Implementasi dan Evaluasi Model Deep Learning untuk Analisis Prediktif Proses Pengolahan Air Limbah

Nugraha, Muhammad Ryan Fikri and Bahalwan, Muammar (2026) Implementasi dan Evaluasi Model Deep Learning untuk Analisis Prediktif Proses Pengolahan Air Limbah. Project Report. [s.n.]. (Unpublished)

Text 5025231013_5053231020-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Pengolahan air limbah merupakan komponen kritis dalam infrastruktur air modern yang berperan melindungi kesehatan masyarakat dan kelestarian lingkungan. Sistem pengolahan air limbah menghasilkan aliran data operasional yang terus-menerus, mencerminkan perilaku dinamis dari proses kimia dan biologis yang kompleks. Metode statistik klasik kerap kesulitan menangkap pola tersebut, sehingga membatasi efektivitasnya untuk pemantauan real-time dan pengambilan keputusan prediktif. Pada Kerja Praktik ini diimplementasikan dan dibandingkan tujuh model prediktif: Linear Regression sebagai baseline statistik, tiga model gradient boosting (CatBoost, LightGBM, XGBoost), satu model rekuren (LSTM), serta dua model deep learning lanjutan (iTransformer dan FRNet). Evaluasi dilakukan pada lima dataset kualitas air dunia nyata dengan metrik MSE, RMSE, MAE, dan R². Pengaruh strategi Adaptive Cycling Learning Rate (ACyLeR) terhadap iTransformer dan FRNet juga dievaluasi pada tiga ukuran jendela input (96, 192, dan 336 langkah waktu). Hasil eksperimen menunjukkan bahwa model deep learning, khususnya iTransformer dan FRNet, secara umum mengungguli baseline pada dataset N2O Treatment Wastewater dan Water Quality Prediction (mis. iTransformer mencapai RMSE 0,0846 dan R² 0,7454 pada N2O), namun tidak secara seragam pada semua dataset; pada River Water Quality dengan resolusi rendah, model gradient boosting justru lebih kompetitif. StrategiACyLeR memberikan perbaikan konsisten pada iTransformer untuk dataset N2O di semua ukuran jendela, namun memberikan hasil beragam pada dataset lain. Temuan ini menunjukkan pentingnya pemilihan model yang sesuai dengan karakteristik data dalam pemantauan cerdas instalasi pengolahan air limbah.
================================================================================================================================
Wastewater treatment is a critical component of modern water infrastructure, playing a vital role in protecting public health and environmental sustainability. Wastewater treatment systems generate continuous streams of operational data that reflect the dynamic behavior of complex chemical and biological processes. Classical statistical methods often struggle to capture these patterns, thereby limiting their effectiveness for real-time monitoring and predictive decision-making. In this internship project, seven predictive models were implemented and compared: Linear Regression as a statistical baseline, three gradient boosting models (CatBoost, LightGBM, XGBoost), one recurrent model (LSTM), and two advanced deep learning models (iTransformer and FRNet). Evaluation was conducted on five real-world water quality datasets using the metrics MSE, RMSE, MAE, and R². The influence of the Adaptive Cycling Learning Rate (ACyLeR) strategy on iTransformer and FRNet was also evaluated across three input window sizes (96, 192, and 336 time steps). Experimental results show that deep learning models, particularly iTransformer and FRNet, generally outperformed the baseline on the N2O Treatment Wastewater and Water Quality Prediction datasets (e.g., iTransformer achieved an RMSE of 0.0846 and an R² of 0.7454 on N2O), but did not do so uniformly across all datasets; on the River Water Quality dataset with low resolution, gradient boosting models were actually more competitive. The ACyLeR strategy provided consistent improvements for iTransformer on the N2O dataset across all window sizes, but yielded mixed results on other datasets. These findings highlight the importance of selecting models that are appropriate to the data characteristics in intelligent monitoring of wastewater treatment plants.

Item Type:	Monograph (Project Report)
Uncontrolled Keywords:	deep learning, pengolahan air limbah, iTransformer, FRNet, time series forecasting, multivariate prediction, ACyLeR
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Ryan Fikri Nugraha
Date Deposited:	07 Jul 2026 06:55
Last Modified:	07 Jul 2026 06:55
URI:	http://repository.its.ac.id/id/eprint/134384

Actions (login required)

View Item