Neural network–based prediction methods for height of water-flowing fractured zone caused by underground coal mining

摘要: The accurate calculation of the height water-flowing fractured zone (WFFZ) in coal mine is a critical factor ensuring safety and protecting surficial eco-environment. In view inapplicability traditional empirical formula for predicting WFFZ, correlation WFFZ influence factors was analyzed firstly based on 82 collected groups coalfield measured data China. Results show that mining thickness depth have significant effect WFFZ. Subsequently, were divided into two parts: 80% training models remaining 20% validation. Two prediction models, i.e., multiple regression (MR) model BP neural network (BPNN) model, established trained. A new merging regression-BP (MR-BPNN) proposed by combining model. accuracy generalization ability three verified recorded testing samples. comparison suggest all had better applicability coalmine, compared with existing methods. More importantly, MR-BPNN combined nonlinear mapping empiric which could provide high-accurate, strong-generalized, practical application coalfield. addition, reason practicability network-based discussed.