在實際應用中，需要不斷將新測的監(jiān)測數(shù)據(jù)模式補充到訓練模式集中，以保持模式識別器的時效性。現(xiàn)實計算條件要求訓練模式數(shù)保持一個適當?shù)囊?guī)模，因此在增加一些訓練模式時要淘汰相應數(shù)量的最“老”的訓練模式。
5．實例分析
作為應用實例，以小浪底水利樞紐出水口高邊坡某斷面一段時間的監(jiān)測數(shù)據(jù)進行分析。
表1列出了進行監(jiān)測資料分析試驗的監(jiān)測時刻，它們對應的監(jiān)測數(shù)據(jù)向量模式列于表2。分析時只需從小浪底水利樞紐安全監(jiān)控系統(tǒng)的原始數(shù)據(jù)庫中調(diào)出這些數(shù)據(jù)即可。
表1 小浪底水利樞紐出水口高邊坡某斷面監(jiān)測資料分析模式識別采樣情況
|
1 |
2000-1-3 14:18 |
|
2 |
2000-1-10 08:24 |
|
3 |
2000-1-17 15:40 |
|
4 |
2000-1-24 08:45 |
|
5 |
2000-1-31 15:00 |
|
6 |
2000-2-7 14:55 |
V1=(16.5, 13.95,8.25,2.85,1.45,0.65,13.35, 13.15, 8.25, 1.65, 2.45, 2.25, 1876.2, 1815.51, 1403.69, 1687.55, 1682.21, 1822.02, -19.82, -38.61, 0, 0, 0, 0)
V2=(16.75,13.8, 8.2, 2.75,1.4, 0.6, 13.3, 13, 8.2, 1.6, 2.4, 2.25, 1875.66, 1817.09, 1397.67, 1686.91, 1694.52, 1822.15, -19.29, -4.21, 0, 0, 0, 0)
V3=(16.95,12.95,8.3, 2.9, 1.35,0.6, 13.25, 12.95, 8.15, 1.6, 2.35, 2.2, 1872.03, 1817.2, 1403.9, 1686.03, 1693.11, 1822.45, -18.75, -4.04, 0, 0, 0, 0)
V4=(15.8, 12.8, 8.2, 3.05,1.5, 0.65,13.35, 13.05, 8.2, 1.65, 2.5, 2.25, 1873.69, 1816.26, 1402.31, 1681.57, 1681.91, 1822.96, -17.33, -4.04, 0, 0, 0, 0)
V5=(15.75,13.15,8.2, 3.03,1.5, 0.6, 13.3, 13.25, 8.15, 1.6, 2.4, 2.2, 1874.54, 1816.32, 1409, 1686.83, 1662.81, 1826.46, -18.93, -6.44, 0, 0, 0, 0)
V6=(15.85,13.25,8.15,2.9, 1.4, 0.6, 13.35, 13.35, 8.2, 1.7, 2.4, 2.25, 1874.68, 1816.27, 1405.01, 1687.63, 1701.47, 1825.66, -19.91, -5.19, 0, 0, 0, 0)
將上述監(jiān)測數(shù)據(jù)向量輸入邊坡監(jiān)測模式識別器，即刻可求出對應的邊坡滑動模式，如表2所示。
這些分析成果可成為邊坡極限分析程序數(shù)據(jù)文件的基礎，應用于邊坡的穩(wěn)定分析中，也可以用來推測邊坡穩(wěn)定的主要因素。
6．結(jié)語
結(jié)合自動監(jiān)測儀器系統(tǒng)的使用，應用人工神經(jīng)網(wǎng)絡模式識別技術和邊坡極限分析理論，可實現(xiàn)邊坡安全監(jiān)測資料分析的自動化。自動化的在線監(jiān)測功能和準確的分析成果將顯著提高水電工程管理部門對邊坡安全和整個水電工程系統(tǒng)運行的可靠性的管理水平。
表2 小浪底水利樞紐出水口高邊坡某斷面監(jiān)測資料模式識別分析情況
|
序號 |
時間 |
滑裂面型式 |
|
1 |
2000-1-3 14:18 |
1 |
|
2 |
2000-1-10 08:24 |
1 |
|
3 |
2000-1-17 15:40 |
1 |
|
4 |
2000-1-24 08:45 |
1 |
|
5 |
2000-1-31 15:00 |
1 |
|
6 |
2000-2-7 14:55 |
3 |
參考文獻：
陸峰，博士學位論文《邊坡監(jiān)測的模式識別和極限分析研究》，中國水利水電科學研究院，2001.8
陳祖煜，《巖質(zhì)高邊坡穩(wěn)定分析和軟件系統(tǒng)》，中國水利水電科學研究院，1995.5
Abhijit S. Pandya, Robert B. Macy 著, 徐勇, 荊濤譯, 神經(jīng)網(wǎng)絡模式識別及其實現(xiàn), 電子工業(yè)出版社, 1999.6
戴葵. 神經(jīng)網(wǎng)絡實現(xiàn)技術. 國防科技大學出版社, 1998.7
Pattern Recognition Method in Slope Monitor
Abstract: A concept of slice patterns is put forward in this paper. By using slice patterns in slope monitor, we can build the mapping relations between slice patterns and monitor information. Introducing the pattern recognition method of Artificial Neural Network, a pattern recognizer for slope monitor is built to judge the safe state patterns of the slopes at any monitor times based on the slope monitor information. The stability of slopes can be estimated based on the above information. As a case, this pattern recognizer is applied in analyzing a section of monitor data of Xiaolangdi Outtake Slope. It is showed that the effects of this pattern recognizer is reliable enough for slope monitor.
Keyword: Slope; Safe Monitor; Pattern Recognition; Artificial Neural Network
