Prediction of diesel generator performance and emissions using minimal sensor data and analysis of advanced machine learning techniques

Abstract

To this day, diesel generator (DG) continues to play an indispensable role in all industries and smart engines and engines with eco-friendly technologies are being developed. However, with the advent of the unmanned automation era, countermeasures are required when DG sensors are non-functional. Therefore, an optimized AI model for backing up sensor data, which is necessary for the safety of smart engines equipped with eco-friendly facilities, was developed. To develop an AI model for this purpose, an experiment was conducted to obtain the engine and emission data to be used and 11 models were created. By predicting 16 variables related to the engine performance and emissions using a total of five sensor data, including three sensors essential for the engine safety, the proposed AI model could back up data when some sensors failed. Moreover, various hyperparameter tunings were applied and compared to maximize the model performance. Consequently, the decision tree (DT)-based models and genetic algorithm showed a good performance, and the weighted average of ensemble (DT) model showed the best performance with R2 value of 0.9981, and a SMAPE value of 0.7244. Additionally, to confirm the generalization performance of the model, the prediction performance of the models was measured using new data, and the blending of ensemble (ALL) model had the best performance with R2 value of 0.9266, and a SMAPE value of 5.585. Finally, the application of the concept used to develop the AI model and the future direction of the work were discussed.