93 / 2021-09-28 15:53:39

BP neural network model optimization for rockburst prediction considering sample sizes, optimization algorithms and dimensionless methods

rockburst prediction,BP neural network,model optimization,sample sizes,optimization algorithm,index dimensionless

Rockburst is a serious threat to mine safety production all over the world, resulting in serious casualties and property losses. Accurate prediction of rockburst is an important premise to prevent and control rockburst. As a classical machine learning algorithm, BP (Back Propagation) neural network has been widely used in rockburst prediction. But there are few reports about the influence study of different sample sizes, optimization algorithms and dimensionless methods on the prediction accuracy of BP neural network model. Therefore, 100 groups of typical rockburst engineering samples were collected at home and abroad, and considering the relevance, scientificity and quantifiability of the prediction indexes, the ratio of maximum tangential stress of surrounding rock to rock uniaxial compressive strength (σ_θ/σ_c), the ratio of rock uniaxial compressive strength to rock uniaxial tensile strength (σ_c/σ_t) and the elastic energy index (W_et) were chosen as the rockburst prediction indexes. When the number of samples was 40, 70 and 100 respectively, sixteen improved BP models were established based on the four optimization algorithms (momentum gradient descent algorithm, quasi-Newton algorithm, conjugate gradient algorithm, Levenberg-Marquardt algorithm) and four index dimensionless methods (unified extreme value processing method, differentiated extreme value processing method, data averaging processing method, normalized processing method). The prediction performance of each improved model was also compared with that of standard BP model. The comparative study results indicate that the sample sizes, optimization algorithms and dimensionless methods have different effects on the prediction accuracy of BP models, which are as follows: (1) The prediction accuracy value A of BP model increases with the addition of sample sizes. The average value A of sixteen improved models under three kinds of sample sizes increases from 69.4% to 77.8%; (2) The value A and comprehensive accuracy value N of BP neural network model based on four optimization algorithms are generally higher than that of standard BP model; (3) The improved BP model based on unified extreme value processing method combined with Levenberg-Marquardt algorithm has the highest value A (97.0%) and value N (194), and the prediction results of engineering cases are completely consistent with the field actual situation, so which is the best BP neural network model selected in this paper.