniques to forecast aviation demand, and developing related models. The results of the study show that the proposed methods exhibit satisfactory accuracy and robustness in predicting air travel demand.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Thermodynamics
Niaz Bashiri Behmiri, Carlo Fezzi, Francesco Ravazzolo
Summary: One of the most controversial issues in mid-term load forecasting is how to treat weather. This article compares three approaches: excluding weather, assuming perfect knowledge of future weather, and including weather forecasts in load forecasting models. The results show that models including future temperature consistently outperform models excluding temperature, but predictions of future temperature weaken the results.
Article
Physics, Multidisciplinary
Kady Sako, Berthine Nyunga Mpinda, Paulo Canas Rodrigues
Summary: This study forecasts stock market indexes and currency exchange rates using Recurrent Neural Networks (RNNs) and its variants, with the Gated Recurrent Unit (GRU) model performing the best overall.
Article
Multidisciplinary Sciences
Yuzhen Zhu, Shaojie Luo, Di Huang, Weiyan Zheng, Fang Su, Beiping Hou
Summary: Recent studies have shown great performance of Transformer-based models in long-term time series forecasting, but they have limitations when training on small datasets. This paper proposes the DRCNN method to utilize the continuity between data by decomposing data into residual and trend terms, and designs the DR-Block to extract features. Additionally, a Multi-head Sequence method is proposed for longer inputs and accurate forecasts.
SCIENTIFIC REPORTS
(2023)
Article
Computer Science, Artificial Intelligence
Zimeng Lyu, Alexander Ororbia, Travis Desell
Summary: Time series forecasting is an important task in data science, but offline-trained models often face data drift issues. To address this, this paper proposes an online neural architecture search algorithm (ONE-NAS) that can automatically design and train recurrent neural networks for online forecasting tasks. Experimental results show that ONE-NAS outperforms traditional statistical methods and using multiple populations of RNNs can significantly improve performance.
APPLIED SOFT COMPUTING
(2023)