Peramalan Output Daya Pembangkit Listrik Tenaga Surya di Nusa Penida Menggunakan Hybrid Seasonal Autoregressive Integrated Moving Average (SARIMA) - Long Short-Term Memory (LSTM)

Lohjati, Sekar Kinanthi Wohing (2026) Peramalan Output Daya Pembangkit Listrik Tenaga Surya di Nusa Penida Menggunakan Hybrid Seasonal Autoregressive Integrated Moving Average (SARIMA) - Long Short-Term Memory (LSTM). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003211106-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Peningkatan kebutuhan energi listrik di Indonesia seiring dengan pertumbuhan ekonomi dan populasi menuntut pengembangan sumber energi yang berkelanjutan dan ramah lingkungan. Pembangkit Listrik Tenaga Surya (PLTS) menjadi salah satu solusi strategis karena memanfaatkan energi matahari yang bersih dan berlimpah. Namun, karakteristik output daya PLTS yang sangat dipengaruhi oleh kondisi cuaca dan faktor lingkungan menyebabkan fluktuasi daya yang tinggi, sehingga diperlukan metode peramalan yang akurat untuk mendukung perencanaan dan pengelolaan sistem pembangkitan. Penelitian ini bertujuan untuk membangun dan mengevaluasi model peramalan output daya PLTS di Nusa Penida menggunakan pendekatan Hybrid Seasonal Autoregressive Integrated Moving Average–Long Short-Term Memory (SARIMA–LSTM). Data yang digunakan merupakan data output daya PLTS dengan interval pencatatan 10 menit selama periode Januari hingga Desember 2025. Data kemudian diolah menjadi rata-rata harian pada rentang waktu pukul 06.00–18.00 untuk merepresentasikan jam operasional utama pembangkitan energi surya. Tahap preprocessing mencakup agregasi data, imputasi data tidak wajar, serta pengujian stasioneritas menggunakan uji Augmented Dickey–Fuller (ADF). Model SARIMA terbaik yang diperoleh adalah SARIMA (2,1,1)(1,0,1)[7] berdasarkan nilai Akaike Information Criterion (AIC) terendah dan hasil uji diagnostik residual yang memenuhi asumsi white noise. Selanjutnya, residual SARIMA dimodelkan menggunakan LSTM untuk menangkap pola non-linier yang tidak dapat dijelaskan oleh model linier. Hasil evaluasi menunjukkan bahwa model Hybrid SARIMA–LSTM memiliki performa yang lebih baik dibandingkan model SARIMA tunggal. Pada data out-sample, model hybrid menghasilkan nilai MAE sebesar 236,52, RMSE sebesar 302,73, dan MAPE sebesar 29,70%, yang menunjukkan peningkatan akurasi peramalan. Peramalan jangka pendek selama 7 hari ke depan menunjukkan hasil yang stabil dan realistis sesuai dengan karakteristik pembangkitan PLTS. Dengan demikian, model Hybrid SARIMA–LSTM dinilai layak digunakan sebagai alat bantu peramalan jangka pendek untuk mendukung pengelolaan dan perencanaan operasional PLTS di Nusa Penida.
==================================================================================================================================
The rapid growth of electricity demand in Indonesia, driven by economic development and population increase, requires the adoption of sustainable and environmentally friendly energy sources. Solar Power Plants (PLTS) represent a strategic solution due to the abundance and cleanliness of solar energy. However, the power output of PLTS is highly dependent on weather conditions and environmental factors, resulting in significant fluctuations that complicate operational planning. Therefore, accurate power output forecasting is essential to support effective energy management and system reliability. This study aims to develop and evaluate a hybrid forecasting model for PLTS power output in Nusa Penida using the Hybrid Seasonal Autoregressive Integrated Moving Average–Long Short-Term Memory (SARIMA–LSTM) approach. The dataset used in this study consists of PLTS power output recorded at 10-minute intervals from January to December 2025. To better represent effective solar generation periods, the data were aggregated into daily averages within the time window of 06:00–18:00. Data preprocessing included aggregation, correction of anomalous values through imputation, and stationarity testing using the Augmented Dickey–Fuller (ADF) test. Based on the lowest Akaike Information Criterion (AIC) value and residual diagnostic tests, the SARIMA (2,1,1)(1,0,1)[7] model was selected as the optimal linear model. The residuals generated from the SARIMA model were subsequently modeled using Long Short-Term Memory (LSTM) networks to capture nonlinear patterns that could not be adequately represented by linear statistical methods. Model performance evaluation demonstrates that the Hybrid SARIMA–LSTM approach outperforms the standalone SARIMA model. On out-of-sample data, the hybrid model achieved a Mean Absolute Error (MAE) of 236.52, a Root Mean Squared Error (RMSE) of 302.73, and a Mean Absolute Percentage Error (MAPE) of 29.70%, indicating improved forecasting accuracy. Furthermore, short-term forecasting for the next seven days produced stable and realistic predictions consistent with the characteristics of solar power generation. These results indicate that the Hybrid SARIMA–LSTM model is a reliable and effective tool for short-term forecasting of PLTS power output and can support operational planning and energy management for solar power systems in Nusa Penida.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Energi Terbarukan, SARIMA, LSTM, Hybrid SARIMA-LSTM, PLTS, Renewable Energy, SARIMA, LSTM, Hybrid SARIMA-LSTM, Photovoltaic Power Plant
Subjects:	Q Science Q Science > Q Science (General) Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics Q Science > QA Mathematics > QA280 Box-Jenkins forecasting Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Sekar Kinanthi Wohing Lohjati
Date Deposited:	30 Jan 2026 07:22
Last Modified:	30 Jan 2026 07:28
URI:	http://repository.its.ac.id/id/eprint/131198

Actions (login required)

View Item