Xinhua Zhang

Northwestern Polytechnical University

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

Zhiyi Song

Beijing Automation Control Equipment Institute

Dale Schuurmans

Google (United States)

Jian Huang

Beijing Automation Control Equipment Institute

Yaoliang Yu

Vector Institute

Xiaodong Sun

Jiangsu University

Xiaoshuai Duan

Beijing Automation Control Equipment Institute

Guan Wang

Dalian University of Technology

Yukun Sun

Dalian Maritime University

Xijian Huo

Beijing Automation Control Equipment Institute

Ankan Saha

Khulna University of Engineering and Technology

Cooperative Institutions

Chinese Academy of Sciences

Jiangsu University

Beijing Automation Control Equipment Institute

Xi'an Jiaotong University

Shanghai Jiao Tong University

Harbin Institute of Technology

Tsinghua University

Southeast University

Peking University

Sun Yat-sen University

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

Distilling Conditional Diffusion Models for Offline Reinforcement Learning through Trajectory Stitching

arXiv (Cornell University) (2024)

Shangzhe Li Xinhua Zhang

Deep generative models have recently emerged as an effective approach to offline reinforcement learning. However, their large model size poses challenges in computation. We address this issue by proposing a knowledge distillation method based on data augmentation. In particular, high-return trajectories are generated from a conditional diffusion model, and they are blended with the original trajectories through a novel stitching algorithm that leverages a new reward generator. Applying the resulting dataset to behavioral cloning, the learned shallow policy whose size is much smaller outperforms or nearly matches deep generative planners on several D4RL benchmarks.

Image stitching

10.48550/arxiv.2402.00807

Cite

Citations (0)

Design of a novel limited angle toque motor with compound Halbach array for electric direct-drive servo system in aerospace vehicle

Xinhua Zhang Tianyi Wang Jian Huang Guan Wang Zhiyi Song

Aerospace vehicle is the next-generation space transportation system with horizontal take-off/Landing and reusable features. However, due to its hypersonic speed, super-maneuver and safety performance, conventional servo system can no longer satisfy the harsh requirements of high dynamic response and reliability in aerospace vehicles. In this paper, an electric direct-drive servo system is proposed based on novel limited angle torque motor to achieve fast frequency response and fault-tolerant simultaneously. For the motor, outer-rotor topology with compound Halbach array is researched to enhance output performance. In addition, multiple single-phase windings with flat wire are designed to facilitate redundancy configuration in the stator. Firstly, the structure of the novel motor and operation principle of the system is presented. Then, focusing on the topology, electromagnetic finite-element (FE) analysis and comparison are conducted to validate its output advantages. Finally, electric redundancy configuration method is demonstrated. And it is expected to provide a new solution to aerospace servo system.

Halbach array

Servomechanism

10.1109/icems.2019.8921441

Cite

Citations (1)

Open Problem: Lower bounds for Boosting with Hadamard

Manfred K. Warmuth Xinhua Zhang

Boosting algorithms can be viewed as a zero-sum game. At each iteration a new column / hypothesis is chosen from a game matrix representing the entire hypotheses class. There are algorithms for which the gap between the value of the sub-matrix (the t columns chosen so far) and the value of the entire game matrix is O( q logn t ). A matching lower bound has been shown for random game matrices for t up to n where 2 (0; 1 ). We conjecture that with Hadamard matrices we can build a certain game matrix for which the game value grows at the slowest possible rate for t up to a fraction of n.

Matrix (chemical analysis)

Boosting

Source

Cite

Citations (0)

Improved Neural Network Based on Genetic Algorithm in the Application for Tuning of PI Controllers

Electrical Automation (2012)

Xinhua Zhang

To overcome the disadvantage of conventional neutral network which is training only offline, an improved neutral network based on genetic algorithm is proposed. Starting from the performance index of self-tuning PI controllers, principle of recurrent neutral network and speed control system model of induction motor are elaborated. Then, the feasibility of by genetic algorithm to seak optimal constants of extended kalman filter is explained. Finally, modeling and simulation are performed by MATLAB/Simulink with control system of induction motors as platform. The results show that the PI parameters obtained by the proposed improved neutral network based on genetic algorithm can reduce the overshoot of system effectively and the system has good dynamic and static performance.

Overshoot (microwave communication)

Neutral network

Source

Cite

Citations (1)

OPTIMAL DESIGN OF MAGNETICALLY LEVITATED PERMANENT-MAGNET PLANAR MOTORS USING THD<sub>2D</sub> METHOD

Progress In Electromagnetics Research M (2015)

Xinhua Zhang Hao Luo Zebin Yang Xiaodong Sun Jinchun Hu

In order to reduce the higher harmonic component of the magnetic field in magnetically levitated permanent-magnet planar motors (PMPMs), a sinusoidal index for 2-dimensional (2D) wave is proposed and verified by calculating the average total harmonic distortion (THD) of the discrete waveform. Based on this, a novel optimal design method using THD for 2-dimensional wave (THD2D) and fundamental amplitude is proposed for PMPMs. The finite element analysis and experimental results show that the higher harmonic component of magnetic field is reduced with average fundamental amplitude invariant, and the back-EMF is better. So the optimal design method can improve the electromagnetic performance of magnetically levitated PMPMs.

10.2528/pierm15020304

Cite

Citations (0)

Superconducting magnetic energy storage controller design based on Hamiltonian system theory

Dianli zidonghua shebei (2003)

Xinhua Zhang

Considering the nonlinear characteristic of power system,a method is proposed to trans-form an affined nonlinear system existing in the power system into a standard Hamiltonian system,based on which the active power controller is designed.The reactive power controller is designed by the traditional method.The PCH(Port Controlled Hamiltonian)model of SMES(Superconducting Magnetic Energy Storage)is established based on the forced Hamiltonian theory.To deal with the problems of outside disturbance and parametric indetermination,the adaptive L 2 -gain controller of SMES in single-machine infinite-bus system is designed.The results of digital simulation manifest that the controller can attenuate disturbance and improve the dynamic performance efficiently.

Hamiltonian (control theory)

Source

Cite

Citations (0)

Risk Analysis of Power Bidding Market Based on Extreme Value Theory

Xinhua Zhang Mingyong Lai

This paper proposes three econometric AR-EGARCH models, whose error is governed by normal, students' t and GED distribution respectively, to nine o'clock spot price serial of Nordpool power market from 2001 to 2005 year. In addition, POT method of extreme value theory is adopted to model the tails of the return distribution, and the parameters of GPD distribution are estimated, then VaR and CVaR are measured under 1% and 5% confidence level. In the end, factual return serial consisted of 175 datas in 2006 year is adopted to test the proposed models, the statistical test shows that CVaR, not VaR, is a good tool to measure risk, and the hypothesis of AR-EGARCH models' error are students' t and GED distribution provides appropriate estimate, which one is better estimate factual return lies on confidence level.

CVAR

Value (mathematics)

10.1109/wicom.2008.2503

Cite

Citations (0)

Ballistic performance of UHMWPE fabrics/EAMS hybrid panel

Journal of Materials Science (2018)

Xiaofei Liu Maoqing Li Xin Li Xiaobin Deng Xinhua Zhang

Armour

Ultra-high-molecular-weight polyethylene

10.1007/s10853-018-2055-4

Cite

Citations (39)

A Research on Medical Image Interpolation Based on Cubic Convolution and Adaptive Interpolation.

International MultiConference of Engineers and Computer Scientists (2007)

Anna Wang Kai Luo Peng Li Xinhua Zhang

Interpolation

Convolution (computer science)

Image scaling

Monotone cubic interpolation

Overlap–add method

Source

Cite

Citations (0)

Bayesian Model of Ice Thickness and Its Analysis Based on Error Correction

Journal of Hunan University (2011)

Xinhua Zhang

A statistical inference from the existing ice thickness data was made.The time series model of ice thickness was established,and the statistical inference from ice thickness was made.Then,the parameter estimation of the inference model by MCMC was conducted on the basis of the Gibbs sampling.At last,error correction of the Markov Chain Monte Carlo(MCMC) model was carried out to solve the error increase problem when the sample data was not enough during the maximum likelihood estimation.A comparison between the Gibbs sampling results by WINBUGS software and the maximum likelihood estimation result was made.By analysis and comparison,it has been proved that the Bayesian method for error correction has a higher accuracy for the estimation of the ice thickness on the cable.

Gibbs sampling

Statistical Inference

Metropolis–Hastings algorithm

Source

Cite

Citations (0)