Essay Sample on Daily Forecasting of Dam Water Levels Using Machine Learning

Introduction

Dam reservoirs are essential components of water resource management systems. Mainly, reservoirs assist in providing and regulating adequate water in irrigation, flood control, generation of hydropower as well as providing other related hydrological functions in everyday life. Awan et al. (2014) asserts that effective operation and management of reservoirs is vital by creating an appropriate schedule for water release policy for effective planning of water management. It is important to predict the level of water in the dam for effective water resource management. Shafaei et al. (2016) states that various hydrological phenomena's including rainfall, river flaws from neighboring basins and evaporation from lakes and oceans, atmospheric temperature and the relationship between the lake and aquifers result to fluctuation of water level in lakes.

Is your time best spent reading someone else’s essay? Get a 100% original essay FROM A CERTIFIED WRITER!

Save 25%

Water is considered a basic requirement that significantly impacts on the economy and lives of individuals. Water resource management is vital in the society in various angles including protection of available water, preserve the available water resource existing and control the risk of flooding. Dams have an fundamental role in water resource management, primarily, forecasting of water short and long term inflows are detrimental in the operation of a dam. Short term forecasts are essential because they provide real time information to prevent flooding and make sure that there is a consistent water supply and required dam levels and appropriately maintained. Long-term forecast is critical to adapt dam water management for use in agriculture, domestic as well as hydropower generation (Awan et al. 2014).

Today, dam inflow prediction is a complex phenomenon because of fluctuating weather patterns and high unpredictability of rainfall. However, efforts to establish the factors that cause uncertain change in weather and climate patterns. Prediction of long-term rainfall is still challenging despite the creation of modern high-resolution climate and weather prediction instruments. Owing to the challenges of uncertainty in quantitative prediction, majority of weather forecasting agencies primarily depend on the qualitative forecasting that are mainly probabilistic estimation of different normal categories such as below, normal or above normal. Awan et al. (2014) asserts that while qualitative forecasting has significantly reduced the doubt compared to quantitative predictions, suitable methods and models are required to interpret and incorporate the various categories to achieve the desired prediction and increase accuracy.

The inflow patterns of dams primarily involve multifaceted processes that are described using the simple predictive models for the reason of nonlinearity, spatial distribution and the fluctuation of data (Valizadeh et al. 2017). In regions that lack fresh water, intense water managing problems intensifies in the various hydrological instances such as heavy rains or famine. A lengthy drought season significantly impacts on the reduction of water levels in reservoirs that directly has an impact on various sectors of a country's economy such as power and agriculture. A consistent level of reservoir water level is vital to provide sufficient water supply and maintained desired marine environment in various rivers. Seckin et al. (2013) states that overflow is a substantial factor that impacts dams, more so ungagged reservoirs. Precise and consistent forecast of inflow level is essential in decision making to reduce adverse impacts of water deficit and surplus. Moreover, establishing the most suitable weather forecasting model is vital in making appropriate decisions and planning and developing effective successful reservoir policies. Cpredictive models are proposed using the time series regression model and the assistance vector machine model for the water level forecast. Several input modelling problems will be addressed. The most essential variables are the daily prediction and water level data of the reservoir. The proposed time series forecast model with or without selection has a better forcused performance than the support vector machine that utilizes the four rating indices. Contextually, creating a suitable model for inflow prediction is of specific interest in the functioning areas of water resource management (Lohani et al. 2012; Allawi et al. 2018).

Hipni et al. (2013) states that forecasting models are critical amongst hydrological communities because they impact in various essential ways. the primary aim of identifying the appropriate forecasting model is to develop effective use of scares fresh water. Reservoir operations operate effectively forecasting of inflows is precise to control water supply, flood prevention, hydropower generation as well as decision support systems. Recently, artificial intelligence (AI) techniques are developing for modeling no-linear hydrologic systems. Particularly, the Adaptive Neuro-Fuzzy Inference System (ANFIS) and Artificial Neural Networks (ANNs) are utilized for modern and multifaceted hydrologic systems. Artificial Neural Networks (ANNs) systems are similar computational models that work comparable to the biological neural network, therefore, they have advanced level of generalization abilities (Seo et al. 2015). Considerable growth in technology in advancement in forecasting models has significantly benefited the weather forecasters, particularly Artificial Neural Network (ANN) modelling is widely used. The ANN model can be run without necessarily knowledge of computational relationships between input and outputs or the explicit categorization of physical characteristics and circumstances (Najah et al. 2013).

Kisi et al. (2012) stated that application of AI methods in forecasting has widely been accepted as a suitable tool for modeling complex nonlinear circumstances in water resource systems that results to broadening of its use. Artificial intelligent (AI) methods such as the Adaptive Neuro-Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN) are essentially new methods and the emerging of Linear Genetic Programming (LGP) is crucial backup that assist to overcome the weaknesses of conventional modelling. Therefore, data-driven methods are mainly based on data mining and re-use of data confined in various hydrological time-series without considering the physical laws that are critical part of the process. During the past years, Artificial Neural Network (ANN) has been successfully applied in modelling various range of hydrological processes because of its ability to excellently model non-linear systems. Though it does not have the various attractive features, ANN has disadvantages such as difficulty in selecting the best appropriate training algorithm and time consuming aspect involved in developing its complex structure. Nonetheless, Jothiprakash et al. (2012) affirms that there are still improvements to advance the ANN network and training algorithm to enhance particular parameters of the network.

In this context, creation of various models for actual prediction of dam water levels is a vital research area that is constantly evolving. The development of improved techniques and efficient forecasting models is a recipe for the improvement and modification of flood management techniques as well as flood protection.

Scope of the Study

To accomplish the research objectives, the following activities will be done:

• A detailed literature review of techniques of inflow forecasting and its use in flood warning and prevention as well as dam operations.
• Review of related documents relevant to the research
• Gathering of measured meteorological statistics, quality control and data analysis.
• Obtaining of meteorological forecast including temperature and rainfall data, downscaling when required as well as conducting bias alterations.
• Apply reservoir water level prediction regarding dam operation as flood warning.
• Validate model and prediction simulation on the basis of forecasted inflow of data concerning reservoir water level.

2.0 Literature Review

2.1 Exponential Smoothing Methods

In current days, the artificial intelligence methods like the adaptive neuro-fuzzy inference system (ANFIS), support vector regression (SVR) have been used as a method to predict difficult non-linear sequences in hydrology occurrences (Shafaei et al. 2016). Artificial intelligence (AI) methods gained significance in hydrological modelling to enhance their capability in learning unseen patterns from historical information and prediction of extremely non-linear systems. Adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN) are usually used AI methods that have been applicable in various domains. The ANNs are parallel computational simulations that are similar to biological neutral network and have better overview abilities (Awan et al. 2014).

Various methods such as lumped, physical based, and autoregressive moving average (ARMA) for in-flow predicting have been recently suggested for transferring function models and numerous recursive approximation methods like Kalman' filtering and ANN models. Introduction of the moving average models was in 1938 as a type of time sequence model that introduced the ARMA and ARIMA models. The various types of moving average techniques include double moving models, weighted moving averages, and simple moving averages; furthermore, moving average models are commonly applied uniquely. Autoregressive moving average (ARMA), which is among the common models that are founded on times sequence analysis, are models founded on the moving average model and blend of autoregressive model to improve effectiveness and accuracy, in contradictory to moving average and autoregressive models. Moreover, the elasticity has been considerably enhanced. The ANN model is an organized group of interconnected artificial neurons that was first introduced by McCullouch and Pitts (1943) in basic. There is more accurate performance of artificial neural network compared to other time series techniques, like MA techniques in kinds of rainfall-runoff subgroups, such as stream flows. Besides, Artificial Neural Networks are more time important with improved response to resilient record tolerance in data groups (Valizadeh et al. 2011)

Regression-based statistical models are usually applied in medium and long-term prediction. The statistical models include autoregressive moving average with exogenous (ARMAX), autoregressive integrated moving average (ARIMA), and autoregressive moving average (ARMA) are founded on the notion of stationarity and linearity that inhibit their capability of resolving non-linear linkage of dam in-flow and related predictor variables (Othman et al. 2011)

Neural networks being elastic mathematical simulations are applied in the model of the river flow. Predicting the river flow in to a reservoir is essential in planning and managing the water resources. Past familiarity of the flooding can be useful in routing the flood safely over the reservoir. The technique leads to reduction of the danger of flood destruction at the downstream area thereby ensuring the safety of the dam. Consequently, the members of water resources recognize and emphasize the necessity of punctual and accurate streamflow forecasting. Real time predictions of natural in-flows to reservoirs are of specific awareness for operation and preparation. Various techniques have been suggested for the real time forecasting which include conceptual (physical) and empirical simulations, though, these methods are not regarded as sole superior simulations. Regarding the problems associated with formulation of rational non-linear watershed simulations, latest efforts ha...