An Artificial Intelligence-Based Prediction Model for Optimum Design Variables of Reinforced Concrete Retaining Walls

Abstract

In this study, a hybrid method was developed to predict optimum dimensions of reinforced concrete (RC) cantilever-type retaining walls. The metaheuristic-based optimization process of RC retaining walls needs iterative analysis, including the consideration of two different design limit states. These limit states defined as design constraints are geotechnical and structural limit states. Since the optimization process is long, it is aimed to generate a prediction model for optimum dimensions of RC cantilever retaining walls. For this purpose, artificial intelligence and machine-learning methods can be combined with metaheuristic algorithms used in the optimization of the problem. The method uses artificial neural networks (ANNs) for prediction results without iterative process and flower pollination algorithm (FPA) to obtain training data for machine learning. The proposed hybrid model is called flower pollination algorithm-based artificial neural network (FPA-ANN). To verify the prediction model, mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE) values are calculated for the optimum data set. Because this is the first proposed prediction model for a real structural engineering application related to RC cantilever-type retaining walls, the model was tested on different input values that are not included in the machine-learning process. Also, the relations between height, unit soil weight, internal friction angle, and surcharge load are investigated via the prediction model.