Data Based Modeling of a Large Scale Water Delivery System

Water is a vital resource and the growing populations and economies around the globe are pushing its demand worldwide. Therefore, the water conveyance operation should be well managed and improved. This paper proposes the development of reliable models able to predict water levels of a real 24.4 km water delivery channel in real time. This is a difficult task because this is a time-delayed dynamical system distributed over a long distance with nonlinear characteristics and external perturbations. Artificial neural networks are used, which are a well-known modeling technique that has been applied to complex and nonlinear systems. Real data is used for the design and validation of the models. The model obtained has the ability to predict water levels along the channel with minimum error, which can result in significant reduction of wasted water when implementing an automatic controller.