artificial neural network combined with imperialist ...

$#

!%

!" !

"# "% %" !"""$ " " " $ !"! -0,11377!-AA-,*,766)*1 )B)>1 ):12)6)),B15)A):37 1

Young Researchers and Elite Club, Ahvaz Branch, Islamic Azad University, Ahvaz, Iran Department of Civil Engineering, Abadan Branch, Islamic Azad University, Abadan, Iran 3 Faculty of Civil Engineering, University of J.J. Strossmayer, Vladimira Preloga 3, 31000, Osijek, Croatia 2

" #"

!" " Estimation of sediment volume transported by a river has become an important water engineering issue. Due to the lack of exact and detailed information on the parameters affecting the non-linear nature of sedimentation process including spatial and temporal variances, a comprehensive sedimentation model cannot be formulated. The new evolving technique of utilizing artificial neural networks, which is based on an optimization algorithm, has found vast applications in different scientific fields, especially in water and river engineering. The Imperialist Competitive Algorithm (ICA) is based on random populations and the idea of the human's socio-political evolution. In this algorithm, a number of imperialist countries and their colonies search for a generalized optimization method for finding optimizing solutions. This research is based on the Feed Forward Artificial Neural Network (FF-ANN) model and attempts to predict and determine sedimentation in rivers. One of the elements used as a new method is employment of ICA for finding the optimized values within ANNs, which is also used for predicting river sedimentations of Karoon River in Ahvaz, Iran. For this purpose, discharge, month of year, height, and density coefficient are the input parameters, and sedimentation estimation is the output value. To determine accuracy of the FF-ICA model, it was compared with genetic and particle swarm group algorithms. This comparison was carried out in three stages of investigation, training, and testing. The results state that the ANN with its weights optimized within ICA, when compared with Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) Algorithm, has greater flexibility and accuracy in predicting river sedimentation.

Determining the exact volume of sediments carried by rivers has great importance in many water management projects. Up to now, many numerical models, which are based on the theory of nonequilibrium sediment transport, have been provided for predicting river sedimentation and are widely used in the pre-design phase for engineering projects. However, the biggest disadvantage is that the simulated results of classic numerical model usually disagree with the measurements, what may be attributed to the imperfection of the model structure and the difficulty in parameter calibration [1]. According to Fang et al. [1], to improve the numerical model accuracy, there are three approaches which can be employed: (1) data assimilation method, which originates from weather prediction; (2) building up coupled numerical models, in which the flow, sediment transport and morphological evolution processes are strongly coupled with one another and (3) to use a series of optimal methods, such as artificial intelligence and neural networks (e.g. [2], [3], [4]). Considering the nonlinear behavior of hydraulic parameters make us to find some more progressive alternatives instead of the older classic methods such as rating curve models which do not have the required accuracy [5]. Recent research has shown that Artificial Neural Network Models can be used as a black-box method in modeling hydraulic parameters. Yitian and Gu [6] modeled river flow and sedimentations in river systems with artificial neural networks. For this purpose, ANN model based on an existing river was developed to predict river flow and sedimentations. Results indicated that artificial neural networks are powerful tools for real-time prediction of this occurrence in a complex river system. Alp and Cigizoglu [7] considered the meteorological data within Artificial Neural Models to predict river sedimentations. They used feed-forward back-propagation (FFBP) method and radial basis functions (RBF) to model the suspended load of sediments in the Catchment River. Dogan et al. [8] used artificial neural networks to predict the total sediment load. Their aim was to create an effective model regarding sediment load, greater slopes, and the size of the load

&% ! Imperialist Competitive Algorithm (ICA), Artificial Neural Network (ANN), Sedimentation, Karoon River, Genetic Algorithm (GA), PSO Algorithm

4658

$#

!%

particles. Results illustrated that the developed neural network determined greater accuracy than the Acaroglu equation and Graf model. Kisi [9] predicted river sedimentations using Fuzzy models. Results showed that fuzzy models could predict the river sediment loads with a very high accuracy. Yand et al. [10] used effective parameters on neural network for sediment determination. They also used four parameters of average flow velocity, water surface slopes, average flow depth, and median particle diameter for sediment determination. Results showed that the ANN model with minimum input parameters is a reliable method for predicting the total sediment. Melesee et al. [11] predicted the suspended sedimentation load of rivers using MLP network. The input parameters in that research were discharge, previous day's discharge, rainfall, and the output was river sedimentation. They also used nonlinear regression and time series models. Results indicated that artificial neural networks have suitable accuracy in predicting the level of river sedimentations. Rajaee [12] used combination of wavelets and artificial neural networks to determine river sedimentation. They evaluated the model using multi linear regression and rating curve. The results showed that in comparison with the other two models, the combinational Wavelet Artificial Neural Network (WANN) model provides better results in predicting the suspended river loads. Ramezani et al. [13] determined sediment on Maron River using both artificial neural network and social-based algorithm. They used social-based algorithm to optimize the artificial neural network weights due to determination of sediment on Idenak, Tang Takab, Cham Nezam and Jookang stations. Likewise, the input data were length, discharge, and height of every station, and the output was sediment. They illustrated that artificial neural network combined by social-based algorithm is more flexible and able to determine the sediment on Maron river. Different methods for solving optimization problems have been introduced. Some of them are repetitive methods, which are gradient based and find the optimized cost function value. Although they have great speed, one of their weakness in concentrating on local optimizations [14]. The main aim of this study is to use the Imperialist Competitive Algorithm for optimizing weights within Artificial Neural Networks as a new optimizing algorithm for predicting river sedimentation. The advantages of this system include: (1) Introducing the new fundamental idea of use of Imperialist Competitive Algorithm as the first optimizing algorithm based on a socio-political process; (2) The ability equal to or even higher than different optimization algorithms in facing different optimization issues and (3) Suitable speed in finding solutions. In this article, Karoon river sedimentation prediction has been carried out using Feed Forward Artificial Neural Network. Existing data from Ahvaz

!" !

station were employed in a suitable model of Artificial Neural Networks. For this purpose, discharge, month of year, height and density coefficient are the input parameters, and sedimentation estimation is the output. For evaluation of accuracy of the FF-ICA model, it was compared with the genetic algorithm and PSO algorithm. This research was carried out in three stages of training, testing and prediction. " # !" ""$ " "# "% ! 58-:1)41;