Forecasting Emisi CO pada Rotary Dryer dengan Pendekatan Temporal Fusion Transformer (TFT)

Ilham, Aulya Sri Utami (2026) Forecasting Emisi CO pada Rotary Dryer dengan Pendekatan Temporal Fusion Transformer (TFT). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025241030-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Rotary dryer pada proses pengolahan nikel menghasilkan emisi gas karbon monoksida (CO) sebagai produk samping dari proses pembakaran pada combustion chamber. Kadar emisi CO yang berlebih berpotensi menimbulkan risiko ledakan pada Electrostatic Precipitator (ESP), sehingga sistem kontrol saat ini menerapkan pemadaman otomatis pada ambang batas tertentu yang dapat menghambat operasi produksi. Penelitian ini bertujuan untuk merancang model forecasting emisi gas CO berbasis Temporal Fusion Transformer (TFT) serta mengembangkan pendekatan prescriptive analytics guna menghasilkan rekomendasi pengendalian parameter operasional rotary dryer secara proaktif. Model TFT dikembangkan melalui beberapa iterasi hyperparameter tuning, menghasilkan performa RMSE sebesar 2,42 ppm, MAPE sebesar 1,58%, dan R-squared sebesar 0,998, serta unggul dibandingkan model pembanding lain, yaitu Generalized Additive Model (GAM), Neural Network Regression (NNR), dan Long Short-Term Memory (LSTM). Model TFT terlatih kemudian dimanfaatkan dalam kerangka prescriptive analytics untuk mencari kombinasi nilai parameter operasi yang dapat meminimalkan prediksi emisi CO, dengan membandingkan pendekatan Random Search dan Variable Selection Network (VSN)-Guided Search. Hasil evaluasi menunjukkan bahwa VSN-Guided Search memberikan rekomendasi yang lebih konsisten menurunkan emisi CO, dengan feasibility rate 100% pada seluruh parameter controllable dan besaran perubahan setpoint yang lebih terkendali dibandingkan Random Search. Penelitian ini diharapkan dapat menjadi solusi inovatif dalam mendukung pengendalian emisi CO secara lebih dini dan proaktif pada operasi rotary dryer.
=======================================================================================================================================
The rotary dryer utilized in the nickel processing facility generates carbon monoxide (CO) emissions as a byproduct of the combustion process within the combustion chamber. The presence of elevated levels of carbon monoxide (CO) emissions has been identified as a factor that could potentially lead to an explosion risk with respect to the Electrostatic Precipitator (ESP). Considering this potential hazard, the existing control system has been programmed to initiate an automatic shutdown at specific thresholds, a measure that could result in the disruption of ongoing production operations. The objective of this study is to design a Temporal Fusion Transformer (TFT)-based carbon monoxide (CO) emission forecasting model and to develop a prescriptive analytics approach to generate proactive recommendations for controlling the rotary dryer's operational parameters. The TFT model was developed through multiple iterations of hyperparameter tuning, yielding an RMSE of 2.42 ppm, a MAPE of 1.58%, and an R-squared of 0.998. It outperformed other comparison models, namely the Generalized Additive Model (GAM), Neural Network Regression (NNR), and Long Short-Term Memory (LSTM). The trained TFT model was then employed within a prescriptive analytics framework to identify combinations of operational parameter values that could minimize predicted CO emissions. This was achieved by comparing the Random Search approach with the Variable Selection Network (VSN)-Guided Search approach. The evaluation results demonstrate that VSN-Guided Search provides more consistent recommendations for reducing CO emissions, with a 100% feasibility rate across all controllable parameters and more controlled setpoint changes compared to Random Search.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep Learning, Forecasting Emisi CO, Prescriptive Analytics, Rotary Dryer, Temporal Fusion Transformer
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Aulya Sri Utami Ilham
Date Deposited:	03 Aug 2026 01:59
Last Modified:	03 Aug 2026 01:59
URI:	http://repository.its.ac.id/id/eprint/140323

Actions (login required)

View Item