边坡岩体力学参数反分析遗传_神经网络算法.pdf

1、 3 4 / bWV o .l 3 2007 M8 Ch inese Journa l o f U nderg round Space and Eng ineer ing A ug. 2007 边坡岩体力学参数反分析遗传) 神经网络算法*邓 勇( i g * u 9 * gy ! 7?K3 , i 409000)K 1: 基于均匀设计a有限元法a人工神经网络和遗传算法建立了新的边坡岩体力学参数反分析方法b按照均匀设计要求,确定数值模拟方案;用有限元程序计算出相应的神经网络训练样本,建立边坡变形的神经网络预测模型, 再利用遗传算法进行反演分析,其中反演过程适应度的计算则采用已训练好的神经网络预测

2、来替代有限元数值仿真, 这样大大缩短了计算时间b通过算例分析,反演结果比较理想,表明该反分析方法是可行性和精确的b1oM: 边坡,反分析,人工神经网络, 遗传算法ms |: TU432; TP183 DS M : B cI|: 1673-0836( 2007) 04-0751-07AGenetic-artificialNeuralNetworkAlgorithmforMechanicalParametersofRockSlopeDENG Yong(Chongqing Q ianjiang City Construction the re latedana lytica l samp les f

3、or neural netw ork are g iven by FEM ca lculations. Thus, a BP neu ra l network wh ich is used tofo recast displacem en t of the slope. s charac ter points is erected and tra ined. T he phys ica l and m echan ica l param eterscan be ana lyzed backwa rds by g enetic a lgor ithm. In this a lgor ithm t

4、he tra ined BP neu ra l netw ork is used to ca lcu latingthe fitness value instead of the FEM me thod and the calculation tim e is m uch reduced. T hrough examp les ana ly sis,the erro r betw een the back analysis resu lts and the theoretica l ones is much less and m eets the requ irem ent of prec-i

5、sion, w hich indicates tha t th is back analysis m ethod is feasib le and accura te.Keywords: slope; back ana ly sis; a rtific ia l neura l netw ork; g enetic algorithm.1 引 言 H 8 H s1o 1b ,8 V 8, r HWr,7 O jr,1 8 d 4, / 0X F s E 5 2bH 8r H saH MH !9 Es1 bB f /, 8r V i = k , 8 f |b i = k |“ a k“j la

6、s #FH* l : 2007-04-04()Te:( 1970-) , 3, +R_ , =,1V Y “5 #M1/ TbE-ma i:l Dy232891 sohu. comqey ,TQ 8 LM Kb 6B H 8 ZE C k,l? C ki s ( a d# aFVE e J;v? C k5i kn5, Otv? C k sH 8 )Y7 4bk i = k C k iZ ,r MQs+ E M EB 5bQsE CH 8 L=M4,YVy s9 , “Sf ,s “Sf Kl,Kl “Sf r r8r bMQsV20 W70M 7 S4,30 M, %5+, s j, y %r

7、5 b| ( !9aKEa * EL. EM,y BH MQsL.) * Eb2 基于均匀设计的数值模拟试验H 9 m1 U, HH, BH ,H +AaBaC,H 8 i g , u i g?Svyp S , |S V1,9 PLAN 42, 5D- P5bm1 9 F ig. 1 Computation model 9 Z ( , ( !9y kZb ( !9 B k !9ZE 3b V kQ , y a y k, ( /Kzy k !9ZEb ( !9 V ( !9Vb k ,H 8V1 H 8SSTable. 1 Characteristic value of soft rock sl

8、opeC( kN /m3 )T C(M Pa)= U( b) E( GPa) 1L23. 525. 00. 250. 5032 38 1. 0 2. 0 0. 25 0. 33CaT Ca = Ua Ea 1LaHHH B7 ky , I nCM l,9 s |C= 24 kN /m 3, 6y (s30 ,5 30,V2 kZVbV2 kZVTab. 2 Simulation schemesZT /M P a= /b /G Pa 1H /mH /b1 0. 250 34. 276 1. 655 0. 311 21. 793 86. 8972 0. 259 36. 759 1. 276 0.

9、289 19. 034 82. 2413 0. 267 32. 828 1. 966 0. 267 16. 276 77. 5864 0. 276 35. 310 1. 586 0. 330 13. 517 72. 9315 0. 284 37. 793 1. 207 0. 308 10. 759 68. 2766 0. 293 33. 862 1. 897 0. 286 8. 000 63. 6217 0. 302 36. 345 1. 517 0. 264 22. 345 58. 9668 0. 310 32. 414 1. 138 0. 327 19. 586 54. 3109 0. 3

10、19 34. 897 1. 828 0. 305 16. 828 49. 65510 0. 328 37. 379 1. 448 0. 283 14. 069 45. 00011 0. 336 33. 448 1. 069 0. 261) 11. 310 88. 44812 0. 345 35. 931 1. 759 0. 324 8. 552 83. 79313 0. 353 32. 000 1. 379 0. 302 22. 897 79. 13814 0. 362 34. 483 1. 000 0. 280 20. 138 74. 48315 0. 371 36. 966 1. 690

11、0. 258 17. 379 69. 82816 0. 379 33. 034 1. 310 0. 322 14. 621 65. 17217 0. 388 35. 517 2. 000 0. 300 11. 862 60. 51718 0. 397 38. 000 1. 621 0. 278 9. 103 55. 86219 0. 405 34. 069 1. 241 0. 256 23. 448 51. 20720 0. 414 36. 552 1. 931 0. 319 20. 690 46. 55221 0. 422 32. 621 1. 552 0. 297 17. 931 90.

12、00022 0. 431 35. 103 1. 172 0. 275 15. 172 85. 34523 0. 440 37. 586 1. 862 0. 253 12. 414 80. 69024 0. 448 33. 655 1. 483 0. 316 9. 655 76. 03425 0. 457 36. 138 1. 103 0. 294 24. 000 71. 37926 0. 466 32. 207 1. 793 0. 272 21. 241 66. 72427 0. 474 34. 690 1. 414 0. 250 18. 483 62. 06928 0. 483 37. 17

13、2 1. 034 0. 313 15. 724 57. 41429 0. 491 33. 241 1. 724 0. 291 12. 966 52. 75930 0. 500 35. 724 1. 345 0. 269 10. 207 48. 103752 地下空间与工程学报 第 3卷vV2 z LZ,ANSYS9 ) M M,TnV3bV3 E9 TTab. 3 Simulation resultsZABCx y x y x y1 0. 7081 - 0. 6030 1. 1367 - 16. 362- 21. 524 - 23. 7811 0. 7081 - 0. 6030 1. 1367

14、 - 16. 362- 21. 524 - 23. 7812 0. 7546 - 0. 3115 1. 6834 - 16. 255- 20. 753 - 23. 4583 0. 3913 - 0. 0233 1. 0278 - 7. 8659- 9. 6081 - 11. 1544 0. 1009 - 0. 2497 0. 7866 - 6. 0264- 7. 5697 - 9. 11575 0. 1127 - 0. 1240 0. 7824 - 5. 0499- 6. 1574 - 7. 67676 0. 0503 - 0. 0146 0. 3149 - 1. 8082- 2. 1434

15、- 2. 73817 0. 5993 0. 1403 3. 4091 - 17. 892- 20. 338 - 27. 118 - 0. 0018 - 0. 4986 2. 9711 - 16. 423- 19. 385 - 26. 8849 0. 0544 - 0. 0987 1. 4773 - 7. 7426- 8. 8273 - 12. 6310 0. 1000 0. 0138 1. 3804 - 6. 9807- 7. 5622 - 11. 411 0. 4790 0. 1023 0. 5751 - 7. 2456- 8. 7405 - 10. 26812 0. 0751 - 0. 0

16、986 0. 1908 - 2. 2861- 2. 9975 - 3. 376713 0. 7811 - 0. 5408 2. 4122 - 21. 357- 26. 749 - 30. 92214 0. 9427 - 0. 2087 3. 2089 - 22. 85 - 27. 827 - 33. 15715 0. 4656 0. 0595 1. 6636 - 10. 246 - 11. 96 - 14. 73416 0. 1038 - 0. 2731 1. 3307 - 8. 4176- 10. 186 - 12. 94217 0. 0677 - 0. 0557 0. 6492 - 3.

17、6501- 4. 3499 - 5. 671818 0. 0666 0. 0054 0. 5190 - 2. 6919- 3. 1275 - 4. 194219 0. 7362 0. 3765 4. 8379 - 23. 866 - 26. 34 - 37. 01320 - 0. 0238 - 0. 2171 2. 0176 - 10. 735- 12. 037 - 18. 0321 0. 6465 - 0. 3566 0. 6951 - 12. 293 - 16. 12 - 17. 54222 0. 6516 - 0. 1451 1. 0661 - 11. 673- 14. 755 - 16

18、. 52123 0. 2931 0. 0167 0. 6018 - 4. 9215- 6. 0565 - 6. 948524 0. 0913 - 0. 1125 0. 4170 - 3. 4067- 4. 3031 - 5. 040725 0. 9086 - 0. 4660 4. 2188 - 28. 425- 34. 447 - 42. 21426 0. 5132 - 0. 0146 2. 3558 - 13. 94 - 16. 41 - 20. 57127 0. 5691 0. 1929 2. 5341 - 13. 388- 15. 559 - 20. 00128 0. 1156 - 0.

19、 2716 2. 1688 - 12. 156- 14. 165 - 19. 17129 0. 0802 - 0. 0286 0. 9649 - 5. 0913- 5. 7161 - 7. 974730 0. 0930 0. 0377 0. 8182 - 4. 0432- 4. 4047 - 6. 46673 边坡位移预测的 BP神经网络模型H M dL ) 1“5, “ - s I ny Y Vb7 * Kv ? dLf r/,H MasZ 4 - 7H My 1,18 aH + +a8 byN, y0:H 8T Ca = Ua Ea1LaH HH B;H +M bBP * yMATLAB

20、Q * s LC,new fff y BP ,4 *.f sYtansigf pure-linf , Levenberg - M arquardt Etrainlm,simf _b c * ,| “s s,sY a kb| sBasB k, ,WZ T S |t bn5| * !12 , a k wL X,YVQ k, * !16 HT X,M wLnm2, | c * 16bm2 aa k wLF ig. 2 T ra ining e rror curve, v alidationerro r curve and test erro r curve B ?,_6 T “S T LBs, M1

21、“ (0. 94 , T 6 z, T 7 ib ( !9 -5Z_ “S nV4,V V A, zb4 边坡位移反分析的遗传) 神经网络S = %Qs5 * ZE,Qs+ * PQsTB V LbW illiam Yeh( 1986);,Qs ab aB7532007年第 4期 邓 勇:边坡岩体力学参数反分析遗传) 神经网络算法V4 T 1 Tab. 4 Comparison between forecastvaluesand expectation valuesZABCx y x y x y10. 708 - 0. 603 1. 137 - 16. 362- 21. 524- 23. 78

22、0. 708 - 0. 603 1. 136 - 16. 367- 21. 555- 23. 75520. 754 - 0. 311 1. 683 - 16. 255- 20. 753- 23. 4580. 755 - 0. 311 1. 684 - 16. 253- 20. 735- 23. 46930. 391 - 0. 023 1. 028 - 7. 866 - 9. 608- 11. 1540. 391 - 0. 023 1. 027 - 7. 872 - 9. 629- 11. 13540. 101 - 0. 250 0. 786 - 6. 026 - 7. 569- 9. 1160

23、. 101 - 0. 249 0. 787 - 6. 036 - 7. 566- 9. 10150. 1137 - 0. 124 0. 782 - 5. 050 - 6. 157- 7. 6770. 112 - 0. 124 0. 782 - 5. 053 - 6. 167- 7. 670V iaB V 8 10bQ5 V bW bW5b T bWBlSM,M bWiBvSM, * Q b/V, .C l Cb9Qs5 1 , ,j. 7 #M_Yy J./ bW, 2005, 1( 4) :642- 645. ( Z eng Y uan, L I Zh i- gao, W ang Y i-

24、b in.R esearch on Influenc ing F actors of D eep Excavation A d-jacent to Subw ay Station . Ch inese Journal o f U nder-ground Space and Eng ineering. 2005, 1( 4): 642- 645.( in Chinese) ) 3 Itasca Consulting G roup, Inc. FLAC ( Fast L ag rang ianA nalysis o f Continua) U ser M anua ls, V ersion 5.

25、0, M in-neapolis, M innesota, 2005. 5 4 o,b. FLA C a L 2 M . Y, 2005. ( L iu Bo, H an Y anhu.i FLACT heo ry, Examp le and A pp licationM . Ch ina Comm un-ications P ress, 2005. ( in Ch inese) ) 5 ,. FLACrMs J.y S. 2005, 21( 6): 95- 100( X ue L iy-ing, Y ang Bin. T he A pplications o f FLA C in D eform ationA na lysis on Com pound So ilN a iling Support J. Bu ild ingScience, 2005, 21 ( 6): 95- 100. ( in Chinese) )7572007年第 4期 邓 勇:边坡岩体力学参数反分析遗传) 神经网络算法