基于粗糙集理论和改进PNN算法的岩爆预测

Rockburst Prediction Based on Rough Set Theory and Improved PNN Algorithm

对岩爆进行准确且有效地预测具有十分重要的意义。根据岩爆的影响因素、特点及成因,选取岩石取样处的埋深H、岩石单轴抗压强度σ_(c)、应力系数σ_(θ)/σ_(c)、脆性系数σ_(c)/σ_(t)和冲击倾向性指数Wet作为预测指标,利用粗糙集属性约简算法获得关键属性,并得到特定地质条件下岩爆的主要影响因素,将主控因素数据归一化后构成概率神经网络的输入向量,减小了计算的复杂度。通过调整平滑因子的大小,建立基于粗糙集理论和概率神经网络(RS-PNN)的岩爆预测模型,并将RS-PNN的预测结果与其它模型的预测结果进行比较。结果表明,RSPNN模型的判别结果准确率较高,且PNN网络的收敛速度通常在几秒钟之内,故基于RSPNN的岩爆预测模型具有合理性和可行性。

Accurate and effective prediction of rockburst is of great significance.According to the influencing factors,characteristics and causes of rockburst,the buried depth H of the rock sampling site,the uniaxial compressive strengthσ_(c),the stress coefficientσ_(θ)/σ_(c),the brittleness coefficientσ_(c)/σ_(t) and the impact tendency index Wet were selected as the predictive indexes.The key attributes were obtained by using rough set attribute reduction algorithm,and the main influencing factors of rockburst under specific geological conditions were obtained.After normalizing the main control factor data,the input vector of the probabilistic neural network was formed,and then the computational complexity was reduced.By adjusting the size of the smoothing factor to determine the optimal PNN model.A rockburst prediction model based on rough set theory and probabilistic neural network(RS-PNN)was established,and the prediction results of RS-PNN were compared with those of other models.The results show that the discrimination results of RS-PNN model are more accurate.In addition,the convergence speed of the PNN network is usually within a few seconds,therefore the rockburst prediction model based on RS-PNN is reasonable and feasible.