Article
Automation & Control Systems
Yunyi Kang, Logan Mathesen, Giulia Pedrielli, Feng Ju, Loo Hay Lee
Summary: Recent advancements in sensing, data analytics, and manufacturing technologies have enabled the production of highly customized products. However, this also increases the complexity of controlling such systems, leading to the need for fast exploration of alternative operation strategies. The simultaneous use of simulation and stochastic models can help achieve better control and optimization of production systems by balancing accuracy and computational efficiency.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL
(2021)
Article
Materials Science, Multidisciplinary
Danial Khatamsaz, Abhilash Molkeri, Richard Couperthwaite, Jaylen James, Raymundo Arroyave, Ankit Srivastava, Douglas Allaire
Summary: This study introduces an adaptive active subspace method to efficiently handle high-dimensional design space problems in material design. This method can accelerate the design process by prioritizing searches in important regions of the high-dimensional design space.
MATERIALS & DESIGN
(2021)
Article
Computer Science, Artificial Intelligence
Angus Kenny, Tapabrata Ray, Hemant Kumar Singh
Summary: Engineering design optimization often involves using numerical simulations to assess candidate designs. Multifidelity optimization methods aim to optimize computationally expensive problems within a limited budget by efficiently managing low-fidelity and high-fidelity evaluations. This article proposes an improved multifidelity approach called MFITS, which uses an iterative, two-stage scheme to search for good candidates for high-fidelity evaluation based on collective information from previously evaluated designs.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2023)
Article
Computer Science, Artificial Intelligence
Patricia Gonzalez, Roberto R. Osorio, Xoan C. Pardo, Julio R. Banga, Ramon Doallo
Summary: The paper introduces a novel parallel ACO strategy that utilizes efficient asynchronous decentralized cooperative mechanisms to accelerate computations and improve convergence. The strategy stimulates diversification in search and cooperation between different colonies, making it suitable for traditional HPC clusters, cloud infrastructures, and environments with highly coupled resources, showing good scalability in solving combinatorial optimization problems.
APPLIED SOFT COMPUTING
(2022)
Article
Engineering, Aerospace
Francesco Di Fiore, Laura Mainini
Summary: To optimize aerodynamic design, a non-myopic multifidelity Bayesian framework is proposed to address the time-consuming evaluations of high-fidelity CFD models. The framework uses a two-step lookahead policy and utility functions to select the aerodynamic model to interrogate, resulting in a significant reduction in drag coefficient compared to other approaches.
Article
Engineering, Chemical
Kiran Siddappaji, Mark G. Turner
Summary: The MDAO framework has become essential in almost all fields, except for mechanical, transportation, and aerospace industries, for efficient energy conversion and more. It allows for rapid iterative interaction among various engineering disciplines using automation tools for design improvement. An advanced framework, ranging from low to high fidelity, has been developed for ducted and unducted turbomachinery blade designs. The key feature is a parametric blade geometry tool that converts low fidelity results into 3D blade shapes, which can be readily used in high fidelity multidisciplinary simulations as part of an optimization cycle.
Article
Computer Science, Artificial Intelligence
Shion Takeno, Hitoshi Fukuoka, Yuhki Tsukada, Toshiyuki Koyama, Motoki Shiga, Ichiro Takeuchi, Masayuki Karasuyama
Summary: A method called MF-MES is proposed to accelerate BO by incorporating lower-fidelity observations with a lower sampling cost; The generalization of MF-MES allows for evaluating information gain when multiple observations are obtained simultaneously; The effectiveness of MF-MES is demonstrated through benchmark functions and real-world applications.
NEURAL COMPUTATION
(2022)
Article
Computer Science, Artificial Intelligence
Mohammad Mohiuddin Mamun, Hemant Kumar Singh, Tapabrata Ray
Summary: This letter proposes a novel approach to treat bilevel optimization problems as multifidelity optimization problems, reducing the number of function evaluations. By learning the appropriate fidelity to evaluate solutions, redundant evaluations can be significantly reduced, improving the efficiency of problem-solving.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Computer Science, Artificial Intelligence
Shogo Sagawa, Hideitsu Hino
Summary: In domain adaptation, when there is a large distance between the source and target domains, the prediction performance will degrade. Gradual domain adaptation is one of the solutions, assuming access to intermediate domains. However, if the number of accessible intermediate domains is restricted, self-training will fail. To solve the trade-off between cost and accuracy, a framework combining multifidelity and active domain adaptation is proposed.
Article
Computer Science, Information Systems
Ibrahim A. Elgendy, Wei-Zhe Zhang, Chuan-Yi Liu, Ching-Hsien Hsu
Summary: A novel framework is proposed to dynamically decide whether tasks on mobile devices are offloaded to the cloud, with additional security measures provided, achieving significant performance improvement.
IEEE TRANSACTIONS ON CLOUD COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Kangjia Qiao, Kunjie Yu, Boyang Qu, Jing Liang, Hui Song, Caitong Yue
Summary: This article presents an evolutionary multitasking-based constrained multiobjective optimization framework for solving CMOPs. It transforms the optimization problem into two related tasks and utilizes a tentative method to discover and transfer useful knowledge. The approach achieves better performance compared to other state-of-the-art algorithms.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Energy & Fuels
Guoqing Hu, Fengqi You
Summary: To enhance energy efficiency and optimize crop growth in the built environment, this study proposes an AI-based control framework that combines a physics-informed neural network and data-driven robust model predictive control. The framework accurately predicts indoor climate and crop states, and uses machine learning techniques to make the best control decisions for actuators, considering unpredictable weather conditions.
Article
Computer Science, Information Systems
Eneko Osaba, Javier Del Ser, Aritz D. Martinez, Jesus L. Lobo, Francisco Herrera
Summary: Transfer Optimization is a new research area focusing on solving multiple optimization tasks simultaneously, with Evolutionary Multitasking being one effective approach. In this paper, a novel Adaptive Transfer-guided Multifactorial Cellular Genetic Algorithm (AT-MFCGA) is introduced to handle Evolutionary Multitasking environments. AT-MFCGA relies on cellular automata for knowledge exchange and can independently explain synergies among tasks. Experimental results show the superior performance of AT-MFCGA compared to other methods in solving multiple optimization tasks.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Information Systems
Yanchi Li, Wenyin Gong, Shuijia Li
Summary: Multitasking optimization is the process of solving multiple tasks simultaneously in one go. Many multitasking evolutionary algorithms (MTEAs) have been developed to tackle this problem. Existing MTEAs usually employ a single solver to handle multiple optimization tasks. However, different tasks have different characteristics, which could be more efficiently addressed by using task-specific solvers. Therefore, we propose a multitasking evolutionary framework called MTEA-SaO, which incorporates an adaptive solver selection mechanism and enables knowledge transfer between tasks. Experimental results demonstrate the effectiveness of solver adaptation and knowledge transfer, and the superior performance of MTEA-SaO overall.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Interdisciplinary Applications
Su Jiang, Louis J. Durlofsky
Summary: This study presents a method using coarsened geomodels and transfer learning to construct data-driven surrogate models, reducing the training costs for data assimilation/history matching. The method achieves nearly the same accuracy as high-fidelity models in predicting subsurface flow dynamics and outperforms low-fidelity simulations used for most of the training.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Computer Science, Information Systems
Guodong Chen, Yong Li, Kai Zhang, Xiaoming Xue, Jian Wang, Qin Luo, Chuanjin Yao, Jun Yao
Summary: The study proposes a novel and efficient hierarchical surrogate-assisted differential evolution (EHSDE) algorithm for high-dimensional expensive optimization problems. By balancing exploration and exploitation using a hierarchical framework and utilizing global and local surrogate models to accelerate convergence speed, the algorithm demonstrates effectiveness and efficiency on benchmark functions and production optimization problems.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Guodong Chen, Kai Zhang, Xiaoming Xue, Liming Zhang, Chuanjin Yao, Jian Wang, Jun Yao
Summary: In this paper, a novel surrogate-assisted algorithm called RSAEH is proposed for high-dimensional expensive optimization problems. The algorithm combines local search with surrogate-guided prescreening to improve convergence speed. Experimental results show that RSAEH achieves the best optimization results on benchmark problems and also performs well on a real-world oil reservoir production optimization problem.
APPLIED SOFT COMPUTING
(2022)
Article
Energy & Fuels
Jian Wang, Xue Pang, Faliang Yin, Jun Yao
Summary: This paper introduces the use of deep neural network methods to solve high-dimensional stochastic partial differential equations. By combining two deep convolutional residual networks and training them with the help of an adaptive functional factor, approximate solutions are constructed. In addition, the paper proposes using the energy functional deduced from the variational principles as the loss function for training the neural network. The results show that the model has low computational cost and high efficiency in solving high-dimensional SPDEs.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)
Article
Energy & Fuels
Joshua Kwesi Desbordes, Kai Zhang, Xiaoming Xue, Xiaopeng Ma, Qin Luo, Zhaoqin Huang, Sun Hai, Yao Jun
Summary: This paper introduces a transfer learning based optimization framework for dynamic production optimization problems, which utilizes methods such as domain adaptation learning, extended boundary constraints, and transfer component analysis to improve the optimization speed and accuracy of production control options.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)
Article
Energy & Fuels
Chao Zhong, Kai Zhang, Xiaoming Xue, Ji Qi, Liming Zhang, Chuanjin Yao, Yongfei Yang, Jian Wang, Jun Yao, Weidong Zhang
Summary: This research introduces a novel approach combining multiple surrogate models that imitate the landscape of the initial production optimization problem and an advanced multitasking optimization method to achieve optimal solutions within a limited time frame.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)
Article
Energy & Fuels
Xiaoming Xue, Guodong Chen, Kai Zhang, Liming Zhang, Xinggang Zhao, Linqi Song, Menghan Wang, Peng Wang
Summary: This paper proposes a production optimization method based on the divide-and-conquer optimization paradigm. By decomposing the large-scale production optimization problem into lower-dimensional subproblems and optimizing the surrogate models of the subproblems cooperatively, the joint scheme optimization of the original problem is achieved, and the method can also serve as an auxiliary means of connectivity analysis.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)
Article
Automation & Control Systems
Xiaoming Xue, Kai Zhang, Kay Chen Tan, Liang Feng, Jian Wang, Guodong Chen, Xinggang Zhao, Liming Zhang, Jun Yao
Summary: Evolutionary multitasking (EMT) is a new research topic that aims to improve convergence across multiple optimization tasks by facilitating knowledge transfer. Existing EMT algorithms are limited to homogeneous problems, and little effort has been made to generalize EMT for solving heterogeneous problems. This article proposes a novel rank loss function to achieve superior intertask mapping and derive an analytical solution for affine transformation. The proposed technique can seamlessly integrate with EMT paradigms, and its effectiveness is demonstrated through experiments on synthetic multitasking and many-tasking benchmark problems.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Energy & Fuels
Guodong Chen, Xin Luo, Jiu Jimmy Jiao, Xiaoming Xue
Summary: The study proposes a GDDE algorithm to reduce the number of simulation runs in well-placement and control optimization problems, utilizing PNN as a classifier to select candidates, and building a local surrogate model with RBF to accelerate convergence. The algorithm shows a promising perspective in reducing simulation runs and improving optimization efficiency.
Article
Computer Science, Artificial Intelligence
Xiaoming Xue, Cuie Yang, Yao Hu, Kai Zhang, Yiu-Ming Cheung, Linqi Song, Kay Chen Tan
Summary: In this paper, the authors explore the transfer of knowledge between source and target tasks in objective-heterogeneous problems. They propose a method that separates decision variables and utilizes convergence and diversity transfer modules to accelerate evolutionary search.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Engineering, Biomedical
Zhi-An Huang, Yao Hu, Rui Liu, Xiaoming Xue, Zexuan Zhu, Linqi Song, Kay Chen Tan
Summary: This study proposes an effective federated multi-task learning framework to identify multiple related mental disorders based on functional magnetic resonance imaging data. The framework extracts high-level features across client models using a federated contrastive learning-based feature extractor and performs joint classification using a federated multi-gate mixture of expert classifier. The framework achieves high diagnosis accuracies on real-world datasets and improves the generalization capability of computer-aided detection approaches.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING
(2023)
Article
Engineering, Multidisciplinary
Wei Jiang, Xiaoming Xue, Nan Zhang, Yanhe Xu, Jie Liu, Yahui Shan
Summary: In this paper, a two-stage and working-condition-robust health measurement method is proposed for rolling bearings. It combines energy entropy theory, a deep-learning approach, and transfer-learning technology. The method includes an adaptive variational mode decomposition (VMD) improved fruit fly optimization algorithm (IFOA) to detect abnormal states, and a hybrid robust auto-encoder to extract valuable and robust fault features.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
Faliang Yin, Weiguo Li, Kai Zhang, Jian Wang, Nikhil R. Pal
Summary: The Broad Learning System (BLS) is a concise and efficient learning system that has gained attention. However, it struggles with high computational costs and memory usage when dealing with complex problems and large networks. This work proposes an improved iterative projection learning method for training BLS without using pseudo-inverse. Additionally, an evolutionary bilevel programming method is presented to optimize the hyperparameters of the network structure. Experimental results demonstrate that the proposed methods improve efficiency, robustness, and generalization ability compared to existing approaches.
INFORMATION SCIENCES
(2023)
Article
Energy & Fuels
Ji Qi, Yanqing Liu, Yafeng Ju, Kai Zhang, Lu Liu, Yuanyuan Liu, Xiaoming Xue, Liming Zhang, Huaqing Zhang, Haochen Wang, Jun Yao, Weidong Zhang
Summary: This paper proposes a novel transfer learning framework for well placement optimization by using the feature extraction capability of a single-layer denoising autoencoder. By establishing a reconstruction mapping between previous and present tasks, the randomly generated well locations can inherit knowledge from optimal well locations, which helps the evolutionary algorithm quickly bias towards the optimal solution and accelerate the solving process of the present task.
GEOENERGY SCIENCE AND ENGINEERING
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Yao Hu, Zhi-An Huang, Rui Liu, Xiaoming Xue, Linqi Song, Kay Chen Tan
Summary: The high prevalence of mental disorders poses pressure on public healthcare services. Researchers have developed a dual-stage pseudo-labeling based classification framework (DSPL) for diagnosing mental disorders. The method achieves high diagnosis accuracy on functional magnetic resonance imaging data.
2022 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN)
(2022)
Article
Engineering, Petroleum
Chao Zhong, Kai Zhang, Xiaoming Xue, Ji Qi, Liming Zhang, Xia Yan, Huaqing Zhang, Yongfei Yang
Summary: This paper proposes a production optimization method called historical window-enhanced transfer Gaussian process (HWTGP), which extracts beneficial reservoir knowledge from historical data and avoids massive time for trial and error, achieving superior performance.