浙江师范大学学报（自然科学版）
浙江師範大學學報（自然科學版）
절강사범대학학보（자연과학판）
JOURNAL OF ZHEJIANG NORMAL UNIVERSITY(NATURAL SCIENCES)
2015年
1期
73-77
,共5页
k-近邻%距离加权%边坡稳定性%回归预测
k-近鄰%距離加權%邊坡穩定性%迴歸預測
k-근린%거리가권%변파은정성%회귀예측
k-nearest-neighbor%distance weighting%slope stability%regression prediction
边坡稳定性估计的精度直接关系到边坡工程的成败。然而，边坡稳定性与其影响因素之间存在复杂的非线性关系。当目标函数很复杂时，如果只建立目标函数的局部逼近，并将其应用于待测实例的邻域，就能获得较高的预测精度。这种局部建模方法的典型代表就是k-近邻及其改进算法。在研究k-近邻算法的基本原理及其改进方法的基础上，提出了应用距离加权的k-近邻方法对由岩石容重、岩石内聚力、内摩擦角、边坡角、边坡高度和孔隙水压力6个特征参数组成的岩土参数进行建模，估计表征边坡稳定性的安全系数。实验中，用82个圆弧破坏边坡实例中的71个实例进行建模，对另外11个实例进行推广预测。实验结果表明：用k-近邻算法进行边坡稳定性预测有较高的精度。
邊坡穩定性估計的精度直接關繫到邊坡工程的成敗。然而，邊坡穩定性與其影響因素之間存在複雜的非線性關繫。噹目標函數很複雜時，如果隻建立目標函數的跼部逼近，併將其應用于待測實例的鄰域，就能穫得較高的預測精度。這種跼部建模方法的典型代錶就是k-近鄰及其改進算法。在研究k-近鄰算法的基本原理及其改進方法的基礎上，提齣瞭應用距離加權的k-近鄰方法對由巖石容重、巖石內聚力、內摩抆角、邊坡角、邊坡高度和孔隙水壓力6箇特徵參數組成的巖土參數進行建模，估計錶徵邊坡穩定性的安全繫數。實驗中，用82箇圓弧破壞邊坡實例中的71箇實例進行建模，對另外11箇實例進行推廣預測。實驗結果錶明：用k-近鄰算法進行邊坡穩定性預測有較高的精度。
변파은정성고계적정도직접관계도변파공정적성패。연이，변파은정성여기영향인소지간존재복잡적비선성관계。당목표함수흔복잡시，여과지건립목표함수적국부핍근，병장기응용우대측실례적린역，취능획득교고적예측정도。저충국부건모방법적전형대표취시k-근린급기개진산법。재연구k-근린산법적기본원리급기개진방법적기출상，제출료응용거리가권적k-근린방법대유암석용중、암석내취력、내마찰각、변파각、변파고도화공극수압력6개특정삼수조성적암토삼수진행건모，고계표정변파은정성적안전계수。실험중，용82개원호파배변파실례중적71개실례진행건모，대령외11개실례진행추엄예측。실험결과표명：용k-근린산법진행변파은정성예측유교고적정도。
The estimation precision of slope stability was directly related to the success or failure of the slope engineering. However, the relationship between slope stability and influencing factors was complex and nonlin-ear. When the target function was complex, approximating the target function locally and differently for each distinct query instance could get higher accuracy. The typical representative of these methods was k-nearest-neighbor ( KNN) and its improved methods. It was firstly studied the basic principle of KNN and its improved methods, and then the distance weighted KNN ( KNNDW) was applied to build the predicting model based on the instance data composed of 6 characteristic parameters and predict safety factors characterizing slope stabili-ty. In the experiments, 71 instances from 82 arc slope destruction instances were used to build the predicting model and the other 11 instances were used for promotion prediction. The experimental results proved that KNNDW was more accurate than the modified BP algorithm, the GA-BP algorithm and the υ-SVR algorithm.
