Evaluasi Potensi Jaringan Saraf Tiruan Dalam Memprediksi Respons Struktur Gedung Bertingkat Berdasarkan Spektra Gempa Di Wilayah Indonesia

Abstract

Design of quake resistance building is very important in Indonesia, where most of Indonesia’s region located in quake zone with low to high intensity. Analysis of structure response can be done by using the help of finite element software, but needs time and is
complicated when designing takes place. One of the solution in analyzing structure response faster and easy to operate is with using Artificial Neural Network (ANN). Deciding transfer
function in ANN analysis is important as it can ease our ANN analysis based on displacement, velocity, and acceleration as input data. This research is intended to evaluate structure
response based on variation of quake load based on location of Indonesia’ capital of province, which are 34 provinces. With the help of finite element software analysis, total data
obtained for JST is 1836 data. In this research, the composition used for training, testing, and validation for whole data is 60%, 20%, and 20% respectively. Input parameters are
quake data, soil type, and elevation, meanwhile for output is structure response for each direction. Testing result showed that the accuracy of ANN in predicting structure response
reach 99%. Evaluation result of ANN analysis showed that transfer function of Tan-Sigmoid and Purelin gives good prediction with Tan-Sigmoid as the best one with 99% accuracy.
ANN analysis result showed that irregular buildings have 99% accuracy for training and testing, meanwhile regular buildings has about 90% for training and testing. This showed
that ANN with function transfer Tan-Sigmoid can be used for predicting structure response fast and accurately, as well can be used as reference for designer in designing quake resistance building.
Key Words: Artificial Neural Network, structure response value, finite element software, response spectrum, Tan-Sigmoid