Raditya, Arva Taqiy (2026) Prediksi Dan Optimasi Efisiensi Boiler Co-Firing Di PT PLN Nusantara Power Berbasis Pareto Front Menggunakan RSM, ANN, Dan LightGBM (NSGA-II & TOPSIS). Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 2039221090-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
2039221090-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy |
Abstract
Pembangkit Listrik Tenaga Uap (PLTU) batubara masih menjadi sumber pembangkitan listrik utama di Indonesia dengan pangsa sekitar 67% dari total kapasitas terpasang. Program co-firing biomassa yang dicanangkan pemerintah bertujuan untuk menurunkan emisi karbon di sektor ketenagalistrikan. Namun, optimasi setpoint operasional boiler PLTU co-firing secara real-time tetap menjadi tantangan karena kompleksitas dinamika pembakaran dan keterbatasan pendekatan manual operator. Penelitian ini mengembangkan kerangka prediksi dan optimasi berbasis data untuk PLTU Paiton 400 MW dengan rasio co-firing sawdust 5%. Dataset terdiri dari 19.440 baris data operasional SCADA/CEMS yang telah dibersihkan sesuai standar ASME PTC-4. Data ini mencakup enam variabel input (tekanan boiler, temperatur superheater, laju alir udara primer dan sekunder, laju alir batubara, kadar oksigen) dan tiga variabel target (efisiensi boiler, emisi SO₂, emisi CO₂). Tiga model surrogat dikembangkan—Response Surface Methodology (RSM), Artificial Neural Network (ANN), dan Light Gradient Boosting Machine (LightGBM)—yang kemudian dikombinasikan menjadi model ensemble berbobot. Model ensemble tersebut mencapai koefisien determinasi R² = 0,9753; 0,9998; dan 0,9997 masing-masing untuk prediksi efisiensi boiler, emisi SO₂, dan emisi CO₂ pada data uji. Algoritma NSGA-II menghasilkan 150 solusi non-dominated yang membentuk Pareto front tiga dimensi. Setpoint optimal hasil seleksi TOPSIS (Peringkat 1) menghasilkan efisiensi boiler 87,22%, emisi SO₂ 67,85 ppm, dan emisi CO₂ 107.167 ppm; nilai SO₂ memenuhi batas regulasi emisi yang berlaku, sedangkan CO₂ dilaporkan sebagai indikator intensitas karbon. Untuk tahap akhir, hasil penelitian diintegrasikan ke dalam Decision Support System (DSS) berbasis web (React + FastAPI) dengan asisten AI lokal (Ollama/llama3:8b). Integrasi DSS ini memungkinkan operator untuk menerima rekomendasi setpoint secara real-time melalui antarmuka berbasis browser
================================================================================================================================
Coal-fired power plants (CFPPs) remain the primary source of electricity generation in Indonesia, accounting for approximately 67% of the total installed capacity. The government's biomass co-firing program aims to reduce carbon emissions in the power sector. However, real-time optimization of operating setpoints in co-firing CFPP boilers remains challenging due to the complexity of combustion dynamics and the limitations of manual operator adjustments. This study develops a data-driven prediction and optimization framework for the 400 MW Paiton CFPP operating with a 5% sawdust co-firing ratio. The dataset comprises 19,440 rows of operational SCADA/CEMS data cleaned according to the ASME PTC-4 standard. It encompasses six input variables (boiler pressure, superheater temperature, primary and secondary air flow rates, coal flow rate, and oxygen content) and three target variables (boiler efficiency, SO₂ emissions, and CO₂ emissions). Three surrogate models were developed— Response Surface Methodology (RSM), Artificial Neural Network (ANN), and Light Gradient Boosting Machine (LightGBM)—which were then combined into a weighted ensemble model. The ensemble model achieved coefficients of determination R² = 0.9753, 0.9998, and 0.9997 for the prediction of boiler efficiency, SO₂ emissions, and CO₂ emissions on the test data, respectively. The NSGA-II algorithm produced 150 non- dominated solutions forming a three-dimensional Pareto front. The optimal setpoint selected by TOPSIS (Rank 1) yields a boiler efficiency of 87.22%, SO₂ emission of 67.85 ppm, and CO₂ emission of 107,167 ppm; the SO₂ value complies with the applicable emission regulation limit, while CO₂ is reported as a carbon-intensity indicator. For the final stage, the research results were integrated into a Decision Support System (DSS) built on a web stack (React + FastAPI) with a local AI assistant (Ollama/llama3:8b). This DSS integration enables operators to receive real-time setpoint recommendations through a browser-based interface.
Actions (login required)
 |
View Item |