Comparative Assessment of the Hybrid Genetic Algorithm–Artificial Neural Network and Genetic Programming Methods for the Prediction of Longitudinal Velocity Field around a Single Straight Groyne

Highlights

• Longitudinal velocity field around single straight groyne is modelled using a novel hybrid method of Genetic Algorithm based artificial neural network (GAA) that has the ability to automatically adjust the number of hidden neurons.

• The results of GAA is measured and compared with genetic algorithm based prediction method (GP).

• Various input combinations are examined in the GAA and GP models.

• The proposed method for prediction of the flow features around the single groyne is never used in literature before.

Abstract

In the present paper, three-dimensional flow fields around single straight groynes with various lengths have been discussed. The dataset of the flow field is measured in the laboratory using Acoustic Doppler Velocimeter (ADV). Then, the longitudinal velocity field is modelled using a novel hybrid method of Genetic Algorithm based artificial neural network (GAA) that has the ability to automatically adjust the number of hidden neurons. To investigate the proposed method’s performance, the results of GAA is measured and compared with one of the most common genetic algorithm based prediction method, namely genetic programming (GP). It is concluded that that GAA model successfully simulates the complex velocity field, and both the velocity magnitudes and isovel shapes are well predicted by this model. The results show that GAA with RMSE of 0.1236 in test data has a significantly better performance than the GP model with RMSE of 0.2342. In addition, it was founded that the transverse coordinate of the measuring point (Y*) is the most important input variable.

Introduction

Groyne or spur dike is a transverse river structure built from the riverbank toward the center to disrupt the approaching flow and divert it away from the side of the river on which the groyne is installed. It is used to control bank erosion [1], provide the fairway of sufficient depth and width [2], as habitat for various species [3], and as a countermeasure against local scour around a bridge abutment [4]. Baisically, groynes are protective structures, but, on the contrary, they may induce local strong flow around the head zone and it’s downstream as a result of the river constriction. The most important flow structures around a single groyne are as follows: (1) the vertical downflow immediately at the leading face of the groyne [5], (2) horseshoe vortex (HSV) and three dimensional necklace vortex around and next to the the groyne tip Koken and Constantinescu [6], (3) recirculating and reattachment zones at the wake of the groyne, and (4) a turbulent mixing layer (TML) bounding the separation zone Koken and Constantinescu [6]. More detailes about the three dimensional flow features around a single straight groyne are reported by Koken and Constantinescu [6], Koken and Constantinescu [7] and [8].

The accelerated flow around the groyne head, the HSV, the vertical downflow, and TML may induce local bed erosion around the groyne Koken and Constantinescu 2008a[6], and as a result of undermining of the foundation, these flow features can affect the stability of the structure. Consequently, it is of great importance to enhance our understanding about the flow physics and its interaction with the river structures. The current knowledge about the flow around the groynes is based on the previous experimental and numerical studies [5], [6], [7], [8], [9], [10], [11], [12].

Experimental studies are costly and they are time consuming. This is why nowadays most of the flow field studies appear more in the form of numerical models, which provide predictions in a cheap and versatile way. Despite performing advanced numerical simulations in the literature (e.g., [13], [14], [15], [16], [17], [18], [19] comprehensive numerical model is not provided to predict all the aspects of the complex 3D flow fields around groynes. This problem is partly related to the assumptions made in turbulence modeling. Advanced eddy-resolving techniques such as Large Eddy Simulation (LES) and Detached Eddy Simulations (DES) have been developed; however, they still remain computationally prohibitive for hydraulic engineers requiring routine calculations of highly turbulent flows around river structures. In addition, implementation of advanced numerical models requires high-performance computer resources.

On the contrary, soft computing methods such as Artificial Neural Network (ANN), Genetic Algorithms (GA), and Fuzzy logics are becoming an important class of efficient tools for developing intelligent systems and providing solutions to complicated engineering problems with less computational costs compared to the advanced numerical methods.

ANN nowadays is recognized as a very efficient tool for relating input data to output data that is representing arbitrarily complex non-linear processes. Consequently, they have been widely utilized in various fields. In water resource management, the ANN are used to forecast the rainfall intensity [20], River flood forecasting [21], forecasting the quality of water for the North Saskatchewan River [22], estimation of maximum local scour depth around river structures [23], [24], prediction of the friction coefficient in open channels [25], prediction of backwater through arched bridge constrictions [26], and discharge prediction in straight compound open channels [27]. Multi-layer feed-forward neural network was used to simulate the onset of river breakup considering the relationship among the streamflow hydraulics, meteorological conditions, and ice mechanics [28]. Assessment of the tranquility of Trabzon Yacht Harbor was performed using ANN [29]. They tested the capability of the ANN for the input values of different wave and breakwater conditions. Length of hydraulic jumps in a rectangular channel was determined using ANN [30]. Capability of wavelet-based data preprocessing method was investigated for improving the performance of the ANN method in hydrologic field non-stationary time series forecasting [31].

Genetic Programing (GP) is one of the new and efficient techniques to solve a wide range of problems in hydraulic engineering field such as velocity prediction in compound open channels [32], rainfall-runoff modeling [33], [34], [31], [35], sediment modeling [36], [37], [38], [39], [40], and open channel characteristics [41]. GP is a technique that automatically solves the problems without needing the user to specify the form of the solution.

Recently, some attempts have been made to use intelligent methods for studying the open channel hydrodynamics. There are two major advantages of modelling the complex velocity field by using the soft computing methods. First, soft computing methods have a high performance in modelling the complex problems. Therefore, the results of these methods could be used efficiently in the practical situations or researches. Second, soft computing methods reconstruct the discrete laboratory velocity measurements into analytical relationships, and thus continuous velocity field description could be developed. The power of the ANN for modeling the velocity profiles and contours and simulating the discharges in 90-degree combined open channels was evaluated by [42]. A 3D flow field in 90-degree curved channel was predicted using CFD, ANN, and GA methods by [43]. The applicability of ANN models for modeling the velocity distribution of combined open channel flows was investigated by [44]. They used CFD data for construction and test of the ANN models. Recently, ANN and GP models are used for modeling the distribution of the longitudinal velocity in open channel junction [45].

Literature review shows that no attempt has been made to predict the 3D complex flow field around groyne structure using intelligent methods. In this research, performance of an innovative hybrid Genetic Algorithm structure recognition–Artificial neural network (GAA) and GP methods are evaluated for simulating the mean flow field around single straight groynes based on highly accurate and large experimental datasets.

The following section of the paper gives details of the experimental study for data generation. Then structures of the soft computing methods are presented and in subsequent sections step by step implementation of the GAA and GP methods for prediction of the longitudinal flow around single straight groyne is given. Finally, we will apply the developed GAA and GP to plot the velocity contours at various cross sections around the groynes to assess their capability in prediction of the important flow features.