Prediksi Indeks K Sebagai Indikator Badai Geomagnetik di Indonesia Menggunakan Algoritma Long Short-Term Memory

Saputra, Aby Rengga (2026) Prediksi Indeks K Sebagai Indikator Badai Geomagnetik di Indonesia Menggunakan Algoritma Long Short-Term Memory. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025232006-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Badai geomagnetik merupakan fenomena cuaca antariksa yang berpotensi mendisrupsi infrastruktur teknologi kritis, termasuk sistem komunikasi dan navigasi di wilayah ekuator seperti Indonesia. Metode prediksi konvensional sering kali memiliki keterbatasan dalam adaptabilitas terhadap fluktuasi cepat, sementara model berbasis data yang ada mayoritas berfokus pada wilayah lintang tinggi. Penelitian ini bertujuan untuk mengembangkan model prediksi Indeks K lokal di wilayah ekuator Indonesia menggunakan pendekatan Pembelajaran Mendalam yang menggabungkan Long Short-Term Memory (LSTM) dan mekanisme Multi-head Attention (MHA). Penelitian ini mengevaluasi empat skema eksperimen yang membandingkan pendekatan univariate dan multivariate, dengan serta tanpa mekanisme atensi. Dataset yang digunakan mencakup parameter fisika angin matahari dan data geomagnetik lokal periode 2020–2025, yang merepresentasikan fase menaik (ascending phase) dari Siklus Matahari ke-25. Untuk memastikan konfigurasi model yang optimal dan objektif, teknik Automated Hyperparameter Optimization (HPO) diterapkan menggunakan kerangka kerja Optuna. Hasil eksperimen menunjukkan bahwa model Multivariate LSTM + Multi-head Attention merupakan arsitektur terbaik, mencapai nilai koefisien determinasi (R2) sebesar 0,5299 dan Root Mean Square Error (RMSE) 0,7040. Integrasi fitur multivariate terbukti krusial, mengonfirmasi bahwa dinamika geomagnetik ekuator sangat dipengaruhi oleh driver eksternal angin matahari. Mekanisme Attention berhasil meningkatkan stabilitas model dan sinkronisasi fasa, memungkinkan deteksi awal badai (storm onset) yang presisi tanpa jeda waktu (lag) yang signifikan. Selain unggul dalam akurasi, model ini juga terbukti efisien secara komputasi dengan waktu inferensi rata-rata 66 ms, menjadikannya layak untuk diimplementasikan sebagai sistem peringatan dini. Penelitian ini memberikan kontribusi kebaruan dalam pemodelan aktivitas geomagnetik spesifik di wilayah ekuator yang memiliki karakteristik unik dibandingkan lintang tinggi.
========================================================================================================================================
Geomagnetic storms are space weather phenomena capable of disrupting critical technological infrastructure, including communication and navigation systems, particularly in equatorial regions such as Indonesia. Conventional prediction methods often lack adaptability to rapid fluctuations, whereas existing data-driven models predominantly focus on high-latitude regions. This study aims to develop a prediction model for the local K-Index in the Indonesian equatorial region using a hybrid Deep Learning approach that combines Long Short-Term Memory (LSTM) and a Multi-head Attention (MHA) mechanism. This research evaluates four experimental schemes comparing univariate and multivariate approaches, both with and without the attention mechanism. The dataset comprises solar wind physical parameters and local geomagnetic data covering the period of 2020–2025, representing the ascending phase of Solar Cycle 25. To ensure optimal and objective model configurations, Automated Hyperparameter Optimization (HPO) is implemented using the Optuna framework. Experimental results demonstrate that the Multivariate LSTM + Multi-head Attention model serves as the best architecture, achieving a coefficient of determination (R2) of 0.5299 and a Root Mean Square Error (RMSE) of 0.7040. The integration of multivariate features proves crucial, confirming that equatorial geomagnetic dynamics are significantly influenced by external solar wind drivers. The attention mechanism successfully enhances model stability and phase synchronization, enabling precise storm onset detection without significant time lag. In addition to superior accuracy, the model demonstrates computational efficiency with an average inference time of 66 ms, confirming its feasibility for implementation as a real-time Early Warning System. This research contributes a novel perspective in modeling specific geomagnetic activities in the equatorial region, which possesses unique characteristics compared to high-latitude regions.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Badai Geomagnetik, Ekuator, Indeks K, LSTM, Multi-head Attention, Pembelajaran Mendalam. Geomagnetic Storm, Deep Learning, Equator, K-Index, LSTM, Multi-head Attention.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Aby Rengga Saputra
Date Deposited:	27 Jan 2026 06:08
Last Modified:	28 Jan 2026 03:40
URI:	http://repository.its.ac.id/id/eprint/130429

Actions (login required)

View Item