Prediksi Kualitas LSF Kiln Feed Menggunakan NARX-XGboost Dengan Input Fungsi Transfer Pada Data Mixed-Frequency Di Unit Blending Silo PT Semen Tonasa

Ardanika, Aura Lovi (2026) Prediksi Kualitas LSF Kiln Feed Menggunakan NARX-XGboost Dengan Input Fungsi Transfer Pada Data Mixed-Frequency Di Unit Blending Silo PT Semen Tonasa. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221165_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Blending silo pada proses produksi semen berfungsi sebagai unit homogenisasi yang menyebabkan perubahan kualitas raw meal tidak langsung tercermin pada kualitas kiln feed, melainkan muncul setelah jeda waktu tertentu akibat proses pencampuran material di dalam silo. Penelitian ini bertujuan mengembangkan model prediksi kualitas kiln feed berdasarkan parameter Lime Saturation Factor (LSF) menggunakan pendekatan NARX–Fungsi Transfer–XGBoost. Data historis PT Semen Tonasa periode Agustus–September 2025 yang meliputi LSF Raw Meal dan LSF Kiln Feed dianalisis menggunakan kerangka Unrestricted Mixed Data Sampling (U-MIDAS) untuk menangani perbedaan resolusi waktu. Struktur hubungan dinamis diidentifikasi menggunakan metode Fungsi Transfer Box–Jenkins melalui proses prewhitening dengan model ARIMA(5,0,0) dan analisis Cross-Correlation Function (CCF) dengan mempertimbangkan batas bawah retention time fisik blending silo sebesar 44 jam. Hasil identifikasi menghasilkan dua kandidat struktur Fungsi Transfer, yaitu TF(22,0,1) dan TF(37,0,1) dengan komponen noise ARMA(3,1). Berdasarkan hasil evaluasi, model dengan delay 22 dipilih sebagai model akhir. Struktur tersebut digunakan untuk membentuk 13 fitur NARX yang selanjutnya dimodelkan menggunakan algoritma XGBoost dengan optimasi hyperparameter melalui Randomized Search. Model menghasilkan nilai RMSE sebesar 1,3136, MAE sebesar 1,0284, dan MAPE sebesar 0,9994% pada data pengujian. Analisis feature importance menunjukkan bahwa fitur LSF Raw Meal hasil pembentukan U-MIDAS memberikan kontribusi sebesar 59,29%, sedangkan komponen autoregresif memberikan kontribusi sebesar 40,71%. Hasil penelitian menunjukkan bahwa pendekatan NARX–Fungsi Transfer–XGBoost mampu merepresentasikan hubungan dinamis antara kualitas raw meal dan kiln feed, serta berpotensi dimanfaatkan sebagai sistem pendukung pengambilan keputusan dalam pemantauan kualitas proses homogenisasi di unit blending silo.
=======================================================================================================================================
The blending silo in the cement manufacturing process functions as a homogenization unit, causing changes in raw meal quality to affect kiln feed quality only after a certain time delay due to the material mixing process inside the silo. This study aims to develop a prediction model for kiln feed quality based on the Lime Saturation Factor (LSF) using the NARX–Transfer Function–XGBoost approach. Historical data from PT Semen Tonasa collected during August–September 2025, consisting of Raw Meal LSF and Kiln Feed LSF, were analyzed using the Unrestricted Mixed Data Sampling (U-MIDAS) framework to address differences in sampling frequencies. The dynamic relationship between the input and output variables was identified using the Box–Jenkins Transfer Function method through a prewhitening process with an ARIMA(5,0,0) model followed by Cross-Correlation Function (CCF) analysis while considering the physical lower bound of the blending silo retention time of 44 hours. The identification produced two candidate Transfer Function structures, TF(22,0,1) and TF(37,0,1), with an ARMA(3,1) noise component. Based on the evaluation results, the 22-period delay model was selected as the final model. The resulting structure was used to construct 13 NARX features, which were modeled using the XGBoost algorithm with hyperparameter optimization through Randomized Search. The final model achieved an RMSE of 1.3136, MAE of 1.0284, and MAPE of 0.9994% on the testing dataset. Feature importance analysis showed that the U-MIDAS-based Raw Meal LSF features contributed 59.29%, while the autoregressive components contributed 40.71%. The proposed approach effectively represents the dynamic relationship between raw meal and kiln feed quality and has the potential to support decision-making in monitoring the homogenization process in the blending silo.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Fungsi Transfer, Kiln Feed, NARX, Raw Meal, XGBoost, Kiln Feed, NARX, Raw Meal, Transfer Function, XGBoost
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Aura Lovi Ardanika
Date Deposited:	04 Aug 2026 01:28
Last Modified:	04 Aug 2026 01:28
URI:	http://repository.its.ac.id/id/eprint/142429

Actions (login required)

View Item