An integrated SRM-multi-gene genetic programming approach for prediction of factor of safety of 3-D soil nailed slopes

https://doi.org/10.1016/j.engappai.2013.12.011Get rights and content

Highlights

  • An integrated SRM-MGGP method for FOS prediction of nailed slope is proposed.

  • Proposed SRM-MGGP is compared to those of MGGP, SVR and ANN.

  • Out of four methods, SRM-MGGP evolves a model with better generalisation ability.

  • SRM-MGGP method also represents explicit formulation of FOS of nailed slope.

Abstract

Soil nailing is one of the slope stabilisation techniques useful for the strengthening of existing slopes. It helps to reinforce the soil with passive inclusions that increase the overall shear strength of the soil slope and also restrains its displacements. The limit equilibrium method is usually employed to estimate factor of safety (FOS) of nailed slopes through either finite element or finite difference methods. Alternatively, soft computing methods such as multi-gene genetic programming (MGGP), support vector regression (SVR) and artificial neural network (ANN) can also be used to predict the FOS for different soil properties. Among these methods, MGGP possesses the ability to evolve the model structure and its coefficients automatically. Although widely used, the MGGP method has the limitation of producing models that perform poorly on testing data. Therefore, in this study, an integrated structural risk minimisation-multi-gene genetic programming (SRM-MGGP) method is proposed to formulate the mathematical relationship between FOS and the six input variables of cohesion, frictional angle, nail inclination angle, nail length, slope height and slope angle of 3-D nailed slope. The results indicate that the SRM-MGGP model outperforms the other three models (MGGP, SVR and ANN) and is able to generalise the FOS satisfactorily for any given input variables conditions. This would be useful for engineers in their design calculations of slopes with different soil, slope and nail conditions based on certain limitations such as ignorance of effect of pore water pressure or overburden pressure.

Introduction

Soil nailing is one of the economic slope stabilisation technique and is particularly useful for the strengthening of existing slopes. The principle behind soil nailing is to reinforce the soil with passive inclusions which increases the overall shear strength of the soil slope and also restrains its displacements. There are many different design methods for soil nailing, which includes limit equilibrium method (LEM), strength reduction method, several working stress design methods, the Davis method (Shen et al., 1981), the German method (Stocker et al., 1979) and the French method (Schlosser, 1982, Schlosser, 1991). Among them, LEM is commonly used for slope stability analysis by many researchers (Donald and Giam, 1988, Matsui and San, 1992, Griffiths and Lane, 1999, Ugai and Leshchinsky, 1995, Song, 1997, Dawson et al., 1999, Cheng et al., 2007, Cheng et al., 2008). The LEM analysis used in nailed slope analysis is based essentially on Spencer's method, where the effect of a soil nail is considered by applying a concentrated load provided by the nail on the slip surface. FOS for slopes using LEM methods are computed using 2-D or 3-D Finite element method (FEM) or Finite difference method (FDM) is generally used to estimate FOS of nailed slopes (Shen et al., 1981, Juran and Elias,, Plumelle et al., 1990, Thompson and Miller, 1990, Srinivasa Murthy et al., 2002, Sivakumar Babu et al., 2002, Wei and Cheng, 2010).

On the other hand, in recent years use of soft computing methods such as artificial neural network (ANN) is becoming increasingly important in geotechnical engineering (Shahin et al., 2008). Many ANN models have been developed to estimate FOS of slopes with different slope geometry and soil conditions (Li and Liu, 2004, Cho, 2009, Lin et al., 2009, Choobbasti et al., 2009, Li and Wang, 2010, Abdalla et al., 2012). However, all those models did not take into account any effect of nails on FOS of slopes. In addition to this, most of these developed models are based on 2-D boundary conditions without incorporating 3-D effects which are more realistic in nature and is essential for soil nail problems. Since, Wei and Cheng (2010) showed that FOS of nailed slope is greatly affected by cohesion, frictional angle, nail inclination angle, nail length, slope height and slope angle of 3-D nailed slope. These parameters must be taken into account while developing any models using soft computing methods.

Other soft computing methods such as support vector regression (SVR) and genetic programming (GP) can also be used to predict the FOS for the different soil properties. Among these methods, GP possesses the ability to evolve models structure and its coefficients automatically (Cevik and Guzelbey, 2007, Cevik and Sonebi, 2008, Gandomi et al., 2010). Most popular variant of GP used recently is multi-gene genetic programming (MGGP) (Gandomi and Alavi, 2011, Garg and Tai, 2012, Garg and Tai, 2013, Garg et al., 2013a). Despite large applications of MGGP in field of structural and civil engineering, MGGP has limitation for producing models that over-fit on the testing data. The reason can be attributed to the large size models produced during the evolutionary stage in MGGP that captures noise along with relationships from the training data. Over-fitting in MGGP is the popular problem among researchers and have been paid less attention (Chan et al., 2011, Gonçalves et al., 2012, Garg et al., 2013b).

Therefore, in the present work, SRM-based-multi-gene genetic programming approach (SRM-MGGP) is proposed for predicting FOS for 3-D nailed slopes using the results from the LEM analysis. FOS under different six input parameters such as cohesion, frictional angle, nail inclination angle, nail length, slope height and slope angle of 3-D nailed slope were used to train and test newly developed SRM-MGGP model. Unlike standard GP, each model participating in SRM-MGGP approach is a set of combination of genes. Structural risk minimisation (SRM) principle is integrated to improve the generalisation ability of the models during the evolutionary stage of the method. The performance of the proposed SRM-MGGP method is compared to that of the other three potential methods: standardized MGGP, SVR and ANN.

Section snippets

Numerical method for estimating FOS

For model development, FOS data for nailed slopes was selected from a comprehensive study conducted by Wei and Cheng (2010). A parametric study was conducted to investigate FOS using LEM analysis under different combinations of cohesion, frictional angle, nail inclination angle, nail length, slope height and slope angle. In their LEM analysis, only the tensile strength and pull-out capacity of the nail were considered. The tensile forces mobilised in the nails were divided into tangential and

Multi-gene genetic programming

In order to understand the concepts of MGGP method, the basics of GP is first discussed in brief.

GP generates computer programs/models automatically based on the given data using Darwinian principle of “Survival of the fittest” (Koza, 1996). Working principle of GP is same as GA but the only difference between them, is that, GA evolves solutions represented by strings (binary or real number) of fixed length, whereas GP generates solutions represented by tree structures of varying sizes (Garg et

Models evaluation and comparison

The results obtained from the four prediction modelling methods SRM-MGGP, MGGP, SVR and ANN are illustrated in Fig. 9, Fig. 10, Fig. 11, Fig. 12 on training and testing data respectively. The best prediction method that gives highest accuracy is determined by comparing these four modelling methods. The correlation coefficient (R) and relative error (%) between the predicted values and the actual values of the FOS are estimated. The correlation coefficient (R) and relative error (%) are given by:

Conclusion and future work

This study presents an alternative approach using four soft computing methods to investigate FOS of 3-D nailed slope based on certain limitations such as ignoring effects of pore water pressure and overburden pressure. Problem of over-fitting in soft computing method, namely, MGGP has been discussed. To overcome this, a SRM-MGGP methodology is proposed for the prediction of FOS of the 3-D nailed slope. The performance of the proposed SRM-MGGP model is compared to those of MGGP, SVR and ANN

Acknowledgement

This work was partially supported by the Singapore Ministry of Education Academic Research Fund through Research Grant RG30/10, which the authors gratefully acknowledge.

References (60)

  • I. Maric

    Optimization of self-organizing polynomial neural networks

    Expert Syst. Appl.

    (2013)
  • T. Matsui et al.

    Finite element slope stability analysis by shear strength reduction technique

    Soils Found.

    (1992)
  • D. Salgado et al.

    An approach based on current and sound signals for in-process tool wear monitoring

    Int. J. Mach. Tools Manuf.

    (2007)
  • K. Ugai et al.

    Three-dimensional limit equilibrium and finite element analysis: a comparison of results

    Soils Found.

    (1995)
  • W.B. Wei et al.

    Soil nailed slope by strength reduction and limit equilibrium methods

    Comput. Geotech.

    (2010)
  • Abdalla, J.A., Attom, M., Hawileh, R., 2012. Artificial neural network prediction of factor of safety of slope...
  • Byrne R.J., Cotton D., Porterfield J., Wolschlag C., Ueblacker G., 1996. Manual for design and construction monitoring...
  • E. Byvatov et al.

    Support vector machine applications in bioinformatics

    Appl. Bioinforma.

    (2003)
  • A. Cevik et al.

    Modelling the performance of self-compacting sifcon of cement slurries using genetic programming technique

    Comput. Concr.

    (2008)
  • V. Cherkassky et al.

    Comparison of model selection for regression

    Neural Comput.

    (2003)
  • A.J. Choobbasti et al.

    Prediction of slope stability using artificial neural network (case study: Noabad, Mazandaran, Iran)

    Arabian J. Geosci.

    (2009)
  • E.M. Dawson et al.

    Slope stability analysis by strength reduction

    Geotechnique

    (1999)
  • Donald, I.B., Giam, S.K. 1988. Application of the nodal displacement method to slope stability analysis. In:...
  • A.H. Gandomi et al.

    Genetic programming and orthogonal least squares: a hybrid approach to modeling the compressive strength of Cfrp-confined concrete cylinders

    J Mech Mater Struct

    (2010)
  • A. Garg et al.

    Review of empirical modeling techniques for modeling of turning process

    Int. J. Model., Identif. Control

    (2013)
  • A. Garg et al.

    A multi-gene genetic programming model for estimating stress dependent soil water retention curves

    Comput. Geosci.

    (2014)
  • A. Garg et al.

    Classification-driven model selection approach of genetic programming in modelling of turning process

    Int. J. Adv. Manuf. Technol.

    (2013)
  • A. Garg et al.

    State-of-the-art in empirical modelling of rapid prototyping processes

    Rapid Prototyp. J.

    (2014)
  • Garg, A., Sriram S., Tai K. 2013b. Empirical analysis of model selection criteria for genetic programming in modeling...
  • A. Garg et al.

    A hybrid M5-genetic programming approach for ensuring greater trustworthiness of prediction ability in modelling of FDM process

    J. Intell. Manuf.

    (2014)
  • Cited by (88)

    • Application of artificial intelligence in geotechnical engineering: A state-of-the-art review

      2022, Earth-Science Reviews
      Citation Excerpt :

      SVM was also used by some researchers with a different combination of inputs. These included cohesion, angle of internal friction, pore water pressure coefficient, unit weight of soil, slope angle, and height of slope (Garg et al., 2014; Hoang and Pham, 2016; Liang and Zhang, 2010; Samui, 2008, 2013; Xue et al., 2015; Zhang et al., 2014). To compare different methods of AI, Moayedi et al. (2019a) used ANN, ANFIS and a hybrid PSO-ANN method to predict the factor of safety.

    • Multigene genetic programming and its various applications

      2022, Handbook of HydroInformatics: Volume I: Classic Soft-Computing Techniques
    • A rigorous model for prediction of viscosity of oil-based hybrid nanofluids

      2020, Physica A: Statistical Mechanics and its Applications
      Citation Excerpt :

      The GP and GA have some similarities, for example, most of the GA operators can be used in the GP method with minor modifications. On the other hand, they also have differences, the GP has a tree structure to present solutions, while the GA uses string to represent solutions [55]. In this study, multi-gene genetic programming (MGGP) is developed as a promising alternative to GP to estimate the viscosity of Newtonian oil-based hybrid nanofluids.

    View all citing articles on Scopus
    View full text