Rancang Bangun Soft Sensor Berbasis LSTM untuk Estimasi Kandungan Etana pada Unit Deethanizer Column di Plant 3 Train G
Ainul Yaqin, Prof. Ir. Nazrul Effendy, S.T., M.Eng., Ph.D., IPM., ASEAN Eng.
2026 | Skripsi | FISIKA TEKNIK
PT Badak LNG Bontang, an oil and gas company, faces fluctuations in feed gas composition following the addition of lean gas from the Merakes Field in 2021. The increase in methane and the decrease in heavy hydrocarbons, including ethane, necessitate efficient gas composition monitoring. The currently used Gas Chromatograph requires high maintenance costs and has analytical latency, prompting the consideration of a soft sensor as a continuous alternative solution..
This research developed a soft sensor model using a Long Short-Term Memory (LSTM) architecture to estimate the ethane content in the Deethanizer Column at Plant 3 Train G. The model was trained using 2,057 historical data observations (March 17, 2022 – January 31, 2024) with 13 process variables as inputs. Data preprocessing utilized linear interpolation imputation, and temporal order was maintained through chronological data splitting and lookback sequences.
The results showed that the LSTM model configured with two layers, a dropout of 0.2, a hidden size of 32, and a 36-timestep lookback delivered optimal performance. This configuration yielded testing metrics of R² at 0.9121, RMSE at 5.1981, MAE at 2.7517, and Bias at 1.2557. The model successfully tracked ethane variation trends, although deviations remained when the actual target values changed abruptly.
Kata Kunci : soft sensor, LSTM, etana, deethanizer column, Badak LNG