ABSTRACT Machine learning methods were employed to predict the detonation heat of energetic compounds containing C, H, O, and N elements with high accuracy, based on structural features and estimated detonation product descriptors. Since detonation heat fundamentally reflects the cumulative contributions of chemical bond cleavage and formation, it is hypothesized to possess a certain degree of linear additivity. Initially, the MLR model was developed for predicting the heat of detonation. The leave‐one‐out cross‐validation results of the training set were R 2 = 0.868, MAE = 459 kJ/kg, and RMSE = 673 kJ/kg. For the test set, the results were R 2 = 0.912, MAE = 334 kJ/kg, and RMSE = 524 kJ/kg. This prediction result indicates that the assumption that linear addition is reasonable. Subsequently, an SVR model with a linear kernel was employed to construct the prediction model. This approach yielded improved performance for the training set, with leave‐one‐out cross‐validation R 2 = 0.884, MAE = 421 kJ/kg, and RMSE = 636 kJ/kg. For the test set, the model achieved R 2 = 0.938, MAE = 309 kJ/kg, and RMSE = 439 kJ/kg. These results suggest that the inherent complexity of the explosion process also introduces nonlinear effects, thereby increasing the challenge of predicting the heat of detonation. These results not only verify the interpretability of the model but also provide a crucial basis for understanding the formation mechanism of the detonation heat.
Hu et al. (Mon,) studied this question.