Water level prediction with artificial neural network on graphical processing unit

Abstract

Water is the most instinctive natural resource available on Earth. To evade the dearth of water, hydrological researchers have to take the crucial step to predict water level and take the inevitable process to block the forthcoming situation of the water crisis. Different hydrological studies have shown the impact of Artificial Neural Networks (ANN) as an emerging field for indicating the level of Groundwater in advance. An unpredictable and inconsistent aspect arises, due to expanded feature and faulty data set for prediction. Stringent action must be taken for Groundwater management to avoid a water crisis. Recent work of researchers shows the effectiveness of Artificial Neural Networks (ANN) in three different areas; it is capable to handle extremely large data sets, immense computing problems, and eminence of training at the discrete level of representation or illustration. Stimulation of deep learning desires to have strong and accelerated hardware. Faulty and erroneous data set of input variable will not adept at accurate prediction with a single-core processor that is Central Processing Unit (CPU), whereas prediction on a multi-core processor makes it more efficient. A Graphical Processing Unit (GPU) is a multi-processor system having thousands of computing units known as cores. The main purpose of this research is to show up with an improvised and parallelized form of the Back Propagation Network (BPN) algorithm on an Artificial Neural Network (ANN) with GPU. The proposed model used, predicts the groundwater level of the district Faridabad, Haryana; India. Parameters that affect the groundwater level for prediction have been identified such as Temperature, Rainfall, and Water level. Twenty years of data from 2001 to 2020 have been chosen for simulating the algorithm. A platform by NVIDIA i.e. Computer Unified Device Architecture (CUDA) has been considered for the implementation of Parallelized Back Propagation Network (PBPN). Thamar Abdul Rahman , 2016 predicted the groundwater level using Geographic Info