Dam break induced flood analysis by soft computing techniques

Abstract

Water structures have been one of the basic components that have served the various needs of societies since ancient times. Fresh water resources, which are examples of these structures and have a great impact on the residents in vicinity, are dams. Dams have a variety of purposes such as fresh water supply, waste storage, agricultural irrigation, electricity generation and flood protection. The management of massive structures such as dams is also a critical issue affecting the well-being of societies. The proper management of dams is important not only for the efficient, but also for the safe operation of the water resource. Otherwise, the dams can collapse, as in many examples in the past, causing enormous material and moral damages and losses. The failure of dams can be caused not only by their proper operation, but also by many various reasons such as earthquakes, extreme precipitations, sabotage and aging. Since dams hold large bodies of water, they cause many life losses and material damage in case of their collapse. History has shown this to humanity in a painful way. It is essential that the dams are properly designed and operated in order to prevent them from collapsing. Since these structures are man-made, they can still be demolished, so preparations for this scenario should be done and ready within the scope of risk and disaster management. It is a clear reason that dam operations will become more difficult in rapidly changing climatic conditions. The aging of dams in the coming years is an additional problem to these difficulties. These reasons assurance that dam failure problems will be a hot topic to work on in the years to come. Unfortunately, dam failure problems are inherently complex. The cause of dam failures may not be known because they are very large engineering structures. There can be many different reasons behind a dam collapse. The dam failure process is also difficult to analyze as it involves many variables. Researchers have proposed many different methods and approaches to illuminate this process. As a result of dam failures, a tsunami wave may occur at the downstream, followed by flooding and rising water level over time. In recent years, problems with complex definition and many variables have been analyzed with soft computing methods. Soft computing methods are a very broad field that includes many algorithms and are frequently used in many different branches of science. Their success, especially in the analysis of major nature problems, has been proven by researchers with various studies. Grey relational analysis (GRA), long-short term memory (LSTM) and artificial neural fuzzy inference system (ANFIS), which are used in this study, can be given as example algorithms of soft computing methods. It has been intended to provide a data set to researchers and local governments and aimed to contribute to risk and disaster management in water resources by presenting numerical simulations of some large and aging dams in Turkey and the possible loss of life they will cause. In this context, GRA, LSTM and ANFIS models from soft computing methods were developed and tested for the analysis of dam failure problems and dam break induced flood. In summary, the main purpose of this study is to prove the usability of soft computing techniques in dam break problems and to conduct risk analysis of some aging dams in Turkey within this framework. In the third section titled "Danger level ranking of possible dam failures in Turkey", the 15 aging dams in Turkey which have potential to cause the most dangerous consequences in case of failure were selected intuitively from engineering judgement in order to perform their failure simulation. The selected 15 aging dams are as follows. Seyhan Dam and Hydroelectric Power Plant (HPP), Borçka Dam and HPP, Ürkmez Dam, Eğrekkaya Dam, Tahtalı Dam, Mamasın Dam, Dim Dam and HPP, Kurtboğazı Dam, Atasu Dam, Alibey Dam, Akköprü Dam and HPP, Suat Uğurlu Dam and HPP, Derbent Dam and HPP, Manavgat Dam and HPP, Kirazlıköprü Dam. The failure simulations of the dams were carried out with the worst-case scenario, the sudden collapse failure mode. The simulations performed were made through the HEC-RAS application and the simulation stages were shared step by step. Maximum water depth maps obtained according to the simulation were given for the impact zone of each dam. Using first wave arrival time and affected population values obtained from the simulastions, possible life losses were calculated with the equations of DeKay and Mclelland (1993) and Brown and Graham (1988) and shared for each dam. Accordingly, the collapse of Seyhan Dam and HPP causes the higest life losses among others. The 15 dams were ranked according to the probable loss of life obtained from the DeKay and Mclelland (1993) equation in case of their failure. The GRA model has been developed for the danger level ranking analysis of dams in order to be a more practical solution, since the danger level ranking of dams by numerical analysis is a labor-intensive task that takes a lot of time. The effective attributes of GRA model were chosen as follows. Surrounding population, distance from that population, elevation relative to that population and reservoir size. These selected attributes have been decided from an engineering point of view. The quantitative values of the attributes were determined by engineering evaluation. The output of the model is to rank the dams according to the possible loss of life that may occur in case of failure. According to results produced by the GRA model developed in this frame, the dam that can cause the highest life losses in case of collapse is Adana Dam and HPP. The rankings produced by the GRA model and by the numerical analysis were compared and interpreted. The results indicate that the GRA model and numerical analysis produce similar ordering. Thus, it has been revealed that the GRA method is an effective and useful tool and can produce practical solutions for risk and disaster management studies on water structures. In addition, simulations of failure of 15 aging dams in Turkey and their possible consequences were brought to the literature. The implications of this section will contribute to the risk and disaster management in water structures. In the fourth section titled "Prediction of dam break induced flood parameters by LSTM network and ANFIS", the analysis of the most vital parameters that determine the lethality level of flood disasters caused by dam failure was carried out with LSTM and ANFIS from soft computing methods. Past events don't provide a very adequate data set, as well as experimental studies also remain very fictional. In real situations, every flood event is unique, which has different forces that drive and effect damage level. Therefore, it is necessary to examine such flood events on a case-by-case basis and to carry out special studies for each region. Performing flood simulations of regions with numerical analysis is the most satisfactory method currently available. Data obtained from simulations of sample regions by numerical analysis were selected for feeding and testing LSTM and ANFIS models developed under this title. The main parameters affecting the lethality level of the flood are water depth, flow velocity, wave damping and first wave arrival time. The data set of these parameters was created by simulating the Alibey Dam break and Froehlich (2008) approach from the breach development methods accepted in the literature defined in the HEC-RAS application was chosen as its breach development method. The information about the measurement locations determined in the impact area of Alibey Dam and the data obtained from these points were given in the relevant section. Since flood parameters such as water depth, flow velocity and wave damping are data sets consisting of time and location series, it was preferred to create LSTM models in the analysis of these parameters. On other hand, ANFIS model was found more convenient for the analysis of the first wave arrival time because it was deemed more appropriate to create a multivariate model. Due to the suitability of the data sets of water depth, flow velocity and wave damping, nonlinear autoregressive LSTM models containing the same following structure were created for the analysis of water depth, flow velocity and wave damping parameters, which does not need any external input and takes the previous output as input. The training, validation and testing ratios were chosen as 20%, 5% and 75%, respectively. The training function was selected as bayesian regularization backpropagation and the dataset dividing function was designed as divedeblock so called considering it is prominent to test the last part of the data. In this case, the last 75% of the data set was exposed testing. Coefficient of correlation and root mean squared error values expressing the success of the LSTM models developed for the analysis of water depth in the rezervoir, water depth in downstream, flow velocity and wave damping parameters in the test phase are 0.9988 and 0.5798, 1.0000 and 0.0192, 0.9995 and 0.0018, 0.9986 and 0.1804, respectively. The inputs of the ANFIS model predictor of first wave arrival time were selected as pool elevation at failure, final breach bottom elevation and distance from the dam. The training/testing ratio was determined as 75/25%. The dataset contains 140 samples which was divided as 105 samples for training phase and 35 samples for testing phase. Coefficient of correlation and root mean squared error values expressing the success of the ANFIS model developed for the analysis of first wave arrival time parameter in the test phase are 0.9828 and 3.6216, respectively. The statistical parameters prove that the analysis success of the models are robust. The success of soft computing methods in the analysis of dam failure problem has been shown to be a useful tool in the management of real disaster events. To put it more clearly, some applications that contain algorithms working in real time can be established for such possible disasters. These applications can reduce the extent of the disaster by using them as practical solution tools in such disaster times. It is possible to learn the failure mode of any dam and the mechanism of breach development as soon as the event occurs. However, this is too late for any flood parameters to be analysed. In such events, there are only minutes for the right actions to be taken. For example, when the Manavgat Dam collapses for some reason, how long the first wave will arrive at any location is important for evacuation, or knowing the possible maximum water elavation at any point means the elevation at which people must climb. In such limited times, it takes time to perform numerical analysis according to the emerging flood hydrograph, but algorithms trained before the event according to various scenarios can provide practical solutions at the time of the event and provide information to the relevant people.