An optimal data analytics model for accurate projection of CI engine performance and emission reduction in hybrid fuel mode

As different fuel types exhibit diverse emission characteristics, and the stringent environmental regulations often drive the development of low-emission fuels. In this work, an optimal data analysis on emission variables and efficiency of single unit CI engine running in a hybrid fuel mode i.e. hydrogen, lemon grass oil, and water is investigated. Initially, to gather the experimental data, simulations of varying fuel modes are conducted by using the Simulink software. The obtained data is subjected to a data analytics process, where a generic preprocessing technique is involved, to eliminate unwanted artefacts present in the input dataset. Subsequently, an optimal extraction method for features is employed to remove most relevant features of pre-processed data to enhance the prediction accuracy and feasibility of the proposed model. A hybrid deep reinforcement learning (DRL) with artificial neural network (ANN) is implemented for analyzing the characteristics of CI engine. This approach leverages the capabilities of deep support and NN, to optimize the prediction model. Finally, the performance of proposed prediction method is assessed in terms of evaluation metrics like mean square error (MSE), R-2, accuracy, recall, precision and F1 score. From analysis, it is interpreted that the accuracy of proposed DRL-ANN model is measured to be 96.523%, 97.563%, 94.587%, 96.235% and 95.623% for BTE, BSFC, CO, HC, NOx and smoke emission. This shows that the proposed DRL-ANN approach is found to be reliable and effective method for efficient assessment of CI engine emission parameters.