Research on Prediction Method of Engine Instantaneous Thermal Efficiency Based on 1DCNN-LSTM Neural Network

To address the challenge of directly calculating engine thermal efficiency under real-vehicle driving conditions, which hinders the evaluation of overall operational status and degradation, a prediction model for instantaneous thermal efficiency based on 1DCNN-LSTM was developed. Bench tests under steady-state and transient conditions were conducted on a high-pressure common-rail diesel engine. Pearson, Spearman, and Grey Relation were applied to select strongly correlated variables as model inputs. The test set was divided into three prediction intervals based on engine speed: 2050–2150 rpm, 2100–2200 rpm, and 2150–2250 rpm. The Pearson method achieved a correlation coefficient R of 0.93881 in the 2150–2250 rpm interval, with average MAE, MSE, and RMSE values of 0.007918, 0.000213, and 0.014602, respectively. Compared to the Spearman method, these metrics were reduced by 9.4%, 9.3%, and 4.8%, and by 9.1%, 6.6%, and 3.2% compared to the Grey Relation method. Training time increased by only 3.2% compared to Spearman and decreased by 1.3% compared to Grey Relation. The proposed method achieved low prediction errors without significantly increasing computational resource consumption. Optimal input variables and prediction intervals were identified, providing a novel solution for engine performance optimization and fault diagnosis under real-vehicle conditions using thermal efficiency as an evaluation indicator.