Estimation of total bottom sediment load in waterways based on Support Vector Regression (SVR) model and Particle Community Optimization (PSO) algorithm

Master's Thesis in Civil Engineering (Trend of Hydraulic Structures)

Abstract

Sediment transfer and sedimentation, consequences such as the creation of sediment islands in the river path and as a result of reducing the transfer capacity. Flood currents include corrosion of river structures and other problems. Also, suspended sediments affect the quality of water for human use. Therefore, in the hydraulics of the river and its geomorphology, it is important to investigate the sediment carrying capacity of the stream and the sediment transport mechanism. Conventional approaches are often based on ideal assumptions and are not able to provide acceptable results of estimating bed sediment transport rates. In this thesis, an effort is made to implement a comprehensive and accurate method using the knowledge of artificial intelligence on the problems of predicting and estimating sediment. Two methods called least squares regression of the support vector and particle community optimization algorithm were used in order to estimate the transfer rate of bed sediments in waterways with much higher accuracy than common methods such as Eckers and White, Engeland and Hansen, Graff and Young methods. The support vector machine approach is based on the theory of constrained optimization and uses the principle of structural error minimization, which leads to an overall optimal solution. The particle community optimization algorithm is included in the category of meta-exploratory methods and the idea is taken from the order in the collective behavior of birds to search for food. The results of the implementation of the support vector least squares regression model on the sets of laboratory and field data have been far better compared to the conventional approaches. Then, in order to improve the model, the input variables were scaled logarithmically and negative concentration values ??were prevented in the model and the results were relatively improved. The results of the particle community optimization algorithm are satisfactory compared to conventional approaches, but the performance of the support vector machine least squares model is more satisfactory, and the support vector regression can provide a comprehensive and accurate method to simulate the transfer rate of bed sediments. It is proportional to the ability to maximize benefits and minimize losses caused by rivers. Rivers are constantly faced with the phenomena of erosion and sediment transfer and adjust their cross section, longitudinal profile, direction and flow pattern through the processes of sediment transfer, scouring and sedimentation. For sustainable economic and cultural development along the river, it is necessary to understand the principles of sediment transport and its estimation. These principles can be used to solve environmental and engineering problems related to natural disasters and human activities. In human activities including; Agriculture, animal husbandry, development of industries and urban development as well as mines, the natural state of soil and plants have undergone changes dramatically and without applying strict control, it usually leads to unnatural soil erosion. Therefore, in the hydraulics of the river and its geomorphology, the investigation of the sediment carrying capacity of the stream and the mechanism of sediment transfer are of particular importance. The science of sediment transport deals with the mutual relationship between water flow and sediment particles. Sediment transfer and sedimentation, consequences such as the creation of sediment islands in the river course and as a result, the reduction of flood flow transfer capacity, the reduction of the useful life of dams and the storage capacity of reservoirs, the corrosion of facilities. It includes river structures and damage to water structures and farms, sedimentation in the channel floor and many other issues and problems. On the other hand, suspended sediments affect the quality of water for human use. Suspended inorganic and organic substances are not only the main factor in water pollution, but also as a carrier of other pollutants such as; Agricultural poisons or harmful microbes work. Also, considering the need to know the amount of sediments carried by the river flow in the design of river structures, the need to estimate the sediment load of rivers is clearly explained. Sediment movement in rivers has been studied by hydraulic engineers and geologists due to its importance for understanding river hydraulics, river engineering, river morphology, and other such topics. Sediment transport is a complex issue and often has experimental or semi-experimental relationships.Most of the theoretical relationships are based on ideal assumptions and simplifications so that the sediment transport rate can be determined by one or two dominant factors such as water discharge, average flow velocity, energy gradient and shear stress. Various approaches have been used to solve engineering problems and various numerical relations have been published. The results obtained from different approaches are often very different from each other and from field observations. As a result, none of the traditional sediment transfer relationships, due to not providing a comprehensive approach and not taking into account all the effective variables in sediment discharge calculations, the sediment estimation is done with very low accuracy.

In this research, artificial intelligence methods are used to estimate the total sediment load of the floor based on the flow and sediment geometric variables as well as the hydraulic variables of the flow. Traditional methods are compared.

1-2- Necessity of conducting research

Considering the many studies that have been conducted in the field of sediment transport issues, some of them require theoretical analysis and some are based on experimental methods, and in many cases a combination of theoretical and experimental approaches is required. Most of the theoretical methods are based on some ideal and simplified assumptions so that the sediment transfer rate is based on one or two dominant factors including; Flow rate, average flow velocity, energy gradient or shear stress are determined. Many equations have been published. Each of these relationships have been obtained by limited laboratory data and in some cases based on field data. The results obtained from different relationships often have significant differences with each other and with the observed values. As a result, none of the sediment transfer relationships have general acceptance, especially in rivers, in predicting the sediment transfer rate, and they provide an acceptable answer only in special conditions. If we want to briefly express the important relations of sediment transport, apart from probabilistic and regression approaches, they can be expressed in one of the following forms, assuming that the rate or concentration of sediment transport can be determined through a dominant variable:

(Formulas are available in the main file) , , and , respectively, are sediment discharge per channel width unit, flow rate, average flow velocity, water surface slope or energy, shear stress, flow power per unit floor area, and flow unit power, and are also parameters related to flow and sediment conditions. The subscript c also indicates critical conditions at the threshold of movement. Many complex aspects of sediment transport require correct understanding and will be a challenging topic for future studies. 1-3- Research Objectives In this research, an effort is made to implement a comprehensive and accurate method based on artificial intelligence knowledge on sediment prediction and estimation issues. The issue of sediment transport has been studied by river engineers and morphologists for several centuries. Various approaches have been used to solve engineering problems. The results obtained from these approaches are drastically different from each other and from field results. In recent years, a number of basic concepts, limitations of application and mutual relationship between them have been clarified for us.

Given that sediment transport is a complex phenomenon and the size of sediment particles has a wide range and on the other hand, the channel bed takes various forms, using data-driven methods for such issues will definitely provide a better answer. In this method, most of the effective parameters in the sediment transfer rate play a role in the problem, and the current problem is not based on only one variable or dominant factor. In this research, an effort is made to compare the results obtained from support vector regression with traditional methods and reach a general conclusion. In the support vector regression method, the calibration parameters are difficult to calibrate. For this reason, he used an optimization algorithm called the particle assembly algorithm so that parameters can be calibrated easily. During this research, a sensitivity analysis will be performed on the input parameters and the important parameters will be introduced in order. This method uses a wide range of laboratory and field data and is used for different flow and sediment conditions and gives good results.