Abstract
River water flow forecast in general and particularly in floods is of worth importance for monitoring operations of floods in canals and rivers. Floods in rivers bring destructions to road, houses, crops and causes human dislocation. The River Chenab is one of the largest rivers in Pakistan and has a historical recording of heavy floods. Prior to heavy floods, in time warning is mandatory to save lives and property. Accordingly, this paper presents an intelligent model to predict an advance alarming water flow from Chenab River. Standard learning algorithm is applied to train the ANN for this task. Inputs to the neural network are taken from the daily discharge values and the output layer composed of four neurons to represent number of predicted days. Moreover, trial and error approach is adopted to select appropriate number of inputs for time-series data. Two different architecture (single and double hidden layers) of neural network are evaluated and compared to find the most suitable one. Additionally, two activation functions are tested. The results thus achieved reveal well in time warning to the surroundings to secure flood victims. However, during low discharge, neural network miscalculated.