Modeling peak ground acceleration for earthquake hazard safety evaluation

This paper presents a ground motion prediction (GMP) model using an artificial neural network (ANN) for shallow earthquakes, aimed at improving earthquake hazard safety evaluation. The proposed model leverages essential input variables such as moment magnitude, fault type, epicentral distance, and soil type, with the output variable being peak ground acceleration (PGA) at 5% damping. To develop this model, 885 data pairs were obtained from the Pacific Engineering Research Center, providing a robust dataset for training and validation. The ANN architecture comprises 4 nodes in the input layer, two hidden layers each containing 25 nodes, and a single-node output layer, resulting in 750 unknown weight and bias values that the model must optimize. Following the model assessment, a genetic algorithm (GA) was integrated with the ANN model to enhance its predictive capabilities. This integration aimed to forecast 20 potential earthquake scenarios, a crucial step in validating the model's effectiveness. The results were promising, as the ANN-GA successfully predicted earthquake occurrences in 15 out of 20 scenarios. These findings underscore the model's potential in accurately forecasting seismic events, thereby contributing to the development of more resilient infrastructure and better-informed urban planning strategies.