Perancangan Sistem Kendali PI Adaptif Berbasis Particle Swarm Optimization (PSO) dan Long Short-Term Memory (LSTM) Untuk Pengendalian Temperatur Pada Vertical Thermal Oxidizer

Laksana, Ligama Putra and Vega, Putri Amelya (2026) Perancangan Sistem Kendali PI Adaptif Berbasis Particle Swarm Optimization (PSO) dan Long Short-Term Memory (LSTM) Untuk Pengendalian Temperatur Pada Vertical Thermal Oxidizer. Diploma thesis, Institut Teknologi Sepuluh Nopember.

This is the latest version of this item.

[thumbnail of 2041221105-Undergraduate_Thesis.pdf] Text
2041221105-Undergraduate_Thesis.pdf
Restricted to Repository staff only
Download (3MB)

Abstract

Unit Vertical Thermal Oxidizer (VTO) memegang peranan krusial dalam fasilitas pengolahan minyak dan gas untuk mendegradasi emisi berbahaya melalui oksidasi termal. Stabilitas temperatur ruang bakar menjadi parameter kritis untuk memenuhi baku mutu lingkungan dan mencegah kerusakan alat. Namun, operasi VTO menghadapi tantangan dinamika proses yang kompleks akibat fluktuasi beban umpan stokastik, non-linearitas kinetika reaksi, serta inersia termal makro yang besar. Strategi pengendalian konvensional menggunakan Proportional-Integral (PI) dengan parameter tetap (fixed-gain) seringkali gagal mempertahankan stabilitas saat terjadi disturbance ekstrem, yang berpotensi menyebabkan overshoot temperatur atau respon pemulihan yang lambat. Proyek akhir ini mengusulkan strategi kendali cerdas menggunakan PI adaptif berbasis Particle Swarm Optimization (PSO) dan Long Short-Term Memory (LSTM). Algoritma PSO diutilisasi secara offline untuk mengekstrak himpunan data latih parameter optimal, sedangkan arsitektur LSTM difungsikan sebagai auto-tuner yang memprediksi parameter PI optimal (Kp dan Ki) secara real-time dengan mempelajari pola historis disturbance laju alir off-gas. Metode proyek akhir dilakukan melalui simulasi dinamis pada perangkat lunak MATLAB/Simulink dengan memodelkan karakteristik VTO menggunakan pendekatan First-Order Transfer Function (FOTF) tanpa waktu tunda. Hasil pelatihan LSTM menunjukkan performa komputasi yang konvergen dengan Validation MSE sebesar 0,001679 serta akurasi pelacakan (R2) mencapai 90,95% untuk Kp dan 90,67% untuk Ki. Kinerja sistem kendali adaptif PI-LSTM dievaluasi secara komparatif terhadap PI konvensional pada step disturbance 29.000 SCFM, di mana model mampu mereduksi overshoot sebesar 49,13% dan mempercepat settling time sebesar 73,11% (dari 119 detik menjadi 32 detik). Pada uji continuous dynamic disturbances, PI-LSTM berhasil menekan Integral Absolute Error (IAE) sebesar 65,99% serta memberikan potensi penghematan konsumsi fuel gas tahunan operasional sebesar Rp1,69 miliar.
===============================================================================================================================
The Vertical Thermal Oxidizer (VTO) unit plays a crucial role in oil and gas processing facilities to degrade hazardous emissions through thermal oxidation. Combustion chamber temperature stability is a critical parameter for meeting environmental standards and preventing equipment damage. However, VTO operations face complex process dynamic challenges due to stochastic feed load fluctuations, reaction kinetics non-linearity, and large macro thermal inertia. Conventional control strategies using fixed-gain Proportional-Integral (PI) controllers often fail to maintain stability during extreme load disturbances, potentially leading to temperature overshoot or sluggish recovery response. This final project proposes an intelligent control strategy using an adaptive PI based on Particle Swarm Optimization (PSO) and Long Short-Term Memory (LSTM). The PSO algorithm is utilized offline to generate optimal parameter datasets, while the LSTM architecture functions as an online auto-tuner that predicts optimal PI parameters (Kp and Ki) in real-time by learning from historical permeate gas flowrate disturbance patterns. The research methodology is conducted through dynamic simulation in MATLAB/Simulink software by modeling VTO characteristics using a First-Order Transfer Function (FOTF) without dead time. The LSTM training results show convergent computational performance with a Validation MSE of 0.001679, tracking accuracy (R2) of 90.95% for Kp, and 90.67% for Ki. The performance of the adaptive PI-LSTM control system is evaluated comparatively against conventional PI under a 29,000 SCFM step disturbance, where it reduces overshoot by 49.13% and accelerates settling time by 73.11% (from 119 seconds to 32 seconds). Under continuous dynamic disturbances, the PI-LSTM successfully suppresses the Integral Absolute Error (IAE) by 65.99% and yields a potential annual fuel gas saving of IDR 1.69 billion.

Item Type: Thesis (Diploma)
Uncontrolled Keywords: Adaptive Control, FOTF, Long Short-Term Memory (LSTM), Particle Swarm Optimization (PSO), PI, Vertical Thermal Oxidizer.
Subjects: T Technology > T Technology (General)
T Technology > T Technology (General) > T57.5 Data Processing
T Technology > T Technology (General) > T57.62 Simulation
T Technology > T Technology (General) > T57.84 Heuristic algorithms.
T Technology > TP Chemical technology
Divisions: Faculty of Vocational > 24305-Industrial Chemical Engineering Technology
Depositing User: Ligama Putra Laksana
Date Deposited: 05 Aug 2026 01:11
Last Modified: 05 Aug 2026 01:11
URI: http://repository.its.ac.id/id/eprint/143426

Available Versions of this Item

Actions (login required)

View Item View Item