Prediksi Respons Struktur Bangunan Berdasarkan Spektra Gempa Indonesia Di Pulau Sumatera Menggunakan Jaringan Saraf Tiruan

Abstract

Sumatera Island is one of the most active seismic area in Indonesia. The high seismic acvitity combined with mostly soft soil condition turn some location in this island into a devastating shaking area especially for highrise building. If not well designed, highrise reinforced concrete (RC) building may experience excessive deformation and endanger its occupants.
Due to these facts, this study aims to predict deformation characteristics of highrise RC building under earthquake loading using artificial neural network (ANN). Prior to ANN
analysis, modal response spectrum analysis is conducted to produce building response data sets. By selecting 8 capital cities and 3 other cities in Sumatera Island as seismic location,
1080 data sets are generated for ANN training and 405 data sets for testing. The ANN analysis uses 3 layers: input layer, hidden layer, and output layer. Building geometry, soil
condition, and seismic load are selected as input parameters, while story-drift, velocity, and acceleration are selected as output parameters. After 6000 iterations at training process,
average mean-squared errors (MSE) of 3x10-4 and 4x10-4 are achieved for training and testing process, respectively. The calculated R2 is ranging from 83% to 95% which is
adequately high for prediction rate. This shows that ANN is a very promising tool to predict story-drift, velocity, and acceleration of highrise RC building under earthquake loading.
Keywords: artificial neural network (ANN), Sumatera Island, modal response spectrum analysis, structural response