Inference of the optimal probability distribution model for geotechnical parameters by using Weibull and NID distributions

• Main Authors: Fengqiang Gong, Tiancheng Wang, Shanyong Wang
• Format: Article
• Language: English
• Published: JVE International 2019-06-01
• Series: Journal of Vibroengineering
• Subjects: reliability analysis; geotechnical parameters; the optimal probability distribution; probability density function (PDF); normal information diffusion; Weibull distribution
• Online Access: https://www.jvejournals.com/article/19758
• ISSN: 1392-8716; 2538-8460

To obtain the optimal probability distribution models of geotechnical parameters, the goodness of fit of the normal information diffusion (NID) distribution and Weibull distribution were investigated and compared for actual engineering samples and Monte Carlo (MC) simulated samples. Two datasets from actual engineering parameters (the strength of a rock mass and the average wind speed) were used to test the fitting abilities of these two distributions. The results show that the parameters of the NID distribution are easily estimated, the Kolmogorov-Smirnov (K-S) test results of the NID distribution are smaller than those of the Weibull distribution, and the NID distribution curves can perfectly reflect the stochastic volatility of geotechnical parameters with small sample sizes. The sample size effects on the fitting accuracy of the NID distribution and Weibull distribution were also investigated in this paper. Eight simulated samples with different sample sizes, namely, 15, 20, 30, 50, 100, 200, 500, and 1000, were produced by using the MC method based on two known Weibull distributions. The results show that with an increase in the sample size, the K-S test results of the NID distribution gradually decrease and tend to converge, while the chi-square test results of the NID distribution remain low and are always lower than those of the Weibull distribution. The cumulative probability values of the NID distribution are larger than those of the Weibull distribution and are always equal to 1.0000. Finally, the comparison of the fitting accuracy between the NID distribution and normalized Weibull distribution was also analyzed.