Elsevier

Journal of Cleaner Production

Volume 276, 10 December 2020, 124267
Journal of Cleaner Production

Invasive weed optimization-based adaptive neuro-fuzzy inference system hybrid model for sediment transport with a bed deposit

https://doi.org/10.1016/j.jclepro.2020.124267Get rights and content

Highlights

  • Non-deposition with deposited bed (NDB) is applied for large channel design.

  • NDB sediment transport is modeled using wide ranges of experimental data.

  • ANFIS-IWO is implemented for NDB sediment transport modeling for the first time.

  • ANFIS-IWO outperforms ANFIS-GA, ANFIS, MGP and PSO models.

  • Ranges of data, parameters used and applied techniques are important in the modeling.

Abstract

Inasmuch as channels are designed to mitigate continues sedimentation, sediment transport models have been developed to calculate flow velocity to keep sediment particles in motion. In order to promote the computation capability of sediment transport models, recently machine learning algorithms have attracted interests, extensively. However, accuracy of such a model is attributed to the range of data and applied technique for model construction. For this purpose, the current study scrutinizes the applicability of “non-deposition with deposited bed” (NDB) concept for design of large channels applying hybrid machine learning algorithms. Through the modeling, firstly, conventional adaptive neuro-fuzzy inference system (ANFIS) technique is applied to develop a stand-alone model. In furtherance of improving the model’s performance, the ANFIS is hybridized with invasive weed optimization (IWO) algorithm to construct a hybrid ANFIS-IWO model. As a benchmark, the ANFIS is further hybridized with classical genetic algorithm (GA) to compare with ANFIS-IWO outcomes. Furthermore, the developed machine learning models are compared to multigene genetic programming (MGP) and particle swarm optimization (PSO) stand-alone machine learning results reported in the literature and classical regression models by means of variety of statistical performance measurements. Hybridization of ANFIS with IWO, enhances its accuracy with a factor of 30%. Respecting to the models performance examination, the ANFIS-IWO model is found superior to its alternatives for sediment transport computation. The thickness of the deposited bed and deposited bed width are found as effective parameters for sediment transport modeling in open channels with a bed deposit.

Introduction

Controlling sediment transport process plays an essential role in designing the lined channels as a fundamental environmental engineering practice. Pollution transport is one of the most important problems resulting from sedimentation. Sediment deposition decreases the transport capacity of the channels due to deduction in the flow cross-section area and increasing the channel bed roughness (Safari et al., 2017a). Based on aforesaid problems, channels are designed by means of self-cleansing concept. Regarding the self-cleansing definition, flow can carry sediments without permanent deposition and erode stationary deposited particles from the channel bed (Safari et al., 2018). Among variety of self-cleansing concepts, non-deposition criteria in which flow keeps the sediment particle in motion are reliable design approach (El- Zaemey, 1991; May 1993; Ab Ghani, 1993; Nalluri et al., 1997; Safari et al., 2017b). Non-deposition concept covers three criteria of “incipient deposition” (ID), “non-deposition with clean bed” (NCB) and “non-deposition with deposited bed” (NDB). The condition where sediment particles are gathered at the bottom of the channels is known as ID (Aksoy et al., 2017). In the NCB condition, sediment particles are transported and the channel has a clean bed (Ackers et al., 1996; May et al., 1996; Butler et al., 2003), while in the NDB sediment transport condition, the small thickness of deposited sediments presents at the channel bottom to reduce the required design velocity.

It is known that, the channel size is an important element in designing the channels and the self-cleansing velocity is reliant on that. Larger channels need higher design self-cleansing velocity; therefore, design of the large channels based on NCB is not an economical process since it requires steeper channel bed slope. Therefore, large channels should be designed based on NDB condition (Ab Ghani, 1993; May 1993; Ota and Nalluri, 2003). Nalluri et al. (1997) and Butler et al. (2003) demonstrated that 1–2% of pipe diameter is acceptable for the thickness of the deposited bed. Alvarez (1990), El-Zaemey (1991), May (1993), Ab Ghani (1993), Nalluri et al. (1994, 1997), Ackers et al. (1996) and Butler et al. (2003) studied NDB sediment transport condition. Safari and Shirzad (2019) showed that the bed thickness of 1–5% of pipe diameter considerably decreases the required design velocity; however, 1% is found appropriate where design self-cleaning velocity of the channel decreases with a factor of 20%.

Studies mentioned above developed self-cleansing models established on non-linear regression analysis. Such models are mostly utilized for describing the effective parameters involved in the sediment transport in channels with a simple structure; however, from the hydraulics point of view, the main deficiency of existing empirical models in the literature for NDB condition of sediment transport can be linked to the limitation of the data used for the model development. It is known that the credibility of a sediment transport model is attributed to the ranges of data used. Most of the studies in the literature utilized limited number of data and more importantly, they were over-fitted on the entire data set which cause their poor performance on alternative data sets. Recently, machine learning methods are applied for different engineering problems. For modeling the sediment transport in NCB condition, several studies were conducted (Safari, 2019). For example, Roushangar and Ghasempour (2017) and Wan Mohtar et al. (2018) used evolutionary algorithm and artificial neural networks respectively, for the same purposes. The successful implementation of the ANFIS technique was documented by Azamathulla et al. (2012) and Ebtehaj and Bonakdari (2014). Aforementioned studies modeled sediment transport in NCB condition. As shown in Table 1, applying machine learning techniques, Safari and Danandeh Mehr (2018) and Safari and Shirzad (2019) used multigene genetic programming (MGP) and particle swarm optimization (PSO) algorithm, respectively, for modeling the NDB condition. Danandeh Mehr and Safari (2020) reported the superiority of MGP to gene expression programing (GEP) and multi-layer perceptron (MLP) for the same purpose. Montes et al. (2020) applied least absolute shrinkage and selection operator (LASSO) for NDB modeling. It must be noticed that they used stand-alone models. Although the application of hybrid techniques is reported for the suspended sediment transport (Shiri and Kişi, 2012), to the best of the authors’ knowledge, the hybrid models are not utilized for modeling sediment transport in NDB condition applicable for large channel design.

Existing empirical equations in the literature are mostly developed on limited number of experimental data. Due to the fact that empirical equations were over-fitted on the limited data ranges, they may provide poor results, once applied on variety of data sources having wider data ranges. To this end, comprising all reported data in the literature, this study recommends models established on wide data ranges to promote their reliability for practical use. As first implementation of hybrid algorithms in NDB condition of sediment transport, this study presents a novel approach through hybridization of adaptive neuro-fuzzy inference system (ANFIS) with invasive weed optimization algorithm (IWO). According to the literature, a few models were established for designing the NDB sediment transport condition and it seems to be an essential task to develop a robust model that can be used for various data sources. For model development, a wide experimental data ranges are utilized, and then, the computation ability of the developed models is compared to their alternatives.

Section snippets

Self-cleansing based on NDB condition

Large channel pipes with D > 500 mm are designed based on NDB condition as shown in Fig. 1. This criterion is economical in comparison with NCB since a small amount of sediment deposited at the channel bed is allowed and, accordingly it decreases the channel design bed slope. Small depth of sediment can increase the flow capacity for bed load sediment transport and particles just transport in the top layers of channel bed (May et al., 1989; Butler et al., 2003; May 1993). As examples of the

Comparison of models

The developed stand-alone ANFIS, and hybrid ANFIS-IWO and ANFIS-GA models developed in the current study, are compared to results of machine learning models and classical regression models existing in the relevant literature with regard to the particle Froude number (Frp) computation in NDB condition of sediment transport. Accordingly, Eq. (1) of El-Zaemey (1991), Eq. (2) of Ab Ghani (1993), Eq. (3) of Nalluri et al. (1997), Eq. (5) and Eq. (6) generated based on MGP and PSO, respectively; are

Discussion

This study investigates the sediment transport at NDB condition, which is suggested as an economic design criterion for large channels. Existing the deposited sediment at the channel bed causes to increase in channel roughness; however, transport capacity of the flow is increased as it is found already in the literature. The difficulty of finding a reliable model which can perform satisfactory on different data sets was emphasized by Safari et al. (2018). Therefore, it is considered as a main

Conclusions

Large channels are designed based on the NDB condition of sediment transport. Utilizing wide ranges of experimental data taken from the literature covering wide ranges of sediment volumetric concentration, pipe and sediment sizes, deposited bed thickness and width, flow depth and velocity, self-cleansing models are recommended in this study. Incorporating the channel, fluid, sediment and flow variables, sediment volumetric concentration, dimensionless grain size parameter, relative particle

Credit author statement

Mir Jafar Sadegh Safari: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Visualization. Babak Mohammadi: Conceptualization, Methodology, Implementation of the computer code and algorithms, Validation, Investigation, Writing - original draft, Visualization. Katayoun Kargar: Writing - original draft, Writing - review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (53)

  • M.J.S. Safari

    Hybridization of multivariate adaptive regression splines and random forest models with an empirical equation for sediment deposition prediction in open channel flow

    J. Hydrol.

    (2020)
  • M.J.S. Safari et al.

    Multigene genetic programming for sediment transport modeling in sewers for conditions of non-deposition with a bed deposit

    Int. J. Sediment Res.

    (2018)
  • A. Ab Ghani

    Sediment Transport in Sewers (Doctoral Dissertation)

    (1993)
  • J.C. Ackers et al.

    Design of Sewers to Control Sediment Problems

    (1996)
  • P. Aghelpour et al.

    Long-term monthly average temperature forecasting in some climate types of Iran, using the models SARIMA, SVR, and SVR-FA

    Theor. Appl. Climatol.

    (2019)
  • H. Aksoy et al.

    Velocity-based analysis of sediment incipient deposition in rigid boundary open channels

    Water Sci. Technol.

    (2017)
  • E.M. Alvarez

    The Influence Cohesion on Sediment Movement in Channels of Circular Cross-Section

    (1990)
  • C.M. Bishop

    Neural Networks for Pattern Recognition

    (1995)
  • D. Butler et al.

    Self-cleansing 442 sewer design based on sediment transport principles

    J. Hydraul. Eng.

    (2003)
  • A. Danandeh Mehr et al.

    Application of soft computing techniques for particle Froude number estimation in sewer pipes

    J. Pipeline Syst. Eng.

    (2020)
  • I. Ebtehaj et al.

    Performance evaluation of adaptive neural fuzzy inference system for sediment transport in sewers

    Water Resour. Manag.

    (2014)
  • A.K.S. El-Zaemey

    Sediment Transport over Deposited Beds in Sewers (Doctoral Dissertation)

    (1991)
  • J. Farhoudi et al.

    Application of the neuro-fuzzy model to estimate the characteristics of local scour downstream of stilling basins

    J. Hydroinf.

    (2010)
  • E. Ghasemi et al.

    A new hybrid ANFIS–PSO model for prediction of peak particle velocity due to bench blasting

    Eng. Comput.

    (2016)
  • S. Haykin

    Neural Networks: A Comprehensive Foundation

    (1999)
  • J.H. Holland

    Adaptation in Natural and Artificial Systems: an Introductory Analysis with Applications to Biology, Control, and Artificial Intelligence

    (1992)
  • Cited by (18)

    • Compensating for the loss of future tree values in the model of Fuzzy knowledge units

      2022, Urban Forestry and Urban Greening
      Citation Excerpt :

      These hybrid models provide the best predictive results in forecast among the three hybrid models with RMSE of 0.127. Also, Safari et al. (2020) recommended ANFIS which is hybridized with invasive weed optimization (IWO) algorithm. Fuzzy expert systems (FES) are a part of hybrid expert systems, which partially solve the given problem.

    • Predictability performance enhancement for suspended sediment in rivers: Inspection of newly developed hybrid adaptive neuro-fuzzy system model

      2022, International Journal of Sediment Research
      Citation Excerpt :

      The fact is that hybrid models aid in finding the optimal results by reducing the drawbacks of standalone soft computing models (Kargar et al., 2021; Meshram et al., 2021; Mohammadi et al., 2021; Panahi et al., 2021; Safari, 2020). Recent literature also showed the effectiveness of the proposed PSOGSA algorithm in finding better optimal values of model parameters, and, thus, improvement of model estimation accuracy in comparison to other optimization algorithms (Liu & Chen, 2020; Safari et al., 2020; Xu & Liu, 2020; Zhu et al., 2018). For a robust and generalized optimization algorithm, it is expected that the algorithm would able to search the best parameter for a machine learning model.

    View all citing articles on Scopus
    View full text