A Novel Deep Learning-based Prediction Approach for Groundwater Salinity Assessment of Urban Areas
2023
Abbasimaedeh, Pouyan | Ferdosian, Nasim
The high amount of Electrical Conductivity (EC) in the groundwater is one of the major negative Geo-environmental problems which has a considerable effect on the quality of drinking water. To address this challenging problem we proposed an intelligent Machine Learning (ML) based approach to predict EC in urban areas. We applied the deep learning technique as one of the most applicable ML techniques with high capabilities for intelligent predictions. Five different deep neural networks (Net 1 to Net 5) were developed in this study and their reliability to predict EC with an emphasis on different settings of inputs, features, functions, and the number of hidden layers was evaluated. The achieved results showed that deep neural networks can predict EC parameters using minimum and economic input parameters. Results showed parameters Cl and SO4 with a high range of correlation and pH with a low range of Pearson correlation properties are influential parameters to be used as the input of neural networks. Activation function Relu, optimization function Adam with a learning rate of 0.0005 and loss function Mean Squared Error with the minimum of two hidden dense layers from Keras laboratory of Tensor Flow developed an efficient and fast network to predict the EC parameter in urban areas. Maximum epochs for developed networks were defined up to 2000 iterations while epochs are reducible up to 200 to drive minimum loss function outcome. The maximum training and testing R2 for developed networks was 0.99 in both the training and testing parts.
显示更多 [+] 显示较少 [-]