Maulana, Akbar (2026) Peramalan Indeks Energi Unit Raw Mill PT Semen Tonasa Menggunakan Neural Network Autoregressive (NNAR) dan Hybrid ARIMA-NNAR. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5003221104-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5003221104-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy |
Abstract
Peramalan indeks energi pada unit raw mill PT Semen Tonasa merupakan instrumen penting dalam upaya optimalisasi biaya produksi dan peningkatan efisiensi energi di industri semen. Penelitian ini berfokus pada peramalan indeks energi harian dengan memanfaatkan data historis periode 01 November 2024 hingga 31 Oktober 2025. Data mengandung 35 observasi yang hilang akibat maintenance operasional sehingga dilakukan imputasi menggunakan metode Linear Interpolation dan Kalman Smoothing. Mengingat hasil uji Terasvirta mengidentifikasi adanya karakteristik nonlinier pada data indeks energi, strategi peramalan diarahkan pada penggunaan metode Neural Network Autoregressive (NNAR) dan pendekatan hybrid ARIMA-NNAR. Proses pencarian kombinasi hyperparameter optimal dilakukan menggunakan dua pendekatan, yaitu random search dan bayesian optimization. Evaluasi model dilakukan secara komprehensif menggunakan metrik Root Mean Square Error (RMSE) dan Mean Absolute Percentage Error (MAPE) pada data out-of-sample. Hasil penelitian menunjukkan bahwa model NNAR terbaik diperoleh pada konfigurasi NNAR(26,13) dengan imputasi kalman smoothing dan tuning bayesian optimization yang menghasilkan RMSE sebesar 1,2468 kWh/Ton dan MAPE sebesar 7,06%. Sementara itu, model hybrid ARIMA-NNAR terbaik diperoleh pada konfigurasi ARIMA([1,4,6,10],1,1)-NNAR(5,3) dengan imputasi kalman smoothing dan tuning bayesian optimization yang menghasilkan RMSE sebesar 0,7460 kWh/Ton dan MAPE sebesar 4,04%. Hasil peramalan indeks energi untuk periode mendatang diharapkan dapat menjadi acuan strategis bagi manajemen PT Semen Tonasa dalam melakukan perencanaan operasional yang lebih efisien dan berkelanjutan.
=======================================================================================================================================
Forecasting energy index in the Raw Mill unit of PT Semen Tonasa constitutes an important instrument in efforts to optimize production costs and improve energy efficiency in the cement industry. This research focuses on forecasting the daily energy index by utilizing historical data from November 1, 2024, to October 31, 2025. A total of 35 missing observations were identified due to operational maintenance, which were addressed through imputation using Linear Interpolation and Kalman Smoothing methods. Given that the Terasvirta test identified nonlinear characteristics in the energy index data, the forecasting strategy was directed towards the Neural Network Autoregressive (NNAR) method and the hybrid ARIMA-NNAR approach. The search for optimal hyperparameter combinations was conducted utilizing two approaches, namely Random Search and Bayesian Optimization. Model evaluation was carried out comprehensively based on Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) metrics on out-of-sample data. The findings indicate that the best NNAR model was obtained with the NNAR(26,13) configuration utilizing Kalman Smoothing imputation and Bayesian Optimization tuning, yielding an RMSE of 1.2468 kWh/Ton and MAPE of 7.06%. Meanwhile, the best hybrid ARIMA-NNAR model was obtained with the ARIMA([1,4,6,10],1,1)-NNAR(5,3) configuration utilizing Kalman Smoothing imputation and Bayesian Optimization tuning, yielding an RMSE of 0,7460 kWh/Ton and MAPE of 4,04%. The energy index forecasting results for the forthcoming period are expected to serve as a strategic reference for PT Semen Tonasa management in conducting more efficient and sustainable operational planning.
Actions (login required)
 |
View Item |