Blasting vibration velocity prediction based on least squares support vector machine with particle swarm optimization algorithm. Issue 5 (3rd September 2019)

• Record Type: Journal Article
• Authors: Yuan, Qing; Zhai, Shihong; Wu, Li; Chen, Peishuai; Zhou, Yuchun; Zuo, Qingjun
• Is Part Of: Geosystem engineering. Volume 22:Issue 5(2019)
• Page Start: 279
• Page End: 288
• Publication Date: 2019-09-03
• Subjects: Particle swarm optimization algorithm -- blasting vibration velocity prediction -- least squares support vector machine -- fuzzy neural network -- characteristic parameter optimization; Mining engineering -- Periodicals; Petroleum engineering -- Periodicals; Gas engineering -- Periodicals; Geology, Economic -- Periodicals; 620
• DOI: 10.1080/12269328.2019.1607570
• Languages: English
• ISSNs: 1226-9328

Abstract

ABSTRACT: In blasting excavation engineering of super-large section underground caverns, blasting vibration velocity prediction accuracy is affected by many factors. In order to improve its accuracy, the key problem is to obtain these affect factors comprehensively. In this paper, we innovatively put forward eight independent factors in the aspect of explosion source conditions, engineering conditions and propagation medium conditions. These factors have complex non-linear relationship with blasting vibration velocity. We consider particle swarm optimization (PSO) algorithm and least squares support vector machine (LS-SVM) method for prediction (PSO-LSSVM). In this way, how to determine the characteristic parameters and calculation rules of PSO-LSSVM method is another key problem, which has been innovatively solved in this paper. Then it is used to predict the blasting vibration velocity of underground water-sealed LPG caverns in Yantai, China, and compared with Sadov's formula (SA), fuzzy neural network (FNN) and LS-SVM methods. The results indicate that relative errors of PSO-LSSVM are significantly less than LS-SVM, FNN and SA. Whether global root mean square relative error for prediction accuracy, or group number meeting requirement of error threshold value for generalization performance, the PSO-LSSVM method is superior to LS-SVM, FNN and SA with best availability and superiority.