Abstract: Underwater blasting is a complicated, nonlinear, and dynamic process of energy release. It is affected by many factors, and its process has not been fully investigated at present. In order to accurately predict the peak particle velocity induced by underwater blasting based on a small amount of field measurements, the GRA-GA-BP model is established based on the grey relational analysis theory combining with the genetic neural network which has the nonlinear mapping and global searching capabilities. In the model, the potential information of the small sample is fully discovered, and the main factors affecting the vibration velocity are reasonably determined based on the grey relational analysis theory. Moreover, the problems of the neural network unable to automatically select and optimize input variables in complicated and multivariate systems are solved, which enhances the adaptability and stability of the genetic neural network. Finally, the GRA-GA-BP model is adopted to predict the peak particle velocity induced by underwater blasting at Dajin Island in the first phase of Taishan nuclear power station. Compared with the results obtained by traditional genetic neural network and the Sadaovsk formula, the prediction error of the GRA-GA-BP model is smaller and more stable. Therefore, the proposed procedure provides an appropriate way to predict the peak particle velocity induced by underwater blasting.

Key words: grey relational analysis, genetic neural network, underwater blasting, peak particle velocity, prediction