Peramalan Cuaca untuk Keselamatan Penerbangan Menggunakan Non-stationary Transformer

Syaefudin, Mohamad Anwar (2026) Peramalan Cuaca untuk Keselamatan Penerbangan Menggunakan Non-stationary Transformer. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025232002-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Peramalan deret waktu memegang peran krusial dalam berbagai aplikasi dunia nyata, termasuk peramalan cuaca demi keselamatan penerbangan. Model Transformer telah menunjukkan potensi dalam tugas ini dengan kemampuan menangkap pola temporal. Namun, data deret waktu di dunia nyata sering kali bersifat non-stasioner, ditandai perubahan properti statistik seiring waktu, yang menjadi tantangan bagi model deep learning. Penelitian ini bertujuan untuk memperoleh model peramalan terbaik untuk data cuaca bandara menggunakan arsitektur yang diusulkan berbasis Non-stationary Transformer (NST), yang dianggap tepat untuk peramalan cuaca penerbangan. Tujuan spesifik meliputi mengidentifikasi mekanisme attention yang sesuai untuk cuaca penerbangan tropis, serta mengeksplorasi dampak penggunaan RevIN dalam model NST. Studi ini mengeksplorasi teknik deep learning canggih dengan mengintegrasikan strategi De-stationary Attention dari NST dan mekanisme Probabilistic Sparse Attention dari Informer. Mekanisme attention yang diusulkan yaitu De-Stationary ProbSparse Attention (DSProb Attention), menggabungkan elemen keduanya. Eksperimen membandingkan DSProb Attention dengan mode max/avg terhadap model dasar NST dan Informer pada dataset publik (ETT, Weather) dan data privat Automatic Weather Observing System (AWOS). Hasil eksperimen menunjukkan bahwa NST cocok untuk data AWOS dibandingkan model Transformer lainnya. Mekanisme DSProb Attention yang diusulkan juga meningkatkan performa NST serta DSProb max mode direkomendasikan untuk peramalan potensi angin kencang dalam mode multivariat. Penggunaan RevIN tidak disarankan untuk data tanpa periodisitas jelas karena menurunkan korelasi, tetapi sangat disarankan untuk data dengan periodisitas jelas karena memperkuat performa.
========================================================================================================================================
Time series forecasting plays crucial role in various real-world applications, including weather forecasting for aviation safety. Transformer models have shown potential in this task due to their ability to capture temporal patterns. However, realworld time series data is often non-stationary, characterized by changes in statistical properties over time, posing a challenge for deep learning models. This study aims to develop the best forecasting model for airport weather data using a proposed architecture based on Non-stationary Transformer (NST), which is considered suitable for aviation weather forecasting. Specific objectives include identifying an appropriate attention mechanism for tropical aviation weather and exploring the impact of using RevIN in the NST model. The study explores advanced deep learning techniques by integrating the De-stationary Attention strategy from NST and the Probabilistic Sparse Attention mechanism from Informer. The proposed attention mechanism is De-Stationary ProbSparse Attention (DSProb Attention), combines elements of both. Experiments compare DSProb Attention in max/avg modes compare with baseline NST and Informer models on public datasets (ETT,Weather) and private Automatic Weather Observing System (AWOS) data. The results demonstrate that NST is more suitable for AWOS data compared to other Transformer models. The proposed DSProb Attention mechanism also enhances NST performance, with DSProb max mode recommended for forecasting potential strong winds in multivariate mode. The use of RevIN is not recommended for data without clear periodicity as it reduces correlation, however it is highly recommended for data with clear periodicity as it significantly improves performance.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Non-stationary Tranformer, keselamatan penerbangan, cuaca bandara, Non-stationary Tranformer, safety flight, aviation weather
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA9.58 Algorithms
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Mohamad Anwar Syaefudin
Date Deposited:	26 Jan 2026 07:24
Last Modified:	26 Jan 2026 07:24
URI:	http://repository.its.ac.id/id/eprint/130320

Actions (login required)

View Item