Concentration Field Estimation on Effluent Mixing and Transport With a Parameter‐Based Field Reconstruction Convolutional Neural Network and Random Forests

Abstract

AbstractEstimating the concentration field of effluent mixing and transport is a challenging task because of the complex mechanisms and three‐dimensional (3D) variability, but it is of great importance for water quality assessment and water resources management. Sophisticated numerical models based on the 3D computational fluid dynamics (CFD) technique can provide accurate predictions, but they are computationally expensive and require high‐level 3D CFD expertise, impeding their widespread usage. This work primarily develops a Parameter‐based Field Reconstruction convolutional neural network (PFR‐CNN) that can predict the complete concentration field of a target variable using the main influencing parameters, and explores the possibility of employing the proposed network to model the process of effluent mixing and transport. A validated numerical solver, TwoLiquidMixingFoam, was utilized to provide extensive data for multiple vertical buoyant effluents. A PFR‐CNN was developed to estimate the concentration fields based on the densimetric Froude number (F) and normalized port spacing (S/D). The network was developed using the training‐validating dataset from the numerical dataset, and further assessed using additional testing data. A random forest model was also developed, and the results showed that both methods can provide good performance, but the PFR‐CNN method was more robust and required less disk storage space. The results of our study demonstrated that the proposed PFR‐CNN approach is a promising tool for estimating effluent mixing and transport, and can be potentially leveraged to estimate other variables in the field of water resources with the governing parameters.