检索结果-南通市图书馆

FSS: algorithm and Neural network accelerator for style transfer

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第2期67卷 253-266页

作者： Yi LING Yujie HUANG Yujie CAI Zhaojie LI Mingyu WANG Wenhong LI Xiaoyang ZENG State Key Laboratory of ASIC & System Fudan University Shanghai Explore X Technology Co. Ltd.

Neural networks(NNs), owing to their impressive performance, have gradually begun to dominate multimedia processing. For resource-constrained and energy-sensitive mobile devices, an efficient NN accelerator is necessary. Style transfer is an important multimedia application. However, existing arbitrary style transfer networks are complex and not well supported by current NN accelerators, limiting their application on mobile devices. Moreover, the quality of style transfer needs improvement. Thus, we design the Fast Style system(FSS), where a novel algorithm and an NN accelerator are proposed for style transfer. In FSS, we first propose a novel arbitrary style transfer algorithm, Fast Style. We propose a light network that contributes to high quality and low computational complexity and a prior mechanism to avoid retraining when the style changes. Then, we redesign an NN accelerator for Fast Style by applying two improvements to the basic NVIDIA deep learning accelerator(NVDLA) architecture. First, a flexible dat FSM and wt FSM are redesigned to enable the original data path to perform other operations(including the GRAM operation)by software programming. Moreover, statistics and judgment logic are designed to utilize the continuity of a video stream and remove the data dependency in the instance normalization, which improves the accelerator performance by 18.6%. The experimental results demonstrate that the proposed Fast Style can achieve higher quality with a lower computational cost, making it more suitable for mobile devices. The proposed NN accelerator is implemented on the Xilinx VCU118 FPGA under a 180-MHz clock. Experimental results show that the accelerator can stylize 512×512-pixel video with 20 FPS, and the measured performance reaches up to 306.07 GOPS. The ASIC implementation in TSMC 28 nm achieves about 22 FPS in the case of a 720-p video.

关键词： Neural network accelerator style transfer Neural network deep learning

Turbo-shaft engine adaptive Neural network control based on nonlinear state space equation

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Chinese Journal of Aeronautics 2024年第4期37卷 493-507页

作者： Ziyu GU Qiuhong LI Shuwei PANG Wenxiang ZHOU Jichang WU Chenyang ZHANG Jiangsu Province Key Laboratory of Aerospace Power System Nanjing University of Aeronautics and AstronauticsNanjing 210016China AECC Hunan Aviation Powerplant Research Institute Zhuzhou 412002China

Intelligent Adaptive Control(AC) has remarkable advantages in the control system design of aero-engine which has strong nonlinearity and uncertainty. Inspired by the Nonlinear Autoregressive Moving Average(NARMA)-L2 adaptive control, a novel Nonlinear State Space Equation(NSSE) based Adaptive Neural network Control(NSSE-AC) method is proposed for the turbo-shaft engine control system design. The proposed NSSE model is derived from a special Neural network with an extra layer, and the rotor speed of the gas turbine is taken as the main state variable which makes the NSSE model be able to capture the system dynamic better than the NARMA-L2 model. A hybrid Recursive Least-Square and Levenberg-Marquardt(RLS-LM) algorithm is advanced to perform the online learning of the Neural network, which further enhances both the accuracy of the NSSE model and the performance of the adaptive controller. The feedback correction is also utilized in the NSSE-AC system to eliminate the steady-state tracking error. Simulation results show that, compared with the NARMA-L2 model, the NSSE model of the turboshaft engine is more accurate. The maximum modeling error is decreased from 5.92% to 0.97%when the LM algorithm is introduced to optimize the Neural network parameters. The NSSE-AC method can not only achieve a better main control loop performance than the traditional controller but also limit all the constraint parameters efficiently with quick and accurate switching responses even if component degradation exists. Thus, the effectiveness of the NSSE-AC method is validated.

关键词： Adaptive control systems Turbo-shaft engine Neural network Nonlinear state space equation NARMA-L2

A new neural-network-based method for structural damage identification in single-layer reticulated shells

在线全文

学校读者我要写书评

暂无评论

Frontiers of Structural and Civil Engineering 2024年第1期18卷 104-121页

作者： Jindong ZHANG Xiaonong GUO Shaohan ZONG Yujian ZHANG College of Civil Engineering Tongji UniversityShanghai 200092China China Construction Eighth Engineering Division Co. Ltd.Shanghai 200135China

Single-layer reticulated shells(SLRSs)find widespread application in the roofs of crucial public structures,such as gymnasiums and exhibition center.In this paper,a new neural-network-based method for structural damage identification in SLRSs is proposed.First,a damage vector index,NDL,that is related only to the damage localization,is proposed for SLRSs,and a damage data set is constructed from NDL data.On the basis of visualization of the NDL damage data set,the structural damaged region locations are identified using convolutional Neural networks(CNNs).By cross-dividing the damaged region locations and using parallel CNNs for each regional location,the damaged region locations can be quickly and efficiently identified and the undamaged region locations can be eliminated.Second,a damage vector index,DS,that is related to the damage location and damage degree,is proposed for SLRSs.Based on the damaged region identified previously,a fully connected Neural network(FCNN)is constructed to identify the location and damage degree of members.The effectiveness and reliability of the proposed method are verified by considering a numerical case of a spherical SLRS.The calculation results showed that the proposed method can quickly eliminate candidate locations of potential damaged region locations and precisely determine the location and damage degree of members.

关键词： single-layer reticulated shell damage identification Neural network convolutional Neural network cross-partitioning method

Beam based alignment using a Neural network

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Nuclear Science and Techniques 2024年第4期35卷 108-118页

作者： Guan-Liang Wang Ke-Min Chen Si-Wei Wang Zhe Wang Tao He Masahito Hosaka Guang-Yao Feng Wei Xu National Synchrotron Radiation Laboratory University of Science and Technology of ChinaHefei 230029China Diamond Light Source Oxfordshire OX110DEUK

Beams typically do not travel through the magnet centers because of errors in storage rings.The beam deviating from the quadrupole centers is affected by additional dipole fields due to magnetic field feed-down.Beam-based alignment(BBA)is often performed to determine a golden orbit where the beam circulates around the quadrupole center axes.For storage rings with many quadrupoles,the conventional BBA procedure is time-consuming,particularly in the commissioning phase,because of the necessary iterative process.In addition,the conventional BBA method can be affected by strong coupling and the nonlinearity of the storage ring optics.In this study,a novel method based on a Neural network was proposed to determine the golden orbit in a much shorter time with reasonable accuracy.This golden orbit can be used directly for operation or adopted as a starting point for conventional BBA.The method was demonstrated in the HLS-II storage ring for the first time through simulations and online experiments.The results of the experiments showed that the golden orbit obtained using this new method was consistent with that obtained using the conventional BBA.The development of this new method and the corresponding experiments are reported in this paper.

关键词： Golden orbit Beam-based alignment Neural network Storage ring

Machine learning and Neural network supported state of health simulation and forecasting model for lithium-ion battery

在线全文

学校读者我要写书评

暂无评论

Frontiers in Energy 2024年第2期18卷 223-240页

作者： Nan QI Kang YAN Yajuan YU Rui LI Rong HUANG Lai CHEN Yuefeng SU Beijing Institute of Technology Chongqing Innovation Center Chongqing 401120China Department of Energy and Environmental Materials School of Materials Science and EngineeringBeijing Institute of TechnologyBeijing 100081China Beijing Electric Vehicle Co. Ltd.Beijing 100176China

As the intersection of disciplines deepens,the field of battery modeling is increasingly employing various artificial intelligence(AI)approaches to improve the efficiency of battery management and enhance the stability and reliability of battery operation.This paper reviews the value of AI methods in lithium-ion battery health management and in particular analyses the application of machine learning(ML),one of the many branches of AI,to lithium-ion battery state of health(SOH),focusing on the advantages and strengths of Neural network(NN)methods in ML for lithium-ion battery SOH simulation and prediction.NN is one of the important branches of ML,in which the application of NNs such as backpropagation NN,convolutional NN,and long short-term memory NN in SOH estimation of lithium-ion batteries has received wide attention.Reports so far have shown that the utilization of NN to model the SOH of lithium-ion batteries has the advantages of high efficiency,low energy consumption,high robustness,and scalable models.In the future,NN can make a greater contribution to lithium-ion battery management by,first,utilizing more field data to play a more practical role in health feature screening and model building,and second,by enhancing the intelligent screening and combination of battery parameters to characterize the actual lithium-ion battery SOH to a greater extent.The in-depth application of NN in lithium-ion battery SOH will certainly further enhance the science,reliability,stability,and robustness of lithium-ion battery management.

关键词： machine learning lithium-ion battery state of health Neural network artificial intelligence

Event-Triggered Optimal Nonlinear Systems Control Based on State Observer and Neural network

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Journal of Systems Science & Complexity 2023年第1期36卷 222-238页

作者： CHENG Songsong LI Haoyun GUO Yuchao PAN Tianhong FAN Yuan Anhui Engineering Laboratory of Human-Robot Integration System and Intelligent Equipment School of Electrical Engineering and AutomationAnhui UniversityHefei 230601China Key Laboratory of Intelligent Computing and Signal Processing of Ministry of Education School of Electrical Engineering and AutomationAnhui UniversityHefei 230601China

This paper develops a novel event-triggered optimal control approach based on state observer and Neural network(NN)for nonlinear continuous-time systems.Firstly,the authors propose an online algorithm with critic and actor NNs to solve the optimal control problem and provide an event-triggered method to reduce communication and computation burdens.Moreover,the authors design weight estimation for critic and actor NNs based on gradient descent method and achieve uniformly ultimate boundednesss(UUB)estimation results.Furthermore,by using bounded NN weight estimation and dead-zone operator,the authors propose a triggering condition,prove the asymptotic stability of closed-loop system from Lyapunov stability perspective,and exclude the Zeno behavior.Finally,the authors provide a numerical example to illustrate the effectiveness of the proposed method.

关键词： Event-triggered control Neural network optimal control state observer

neural-network-based fully distributed formation control for nonlinear multi-agent systems with event-triggered communication

在线全文

学校读者我要写书评

暂无评论

Science China(Technological Sciences) 2024年第1期67卷 209-220页

作者： ZHU GuoLiang LIU KeXin GU HaiBo LÜJinHu School of Automation Science and Electrical Engineering Beihang UniversityBeijing 100191China Zhongguancun Laboratory Beijing 100094China

This paper investigates the consensus-based formation control problem for multi-agent systems with unknown nonlinear dynamics.To achieve the desired formation,we propose two formation controllers to achieve the desired formation,one based on system states and the other on system outputs.The proposed controllers utilize adaptive gains to avoid global information and Neural networks to estimate and compensate for nonlinearities.The proposed event-triggered schemes avoid continuous communication among agents and exclude the Zeno behavior.Stability analysis reveals that formation errors are bounded,and numerical simulations are used to validate the effectiveness of the proposed approaches.

关键词： formation control Neural network adaptive control event-triggered communication multi-agent systems

Combining Neural network-based method with heuristic policy for optimal task scheduling in hierarchical edge cloud

在线全文

学校读者我要写书评

暂无评论

Digital Communications and networks 2023年第3期9卷 688-697页

作者： Zhuo Chen Peihong Wei Yan Li College of Computer Science and Engineering Chongqing University of TechnologyChongqing200433China School of Artificial Intelligence Chongqing University of TechnologyChongqing200433China

Deploying service nodes hierarchically at the edge of the network can effectively improve the service quality of offloaded task requests and increase the utilization of resources.In this paper,we study the task scheduling problem in the hierarchically deployed edge cloud.We first formulate the minimization of the service time of scheduled tasks in edge cloud as a combinatorial optimization problem,blue and then prove the NP-hardness of the problem.Different from the existing work that mostly designs heuristic approximation-based algorithms or policies to make scheduling decision,we propose a newly designed scheduling policy,named Joint Neural network and Heuristic Scheduling(JNNHSP),which combines a Neural network-based method with a heuristic based solution.JNNHSP takes the Sequence-to-Sequence(Seq2Seq)model trained by Reinforcement Learning(RL)as the primary policy and adopts the heuristic algorithm as the auxiliary policy to obtain the scheduling solution,thereby achieving a good balance between the quality and the efficiency of the scheduling solution.In-depth experiments show that compared with a variety of related policies and optimization solvers,JNNHSP can achieve better performance in terms of scheduling error ratio,the degree to which the policy is affected by re-sources limitations,average service latency,and execution efficiency in a typical hierarchical edge cloud.

关键词： Edge cloud Task scheduling Neural network Reinforcement learning

Neural network Based Feasible Region Approximation Model for Optimal Operation of Integrated Electricity and Heating System

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

CSEE Journal of Power and Energy Systems 2023年第5期9卷 1808-1819页

作者： Xuewei Wu Bin Zhang Mads Pagh Nielsen Zhe Chen Aalborg University Aalborg 9220Denmark

This paper proposes a Neural network based feasible region approximation model of a district heating system(DHS),and it is intended to be used for optimal operation of integrated electricity and heating system(IEHS)considering privacy protection.In this model,a Neural network is trained to approximate the feasible region of the DHS operation and then is reformulated as a set of mixed-integer linear constraints.Based on the received approximation models of DHSs and detailed electricity system model,the electricity operator conducts centralized optimization,and then sends specific heating generation plans back to corresponding heating operators.Furthermore,subsequent optimization is formulated for each DHS to obtain detailed operation strategy based on received heating generation plan.In this scheme,optimization of the IEHS could be achieved and privacy protection requirement is satisfied since the feasible region approximation model does not contain detailed system parameters.Case studies conducted on a small-scale system demonstrate accuracy of the proposed strategy and a large-scale system verify its application possibility.

关键词： Artificial intelligence district heating system integrated energy system machine learning multi-energy systems Neural network optimal operation wind power