Novriansyah, Sangsaka Wira Utama (2021) Perancangan Model Sistem Prediksi Kapasitas Generator Turbin Gas Menggunakan Machine Learning Berbasis Decision Tree Regressor. Masters thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
Buku_Laporan_Thesis_Sangsaka_Wira_1_September_2021.pdf - Accepted Version Download (4MB) | Preview |
Abstract
Kinerja model sistem prediksi berbasis Machine Learning di sistem prediksi generator turbin gas di PT SAKA Energi yang menggunakan JST tergolong cukup baik. Namun dalam kenyataannya penerapan JST di plant masih memiliki beberapa kekurangan terkait dengan keakurasian dan kepresisian model prediksi. Dalam tesis ini dilakukan perancangan model Machine Learning sebagai upaya untuk meningkatkan kinerja dari model prediksi sebelumnya. Beberapa model selain JST yang dirancang adalah DTR, Elastic Net dan Logistic Regression. Pemodelan ini dilakukan untuk mengatasi permasalahan dibidang Predictive Maintenance, dimana model diharapkan dapat mendeteksi anomali sebelum terjadi kerusakan yang permanen. Masing – masing model dirancang dengan 3 tipe yang berbeda dengan rasio data training dan testing sebesar 50:50 untuk tipe-1, 60:40 untuk tipe-2 dan 70:30 untuk tipe-3. Hasil analisa menunjukkan model terbaik untuk masing – masing model berbasis JST Tipe-3 dengan nilai R2 Score sebesar 0,93581, Decision Tree Regressor Tipe-3 sebesar 0,99823, Logistic Regression Tipe-3 sebesar 0,89591 dan Elastic Net Tipe-1 sebesar 0,91405. Nilai R2 Score terbaik ditunjukkan oleh model berbasis Decision Tree Regressor Tipe-3 dengan nilai R2 Score sebesar 0,99823, MAE (Mean Absoulute Error) sebesar 0,82142, MSE (Mean Squared Error) sebesar 1,32894, EVS (Evaluate Value Score) 0,99834 dan RMSE (Root Mean Squared Error) sebesar 1,11527 sehingga model berbasis Decision Tree Regressor Tipe-3 dapat digunakan sebagai model inference pada aplikasi prediksi generator turbin di PT SAKA Energi.
================================================================================================
Machine learning model system prediction model using gas turbin generator
capacity PT SAKA Energi which are used using ANN. The using of ANN in the plant
has several drawbacks for example accuracy and precission prediction model. In this
research machine learning modelling are researched to improve performance from
previous model. Several models except ANN implemented were DTR, Elastic Net and
Logistic Regression. This modelling is used to solve predictive maintenance problems
where the model expected to be able to detect anomly occurred before permanent
handicap for the system. Each of model has 3 different types with 3 different ratios of
training and testing for 50:50 for type-1, 60:40 for type-2, 70:30 for type-3. The best
models for each type for ANN are type-3 with R2 Score of 0,93581, Decision Tree
Regressor type-3 of 0,99823, Logistic Regression type-3 of 0,89591 and Elastic Net
type-1 of 0,91405. The best R2 Score is shown by Decision Tree Regressor model type-
3 with R2 Score of 0,99823, MAE (Mean Absoulute Error) of 0,82142, MSE (Mean
Squared Error) of 1,32894, EVS (Evaluate Value Score) 0,99834 and RMSE (Root
Mean Squared Error) of 1,11527. The conclusion for this research is Decision Tree
Regressor type-3 model has the best performance and can be implemented as
inference/production model for generator anomaly prediction application.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Model Sistem Prediksi, Generator Gas Turbin, Machine Learning, Decision Tree Regressor, Prediction System Model, Generator Turbine Gas |
| Subjects: | T Technology > TA Engineering (General). Civil engineering (General) > TA169.5 Failure analysis T Technology > TJ Mechanical engineering and machinery > TJ778 Gas turbines |
| Divisions: | Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30101-(S2) Master Thesis |
| Depositing User: | Sangsaka Wira Utama N |
| Date Deposited: | 04 Sep 2021 15:42 |
| Last Modified: | 12 Jun 2024 06:13 |
| URI: | http://repository.its.ac.id/id/eprint/91707 |
Actions (login required)
 |
View Item |