Peramalan Beban Jangka Pendek Menggunakan Machine Learning Lstm (Long Short Term Memory) Pada Subsistem Kelistrikan Di Jakarta

Premiesyani, Hadeta (2026) Peramalan Beban Jangka Pendek Menggunakan Machine Learning Lstm (Long Short Term Memory) Pada Subsistem Kelistrikan Di Jakarta. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241069-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (18MB) | Request a copy

Abstract

Peramalan beban listrik jangka pendek merupakan komponen penting dalam mendukung otomatisasi dispatch, pengelolaan sistem tenaga berbasis smart grid, dan akurasi pengambilan keputusan operasional. Pada sistem interkoneksi dengan skala besar, peramalan pada tingkat subsistem diperlukan untuk menangkap karakteristik beban yang berbeda pada setiap wilayah operasi. Penelitian ini mengusulkan kerangka peramalan multi-horizon berbasis Stacked Long Short-Term Memory (LSTM) untuk sebelas subsistem pada jaringan tenaga listrik DKI Jakarta. Kerangka penelitian yang dikembangkan mencakup optimasi panjang sliding window, optimasi arsitektur model, serta retraining menggunakan data operasional beban listrik dengan resolusi 30 menit selama periode Agustus 2023 hingga Juli 2025. Pengujian peramalan beban pada model hasil optimasi di Juli 2025 menghasilkan nilai Mean Absolute Percentage Error (MAPE) yang berkisar antara 3,34% hingga 11,05%, dimana peningkatan akurasi peramalan tercatat pada sepuluh dari sebelas subsistem saat dibandingkan model baseline. Perbaikan akurasi terbesar terjadi pada subsistem BKSCWGPRK, dengan penurunan MAPE dari 13,77% menjadi 7,77%. Pengujian analisa sensitivitas cuaca yang dilakukan tiga subsistem terpilih menunjukkan hasil bahwa pengaruh variabel cuaca berbeda-beda bergantung pada karakteristik masing-masing subsistem. Pengujian statistik lanjutan dilakukan dengan menambahkan variabel cuaca berupa temperatur, kelembaban relatif, dan curah hujan pada tiga subsistem terpilih. Uji Wilcoxon Signed-Rank menunjukkan bahwa peningkatan akurasi tersebut belum signifikan secara statistik. Maka dapat disimpulkan optimasi parameter yang spesifik untuk setiap subsistem berperan penting dalam meningkatkan akurasi peramalan beban listrik jangka pendek, sementara kontribusi variabel cuaca bersifat subsistem-spesifik.
=======================================================================================================================================
Short-term load forecasting is a critical component for supporting automated dispatch, smart grid-based power system operation, and accurate operational decision-making. In large-interconnected power systems, subsystem-level forecasting is required to capture the diverse load characteristics of individual operational areas. This study proposes a multi-horizon forecasting framework based on Stacked Long Short-Term Memory (LSTM) networks for eleven power subsystems in the DKI Jakarta power network. The proposed framework incorporates sliding window optimization, architecture tuning, and model retraining using operational load data with a 30-minute resolution collected from August 2023 to July 2025. A chronological data partitioning strategy is adopted, where model optimization is performed using validation data and final evaluation is conducted on an independent test set from July 2025. The optimized models achieve Mean Absolute Percentage Error (MAPE) values ranging from 3.34% to 11.05%. Compared with a fixed-configuration baseline model, the optimized models improve forecasting accuracy in ten out of eleven subsystems. The largest improvement is observed in the BKSCWGPRK subsystem, where MAPE is reduced from 13.77% to 7.77%. Additional experiments are conducted by incorporating weather variables, including temperature, relative humidity, and rainfall, into selected subsystems. The results indicate that the impact of weather variables depends on the characteristics of each subsystem. Statistical evaluation using the Wilcoxon Signed-Rank Test shows that the accuracy improvement obtained from weather integration is not statistically significant. The findings demonstrate that subsystem-specific parameter optimization plays an important role in improving short-term load forecasting accuracy, while the contribution of weather variables is highly dependent on subsystem characteristics.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Peramalan beban listrik jangka pendek, Stacked LSTM, sliding window, smart grid, forecasting subsistem, peramalan multi-horizon Short-term load forecasting, optimization, subsystem forecasting, multi-horizon forecasting.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1001 Production of electric energy or power
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Hadeta Premiesyani
Date Deposited:	30 Jul 2026 07:58
Last Modified:	30 Jul 2026 07:58
URI:	http://repository.its.ac.id/id/eprint/140368

Actions (login required)

View Item