Article
Computer Science, Information Systems
Ruijin Wang, Xikai Pei, Juyi Zhu, Zhiyang Zhang, Xin Huang, Jiayi Zhai, Fengli Zhang
Summary: This paper proposes a model fusion-based time series forecasting method to improve the accuracy and efficiency of predictions using multivariate grey model and artificial fish swarm algorithm. Two fusion models based on data decomposition and weighted summation achieve good prediction results in different scenarios.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Fan Zhang, Tiantian Guo, Hua Wang
Summary: The DFNet model proposed in this study accurately obtains the trend, seasonal, and irregular components of time series, improves model performance, handles negative data loss and reduces model computation, and enhances the coupling between sequences. Experimental results demonstrate the significant superiority of the proposed model for long time-series prediction.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Automation & Control Systems
Jiajia Li, Feng Tan, Cheng He, Zikai Wang, Haitao Song, Pengwei Hu, Xin Luo
Summary: This article proposes a novel deep-saliency-aware dual embedded attention network for forecasting aperiodic multivariate time series. The network achieves impressive high performance by representing saliency, capturing long-term dependence features, and fusing latent representations.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Automation & Control Systems
Keyong Wan, Bin Li, Weijie Zhou, Haicheng Zhu, Song Ding
Summary: A novel time-power based grey model is proposed to deal with nonlinear issues, optimizing the time-power parameter using the Particle Swarm Optimization algorithm for higher and more reliable predicting accuracy. Verification through numerical simulations and experimental studies confirms the effectiveness and practicality of the model. The use of probability density prediction method for the first time on settlement prediction further enhances the reliability and stability of the proposed model.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Mathematics, Applied
Waddah Saeed
Summary: This paper introduces the model used in the M4 forecasting competition, which combines several statistical methods and outperforms the benchmarks in terms of forecasting accuracy. The proposed model is also compared with other forecasting methods and shown to produce accurate results.
COMPUTATIONAL & APPLIED MATHEMATICS
(2022)
Article
Thermodynamics
Xiaoping Xiong, Guohua Qing
Summary: This paper proposes a new hybrid forecasting framework to improve the forecasting accuracy of day-ahead electricity prices. The proposed model consists of three valuable strategies: an adaptive copula-based feature selection algorithm, a new method of signal decomposition technique based on decomposition denoising strategy, and a Bayesian optimization and hyperband optimized long short-term memory model. Cross-validation using five datasets from the PJM electricity market shows that the proposed hybrid algorithm is more effective and practical for day-ahead electricity price forecasting.
Article
Mathematics
Kaijian He, Qian Yang, Lei Ji, Jingcheng Pan, Yingchao Zou
Summary: With the continuous development of financial markets worldwide, there has been increasing recognition of the importance of financial time series forecasting in operation and management. This paper proposes a new financial time series forecasting model based on the deep learning ensemble model, combining CNN, LSTM, and ARMA. Empirical results show that the proposed model achieved superior performance in terms of accuracy and robustness compared to benchmark individual models.
Article
Computer Science, Artificial Intelligence
Yajiao Tang, Zhenyu Song, Yulin Zhu, Huaiyu Yuan, Maozhang Hou, Junkai Ji, Cheng Tang, Jianqiang Li
Summary: This paper provides a timely review of the adoption of machine learning in financial time series (FTS) forecasting. The progress and application of machine learning methods in FTS forecasting models are systematically summarized, providing a relevant reference for researchers and practitioners. The paper identifies commonly used models and discusses their merits and demerits, as well as the limitations and future research directions of machine learning models in FTS forecasting.
Article
Computer Science, Information Systems
Zineb Bousbaa, Javier Sanchez-Medina, Omar Bencharef
Summary: Data stream mining can be used to forecast financial time series exchange rate. Traditional static machine learning models are not suitable for the cyclical patterns in financial historical data. This paper proposes a possible methodology that uses incremental and adaptive strategy to cope with instability. The proposed algorithm utilizes online learning and statistical techniques to detect and respond to pattern shifts in the data trend.
Article
Computer Science, Artificial Intelligence
Pengyu Zeng, Guoliang Hu, Xiaofeng Zhou, Shuai Li, Pengjie Liu, Shurui Liu
Summary: This paper proposes an efficient transformer-based predictive model called Muformer. It solves the problem of redundant input information in long sequence time-series forecasting through multiple perceptual domain processing and multi-granularity attention head mechanism, and achieves significant advantages in experiments.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Information Systems
M. A. Castan-Lascorz, P. Jimenez-Herrera, A. Troncoso, G. Asencio-Cortes
Summary: This research proposes a new algorithm based on a combination of clustering, classification, and forecasting techniques for predicting both univariate and multivariate time series. The algorithm groups similar patterns of time series values using clustering and builds specific forecasting models for each pattern. Experimental results show that the proposed algorithm outperforms classical and recent forecasting methods in terms of prediction accuracy.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Hardware & Architecture
Arsalan Dezhkam, Mohammad Taghi Manzuri, Ahmad Aghapour, Afshin Karimi, Ali Rabiee, Shervin Manzuri Shalmani
Summary: This paper presents a classification approach for financial time series patterns, using machine learning and deep learning models, and the performance of the algorithm is enhanced by applying Bayesian optimization technique. The results show excellent trading performance of the framework.
JOURNAL OF SUPERCOMPUTING
(2023)
Article
Operations Research & Management Science
Tai Vovan, Luan Nguyenhuynh, Thuy Lethithu
Summary: This article presents a new fuzzy time series model that improves historical data interpolation and future forecasting. By using an enhanced fuzzy cluster analysis algorithm, the efficiency and accuracy of the algorithm is improved, and its superiority is demonstrated across multiple datasets.
ANNALS OF OPERATIONS RESEARCH
(2022)
Article
Computer Science, Information Systems
Jie Hu, Zhanao Hu, Tianrui Li, Shengdong Du
Summary: Time series forecasting has wide applications in our daily lives, and traditional supervised models have limitations due to a lack of real-time annotated data. Self-supervised methods, particularly contrastive learning, are proposed as a solution to this problem, but the direct transfer of data augmentation techniques from computer vision is not suitable for the time domain. In this paper, we introduce a novel time series forecasting model called ACST, which utilizes disentangled seasonal-trend representation and an improved generative adversarial data augmentation method for contrastive loss. Experimental results show that ACST achieves an average improvement of 26.8% on six benchmarks.
INFORMATION SCIENCES
(2023)
Article
Mathematics
Ana Lazcano, Pedro Javier Herrera, Manuel Monge
Summary: Accurate and real-time forecasting of oil prices is crucial in the global economy. This study combines Graph Convolutional Networks (GCN) with Bidirectional Long Short-Term Memory (BiLSTM) networks to improve the performance of existing models in time series forecasting. The results demonstrate that the combined BiLSTM-GCN approach outperforms the individual BiLSTM and GCN models, as well as traditional models, with lower errors in all the metrics used.