Machine Learning–Based Prediction and Analysis of Engine Performance for a Direct-Port Injection Compressed Natural Gas Fueled Engine

Compressed natural gas (CNG) is one of the most potential alternative fuels for internal combustion engines (ICEs). Investigation of injection strategies is crucial to improve CNG engine performance and extend flammability limit, especially for direct-port combined injection (DI-PI) CNG engine. In this study, engine performance characteristics of a DI-PI CNG engine are predicted and optimized using machine learning techniques. The engine performance data are collected from an experimental single-cylinder CNG engine with combined direct-port injection systems. The performance parameters (BTE, VE, BSFC, and EGT) are predicted using different machine learning algorithms including ANN, LSTM, LR, PR, RFR, and SVR. All predicted models show great performance in fitting the predicted values to experimental results within an error range of 10%. The prediction R-squared values are achieved higher than 0.90 for all performance parameters in most cases. PR, ANN, and RFR models exhibit great accuracy for BTE prediction; ANN and LSTM models show better accuracy for VE prediction; PR shows the best prediction performance for BSFC, while RFR shows greatest accuracy for EGT prediction, with R-squared scores higher than 0.99. Furthermore, the response surface methodology (RSM) is applied to analyze and optimize the output response of BTE, VE, BSFC, and EGT. The optimization R-squared values are achieved as 0.9735, 0.9853, 0.9376, and 0.8993 for BTE, VE, BSFC, and EGT, respectively. The optimum conditions are found at 80.16 L per minute (LPM) of injected fuel, 0.91 of DI/PI ratio, and 98.12 (Nm) of engine torque. The optimum performance characteristics are obtained as 32.10 (%), 82.79 (%), 246.2274 (g/kWh), and 352.50 (°C) for BTE, VE, BSFC, and EGT, respectively. The developed models are beneficial for predicting and optimizing the performance of DI-PI CNG engine and have potential for further development of CNG engine.