Investigating the application of the artificial neural network method in estimating the annual runoff of watersheds (a case study of the entire watershed, Fars province)

Dissertation for Master's degree (M.Sc.)

Investigating the application of artificial neural network method in estimating the annual runoff of watersheds

(Case study: Total watershed, Fars province)

By: Jalal Zarei

Abstract:

The importance and place of water in human life, especially in today's advanced and overpopulated world, is not hidden from anyone, and industrial and urban life has multiplied the per capita consumption of water compared to traditional living conditions. Today, this natural resource is considered as one of the most decisive parameters of social and even political planning, to the extent that they predict the future conflict of different countries around each country's water rights, and this is more visible in countries that benefit from common water resources.  In this regard, the purpose of this research is to determine the efficiency of the artificial neural network method in estimating the runoff in 3 stations of Band Bahman, Chamriz and Darb Qala, located in Fars province. The method of collecting information in this research is field and library. The estimation of the amount of runoff of Qara Aghaj, Kor and Rod Bal rivers, which are related to all 3 mentioned stations, was done by the measurement data such as rainfall, temperature and evaporation. The results were that, by using artificial neural network method of multilayer perceptron and Gossin-Gossin type stimulation function, with 3 neurons in the input layer, 2 neurons in the hidden layer and 1 neuron in the output layer, here, the error back propagation learning law with 50,000 repetitions were able to successfully complete the training procedure and bring the average error of the network to an acceptable level and prove the H1 hypothesis regarding the mentioned stations.

Key words: precipitation, air temperature, runoff, avalanche stop station, neural network model

Chapter One

Introduction and Generalities

The flow-precipitation for a water area is one of the most important issues related to hydrologists and engineers. Information in this field is necessary for the design and management purposes of engineers. This relationship is a non-linear and complex relationship. The use cases of this Harami forecast can be attributed to the correct management of agricultural plans, flood warning systems and exploitation of dam reservoirs. Precipitation- Runoff is one of the most complex hydrological processes. For many years, hydrologists have been concerned with understanding how precipitation is converted to runoff to predict floods. Among the goals of this knowledge are water storage, flood control, irrigation, drainage, water quality, energy production, recreation centers, fish farming, and wildlife expansion. pointed out The number of parameters, the instability of catchment basin characteristics and precipitation models complicate the problem even more. The use of statistical, hydraulic and hydrological models has a long history in the discussion of rainfall-runoff modeling. Considering the problems and weaknesses that exist in conceptual and statistical models, the need for a model that can perform mapping operations with input and output parameters seems essential. During the last decade, the nonlinear mathematical model of artificial neural networks [1] (ANN) has been added to predictive tools and various researches have been conducted in the field of rainfall-runoff modeling using this tool. Artificial neural networks are a powerful tool that is made by simple imitation of the human biological nervous system. Neural networks are divided into two types: artificial and natural neural networks. Research in the field of neural networks began when the brain was a dynamic system with a parallel structure and A new approach to the human brain, which led to the expression of the artificial neural network, was proposed by a person named Segal. Also, the neuron, as the smallest unit of data processing in an artificial neural network, is made up of several neurons that perform a specific function in the network.

Neuron cells in different layers determine the structure of the network, which is called network architecture. The neural network consists of several layers. Layers are responsible for receiving data.. The layers are responsible for receiving data, processing and producing the output quantity.                                                                                                                                                                                                                                                                    

? Input layer: This layer has no neurons and is the layer that transfers the input variables to the neurons of the hidden layer without any changes. Generally, the number of neurons in these layers is a function of the number of input variables to the neural network.                   

? Output layer: this layer has one or more neurons and the number of neurons is a function of the number of the output function.

The layers of the network are connected by attachments with different weights. It is generally called the Perceptron multilayer neural network [2] ((MLP) and the communication between neurons and the adjustment of weights in it follow the learning rules. rtl;"> Estimating the amount of runoff in watersheds is the basis for designing the dimensions of the openings of bridges and other water facilities and is one of the most effective ways to reduce flood damage. This issue has long been the concern of water and river engineering specialists. Determining the amount of runoff is important from the point of view that in some countries of the world, including our country, due to the lack of information about the condition of rivers and the lack of data related to the amount of flow, figures or quantities that represent the amount of monthly runoff or It is not possible to have a river every year, and because the importance of any plan and project that directly or indirectly depends on the amount of water in that area is proportional to the amount of runoff in the watershed, so it is necessary to search for ways to get an estimate of the amount of runoff. Since many watersheds in the country do not have hydrometric stations, therefore, in order to estimate the height of the runoff, it is necessary to use experimental methods so that their results can be used in the management of watersheds. The use of experimental methods to estimate surface runoff in watersheds of arid and semi-arid regions, which mostly lack hydrometric stations, has long been recommended in hydrological studies, although the efficiency and appropriateness of these methods in different geographical areas depends on obtaining the required information and relevant tests. One of the new methods in many hydrology studies, especially in determining the amount of runoff, is the use of artificial neural networks. Currently, neural networks are widely used in various water engineering sciences, including hydrology and meteorology, hydraulics, hydrodynamics of rivers and seas, river engineering, sedimentation and erosion, groundwater, water quality and pollution, flood trends, etc. has Although artificial neural networks are not comparable to the natural nervous system, they have features that distinguish them in some applications such as pattern separation or wherever learning with a linear or non-linear mapping is required. These networks are a very simple imitation of the biological nervous system and the human brain. This imitation is based on a mathematical configuration that consists of several layers and also several neurons (nodes) in each layer. What adds to the importance of the subject is the existence of non-linear relationships between factors affecting different hydrological phenomena and the artificial neural network model has the necessary ability to adapt itself to these non-linear relationships. The idea of ??using this model is that all the important information is hidden inside the data and through this method one can find out the hidden relationships between the data. The purpose of this research is to determine the effectiveness of artificial neural network in estimating runoff in some watersheds of Fars province.