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

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. © 2020 Elsevier B.V. All rights reserved.