首页|Computational intelligence models for predicting the effective stress of unsaturated soil
Computational intelligence models for predicting the effective stress of unsaturated soil
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NSTL
Elsevier
In the design of underground spaces, the value of the effective stress (chi) parameter is important while dealing with unsaturated soils. The estimating procedure might be complex due to the non-linearity of the connection between chi parameter and dependant parameters. As a result, two novel hybrid optimization models are proposed in this paper: relevance vector regression (RVR) optimized using artificial bee colony (ABC) and harmony search (HS) algorithms (RVR-ABC and RVR-HS models). The RVR-HS and RVR-ABC models' prediction abilities were demonstrated using 120 datasets from open-source literatures. In this dataset, characteristics of unsaturated soils such as bubbling pressure, soil-water characteristic curve fitting parameter, saturated volumetric water content, net conning pressure, residual water content, and suction were used as the inputs, while chi parameter was used as the output. The proposed models performances were assessed by comparing them regarding some statistical metrics, e.g., correlation coefficient (R-2). For the RVR-ABC and RVR-HS models, the R2 of 0.86 and 0.89 were attained, respectively. The findings showed that the RVR-ABC and RVR-HS models were effective in predicting the chi parameter. However, the RVR-HS model, outperformed the RVR-ABC model as this issue shows, the HS algorithm contributed significantly to the optimization of the suggested model.
NSTL
Elsevier
