Author(s): G. Kopsiaftis; E. Protopapadakis; N. Doulamis; A. Mantoglou; D. Kalogeras

Linked Author(s):

Keywords: Seawater intrusion; Machine learning; Metamodels; Variable density models; Surrogate models

Abstract: The ability of machine learning techniques to approximate computationally intensive simulations of seawater intrusion and specifically variable density (VD) models is examined in this paper. The concept of utilizing computationally efficient surrogates of high-fidelity models has become very popular in water resources sciences. Surrogate models or metamodels have been also widely used in the field of Seawater Intrusion (SI), mostly coupled with global optimization routines. In this paper a comparative analysis of seawater intrusion (SI) surrogate models is performed, considering four multivariate machine learning algorithms: 1) Artificial Neural Networks (ANNs), 2) Multivariate Linear Regression (MLR), 3) Harmonic Function (HF) and 4) Anchor Graph (AG) and three univariate methods: 1) Linear Regression (LG), 2) Tree Regression (TR), 3) Ensemble Tree Regression (ETR) and 4) Support Vector Regression (SVR). The objective of this work is to test the examined metamodels with regards to their efficiency and accuracy, and determine which model is more preferable for a simulation-optimization framework.

DOI: https://doi.org/10.3850/978-981-11-2731-1_336-cd

Year: 2018