دانلود رایگان مقاله انگلیسی تخمین پتانسیل روانگرایی خاک بر اساس انرژی با استفاده از الگوریتم GMDH - اشپرینگر 2017

Abstract

Reliable determination of seismic liquefaction potential of soil is an important obligation in earthquake engineering. In this study, neuro-fuzzy group method of data handling (NF-GMDH) was employed for prediction of strain energy required to induce liquefaction. The NFGMDH-based model was developed using particle swarm optimization. A wide-ranging database of soil element tests was used to develop an advanced model, capable of predicting soil liquefaction resistance accurately. Input variables of the model were chosen based on the previous studies on the liquefaction potential assessment. Results of geotechnical centrifuge tests were also involved during the training process for adequate generalization of the proposed model for future predictions. A parametric analysis was then performed to evaluate sensitivity of the proposed model to variations of the influencing parameters. A comparison between performance of the developed model and previously recommended relationships was done. The results clearly demonstrate that the proposed model, which was derived based on laboratory results, can be successfully utilized for strain energy-based estimation of liquefaction potential.

5 Summary and Conclusions

In this paper, a relatively large database including laboratory tests including cyclic triaxial, cyclic torsional shear and simple shear tests on sand and silty sands was used. Powerful intelligent tool (i.e., neuro-fuzzy group method of data handling, NF-GMDH) was utilized to develop a model, for prediction of strain energy required for liquefaction onset (W). Also, the particle swarm optimization (PSO) algorithm is applied in topology design of the NFGMDH model. Based on the experimental observations in the gathered experimental database as well as the previous studies on sandy soils, six parameters: initial mean effective confining pressure (r’0), relative density (Dr), fines content (FC), mean grain size (D50), uniformity coefficient (Cu) and coefficient of curvature (Cc), were used as input parameters to develop the NF-GMDH-based model. In addition, results of several centrifuge tests, which were not used during model development, were employed for further validation of the proposed model. The proposed model showed a reasonably good performance for all element tests (R2 = 0.891, MAPE = 1.896, RMSE = 0.074) and centrifuge datasets (R2 = 0.801, MAPE = 2.101, RMSE = 0.101). The relative error of strain energy (W) values of the developed NF-GMDH-based model is approximately below ±0.3 J/m3 .