MODEL OPTIMIZE FOR DEEP LEARNING ALGORITHM IN DETERMINE THE ARRIVAL TIME OF P AND S WAVES IN MICROSEISMIC DATA OF

• Main Author: Rahadi Pangestu, Rayden
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87880
• Institution: Institut Teknologi Bandung

In recent years, the amount of data generated by seismograms has been increasing, especially in geothermal areas. This poses a big challenge in seismic data processing, as the amount of data that must be analysed is not proportional to the human resources available to do it manually. Therefore, algorithms or tools are needed that are able to assist the data processing process efficiently and accurately. One algorithm that is widely used in seismic data analysis is PhaseNet, a deep learning algorithm designed to identify the arrival times of P and S waves in tectonic earthquakes. Although PhaseNet has proven effective in the context of tectonic earthquakes, its application to geothermal areas that have different seismic characteristics is often less than optimal in terms of prediction accuracy. This study aims to improve the performance of the PhaseNet algorithm in detecting wave arrival times in geothermal areas through training a model that is more suitable for the characteristics of microseismic data from geothermal fields. In this study, a comparison was made between three models: the PhaseNet model that has been retrained with microseismic data, the transfer learning model, and the initial PhaseNet model. Evaluation of the prediction results of each model was carried out by comparing them against manual catalogue picking data from the ‘RR’ geothermal field to obtain differences in prediction performance. Furthermore, an analysis using the confusion matrix concept was conducted to determine the best model based on evaluation metrics such as precision, recall, and F1 score. With this approach, it is hoped that this research can contribute to improving the accuracy of wave arrival time prediction in geothermal area seismic data, as well as reducing dependence on manual processes in seismic data analysis.