• lilin2
    了解作者
  • C/C++
    开发工具
  • 914KB
    文件大小
  • rar
    文件格式
  • 0
    收藏次数
  • 10 积分
    下载积分
  • 2
    下载次数
  • 2021-03-23 16:29
    上传日期
关于迭代学习算法的简单例程,其中迭代次数可自行更改,次数越高跟随效果越好
ILC.rar
  • ILC
  • ILC_2.m
    1.3KB
  • ILC_1.rar
    702B
  • A survey of iterative learning control.pdf
    1.2MB
  • ILC_1.m
    1.4KB
内容介绍
<html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css"> <link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css"> <link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/625035886caf596192fda39e/raw.css"> <script src="https://static.pudn.com/base/js/compatibility.min.js"></script> <script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/625035886caf596192fda39e/bg1.jpg"><div class="c x1 y1 w2 h2"><div class="t m0 x2 h3 y2 ff1 fs0 fc0 sc0 ls0 ws0">I</div><div class="t m0 x3 h4 y3 ff2 fs1 fc1 sc0 ls1 ws1">terative learning control (ILC) is based on the notion</div><div class="t m0 x3 h4 y4 ff2 fs1 fc1 sc0 ls2 ws2">that the performance of a system that executes the</div><div class="t m0 x3 h4 y5 ff2 fs1 fc1 sc0 ls3 ws3">same task multiple times can be improved by learning</div><div class="t m0 x3 h4 y6 ff2 fs1 fc1 sc0 ls4 ws4">from previous executions (trials, iterations, passes).</div><div class="t m0 x3 h4 y7 ff2 fs1 fc1 sc0 ls3 ws5">For instance, a basketball player shooting a free throw</div><div class="t m0 x4 h4 y8 ff2 fs1 fc1 sc0 ls5 ws6">from a fixed position can improve his or her ability to</div><div class="t m0 x4 h4 y9 ff2 fs1 fc1 sc0 ls3 ws7">score by practicing the shot repeatedly. During each shot,</div><div class="t m0 x4 h4 ya ff2 fs1 fc1 sc0 ls6 ws8">the basketball player observes</div><div class="t m0 x4 h4 yb ff2 fs1 fc1 sc0 ls7 ws9">the trajectory of the ball and</div><div class="t m0 x4 h4 yc ff2 fs1 fc1 sc0 ls3 wsa">consciously plans an alteration</div><div class="t m0 x4 h4 yd ff2 fs1 fc1 sc0 ls8 wsb">in the shooting motion for the</div><div class="t m0 x4 h4 ye ff2 fs1 fc1 sc0 ls9 wsc">next attempt. As the player</div><div class="t m0 x4 h4 yf ff2 fs1 fc1 sc0 lsa wsd">continues to practice, the cor-</div><div class="t m0 x4 h4 y10 ff2 fs1 fc1 sc0 lsb wse">rect motion is learned and becomes ingrained into the</div><div class="t m0 x4 h4 y11 ff2 fs1 fc1 sc0 ls3 wsf">muscle memory so that the shooting accuracy is iteratively</div><div class="t m0 x4 h4 y12 ff2 fs1 fc1 sc0 lsc ws10">improved.<span class="fc2 sc0"> The converged muscle motion profile is an</span></div><div class="t m0 x4 h4 y13 ff2 fs1 fc1 sc0 lsd ws11"><span class="fc2 sc0">open-loop control generated through repetition and learn-</span></div><div class="t m0 x4 h4 y14 ff2 fs1 fc1 sc0 ls3 ws12"><span class="fc2 sc0">ing. This type of learned open-loop control strategy is the</span></div><div class="t m0 x4 h4 y15 ff2 fs1 fc1 sc0 ls3 ws0"><span class="fc2 sc0">essence of ILC.</span></div><div class="t m0 x5 h4 y16 ff2 fs1 fc1 sc0 lse ws13">We consider learning controllers for systems that per-</div><div class="t m0 x4 h4 y17 ff2 fs1 fc1 sc0 lsf ws14">form the same operation repeatedly and under the same</div><div class="t m0 x4 h4 y18 ff2 fs1 fc1 sc0 ls10 ws15">operating conditions. For such systems, a nonlearning con-</div><div class="t m0 x6 h4 y19 ff2 fs1 fc1 sc0 ls10 ws16">troller yields the same tracking error on each pass. Although</div><div class="t m0 x6 h4 y1a ff2 fs1 fc1 sc0 ls11 ws17">error signals from previous iterations are information rich,</div><div class="t m0 x6 h4 y1b ff2 fs1 fc1 sc0 ls10 ws18">they are unused by a nonlearning controller.<span class="fc2 sc0"> The objective</span></div><div class="t m0 x6 h4 y1c ff2 fs1 fc1 sc0 ls12 ws19"><span class="fc2 sc0">of ILC is to improve performance by incorporating error</span></div><div class="t m0 x6 h4 y1d ff2 fs1 fc1 sc0 ls13 ws1a"><span class="fc2 sc0">information into the control for subsequent iterations. </span>In</div><div class="t m0 x6 h4 y1e ff2 fs1 fc1 sc0 ls10 ws1b">doing so, high performance can be achieved with low tran-</div><div class="t m0 x6 h4 y1f ff2 fs1 fc1 sc0 ls14 ws1c">sient tracking error despite large model uncertainty and</div><div class="t m0 x7 h4 y20 ff2 fs1 fc1 sc0 ls10 ws0">repeating disturbances.</div><div class="t m0 x8 h4 y21 ff2 fs1 fc1 sc0 ls15 ws1d">ILC differs from other</div><div class="t m0 x7 h4 y22 ff2 fs1 fc1 sc0 ls16 ws1e">learning-type control strate-</div><div class="t m0 x7 h4 y23 ff2 fs1 fc1 sc0 ls17 ws1f">gies, such as adaptive control,</div><div class="t m0 x7 h4 y24 ff2 fs1 fc1 sc0 ls18 ws20">neural networks, and repeti-</div><div class="t m0 x7 h4 y25 ff2 fs1 fc1 sc0 ls19 ws21">tive control (RC). <span class="fc2 sc0">Adaptive</span></div><div class="t m0 x6 h4 y26 ff2 fs1 fc1 sc0 ls3 ws22"><span class="fc2 sc0">control strategies modify the controller, which is a system,</span></div><div class="t m0 x6 h4 y27 ff2 fs1 fc1 sc0 ls3 ws23"><span class="fc2 sc0">whereas ILC modifies the control input, which is a signal</span></div><div class="t m0 x6 h4 y28 ff2 fs1 fc1 sc0 ls3 ws24"><span class="fc2 sc0">[1]. </span>Additionally, adaptive controllers typically do not take</div><div class="t m0 x6 h4 y29 ff2 fs1 fc1 sc0 ls3 ws25">advantage of the information contained in repetitive com-</div><div class="t m0 x6 h4 y2a ff2 fs1 fc1 sc0 ls3 ws26">mand signals. Similarly, neural network learning involves</div><div class="t m0 x6 h4 y2b ff2 fs1 fc1 sc0 ls1a ws27">the modification of controller parameters rather than a</div><div class="t m0 x6 h4 y2c ff2 fs1 fc1 sc0 ls1b ws28">control signal; in this case, large networks of nonlinear</div><div class="t m0 x6 h4 y2d ff2 fs1 fc1 sc0 lsd ws29">neurons are modified. These large networks require exten-</div><div class="t m0 x6 h4 y2e ff2 fs1 fc1 sc0 ls3 ws2a">sive training data, and fast convergence may be difficult to</div><div class="t m0 x9 h5 y2f ff3 fs2 fc1 sc0 ls1c ws0">DOUGLAS A. BRISTOW<span class="_ _0"></span>, MARINA THARA<span class="_ _0"></span>YIL, </div><div class="t m0 xa h5 y30 ff3 fs2 fc1 sc0 ls1c ws2b">and<span class="_"> </span>ANDREW G. ALLEYNE</div><div class="t m0 xb h6 y31 ff4 fs3 fc1 sc0 ls1d ws0">A LEARNING-BASED METHOD</div><div class="t m0 xc h6 y32 ff4 fs3 fc1 sc0 ls1d ws0">FOR HIGH-PERFORMANCE </div><div class="t m0 xd h6 y33 ff4 fs3 fc1 sc0 ls1d ws0">TRACKING CONTROL</div><div class="t m0 xe h7 y34 ff5 fs4 fc1 sc0 ls1e ws0">&#169;D<span class="_ _1"></span><span class="ls1f">IGITALVISION &amp; ARTVILLE</span></div><div class="t m0 x4 h8 y35 ff1 fs5 fc1 sc0 ls20 ws0">96<span class="_ _2"> </span><span class="ff6 fs6 ls21 ws2c">IEEE CONTROL SYSTEMS MAGAZINE<span class="_ _3"> </span></span><span class="ff7 fs1 ls0">&#187;<span class="_ _3"> </span><span class="ff5 fs6 ls21">JUNE 2006<span class="_ _4"> </span>1066-033X/06/$20.00&#169;2006IEEE</span></span></div></div></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div> </body> </html>
评论
    相关推荐