Forecasting the Change of Reservoir Water Level Stage Using Neural Network Nur Athirah Ashaary 1 , Wan Hussain Wan Ishak 2 and Ku Ruhana Ku-Mahamud 3 1 School of Computing, College Arts and Sciences, Universiti Utara Malaysia, athirah_2501@yahoo.com, 2 School of Computing, College Arts and Sciences, Universiti Utara Malaysia, hussain@uum.edu.my, 3 School of Computing, College Arts and Sciences, Universiti Utara Malaysia, ruhana@uum.edu.my, Abstract: The reservoir is one of the structural defenses that function to reduce the negative effect of flood and drought. The decision of water release is very critical in both situations where it is influenced by changes of reservoir water level. In this study, the potential of neural network model for forecasting the change of reservoir water level stage is studied. Six neural network models have been developed and tested. Sliding windows have been used to segment the data into various ranges. The finding shows that 2 days of delay have affected the change in stage of the reservoir water level. The finding was achieved through 4-17-1 neural network architecture. Keywords: Model Forecasting, Reservoir Water Level, Neural Network. 1. Introduction Reservoir system is one of the significant components and a part of water resource management. A reservoir is a natural or artificial lake or large tank, where it functions to impound and regulate the water for supplies and irrigation. Moreover, the reservoir can also act as a defense mechanism for flood and drought situations [15; 17; 33]. Reservoir can be classified into two types, which are single reservoir and multipurpose reservoir systems. A multireservoir system is more complex compared to a single reservoir system. The reservoir can also be classified based on its purposes and functions. A reservoir with one purpose or function is called a single purpose reservoir, while a reservoir with more than one purpose is a multipurpose reservoir. According to [16], a single purpose single unit reservoir is the simplest system and a multipurpose multiunit reservoir system is the most complex system. The simplest system is developed for the single purpose operation such as hydropower generation or flood protection. The most complex reservoir system is created for the multiple purpose operation such as flood protection, navigation, hydropower generation and recreation [20]. Reservoir system can be separated into four parts, which are upstream, reservoir, spillway gate and downstream. Upstream is the water source or inflow of the reservoir. The spillway gate is a structure which functions to control the release of the reservoir water. Some reservoirs are equipped with an ungated spillway and some others have both types of spillways. The gated spillway requires several numbers of decisions such as the number of gates to be opened, the duration and size of the opening. Flood is one of the natural disasters that could strike repeatedly. It can indirectly or directly cause extreme losses to the public such as properties, homes and innocent souls. Flood is directly associated with the reservoir as the latter is one of the flood mitigation mechanisms. According to [31], a flood situation usually occurs in the lowland areas which are much more fertile and crowded with human activities, especially in the agriculture sector. This area is typically the downstream area of the reservoir. It usually occurs in areas which are dry and obviously interferes and disrupts social activities and human. An early decision on the reservoir water release can help to reduce and prevent losses to the flood affected areas as the water flow can be controlled in accordance to the river capacity. Another natural disaster that is related with reservoirs is water shortage (drought). It cannot be viewed solely as a physical phenomenon, but the impacts on society and surrounding area must be considered. Drought is a critical situation causing more deaths compared to other natural disasters [28]. This paper discusses the findings of the forecasting model for the change of reservoir water level stage using neural network. The next section presents some related literature on Artificial Neural Network (ANN) and studies on forecasting of water level. This is followed by the methodology and findings of this study. 2. Literature Review Reservoir operating policy is important in reservoir operation as the impact of the reservoir operation of society and economy is huge [32]. Reservoir operators and planners are required to plan a strategy that can be used to determine the decision [1]. The operating policies for reservoirs are usually developed based upon previous meteorological and hydrological data [9]. In certain conditions, operating policies, also known as operating rules, are commonly used in the early or planning level of the proposed reservoir. Moreover, reservoir water release decisions are guided by the reservoir operating policy [2; 24]. In the reservoir operation, decision making is one of the vital procedures that needs to be implemented wisely in order to balance the demand and supply of water for optimal social, economic and environmental benefits. The problems in early decision making of reservoir water release usually occur in unpredicted weather conditions. Therefore, several decision making procedures such as simulation and optimization techniques have been developed to identify optimal operating rules in reservoir systems. As referred to [17], there are four optimization strategies for the multiple reservoir system that have multiple objectives, namely, implicit stochastic optimization, explicit stochastic The 2nd International Conference on Mathematical Sciences and Computer Engineering (ICMSCE 2015) 103