A Simulation of Shoreline Evolution with a Groin Structure Using an Alternative Machine Learning Algorithm

Abstract

Uneven sediment transport is a major cause of coastal erosion. Using groin structures is one method to help slow the outflow of sediment from the shoreline. Studying coastal behavior and forecasting future shoreline changes are crucial for managing and assessing the viability of remediation strategies. This research presents simulations of shoreline evolution with a single groin structure using two different methods, such as mathematical modeling and an alternative machine learning. A mathematical model is a representation of a real-world shoreline evolution phenomenon using partial differential equations. A machine learning algorithm is designed to learn patterns and relationships directly from real data. In this research, an alternative machine learning algorithm is designed to learn patterns and relationships directly from mathematical simulation data and let the machine make a decision in a situation that it has never learned before. For mathematical modeling, we introduced a one-dimensional model to predict the shoreline evolution. The initial and the boundary conditions with related parameter settings are introduced. The Saulyev finite difference method is used to obtain the approximated solution. An alternative machine learning algorithm for unexpected shoreline evolution prediction is also proposed. For alternative machine learning simulations, we identified six suitable features for the training dataset and developed an alternative K-nearest neighbor algorithm. It provides a way of predicting the evolution of the shoreline with a single groin structure. Additionally, an exact solution in an ideal scenario is used to test the precision of the simulation as well. The results show that the Saulyev technique outperforms an alternative K-nearest neighbor algorithm due to the lower root mean square error value. Both results of them are closed together. According to the research, mathematical modeling outperforms the KNN regression technique in terms of computational effectiveness during time periods of 0.5, 1, 5, 10, 15, and 20 years. Based on the modeling configuration and parameter simplicity, the KNN algorithm is still a good option for non-expert users.