Evaluation methods for performance of post-construction settlement prediction models in thick loess filled ground

Abstract

The prediction of post-construction settlement is an important reference for the evaluation of deformation stability evaluation and building layout planning in thick loess filled ground. To choose suitable models for predicting post-construction settlement in thick loess filled grounds, the characteristics of post-construction settlement curves are analyzed based on the measured settlement of a thick loess fill ground project. Seventeen regression parameter models are established, and some evaluation indexes and methods for models are proposed. The best prediction models for post-construction settlement prediction are optimized. The results indicate that the post-construction settlement curves of the filling area change slowly, with no steep increase in the initial stage of earthwork filling. The settlement rate gradually decreases with time, and there is no horizontal section where the settlement tends to be stable. The optimal regression parameter model can be selected by minimizing the extrapolation prediction error, the internal fitting error, and the posteriori error ratio as the comprehensive control objective. The MMF model （TypeⅡ） and hyperbolic model show high prediction accuracy, good stability, and strong adaptability, with the prediction effect being the best among the 17 models. The more stable the settlement data changes, the better the model prediction effect. Increasing the time span of modeling data would improve the prediction accuracy, but the improvement effect on prediction accuracy would no longer be significant after reaching a certain value.

Keywords

• loess
• thick loess fill ground
• post-construction settlement
• prediction model
• evaluation index