[研究] RLS, Recursive Least Square

看板NTUEE_LAB206作者 (湯尼阿塔.LT127)時間14年前 (2010/04/30 10:12), 編輯推噓1(1017)
留言18則, 2人參與, 最新討論串1/1
通常用在Filter Desing, Adaptive Filter Design 不過剛好我同學問到這個,他只看到別人用在馬達決定參數 簡單的來說,整個系統有 desire y_d System y(n)= sigma[w(k)x(n-k)] x,y 都可以量測,設計w(k)的filter parameter將y,y_d的Square Error減小 Wiki上的圖比較清楚 http://en.wikipedia.org/wiki/Recursive_least_squares_filter 設計wn,讓d^(n)=sigma[w(k)x(n-k)]儘量接近d(n) [此處d(n)及前面提到的輸出y(n)] 當然,馬達當中會有實際的馬達,以及演算中的Model Motor 實際的馬達輸出會是d(n),而演算中作系統識別Model輸出會是d^(n) 而x則是 id,iq,vd,vq的四個state,都是可以量測得到‧ 所以可以反過來把filter當作馬達參數,而如何取得正確的馬達參數,minimize模擬 系統的輸出‧就可以用到RLS WIKI上也有提到: The benefit of the RLS algorithm is that there is no need to invert matrices, thereby saving computational power. Another advantage is that it provides intuition behind such results as the Kalman filter. 因為RLS相對於Kalman Filter而言,少了y=Cx+Dv這條某些state可能量不到的問題 因此KF是量不到的,就用其他的部分資訊與精細的Model來猜 而RLS則是要把係數調整到與Desire Output相同‧ http://www.cs.tut.fi/~tabus/course/ASP/LectureNew10.pdf 該門ASP的課程 http://www.cs.tut.fi/~tabus/course/ASP/Lectures_ASP.html -- ▄▅▆▆▆ ▆▆▆▆▆ ▆▆▆▅▄▃ Nissan GT-R █ ▅▅▅ █ █▄▄▄▄█ The Legend is REAL. █ █ █ ██ █▇▅▄ Nurburgring North: 7:38 ◥▆▆▆█ █ ▅▃ http://www.plurk.com/TonyATTA/invite -- ※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 140.112.20.20 ※ 編輯: tonyatta 來自: 140.112.20.20 (04/30 10:12)

05/04 18:50, , 1F
基本上,ANFIS的參數訓練就是用類似的方法。
05/04 18:50, 1F

05/04 22:22, , 2F
不過NN不是Gd嗎? 這也算Gd呀
05/04 22:22, 2F

05/05 00:23, , 3F
不一樣的東西,ANFIS有線性訓練的部份
05/05 00:23, 3F

05/10 19:58, , 4F
test 58
05/10 19:58, 4F

05/10 19:58, , 5F
test 55
05/10 19:58, 5F

05/10 19:59, , 6F
test 50
05/10 19:59, 6F

05/10 20:13, , 7F
test 51
05/10 20:13, 7F

05/10 20:14, , 8F
test 51
05/10 20:14, 8F

05/10 20:15, , 9F
test 52 (15)
05/10 20:15, 9F

05/10 20:18, , 10F
test 53 (17)
05/10 20:18, 10F

05/10 20:18, , 11F
56
05/10 20:18, 11F

05/10 20:18, , 12F
test 53 (18)
05/10 20:18, 12F

05/10 20:19, , 13F
test 54 (19)
05/10 20:19, 13F

05/10 20:20, , 14F
test 55 (20)
05/10 20:20, 14F

05/10 20:22, , 15F
test 56 (21)
05/10 20:22, 15F

05/10 20:22, , 16F
above tested in NTU WIFI
05/10 20:22, 16F

05/10 20:24, , 17F
ping 10ms to 東海 jitter 3ms..
05/10 20:24, 17F

05/10 20:24, , 18F
when back to Dorm, need further test
05/10 20:24, 18F
文章代碼(AID): #1BsZnybC (NTUEE_LAB206)