Optimization of BP Neural Network Model for Rockburst Prediction under Multiple Influence Factors

Rockbursts are serious threats to the safe production of mining, resulting in great casualties and property losses. The accurate prediction rockburst is an important premise that influences safety health miners. As a classical machine learning algorithm, back propagation (BP) neural network has been widely used prediction. However, there few reports about influence study different training sample sizes, optimization algorithms index dimensionless methods on accuracy BP models. Therefore, 100 groups typical engineering samples were collected locally abroad, considering relevance, scientificity quantifiability indexes, ratio maximum tangential stress surrounding rock uniaxial compressive strength (σθ/σc), tensile (σc/σt) elastic energy (Wet) chosen as indexes. When number was 40, 70 100, sixty improved models established based standard gradient descent algorithm four (momentum quasi-Newton conjugate Levenberg–Marquardt algorithm) (unified extreme value processing method, differentiated data averaging normalized method). performances each model compared with those comparative results indicate size, method have effects models, which described follows: (1) A increases addition size. average Aave twenty under three kinds sizes from (40) = 69.7% (100) 75.3%, maximal Amax 85.0% 97.0%. (2) comprehensive C generally higher than model. (3) unified combined highest 97.0% 194, five cases completely consistent actual situation at site, so this best selected paper.