Prediksi Curah Hujan Berbasis Multivariate Long Short-Term Memory Untuk Mitigasi Risiko Operasional Dan Peningkatan Akurasi Perencanaan Produksi Pada Sektor Pertambangan Terbuka

Darmawan, Ilham (2026) Prediksi Curah Hujan Berbasis Multivariate Long Short-Term Memory Untuk Mitigasi Risiko Operasional Dan Peningkatan Akurasi Perencanaan Produksi Pada Sektor Pertambangan Terbuka. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032241096-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Kegiatan operasional pertambangan terbuka sangat dipengaruhi oleh curah hujan yang berpotensi menyebabkan deviasi target produksi secara signifikan, sehingga kemampuan prediksi yang akurat menjadi krusial untuk mitigasi risiko operasional. Penelitian ini membangun dan mengevaluasi tiga skenario model prediksi curah hujan bulanan berbasis Long Short-Term Memory (LSTM) dengan studi kasus di PT Gunungbayan Pratama Coal (GBPC) Blok II, Kutai Barat, Kalimantan Timur, yaitu Univariate LSTM, Multivariate LSTM, dan Hybrid Late Fusion Multi-Headed LSTM. Data yang digunakan mencakup 144 sampel bulanan periode Januari 2014 hingga Desember 2025 dari Open-Meteo Historical Weather API berbasis reanalisis ERA5- Land. Seleksi fitur melalui konsensus enam uji statistik menghasilkan sembilan variabel meteorologi terpilih sebagai input model Multivariate. Pengujian hyperparameter dilakukan terhadap jumlah lapisan, node, lookback, dan epoch menggunakan pendekatan direct multi-step forecasting berhorizon tiga bulan. Model Multivariate LSTM dengan konfigurasi dua lapisan, enam belas node, lookback sembilan bulan, dan seratus epoch ditetapkan sebagai model terbaik dengan R² sebesar 0,1688, RMSE 97,66 mm, dan MAE 80,02 mm. Novelty penelitian ini terletak pada transformasi hasil prediksi menjadi estimasi Rain Hours dan Slippery Hours untuk penyusunan Schedule Production Operation (SPO) tahun 2026. Validasi terhadap realisasi Januari hingga Mei 2026 menunjukkan model LSTM menghasilkan MAE 33,09 jam dan RMSE 35,84 jam, lebih baik dibandingkan metode rata-rata historis GBPC dengan MAE 46,49 jam dan RMSE 55,32 jam.
=======================================================================================================================================
Open-pit mining operations are highly influenced by weather conditions, where high rainfall causes disruptions to heavy equipment activities, reduced haul road bearing capacity (slippery road), and potential pit flooding that leads to significant production target deviations. This research develops and evaluates three monthly rainfall prediction model scenarios based on Long Short-Term Memory (LSTM) architecture, with a case study at PT Gunungbayan Pratama Coal (GBPC) Block II, Kutai Barat, East Kalimantan: Univariate LSTM, Multivariate LSTM, and Hybrid Late Fusion Multi-Headed LSTM. The dataset comprises 144 monthly samples from January 2014 to December 2025, with supporting atmospheric variables sourced from the Open-Meteo Historical Weather API based on ERA5-Land reanalysis. Feature selection was conducted using a six-test statistical consensus approach (Pearson, Spearman, Mutual Information, OLS, VIF, and Granger Causality), yielding nine selected meteorological variables. Hyperparameter optimization was performed systematically using a direct multi-step forecasting approach with a three-month horizon. Based on comparative evaluation using MAE, RMSE, and R² metrics, the Multivariate LSTM model was established as the best model with R² of 0.1688, RMSE of 97.66 mm, and MAE of 80.02 mm. The novelty of this research lies in transforming rainfall predictions into Rain Hours and Slippery Hours estimates for the 2026 Schedule Production Operation (SPO). Validation against the first five months of 2026 shows the LSTM model achieves MAE of 33.09 hours and RMSE of 35.84 hours, outperforming GBPC's five-year historical average method which recorded MAE of 46.49 hours and RMSE of 55.32 hours.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Long Short-Term Memory (LSTM), Prediksi Curah Hujan, Pertambangan Terbuka, Schedule Production Operation, Mitigasi Risiko Operasional.Rainfall Prediction, Surface Mining, Schedule Production Operation, Operational Risk Mitigation.
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics
Divisions:	Faculty of Creative Design and Digital Business (CREABIZ)
Depositing User:	Ilham Darmawan
Date Deposited:	25 Jul 2026 16:05
Last Modified:	25 Jul 2026 16:05
URI:	http://repository.its.ac.id/id/eprint/138432

Actions (login required)

View Item