检索结果-内蒙古大学图书馆

A critical look at deep neural network for dynamic system modeling

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Zhou, Jinming Zhu, Yucai State Key Laboratory of Industrial Control Technology College of Control Science and Engineering Zhejiang University Hangzhou310027 China

Neural network models become increasingly popular as dynamic modeling tools in the control community. They have many appealing features including nonlinear structures, being able to approximate any functions. While most researchers hold optimistic attitudes towards such models, this paper questions the capability of (deep) neural networks for the modeling of dynamic systems using input-output data. For the identification of linear time-invariant (LTI) dynamic systems, two representative neural network models, Long Short-Term Memory (LSTM) and Cascade Foward Neural Network (CFNN) are compared to the standard Prediction Error Method (PEM) of system identification. In the comparison, four essential aspects of system identification are considered, then several possible defects and neglected issues of neural network based modeling are pointed out. Detailed simulation studies are performed to verify these defects: for the LTI system, both LSTM and CFNN fail to deliver consistent models even in noise-free cases;and they give worse results than PEM in noisy cases. Copyright © 2023, The Authors. All rights reserved.

关键词： Deep neural networks

IG2: Integrated Gradient on Iterative Gradient Path for Feature Attribution

学校读者我要写书评

暂无评论

TechRxiv

TechRxiv 2023年

作者： Zhuo, Yue Ge, Zhiqiang The State Key Laboratory of Industrial Control Technology College of Control Science and Engineering Zhejiang University Hangzhou310027 China

Feature attribution explains Artificial Intelligence (AI) at the instance level by providing importance scores of input features’ contributions to model prediction. Integrated Gradients (IG) is the prevalent path attribution method for deep neural networks, involving the integration of gradients along a path from the explained input (explicand) to a counterfactual instance (baseline). Nonetheless, current IG variants exclusively consider the gradient of explicand’s output, but we have discovered that the gradient of the counterfactual output also exerts a substantial influence on feature attribution. To achieve this, we propose Iterative Gradient path Integrated Gradients (IG2), considering both two gradients. IG2 incorporates the counterfactual gradient iteratively into the integration path, consequently obtaining a novel path (GradPath) and a novel baseline (GradCF). These two novel IG components effectively address the issues of attribution noise and arbitrary baseline choice in earlier IG methods. IG2, as a path method, stratifies many desirable axioms, which are theoretically justified in the paper. Experimental results on XAI benchmark, ImageNet, MNIST, TREC questions answering, wafer-map failure patterns, and CelebA face attributes validate that IG2 delivers superior feature attributions compared to the state-of-the-art techniques. © 2023, CC BY.

关键词： Deep neural networks

Reproductive Influences of Erythromycin and Sulfamethoxazole on Caenorhabditis elegans over Generations Mediated by Lipid Metabolism

学校读者我要写书评

暂无评论

Chemical Research in Chinese Universities 2023年第3期39卷 434-440页

作者： YU Zhenyang WANG Lei LI Gaotian ZHANG Jing Ecological Technique and Engineering College Shanghai Institute of TechnologyShanghai201418P.R.China Jiaxing Tongji Institute for Environment Jiaxing314051P.R.China State Key Laboratory of Pollution Control and Resource Reuse Key Laboratory of Yangtze River Water EnvironmentMinistry of EducationCollege of Environmental Science and EngineeringTongji UniversityShanghai200092P.R.China

Antibiotics are emerging pollutants with ***,their reproductive toxicities over generations and underlying mechanisms were seldom *** the present study,reproductive effects of erythromycin(ERY)and sulfamethoxazole(SMZ)were studied on Caenorhabditis elegans(***)with multi-generational exposure from F1 to F4,and their effects in the non-direct exposed offspring of F1 and F4(T4 and T4’)were explored to investigate trans-generational *** high concentrations,ERY and SMZ inhibited the reproduction and the inhibition decreased with more generations’*** low concentrations,SMZ stimulated reproduction in T4,opposite to its inhibition in *** analysis on ten enzymes and proteins in lipid and glucolipid metabolism showed that ERY commonly inhibited fatty acid synthetase(FAS)in F1 to F4,while ERY and SMZ showed stimulation and inhibition on other *** biomarker response(IBR)index values showed that ERY and SMZ stimulated the overall metabolism in F1,and the stimulation decreased and even altered to inhibition in later *** comparing multi-and trans-generational effects,ERY and SMZ at high concentration caused inhibition on reproduction in T4 and stimulation in T4’,which were opposite to those in F1 and F4.

关键词： Multi-generational effect Antibiotic Reproduction Lipid metabolism

Failure evolution and instability prediction of fiber-reinforced polymer-confined cement mortar specimens under axial compression

学校读者我要写书评

暂无评论

Environmental science and Pollution Research 2024年第20期31卷 29730-29748页

作者： Li, Jian Bai, Jinwen Han, Yanna Feng, Guorui Shi, Xudong Ma, Junbiao Cui, Boqiang Wang, Kai Mi, Jiachen College of Mining Engineering Taiyuan University of Technology Shanxi Province Taiyuan030024 China Key Laboratory of Shanxi Province for Mine Rock Strata Control and Disaster Prevention Taiyuan University of Technology Taiyuan030024 China Shanxi-Zheda Institute of Advanced Materials and Chemical Engineering Taiyuan030024 China Department of Mining Engineering Shanxi Institute of Energy Jinzhong030600 China College of Safety and Emergency Management Engineering Taiyuan University of Science and Technology Taiyuan030024 China

In geotechnical engineering, a large number of pillars are often left in underground space to support the overlying strata and protect the surface environment. To enhance pillar stability and prevent instability, this study proposes an innovative technology for pillar reinforcement. Specifically, local confinement of the pillar is achieved through fiber-reinforced polymer (FRP) strips, resulting in the formation of a more stable composite structure. In order to validate the effectiveness of this structural approach, acoustic emission characteristics and surface strain field characteristics were monitored during failure processes, while mathematical models were employed to predict specimen instability. The test results revealed that increasing FRP strip confinement width led to heightened activity in acoustic emission events during failure processes, accompanied by a decrease in shear cracks but an increase in tensile cracks. Moreover, ductility was improved and deformation resistance capacity was enhanced within specimens. Notably, initial crack generation occurred within unconfined regions of specimens during failures;however, both length and width as well as overall numbers of cracks significantly decreased due to implementation of FRP strips. Consequently, specimen failure speed was slowed down accordingly. Finally, the instability of the partial FRP-confined cement mortar could be more accurately predicted based on the model of FRP-confined concrete. It was verified by the test results. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.

关键词： Stability

Robust Nonlinear Self-Triggered control Policy for a Novel Fully Actuated UAVs

学校读者我要写书评

暂无评论

IFAC-PapersOnLine 2023年第2期56卷 2506-2512页

作者： Zhangzhen Zhu Yongliang Lin Yu Zhang State Key Laboratory of Industrial Control Technology College of Control Science and Engineering Zhejiang University Hangzhou China Key Laboratory of Collaborative Sensing and Autonomous Unmanned Systems of Zhejiang Province Hangzhou China

This paper proposes a self-triggered control policy for a general kind of nonlinear system, ensuring robust performance under various perturbations and alleviating communication burden simultaneously. A novel fully actuated multirotors applied this method shows great robustness to perturbations and is able of tracking six dimensional decoupled trajectories without continuous background sensor monitoring. Rigorous input-to-state stability analysis is deduced and numerical simulations confirm the validity of the policy.

关键词： UAVs fully actuated event-triggered robustness nonlinear systems

Directed Acyclic Graphs With Tears

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Chen, Zhichao Ge, Zhiqiang State Key Laboratory of Industrial Control Technology College of Control Science and Engineering Zhejiang University Hangzhou310027 China

Bayesian network is a frequently-used method for fault detection and diagnosis in industrial processes. The basis of Bayesian network is structure learning which learns a directed acyclic graph from data. Since the search space in structure learning will scale super-exponentially with the increase of process variables, data-driven structure learning is a challenging problem. As a novel method for structure learning, DAGs with No Tears methods are being well studied in recent years due to their compatibility with deep learning framework. However, the DAGs with No Tears methods is far from application in industrial scenario due to problems in the gradient descent based solving stage and the post-processing stage. In this work, those problems are theoretically analyzed in detail by mathematical derivations. To solve these problems, the DAGs with Tears method is proposed by using mix-integer linear programming under the deep learning framework. In addition, prior knowledge is able to incorporate into the new proposed method, making structure learning more practical and useful in industrial processes. Finally, a numerical example and an industrial simulation example are adopted as case studies to demonstrate the superiority of the developed method. © 2023, CC BY.

关键词： Bayesian networks

TMoE-P: Towards the Pareto Optimum for Multivariate Soft Sensors

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Pan, Licheng Wang, Hao Chen, Zhichao Huang, Yuxing Liu, Xinggao The State Key Laboratory of Industrial Control Technology College of Control Science and Engineering Zhejiang University Hangzhou310027 China

Multi-variate soft sensor seeks accurate estimation of multiple quality variables using measurable process variables, which have emerged as a key factor in improving the quality of industrial manufacturing. The current progress stays in some direct applications of multitask network architectures;however, there are two fundamental issues remain yet to be investigated with these approaches: (1) negative transfer, where sharing representations despite the difference of discriminate representations for different objectives degrades performance;(2) seesaw phenomenon, where the optimizer focuses on one dominant yet simple objective at the expense of others. In this study, we reformulate the multi-variate soft sensor to a multi-objective problem, to address both issues and advance state-of-the-art performance. To handle the negative transfer issue, we first propose an Objective-aware Mixture-of-Experts (OMoE) module, utilizing objective-specific and objective-shared experts for parameter sharing while maintaining the distinction between objectives. To address the seesaw phenomenon, we then propose a Pareto Objective Routing (POR) module, adjusting the weights of learning objectives dynamically to achieve the Pareto optimum, with solid theoretical supports. We further present a Task-aware Mixture-of-Experts framework for achieving the Pareto optimum (TMoE-P) in multi-variate soft sensor, which consists of a stacked OMoE module and a POR module. We illustrate the efficacy of TMoE-P with an open soft sensor benchmark, where TMoE-P effectively alleviates the negative transfer and seesaw issues and outperforms the baseline models. © 2023, CC BY.

关键词： Multiobjective optimization

Dyna-DepthFormer: Multi-frame Transformer for Self-Supervised Depth Estimation in Dynamic Scenes

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Zhang, Songchun Zhao, Chunhui State Key Laboratory of Industrial Control Technology College of Control Science and Engineering Zhejiang University Hangzhou310027 China

— Self-supervised methods have showed promising results on depth estimation task. However, previous methods estimate the target depth map and camera ego-motion simultaneously, underusing multi-frame correlation information and ignoring the motion of dynamic objects. In this paper, we propose a novel Dyna-Depthformer framework, which predicts scene depth and 3D motion field jointly and aggregates multi-frame information with transformer. Our contributions are twofold. First, we leverage multi-view correlation through a series of self- and cross-attention layers in order to obtain enhanced depth feature representation. Specifically, we use the perspective transformation to acquire the initial reference point, and use deformable attention to reduce the computational cost. Second, we propose a warping-based Motion Network to estimate the motion field of dynamic objects without using semantic prior. To improve the motion field predictions, we propose an iterative optimization strategy, together with a sparsity-regularized loss. The entire pipeline achieves end-to-end self-supervised training by constructing a minimum reprojection loss. Extensive experiments on the KITTI and Cityscapes benchmarks demonstrate the effectiveness of our method and show that our method outperforms state-of-the-art algorithms. Copyright © 2023, The Authors. All rights reserved.

关键词： Semantics

MODEL PREDICTIVE control OF DISCRETE-TIME LINEAR SYSTEMS BY ADMM WITH APPLICATIONS TO TURBOFAN ENGINE control PROBLEMS

学校读者我要写书评

暂无评论

Journal of Applied and Numerical Optimization

Journal of Applied and Numerical Optimization 2023年第2期5卷 201-217页

作者： Teng, Jiao Du, Xian Wang, Lei Wang, Xinwei Wu, Junmei Department of Applied Mathematics The Hong Kong Polytechnic University Hong Kong Hong Kong School of Control Science and Engineering Dalian University of Technology Dalian China School of Mathematical Sciences Dalian University of Technology Dalian China Department of Engineering Mechanics State Key Laboratory of Structural Analysis for Industrial Equipment Dalian University of Technology Dalian China

In this paper, we consider optimal control problems with linear discrete state space model, which originate from a class of turbofan engines. The optimization problem associated with each moving horizon estimation (MHE) in classical model predictive control (MPC) is a quadratic programming (QP) problem, and the general QP algorithms does not exploit the structural features of the turbofan engine itself to improve the computational efficiency of the algorithm. In the framework of model predictive control, the turbofan engine model makes the rolling optimization subproblem exhibit a sparse structure. Based on this feature, the alternating direction method of multipliers (ADMM) is employed to solve each optimization sub-problem and design an improved MPC-ADMM algorithm for solving this class of optimal control problems. The simulation results are compared with the MPC-QP algorithm by numerical examples to show the effectiveness and superiority of the novel algorithm. © 2023 Journal of Applied and Numerical Optimization.

关键词： Model predictive control