Peramalan Jumlah Sambaran Petir Pada Jaringan Transmisi PT PLN (Persero) UPT Surabaya Menggunakan Pendekatan Hierarchical Time Series

Rachmanda, Calista Rana (2026) Peramalan Jumlah Sambaran Petir Pada Jaringan Transmisi PT PLN (Persero) UPT Surabaya Menggunakan Pendekatan Hierarchical Time Series. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221149-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Sambaran petir merupakan salah satu faktor eksternal yang berpotensi menyebabkan gangguan pada jaringan transmisi tenaga listrik. Peramalan jumlah sambaran petir yang akurat dan koheren diperlukan sebagai dasar perencanaan dan mitigasi risiko gangguan sistem ketenagalistrikan. Data sambaran petir pada jaringan transmisi tersusun berdasarkan wilayah geografis sehingga membentuk struktur hierarki tiga level, yaitu tingkat Surabaya, tingkat wilayah, dan tingkat koridor. Penelitian ini bertujuan melakukan peramalan jumlah sambaran petir pada jaringan transmisi di Surabaya menggunakan pendekatan hierarchical time series forecasting dengan forecast reconciliation. Data yang digunakan berupa jumlah sambaran petir bulanan periode Januari 2018 hingga Desember 2025 dari Unit Pelaksana Transmisi (UPT) PLN Surabaya, mencakup 17 koridor transmisi yang membentuk 20 deret waktu hierarkis. Metode peramalan dasar yang digunakan adalah Prophet dan TimeGPT sebagai pendekatan automatic forecasting. Konfigurasi Prophet dipilih melalui grid search 24 kombinasi dari enam nilai Fourier order dan empat skenario changepoint, sedangkan TimeGPT dipilih antara pendekatan zero-shot dan fine-tuning. Pemilihan kedua konfigurasi dievaluasi menggunakan rolling window forecasting dengan tiga skenario jendela waktu pada seluruh 20 deret. Hasil seleksi menunjukkan konfigurasi terbaik Prophet adalah Fourier order 1 tanpa changepoint, sedangkan TimeGPT zero-shot jauh lebih unggul dibandingkan fine-tuning. Base forecast kedua model direkonsiliasi menggunakan enam metode Bottom-Up, Top-Down, Middle-Out, Ordinary Least Squares, Weighted Least Squares, dan Minimum Trace—serta dievaluasi menggunakan MAE, RMSE, MASE, dan RMSSE. Hasil menunjukkan metode Top-Down memberikan performa rekonsiliasi terbaik pada kedua model, dengan kombinasi TimeGPT-Top-Down menghasilkan MASE terkecil sebesar 0,925 sehingga dipilih sebagai model terbaik untuk peramalan 2026. Hasil peramalan menunjukkan pola musiman konsisten pada seluruh level hierarki, dengan intensitas tertinggi pada Februari sebesar 5.077 sambaran dan terendah pada Agustus sebesar 225 sambaran di tingkat kota. Hasil penelitian diharapkan mendukung PT PLN (Persero) UPT Surabaya dalam perencanaan pemeliharaan dan mitigasi risiko gangguan jaringan transmisi akibat sambaran petir.
=====================================================================================================================================
Lightning strikes are an external factor with the potential to cause disturbances in electrical power transmission networks. Accurate and coherent forecasting of lightning strikes is required as a basis for planning and mitigating power system disturbance risks. Lightning strike data on the transmission network are organized by geographical region, forming a three-level hierarchy: the Surabaya city level, the regional level, and the transmission corridor level. This research forecasts the number of lightning strikes on Surabaya's transmission network using a hierarchical time series forecasting approach with forecast reconciliation. The data consist of monthly lightning strike counts from January 2018 to December 2025, obtained from the Transmission Execution Unit (UPT) of PLN Surabaya, covering 17 transmission corridors that form 20 hierarchical series. Prophet and TimeGPT are employed as automatic forecasting methods for the base forecast. The Prophet configuration is selected through a grid search of 24 combinations from six Fourier order values and four changepoint scenarios, while the TimeGPT configuration is selected between zero-shot and fine-tuning approaches. Both selections are evaluated using rolling window forecasting with three time-window scenarios across all 20 series. Results show the best Prophet configuration is a Fourier order of 1 without changepoints, while zero-shot TimeGPT outperforms fine-tuning by a wide margin. The base forecasts are then reconciled using six methods Bottom-Up, Top-Down, Middle-Out, Ordinary Least Squares, Weighted Least Squares, and Minimum Trace evaluated using MAE, RMSE, MASE, and RMSSE. Top-Down yields the best reconciliation performance for both models, with TimeGPT-Top-Down producing the lowest MASE of 0.926 and thus selected as the best model for forecasting 2026 lightning strikes. The 2026 forecast shows a consistent seasonal pattern across all hierarchical levels, with the highest intensity predicted in February at 5,077 strikes and the lowest in August at 225 strikes at the city level. These findings are expected to support PT PLN (Persero) UPT Surabaya in planning maintenance and mitigating transmission network disturbance risks caused by lightning strikes.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Sambaran Petir, Hierarchical Time Series, Automatic Forecasting, Forecast Reconciliation. Lightning Strikes, Hierarchical Time Series, Automatic Forecasting, Forecast Reconciliation.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Calista Rana Rachmanda
Date Deposited:	31 Jul 2026 08:11
Last Modified:	31 Jul 2026 08:11
URI:	http://repository.its.ac.id/id/eprint/140903

Actions (login required)

View Item