인공신경망 기법을 이용한 국내 해성점토의 압축특성 분석

Abstract

New empirical correlations for compression index of marine clay were proposed in this study using the Back Propagation Neural Network (BPNN) and laboratory test results of marine clay sampled at coastal regions of the country in Korea. The compression index determined from laboratory consolidation tests was from 0.13 to 1.86 for marine clay, which sampled a great variation between locations. The compression characteristics estimated by existing empirical equations from single or multiple index parameters was compared with those of consolidation test results, which indicate still large uncertainties in estimation of compression index of soil. However, the compression index from empirical equation based on multiple index parameters proposed in this paper was closed to the measured value more than that of existing empirical equations. The result of BPNN shows that correlation coefficient and standard error between laboratory test and neural network results is nearly close to those of proposed empirical correlation. The compression index obtained from neural network using the multi-variables including natural water content, dry unit weight, liquid limit, in-situ void ratio or plastic index among various factors was close to the measured, and the standard errors between the estimated and the measured were from 0.09 to 0.14. This verifies a possibility that if BPNN be used to determine compression index, a reliable compression index can be estimated using multiple index parameters of soils.