Research on the prediction of compression pressure of 2-stroke marine diesel engine with data-driven models and analysis based on the SHAP
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
In marine diesel engines, the compression pressure in the cylinder is important because it affects combustion performance. However, as the sensor measuring compression pressure is installed in relation to the combustion chamber, its performance may be decreased, reliability may be difficult to guarantee, and there is a possibility of failure. Therefore, it is necessary to develop a model to predict compression pressure. In this study, engine data from the training ship was acquired, preprocessed, and then trained on GBDTbased CatBoost and LightGBM models. The results of the two models were confirmed using five performance metrics and scatter and line plots. LightGBM showed better performance than CatBoost with the train and test sets for all performance metrics, but the difference was minimal. Analysis of the two trained models using SHAP revealed that the 'M/E SCAV AIR RECEIVER IN PRESS' and 'M/E T/C RPM' variables had the greatest impact on prediction of cylinder compression pressure.
Abstract
In marine diesel engines, the compression pressure in the cylinder is important because it affects combustion performance. However, as the sensor measuring compression pressure is installed in relation to the combustion chamber, its performance may be decreased, reliability may be difficult to guarantee, and there is a possibility of failure. Therefore, it is necessary to develop a model to predict compression pressure. In this study, engine data from the training ship was acquired, preprocessed, and then trained on GBDTbased CatBoost and LightGBM models. The results of the two models were confirmed using five performance metrics and scatter and line plots. LightGBM showed better performance than CatBoost with the train and test sets for all performance metrics, but the difference was minimal. Analysis of the two trained models using SHAP revealed that the 'M/E SCAV AIR RECEIVER IN PRESS' and 'M/E T/C RPM' variables had the greatest impact on prediction of cylinder compression pressure.