Article
Computer Science, Artificial Intelligence
Haoyi Zhou, Jianxin Li, Shanghang Zhang, Shuai Zhang, Mengyi Yan, Hui Xiong
Summary: This study proposes an efficient model for long sequence time-series forecasting called Informer, which addresses the limitations of the traditional Transformer in terms of complexity, memory usage, and inference speed. Informer improves the prediction capacity by introducing the ProbSparse self-attention mechanism, attention distilling with convolutional operators, and a generative style decoder. Extensive experiments on large-scale datasets demonstrate that Informer outperforms existing methods and provides a new solution to the long sequence time-series forecasting problem.
ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Fabio Giampaolo, Federico Gatta, Edoardo Prezioso, Salvatore Cuomo, Mengchu Zhou, Giancarlo Fortino, Francesco Piccialli
Summary: This study proposes a novel ensemble approach for generating predictions in a multivariate framework. It reduces data dimensionality through an encoding technique, extracts useful information via single predictive procedures, and combines the processed data to produce the final forecast. Extensive experiments demonstrate the higher accuracy and robustness of the proposed ensemble compared to conventional methods and state-of-the-art strategies.
INFORMATION FUSION
(2023)
Article
Economics
Jian-Wu Bi, Tian-Yu Han, Yanbo Yao
Summary: Compared with coarse-grained forecasting, fine-grained tourism demand forecasting is a more challenging task, but research on this issue is very scarce. To address this issue, a decomposition ensemble deep learning model is proposed by integrating CEEMDAN, CNNs, LSTM networks, and AR models. The effectiveness of the proposed model is verified by comparing with five benchmark models using real-time data on tourist volumes at two attractions.
Article
Computer Science, Artificial Intelligence
Dawei Cheng, Fangzhou Yang, Sheng Xiang, Jin Liu
Summary: Financial time series analysis is essential for hedging market risks and optimizing investment decisions. The proposed multi-modality graph neural network (MAGNN) leverages multimodal inputs for financial time series prediction by constructing a heterogeneous graph network and utilizing a two-phase attention mechanism for joint optimization. Extensive experiments demonstrate the superior performance of MAGNN in financial market prediction, providing investors with profitable and interpretable options to make informed investment decisions.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Interdisciplinary Applications
Wenyan Hu, Stephan Winter, Kourosh Khoshelham
Summary: In this paper, a method for tailored vehicle selection based on forecast fine-grained sensing coverage is proposed without trajectory data. A model is proposed to forecast fine-grained sensing coverage using coarse-grained information of candidate vehicles and a vehicle selection algorithm is developed to maximize the sensing quality. Results show that the selected vehicles based on this method achieve higher sensing quality than two other baselines. This research provides fundamental guidelines for coverage estimation and vehicle selection in urban vehicular sensing applications.
COMPUTERS ENVIRONMENT AND URBAN SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Amal Saadallah, Matthias Jakobs, Katharina Morik
Summary: The complexity and evolution of time series data make forecasting one of the most challenging tasks in machine learning. Combining a diverse set of forecasters in heterogeneous ensembles and using gradient-based saliency maps for online ensemble pruning can lead to excellent results in prediction tasks. Empirical studies show that this method outperforms state-of-the-art approaches and baselines.
Article
Chemistry, Multidisciplinary
Hsun-Ping Hsieh, Su Wu, Ching-Chung Ko, Chris Shei, Zheng-Ting Yao, Yu-Wen Chen
Summary: Urban air pollution has a severe impact on economic development and human health, making the observation and prediction of AQI increasingly important. This study introduces a spatial-temporal model to predict long-term AQI in cities without monitoring stations. Experimental results demonstrate the superiority of the proposed model over all baseline models.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Artificial Intelligence
Vilde Jensen, Filippo Maria Bianchi, Stian Normann Anfinsen
Summary: This article presents a novel probabilistic forecasting method called ensemble conformalized quantile regression (EnCQR), which constructs distribution-free and approximately marginally valid prediction intervals (PIs) suitable for nonstationary and heteroscedastic time series data. By utilizing bootstrap ensemble estimator and generic machine learning algorithms, EnCQR outperforms models based only on quantile regression (QR) or conformal prediction (CP), delivering sharper, more informative, and valid PIs.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Borui Cai, Shuiqiao Yang, Longxiang Gao, Yong Xiang
Summary: This paper introduces a novel hybrid variational autoencoder (HyVAE) for forecasting time series by jointly learning the local patterns and temporal dynamics. Experimental results demonstrate that the proposed HyVAE achieves better results compared to other methods.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Information Systems
Renzhuo Wan, Chengde Tian, Wei Zhang, Wendi Deng, Fan Yang
Summary: This study introduces a novel multivariate time-series forecasting model based on self-attentive mechanisms, which shows significantly improved prediction accuracy and generalization compared to traditional models.
Article
Computer Science, Information Systems
Shuai Zhang, Yong Chen, Wenyu Zhang, Ruijun Feng
Summary: In this study, a novel ensemble deep learning model is proposed for accurate and stable time series forecasting by generating various basic predictors and enhancing them through a new dynamic error correction method. The model combines basic predictors using a stacking-based ensemble method with kernel ridge regression as the meta-predictor, and an enhanced genetic algorithm is used for ensemble pruning to increase forecasting accuracy and stability. Experimental results showed the superiority of the proposed model in dealing with time series forecasting tasks.
INFORMATION SCIENCES
(2021)
Article
Economics
Bryan Lim, Sercan O. Arik, Nicolas Loeff, Tomas Pfister
Summary: This paper introduces the Temporal Fusion Transformer (TFT), a novel attention-based architecture that combines high-performance multi-horizon forecasting with interpretable insights into temporal dynamics. TFT utilizes recurrent layers for local processing and interpretable self-attention layers for long-term dependencies, achieving high performance in a wide range of scenarios. By selecting relevant features and suppressing unnecessary components, TFT demonstrates significant performance improvements over existing benchmarks on various real-world datasets.
INTERNATIONAL JOURNAL OF FORECASTING
(2021)
Article
Computer Science, Artificial Intelligence
Fatemeh Nazarieh, Mohammad Naderi Dehkordi
Summary: This research develops a novel decomposition ensemble-based network named VMD-DENetwork for time series forecasting over different horizons. A robust decomposition technique called variational mode decomposition (VMD) is applied to decompose the input sequence into several intrinsic modes. The proposed DENetwork combines multiple learners to model the nonlinear and complex relationships, and a firefly optimization algorithm is adopted to enhance the efficiency of VMD-DENetwork. The experimental results confirm the outstanding prediction performance of the proposed model.
Article
Engineering, Environmental
Subin Lin, Jiwoong Kim, Chuanbo Hua, Mi-Hyun Park, Seoktae Kang
Summary: This study proposes a deep learning approach using long-term data to determine coagulant dosage and settled water turbidity in water treatment processes. The graph attention multivariate time series forecasting (GAMTF) model outperforms conventional machine learning and deep learning models and successfully predicts both coagulant dosage and settled water turbidity. The GAMTF model improves prediction accuracy by considering hidden interrelationships and past states of features.
Article
Computer Science, Artificial Intelligence
Chenrui Yin, Qun Dai
Summary: The paper presents a novel Deep Multivariate Time Series Multistep Forecasting Network DualMNet based on the Encoder-Decoder model, which utilizes temporal and spatial patterns modules for prediction, addressing the issues of error accumulation and inadequate prediction accuracy found in traditional methods. Extensive experimental results demonstrate that DualMNet significantly outperforms comparison models in multistep prediction performance, highlighting the importance of integrating all components together for robust forecasting.
APPLIED INTELLIGENCE
(2022)
Article
Automation & Control Systems
Carmen Bisogni, Lucia Cimmino, Michele Nappi, Toni Pannese, Chiara Pero
Summary: This paper presents a gait-based emotion recognition method that does not rely on facial cues, achieving competitive performance on small and unbalanced datasets. The proposed approach utilizes advanced deep learning architecture and achieves high recognition and accuracy rates.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Soung Sub Lee
Summary: This study proposed a satellite constellation method that utilizes machine learning and customized repeating ground track orbits to optimize satellite revisit performance for each target.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Jian Wang, Xiuying Zhan, Yuping Yan, Guosheng Zhao
Summary: This paper proposes a method of user recruitment and adaptation degree improvement via community collaboration to solve the task allocation problem in sparse mobile crowdsensing. By matching social relationships and perception task characteristics, the entire perceptual map can be accurately inferred.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Yuhang Gai, Bing Wang, Jiwen Zhang, Dan Wu, Ken Chen
Summary: This paper investigates how to reconfigure existing compliance controllers for new assembly objects with different geometric features. By using the proposed Equivalent Theory of Compliance Law (ETCL) and Weighted Dimensional Policy Distillation (WDPD) method, the learning cost can be reduced and better control performance can be achieved.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Zhihao Xu, Zhiqiang Lv, Benjia Chu, Zhaoyu Sheng, Jianbo Li
Summary: Predicting future urban health status is crucial for identifying urban diseases and planning cities. By applying an improved meta-analysis approach and considering the complexity of cities as systems, this study selects eight urban factors and explores suitable prediction methods for these factors.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Yulong Ye, Qiuzhen Lin, Ka-Chun Wong, Jianqiang Li, Zhong Ming, Carlos A. Coello Coello
Summary: This paper proposes a localized decomposition evolutionary algorithm (LDEA) to tackle imbalanced multi-objective optimization problems (MOPs). LDEA assigns a local region for each subproblem using a localized decomposition method and restricts the solution update within the region to maintain diversity. It also speeds up convergence by evolving only the best-associated solution in each subproblem while balancing the population's diversity.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Longxin Zhang, Jingsheng Chen, Jianguo Chen, Zhicheng Wen, Xusheng Zhou
Summary: This study proposes a lightweight PCB image defect detection network (LDD-Net) that achieves high accuracy by designing a novel lightweight feature extraction network, multi-scale aggregation network, and lightweight decoupling head. Experimental results show that LDD-Net outperforms state-of-the-art models in terms of accuracy, computation, and detection speed, making it suitable for edge systems or resource-constrained embedded devices.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Kemal Ucak, Gulay Oke Gunel
Summary: This paper introduces a novel adaptive stable backstepping controller based on support vector regression for nonlinear dynamical systems. The controller utilizes SVR to identify the dynamics of the nonlinear system and integrates stable BSC behavior. The experimental results demonstrate successful control performance for both nonlinear systems.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Dexuan Zou, Mengdi Li, Haibin Ouyang
Summary: In this study, a photovoltaic thermal collector is integrated into a combined cooling, heating, and power system to reduce primary energy consumption, operation cost, and carbon dioxide emission. By applying a novel genetic algorithm and constraint handling approach, it is found that the CCHP scenarios with PV/T are more efficient and achieve the lowest energy consumption.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Abhinav Pandey, Litton Bhandari, Vidit Gaur
Summary: This research proposes a novel model-agnostic framework based on genetic algorithms to identify and optimize the set of coefficients of the constitutive equations of engineering materials. The framework demonstrates solution convergence, scalability, and high explainability for a wide range of engineering materials. The experimental validation shows that the proposed framework outperforms commercially available software in terms of optimization efficiency.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Zahra Ramezanpoor, Adel Ghazikhani, Ghasem Sadeghi Bajestani
Summary: Time series analysis is a method used to analyze phenomena with temporal measurements. Visibility graphs are a technique for representing and analyzing time series, particularly when dealing with rotations in the polar plane. This research proposes a visibility graph algorithm that efficiently handles biological time series with rotation in the polar plane. Experimental results demonstrate the effectiveness of the proposed algorithm in both synthetic and real world time series.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
ChunLi Li, Qintai Hu, Shuping Zhao, Jigang Wu, Jianbin Xiong
Summary: Efficient and accurate diagnosis of rotating machinery in the petrochemical industry is crucial. However, the nonlinear and non-stationary vibration signals generated in harsh environments pose challenges in distinguishing fault signals from normal ones. This paper proposes a BP-Incremental Broad Learning System (BP-INBLS) model to address these challenges. The effectiveness of the proposed method in fault diagnosis is demonstrated through validation and comparative analysis with a published method.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Fatemeh Chahkoutahi, Mehdi Khashei
Summary: The classification rate is the most important factor in selecting an appropriate classification approach. In this paper, the influence of different cost/loss functions on the classification rate of different classifiers is compared, and empirical results show that cost/loss functions significantly affect the classification rate.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Automation & Control Systems
Jicong Duan, Xibei Yang, Shang Gao, Hualong Yu
Summary: The study proposes a novel partition-based imbalanced multi-label learning algorithm, MLHC, which divides the original label space into disconnected subspaces using hierarchical clustering. It successfully tackles the class imbalance problem in multi-label data and outperforms other class imbalance multi-label learning algorithms.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Review
Automation & Control Systems
Qing Qin, Yuanyuan Chen
Summary: This paper offers a comprehensive review of retinal vessel automatic segmentation research, including both traditional methods and deep learning methods. In particular, supervised learning methods are summarized and analyzed based on CNN, GAN, and UNet. The advantages and disadvantages of existing segmentation methods are also outlined.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)