Prediksi Temperatur dan Curah Hujan Menggunakan Fast Fourier Transform dan Model Hibrid HHO-SATCN-LSTM

Rahmawati, Alvian Dwi (2026) Prediksi Temperatur dan Curah Hujan Menggunakan Fast Fourier Transform dan Model Hibrid HHO-SATCN-LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221150-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Perubahan pola cuaca akibat pemanasan global menyebabkan variabel iklim menjadi tidak stabil, termasuk temperatur dan curah hujan. Perubahan pola cuaca tersebut juga menyebabkan peningkatan jumlah kejadian cuaca ekstrem, tidak terkecuali di wilayah Lombok. Kondisi topografi wilayah Lombok yang bervariasi menyebabkan variabel cuaca seperti curah hujan dan temperatur dapat berubah-ubah sehingga cukup sulit untuk diprediksi. Beberapa model seperti machine learning dan deep learning mampu meningkatkan prediksi temperatur dan curah hujan. Sehingga digunakan model hibrid harris hawks optimization- self attention temporal convolutional network- long short term memory (HHO-SATCN-LSTM) untuk prediksi temperatur dan curah hujan di Lombok. Model SATCN-LSTM sangat dipengaruhi oleh kombinasi hyperparameter-nya, sedangkan penentuan hyperparameter dengan trial and error kurang efektif. Sehingga HHO ditambahkan dengan tujuan menentukan konfigurasi hyperparameter optimal model hibrid SATCN-LSTM untuk meningkatkan kemampuan generalisasi model. Peran self attention-temporal convolutional network (SA-TCN) untuk mengekstrak fitur deret waktu dan menangkap ketergantungan global pada data deret waktu, dan long short term memory (LSTM) menangkap ketergantungan jangka panjang. Selain itu, di preprocessing ditambahkan Fast Fourier Transform (FFT) untuk menangani noise data curah hujan dan
temperatur. Penelitian ini menggunakan data cuaca harian di Lombok dari tahun 2005 hingga 2024. Hasil model prediksi untuk temperatur menunjukkan MAE = 0.0527 °C, RMSE = 0.0878 °C, MAPE = 0.19%, dan R2 = 0.9898. Sedangkan hasil model prediksi untuk curah hujan juga menunjukkan MAE = 0.7831 mm, RMSE = 1.1330 mm, MAPE = 29.78%, dan R2 = 0.9829. Hasil tersebut menunjukkan model HHO-SATCN-LSTM mampu lebih baik dibandingkan model SATCN-LSTM.
====================================================================================================================================================
Changes in weather patterns due to global warming have led to the instability of climate variables, including temperature and rainfall, driving an increase in extreme weather events, particularly in Lombok where a varied topography makes these variables fluctuate significantly and difficult to predict. To address this, a hybrid model combining Harris Hawks Optimization, Self-Attention Temporal Convolutional Network, and Long Short Term Memory (HHO-SATCN-LSTM) was utilized to predict temperature and rainfall in Lombok. The SATCN-LSTM model is heavily influenced by its hyperparameter combination, whereas determining hyperparameters through trial and error is less effective. HHO was incorporated to determine the optimal hyperparameter configuration for the SATCN-LSTM hybrid model, with the aim of enhancing the model’s generalization capability. In this framework, SA-TCN extracts time-series features and captures global dependencies, LSTM captures long-term dependencies, and Fast Fourier Transform (FFT) is integrated into preprocessing to handle data noise. Using daily weather data from Lombok spanning from 2005 to 2024, the prediction results for temperature achieved an MAE of 0.0527 °C, RMSE of 0.0878 °C, MAPE of 0.19%, and R2 of 0.9898, while the rainfall predictions yielded an MAE of 0.7831 mm, RMSE of 1.1330 mm, MAPE of 29.78%, and R2 of 0.9829, demonstrating that the proposed HHO-SATCN-LSTM model out performs the standard SATCN-LSTM model.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Prediksi Temperatur dan Curah Hujan, Self Attention Temporal Convolutional Network (SATCN), Long Short Term Memory (LSTM), Harris Hawks Optimization (HHO), Temperature and Rainfall Prediction, Self Attention Temporal Convolutional Network (SATCN), Long Short Term Memory (LSTM), Harris Hawks Optimization (HHO)
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Alvian Dwi Rahmawati
Date Deposited:	03 Aug 2026 08:49
Last Modified:	03 Aug 2026 08:49
URI:	http://repository.its.ac.id/id/eprint/142750

Actions (login required)

View Item