Application of metaheuristic algorithms in prediction of earthquake peak ground acceleration

The seismic resilience of a structure has been evaluated using peak ground acceleration (PGA). Ground motion parameters such as source characteristics, local site conditions are used to forecast the PGA of the ground motion. This paper aims to develop an Artificial Neural Network (ANN) based model to predict the PGA. Here, hypocentral distance (Rhypo${R}_{hypo}$), shear wave velocity (Vs30${V}_{s30}$), and moment magnitude (Mw${M}_w$), are used as input parameters. The model uses 12,706 ground motion recordings from 283 earthquakes from the revised NGA-West2 database supplied by Pacific Engineering Research Centre. Among the whole data, 70% of the data is set for training, 15% for validation, and 15% for testing the network. The R value derived from the testing dataset is 0.952, indicating the excellent performance of a network. An extensive parametric study is conducted with the PGA values, and the results indicate that the PGA increases with the magnitude and decreases with the hypocentral distance. The predicted PGA values from the present study are comparable with those from the existing relationships in the global database. The generated ANN model is further verified by comparing the predicted and recorded PGA values of an actual recorded event.