APPLICATION OF DEEP NEURAL NETWORK (DNN) TO GEOELECTRIC DATA IN WAYANG WINDU GEOTHERMAL FIELD

• Main Author: Togap Zisochi Lase, Fanzly
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78664
• Institution: Institut Teknologi Bandung

Inversion in geophysics is trying to estimate the distribution of physical properties in the Earth's interior from collected observational data or from above the surface. Some conventional inversion techniques proposed in developing problem solutions have mainly used least-square, simulated annealing, Levenberg-Marquardt, and so on. However, the inversion methods still have great potential in uncertainty quantification and make the computational task heavy. In this paper, a method is proposed using Deep Neural Network (DNN) for geoelectric data to provide effective and accurate solutions. The purpose of this study is to apply inversion using DNN to geoelectric data and compare it with conventional inversion, namely the Levenberg-Marquardt (LM) algorithm. The Wayang Windu geothermal field is designed by combining the components of the geothermal system into a synthetic model and field measurements that have been made. Processing of models and field measurements is done in the Python programming language, where conventional inversion uses the Levenberg-Marquardt algorithm and DNN uses six hidden layers with activation functions such as sigmoid and ReLU. The results obtained from conventional inversion in the first, second, fifth, and sixth layers do not show much difference with the measurement data, but in the third and fourth layers there are significant thickness differences. However, the products of DNN in estimating the resistivity and thickness in the synthetic model are similar to what is being predicted. The output of conventional and DNN inversion was also performed on the Wayang Windu geothermal field measurement data, where DNN in predicting thickness has a much smaller error than conventional inversion. DNN predictions are compared with borehole data) from the Wayang Windu geothermal field, showing precise thickness and resistivity-type predictions so that geoelectric interpretation can be done effectively and accurately.