Prediksi Inflow Waduk PLTA Saguling, Cirata, dan Jatiluhur dengan Menggunakan Machine Learning

Yaskie, Herrizon Okto (2026) Prediksi Inflow Waduk PLTA Saguling, Cirata, dan Jatiluhur dengan Menggunakan Machine Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241059_Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Prediksi aliran masuk (inflow) waduk harian yang akurat sangat penting bagi pengoperasian waduk pembangkit listrik tenaga air dan pengelolaan sumber daya air, namun dinamika aliran masuk bersifat sangat nonlinier dan nonstasioner. Penelitian ini membangun model Long Short-Term Memory (LSTM) dan K-Nearest Neighbors (kNN) dengan variabel hidrometeorologi meliputi curah hujan, kelembapan, suhu dan kecepatan angin untuk meramalkan inflow pada waduk PLTA Saguling, Cirata dan Jatiluhur yang masuk dalam area kaskade Citarum. Data inflow dan hidrometeorologi dari tahun 2014 hingga 2023 dibagi secara kronologis menjadi set pelatihan (2014–2020), validasi (2021), dan pengujian (2022–2023). Hasil menunjukkan bahwa model LSTM dengan data hidrometeorologi memberikan hasil terbaik pada masing-masing pembangkit, PLTA Saguling dengan NRMSE 36,61%, NMAE 23,22%, dan R2 0,811, PLTA Cirata dengan NRMSE 42,52%, NMAE 27,25%, dan R2 0,675, dan PLTA Jatiluhur dengan NRMSE 131,86%, NMAE 86,35%, dan R2 0,130. Simulasi energi yang dapat dihasilkan pada periode Januari 2023 dengan hasil peramalan inflow dan Tinggi Muka Air awal Saguling 637,01 m, Cirata 216,14 m dan Jatiluhur 101,64 m serta kebutuhan pengairan Jatiluhur ditetapkan sebesar 205,77 m³/s adalah sebesar 432,05 GWh.
========================================================================================================================================
Accurate daily reservoir inflow predictions are crucial for the operation of hydroelectric power reservoirs and water resource management; however, inflow dynamics are highly nonlinear and nonstationary. This study develops Long Short-Term Memory (LSTM) and K-Nearest Neighbors (kNN) models using hydrometeorological variables—including precipitation, humidity, temperature, and wind speed—to forecast inflow at the Saguling, Cirata, and Jatiluhur hydroelectric reservoirs within the Citarum cascade area. Inflow and hydrometeorological data from 2014 to 2023 were chronologically divided into training (2014–2020), validation (2021), and testing (2022–2023) sets. The results show that the LSTM model using hydrometeorological data yields the best results for each power plant: the Saguling hydroelectric power plant with an NRMSE of 36.61%, an NMAE of 23.22%, and an R² of 0.811; the Cirata hydroelectric power plant with an NRMSE of 42.52%, NMAE of 27.25%, and R² of 0.675; and the Jatiluhur Hydroelectric Power Plant with an NRMSE of 131.86%, NMAE of 86.35%, and R² of 0.130. The simulation of energy that can be generated in January 2023, based on the forecast results for inflow and initial water levels of Saguling at 637.01 m, Cirata at 216.14 m, and Jatiluhur at 101.64 m, as well as the irrigation requirement for Jatiluhur set at 205.77 m³/s, is 432.05 GWh.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Inflow, Hidrometeorologi, K-Nearest Neighbors, Long Short-Term Memory Inflow, hydrometeorology, K-Nearest Neighbors, Long Short-Term Memory
Subjects:	T Technology > TC Hydraulic engineering. Ocean engineering > TC424 Water levels
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Herrizon Okto Yaskie
Date Deposited:	30 Jul 2026 06:55
Last Modified:	30 Jul 2026 06:55
URI:	http://repository.its.ac.id/id/eprint/140193

Actions (login required)

View Item