王灏,王爱科,杨青巍,丁玄同 核工业西南物理研究院 HL-2A 装置上关于破裂预测的一些尝试 第十三届全国等离子体科学技术会议 2007, 8, 20-22, 成都 核工业西南物理研究院 王灏,等. 第十三届全国等离子体科学技术会议 2007, 8, 20-22,

Slides:



Advertisements
Similar presentations
A Brief Overview of Neural Networks By Rohit Dua, Samuel A. Mulder, Steve E. Watkins, and Donald C. Wunsch.
Advertisements

Digital signal processing -G Ravi kishore. INTRODUCTION The goal of DSP is usually to measure, filter and/or compress continuous real-world analog signals.
第十四届全国核结构大会暨第十次全国核结构专题讨论会 浙江 · 湖州 · Nuclear matter with chiral forces in Brueckner-Hartree-Fock approximation 李增花 复旦大学核科学与技术系(现代物理研究所)
Artificial Intelligence (CS 461D)
An introduction to: Deep Learning aka or related to Deep Neural Networks Deep Structural Learning Deep Belief Networks etc,
Prénom Nom Document Analysis: Artificial Neural Networks Prof. Rolf Ingold, University of Fribourg Master course, spring semester 2008.
Neural Networks. R & G Chapter Feed-Forward Neural Networks otherwise known as The Multi-layer Perceptron or The Back-Propagation Neural Network.
Introduction to Neural Network Justin Jansen December 9 th 2002.
Introduction to Neural Networks Simon Durrant Quantitative Methods December 15th.
Some Applications of Neural Networks in Plasma Physics and Fusion Materials Modelling R. Kemp 1, G. Cottrell 2, and H. K. D. H Bhadeshia 1 University of.
-Artificial Neural Network- Chapter 5 Back Propagation Network
Artificial Neural Networks
© Negnevitsky, Pearson Education, Lecture 7 Artificial neural networks: Supervised learning Introduction, or how the brain works Introduction, or.
Multiple-Layer Networks and Backpropagation Algorithms
Cascade Correlation Architecture and Learning Algorithm for Neural Networks.
Artificial Neural Networks
Using Neural Networks in Database Mining Tino Jimenez CS157B MW 9-10:15 February 19, 2009.
1 Artificial Neural Networks Sanun Srisuk EECP0720 Expert Systems – Artificial Neural Networks.
IPP - Garching Reflectometry Diagnostics and Rational Surface Localization with Fast Swept Systems José Vicente
IE 585 Introduction to Neural Networks. 2 Modeling Continuum Unarticulated Wisdom Articulated Qualitative Models Theoretic (First Principles) Models Empirical.
© Copyright 2004 ECE, UM-Rolla. All rights reserved A Brief Overview of Neural Networks By Rohit Dua, Samuel A. Mulder, Steve E. Watkins, and Donald C.
-Artificial Neural Network- Hopfield Neural Network(HNN) 朝陽科技大學 資訊管理系 李麗華 教授.
Machine Learning Chapter 4. Artificial Neural Networks
Chapter 3 Neural Network Xiu-jun GONG (Ph. D) School of Computer Science and Technology, Tianjin University
L. Chen US TTF meeting, 2014 April 22-25, San Antonio, Texas 1 Study on Power Threshold of the L-I-H Transition on the EAST Superconducting Tokamak L.
The study of MARFE during long pulse discharges in the HT-7 tokamak W.Gao, X.Gao, M.Asif, Z.W.Wu, B.L.Ling, and J.G.Li Institute of Plasma Physics, Chinese.
NEURAL NETWORKS FOR DATA MINING
Classification / Regression Neural Networks 2
Neural Networks for Protein Structure Prediction Brown, JMB 1999 CS 466 Saurabh Sinha.
Artificial Neural Networks. The Brain How do brains work? How do human brains differ from that of other animals? Can we base models of artificial intelligence.
1 Introduction to Neural Networks And Their Applications.
1 托卡马克位形优化 (2) 高庆弟 SWIP 核工业西南物理研究院 成都. 2 Nonlinearity of LH wave absorption  The plasma temperature in HL-2A is much lower than that in future reactor.
Learning to perceive how hand-written digits were drawn Geoffrey Hinton Canadian Institute for Advanced Research and University of Toronto.
Neural Networks II By Jinhwa Kim. 2 Neural Computing is a problem solving methodology that attempts to mimic how human brain function Artificial Neural.
J M Lopez 1 (of 17) 7 th Workshop on Fusion.., Frascati March, Simulator of the JET real-time disruption predictor J.M. Lopez*, S. Dormido-Canto,
Artificial Neural Networks (ANN) modeling of the pulsed heat load during ITER CS magnet operation L. Savoldi Richard 1, R. Bonifetto 1, S. Carli 1, A.
Akram Bitar and Larry Manevitz Department of Computer Science
MHD Suppression with Modulated LHW on HT-7 Superconducting Tokamak* Support by National Natural Science Fund of China No J.S.Mao, J.R.Luo, B.Shen,
ASIPP HT-7 The effect of alleviating the heat load of the first wall by impurity injection The effect of alleviating the heat load of the first wall by.
1 基于轻核衰变实验的考虑 华 辉 北京大学物理学院 HIRFL-RIBLL1 合作组会议, 中科院近代物理研究所,兰州,
Introduction to Neural Networks Introduction to Neural Networks Applied to OCR and Speech Recognition An actual neuron A crude model of a neuron Computational.
Neural Networks Presented by M. Abbasi Course lecturer: Dr.Tohidkhah.
Dr.Abeer Mahmoud ARTIFICIAL INTELLIGENCE (CS 461D) Dr. Abeer Mahmoud Computer science Department Princess Nora University Faculty of Computer & Information.
Back-propagation network (BPN) Student : Dah-Sheng Lee Professor: Hahn-Ming Lee Date:20 September 2003.
Neural Network and Deep Learning 王强昌 MLA lab.
CHEE825 Fall 2005J. McLellan1 Nonlinear Empirical Models.
Artificial Neural Networks (ANN). Artificial Neural Networks First proposed in 1940s as an attempt to simulate the human brain’s cognitive learning processes.
Perceptrons Michael J. Watts
Dispersion interferometer using modulation amplitudes on LHD 中科院等离子体物理研究所 王兴立,曹骑佛,孙兆轩,周凡,魏然,江堤 T.Akiyama,R. Yasuhara,K.Kawahata,S.Okajima,and K.Nakayama,Rev.
Recent Progress of the Control System on HL-2A * HL-2A control group, Presented by: Li Qiang( 李强 ) Southwestern Institute of Physics, P. O. Box 432, Chengdu,
IEEE AI - BASED POWER SYSTEM TRANSIENT SECURITY ASSESSMENT Dr. Hossam Talaat Dept. of Electrical Power & Machines Faculty of Engineering - Ain Shams.
2015 年工作汇报 钟 平. 6 一、校 本 研 修 2 听 评 课听 评 课听 评 课听 评 课 1.
ASIPP/ EAST Presented by Zuchao Zhang on behalf of EAST Central Control Team 1 Institute of Plasma Physics, Chinese Academy of Sciences Implementation.
Soft X-Ray Tomography in HSX V. Sakaguchi, A.F. Almagri, D.T. Anderson, F.S.B. Anderson, K. Likin & the HSX Team The HSX Plasma Laboratory University of.
Big data classification using neural network
an introduction to: Deep Learning
ANN-based program for Tablet PC character recognition
Artificial Intelligence (CS 370D)
CSE 473 Introduction to Artificial Intelligence Neural Networks
MLP Based Feedback System for Gas Valve Control in a Madison Symmetric Torus Andrew Seltzman Dec 14, 2010.
Introduction to Neural Networks And Their Applications
Artificial Intelligence Methods
Neural Networks Geoff Hulten.
The p-wave scattering of spin-polarized Fermi gases in low dimensions
Search for h→ at Dzero
郑 公 平 河南师范大学 第五届全国冷原子物理和量子信息青年学者学术讨论会
An introduction to: Deep Learning aka or related to Deep Neural Networks Deep Structural Learning Deep Belief Networks etc,
Lecture 09: Introduction Image Recognition using Neural Networks
Areas under the receiver operating characteristic (ROC) curves for both the training and testing data sets based on a number of hidden-layer perceptrons.
Akram Bitar and Larry Manevitz Department of Computer Science
Presentation transcript:

王灏,王爱科,杨青巍,丁玄同 核工业西南物理研究院 HL-2A 装置上关于破裂预测的一些尝试 第十三届全国等离子体科学技术会议 2007, 8, 20-22, 成都 核工业西南物理研究院 王灏,等. 第十三届全国等离子体科学技术会议 2007, 8, 20-22, 成都 Introduction Disruption in a tokamak is a sudden loss of confinement and subsequent transfer of plasma energy to the surrounding structure component. The force and heat loads, induced by disruption, may damage the machine walls and support structure. Thus, disruption prediction at a sufficient early time is important for taking measures to avoid the disruption or mitigate the unavoidable ones. Physically, disruptions have different causes, most of which have been identified but whose dynamics is not known in detail. For instance, the relations of various instabilities to disruptions remain an active research area. Therefore, some new methods, includes ANN method and other methods, are designed to forecast the plasma disruption. In our works, we have tried two methods, these are ANN method and MHD method. What is ANN ANN has a complex structure, including many neurons and layers, and Signals transmitted to the neural network through the input layer. By calculating, the output of Layer 1 has been got, and then, the output of Layer 1 became the input of Layer 2. This process can be repeated until a final output has been got. How to train an ANN Our ANN model and database When we trainning, the output have only two states: 0 and 1. When we discharge, the plasma will have two different states, stable state and unstable state. When unstable, we set the standard output equals 0, and Correspondingly, when stable, the output eaquals 1. Therefore, if the output is less than 0 or close to 0, we must be carefully. Because This means that the plasma may disrupte. 13 experimental diagnostic signals Six Mirnov probes, one H-alpha line intensity, one channel of Soft X ray detector (SXR), one line integrated density ne, one main plasma impurity detector, one bremsstrahlung signal, one bolometer, and one vacuum ultraviolet spectrum signal (VUV). Architecture 13:15:30:10:1 After compared many different architectures, we believe this architecture is the best for our database. Our results This figure shows the prediction for the disruptive pulse The dashed line is the plasma current, and another line is the output of ANN. It disrupts at 980ms, while the network gives the disruption alarm at 945ms, 35ms in advance. Advantages and disadvantages of ANN method Relax the resolution requirement If we want to detect and controll disruption directly, we need very good diagnosis and controll technology. A neural network includes biases, sigmoid activation functions and linear activation functions and, can carry out complex non-linear mapping, so we can use ANN to predict. Two many calculations Trainning will take two many times, and sometimes, the error function will not be convergent. If we apply ANN as an online predictor, perhaps we need high performance computer. We can try some other methods For example, MHD method. Principle of MHD method When plasma is in stable state, both amplitude and period of MHD perturbation are small, while before disrupt, there are strong MHD activities. (Yang Q W et al 2006 Chin. Phys. Lett ) Prediction on the pulse 4207 The blue line meas plasma current, and the red line is output of our prediction systerm. It disrupts at 1470ms, and the prediction systerm gives an alarm at 1450ms, 20ms in advance. Advantages and disadvantages of ANN method Using only a few singals We use only one Mirnov signal, so it’s fast and flexible. Perhaps, we can design an online predictor according to this method. High successful rate We can get about 80% successful rate with only a few false alarm, or near 90% successful rate with 25% false alarm. Unable to predict short pulse In our database, all of the discharge time are more than 1000ms. If we apply this method into short pulse, the successful rate will decrease. Summary and future plan Artificial neural networks (ANN) are trained to forecast the plasma disruptions in HL-2A tokamak. The optimized network architecture is obtained. Relation between Mirnov signals and disruptions is used, and then a new attempt, considering periods and swings of Mirnov signals, is carried out. These two methods can successfully detect the disruptive pulses of HL-2A tokamak. The results show the possibility of developing a disruption predictor that intervenes well in advance in order to avoid plasma disruptions or mitigate their effects. In the future, we want to do more works about disruption prediction. We want to use more Mirnov signals and then, we can predict short discharges. We want to consider more signals about period and amplitude, such as soft X ray, bolometer, density, and so on. We will unite MHD method with ANN, may be it will improve the successful rate. And at last, perhaps we can try to predict disruption on-line.