(湖南大學 電氣與信息工程學院，長沙 410082)
摘 要: 針對離散單元法(DEM)仿真非球形散體顆粒休止角計算量大、耗時長的問題，本文基于DEM歷史仿真數(shù)據(jù)，采用數(shù)據(jù)驅(qū)動的智能建模方法—BP、RBF神經(jīng)網(wǎng)絡(luò)建立非球形散體顆粒的休止角模型，并與傳統(tǒng)克里金回歸方法進行比較。結(jié)果表明，智能模型的運算速度相比DEM計算速度有很大提升；智能模型相比傳統(tǒng)克里金回歸模型具有更佳的預(yù)測性能，其中BP神經(jīng)網(wǎng)絡(luò)模型綜合性能最優(yōu)。最后，采用BP神經(jīng)網(wǎng)絡(luò)模型分析顆粒形狀及摩擦因數(shù)對休止角的影響，發(fā)現(xiàn)休止角隨顆粒形狀變量、摩擦因數(shù)的增加都呈現(xiàn)增大的趨勢，與現(xiàn)有研究結(jié)果一致，進一步證明了智能模型進行休止角預(yù)測的可靠性。
關(guān)鍵字: 非球形散體顆粒；休止角；智能模型；離散單元法
（College of Electrical and Information Engineering, Hunan University, Changsha 410082, China）
Abstract:The discrete element method (DEM) simulation of the angle of repose (AoR) of non-spherical is computationally intensive and time consuming. Based on the obtained DEM simulation data, the data driven intelligent modeling methods—the BP neural network and RBF neural networ were used to model the AoR of non-spherical discrete particles, and were compared with the traditional Kriging regression methods. The results show that the speed of the intelligent models is dramatically faster than the speed of the DEM simulation; the intelligent model has better predictive performance than the traditional Kriging regression model, and the BP neural network model has the best overall performance. Finally, based on the BP neural network model, the influences of particle shape and friction coefficient on the AoR were analyzed. It is found that the AoR increases with the increase of particle shape variable and friction coefficient, which further indicates the credibility of the intelligent model.
Key words: non-spherical discrete particles; angle of repose; intelligent model; discrete element method
