SVR-Optimized ANN Model for Predicting Earthquake Risk in Electrical Substations Based on Disaster Datasets in the Aceh Region, Indonesia

This paper proposes a predictive model based on an Artificial Neural Network (ANN) optimized with Support Vector Regression (SVR) to predict seismic parameters as part of earthquake risk mitigation for substations. The proposed model aims to improve the accuracy of seismic parameter predictions, including seismic energy, Peak Ground Acceleration (PGA), and Modified Mercalli Intensity (MMI). A case study was conducted in the Aceh region, Indonesia, which is located near the subduction zone between the Indo-Australian Plate and the Eurasian Plate, making it highly seismically active. Therefore, this model is developed to identify substations most at risk of earthquake impacts while supporting disaster mitigation planning through early warning systems. The dataset used in the modeling includes information on the epicenter and substation coordinates, earthquake magnitude, depth, and the distance between the two points. The ANN training process uses backpropagation with sigmoid and tanh activation functions. The ANN outputs are then used as input features for the SVR model, applying a Radial Basis Function (RBF) kernel to capture the non-linear relationships between variables. Evaluation results show that the hybrid SVR-ANN model significantly improves prediction accuracy compared to conventional ANN. This improvement is indicated by a reduction in error and an increase in the coefficient of determination ( $R^{2}$ ). The findings of this paper can serve as a foundation for developing more effective mitigation systems to reduce earthquake impacts on electrical infrastructure.