(3.238.96.184) 您好!臺灣時間:2021/05/08 04:34
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果

詳目顯示:::

: 
twitterline
研究生:張育誠
研究生(外文):Chang, Yucheng
論文名稱:使用Echo State Network預測感測器輸入
論文名稱(外文):Prediction of Sensory Input Using Echo State Network
指導教授:許宏銘許宏銘引用關係
指導教授(外文):Norbert Michael Mayer
口試委員:黃國勝蔡清池李祖聖
口試委員(外文):Kao-Shing HwangChing-Chi TsaiTzuu-Hseng S. Li
口試日期:100年7月27日
學位類別:碩士
校院名稱:國立中正大學
系所名稱:電機工程研究所
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2011
畢業學年度:99
語文別:英文
論文頁數:52
中文關鍵詞:回響狀態網路感測器預測
外文關鍵詞:Echo State Networksensoryprediction
相關次數:
  • 被引用被引用:0
  • 點閱點閱:492
  • 評分評分:系統版面圖檔系統版面圖檔系統版面圖檔系統版面圖檔系統版面圖檔
  • 下載下載:50
  • 收藏至我的研究室書目清單書目收藏:0
Echo State Network是一種近期被提出來的計算模型。它包含一個可以提供豐富動態的大規模隱藏層以及與其他類神經網路不同的特別訓練機制。這些元素使得它適合用於重建非線性信號。在本論文中提出了一個基於滑動窗口 (Sliding Window) 概念的訓練法,並與其他兩種已知之訓練演算法一起進行了實驗。本論文實驗使用訊息語法分析程式,將機器人足球模擬器SimSpark中之人形機器人陀螺儀感測器之值取出並紀錄成序列,在MATLAB中測試使用不同訓練法的網路之效能。實驗目的為使用人形機器人感測器之過去的輸入信號來預測下一時點之輸入信號。
Echo State Network is a computing model proposed recently. It contains a big sized internal hidden layer that provides rich dynamics, together with the special training mechanism that differs from other neural networks. These features make it suitable for modeling nonlinear signals. In this thesis I proposed a training method based on the concept of sliding window, and made experiments together with two other known training methods. I get the sensory value sequence from simulated humanoid robots of a robot soccer simulator SimSpark by a parser, record them, and then test them in MATLAB, in order to predict the sensory input signals on next iteration from the past input signals of simulated humanoid robots.
ACKNOWLEDGEMENTS I
摘要 II
ABSTRACT III
TABLE OF CONTENTS IV
LIST OF FIGURES VI
LIST OF TABLES VII
I. INTRODUCTION 8
1.1 RESERVOIR COMPUTING 8
1.2 MOTIVATIONS AND OBJECTIVES 9
1.3 STRUCTURE OF THE THESIS 10
II. ECHO STATE NETWORKS 11
2.1 ECHO STATE PROPERTY 12
2.1.1 Concept of Echo States 12
2.1.2 Condition of Echo States 13
2.2 SUMMARY 15
III. IMPLEMENTATION OF ECHO STATE NETWORKS AND TRAINING ALGORITHM 16
3.1 THE NETWORK MODEL 16
3.1.1 The Units 16
3.1.2 The Weight Matrices 16
3.1.3 The Activation Functions 17
3.1.4 Weight Matrices Setup 17
3.2 THE TRAINING PROCEDURE 18
3.2.1 Offline Training 18
3.2.2 Online Training: Recursive Least Square Algorithm 19
3.2.3 Standard Online Training 22
3.3 SUMMARY 24
IV. SIMULATION ENVIRONMENT 25
4.1 SIMSPARK ROBOT SOCCER SIMULATOR 25
4.1.1 System Overview: Server 25
4.1.2 System Overview: Agents 26
4.2 MOTION DESIGN TOOL: QMOTION2 28
4.2.1 GUI Overview 28
4.2.2 Motion Design and Porting 29
4.3 SUMMARY 32
V. RESULTS 35
5.1 THE PREDICT TARGET 35
5.2 BATCH TRAINING 36
5.3 STANDARD ONLINE TRAINING 37
5.4 RECURSIVE LEAST SQUARE LEARNING 38
5.5 DISCUSSION TO HIDDEN NEURONS 39
VI. CONCLUSION AND FUTURE WORKS 48
6.1 CONCLUSION 48
6.2 FUTURE WORKS 49
REFERENCES 50
VITA 52
[1]Herbert Jaeger, The “echo” state approach to analyzing and training recurrent neural networks, German National Research Center for Information Technology Bremen, 2001.
[2]Herbert Jaeger, A tutorial on training recurrent neural networks, covering BPPT, RTRL, EKF and the “echo state network” approach, German National Research Center for Information Technology Bremen, 2002.
[3]Georg Holzmann, Echo State Networks with Filter Neurons and a Delay & Sum Readout with Applications in Audio Signal Processing, Institute for Theoretical Computer Science, Graz University of Technology A-8010 Graz, Austria, 2008.
[4]Tanguy Mezzano, Echo State Networks application on maze problems, Master Thesis, Katholieke Universiteit Leuven, 2007.
[5]Herbert Jaegar, Adaptive nonlinear system identification with echo state networks, Fraunhofer Institute for Autonomous Intelligent Systems, 2003.
[6]Norbert M. Mayer and Matthew Browne, Echo State Networks and Self-Prediction, GMD- Japan Research Laboratory, Collaboration Centre, 2004.
[7]B. Farhang-Boroujeny, Adaptive Filters: Theory and Application, John Wiley & Sons, Inc., New York, NY, USA. ISBN 0471983373, 1998.
[8]Herbert Jaeger, Discovering multiscale dynamical features with hierarchical Echo State Networks, International University Bremen, 2007.
[9]Herbert Jaeger and Harald Hass, Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication, Science, 304, 2004.
[10] Jochen J. Steil, Backpropagation-Decorrelation: online recurrent learning with O(N) complexity, In Proc. IJCNN, volume 1, pages 843–848, 2004.
[11] Joschka Boedecker et al., SimSpark User's Manual v1.2, http://simspark.sourceforge.net/wiki/images/a/ad/User-manual.pdf, 1998.
[12] http://simspark.sourceforge.net/wiki/index.php/About_SimSpark
[13] Norbert M. Mayer, Joschka Boedecker and Minoru Asada, Robot motion description and real-time management with the Harmonic Motion Description Protocol, Robotics and Autonomous Systems Journal, 2009.
[14] http://en.wikipedia.org/wiki/Neural_network
[15] http://www.csie.nctu.edu.tw/~kensl/AIrpt.htm
[16] Kazuhiro Masui, Using Echo State Networks to Classify and to Predict Sensory Input, Master Thesis, Dept. of Adaptive Machine Systems, Osaka University, Japan, 2008.
[17]N. Michael Mayer and Oliver Obst, Time Series Causality Inference Using Echo State Networks, Department of Electrical Engineering, National Chung Cheng University, Taiwan, 2010.
[18]R.J.Frank, N.Davey, S.P.Hunt, Time Series Prediction and Neural Networks, Department of Computer Science, University of Hertfordshire, Hatfield, UK., 2000.
[19]Juris Klonovs and Lin Xiang Zhang, Qmotion2 User's Manual, 2009.

QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
系統版面圖檔 系統版面圖檔