【摘要】 以矿山微震监测数据为基础,结合地质、开采条件等因素,建立矿震危险性神经网络预测模型。首先,利用神经网络变量选择的方法,根据BP神经网络的权值和阈值,对多因素诱导的矿震危险性进行分析;然后,通过灰色关联分析消除输出指标的偶然性,建立基于矿震能量消噪值的神经网络,对矿震危险性进行预测;最后,用基于杂交的粒子群算法优化神经网络。研究结果表明:矿震危险性受爆破药量、岩层强度比、开采深度和空区体积的影响非常大,占总影响值的87.35%,而受其他因素的影响较小,说明矿震与开采深度和空区规模之间呈高度的非线性关系;采用该预测模型使BP网络的收敛速度加快,训练精度较高,预测结果的相对误差下降52.64%。该模型对硬岩金属矿山的矿震活动研究有一定的参考价值。更多还原

【Abstract】 Based on the data of micro-seismic system of mine,combining with mining and geological conditions and other factors,the prediction model of neural network was established.Firstly,using the method of variable selection based on neural network,according to the weights and thresholds of BP neural network,risk of multi-factor-induced mine earthquake was analyzed.Then contingency of output index was eliminated through grey correlation analysis,and the neural network based on de-nosing value for energy of mining tremor was established to predict risk of mine earthquake.Finally,particle swarm optimization algorithm was used to optimize neural network based on hybridization.The results show that mine earthquake is affected mainly by blasting capacity,ratio to intensity of strata,mining depth and gap volume,accounting for 87.35% of the total impact value,but it is less affected by other factors,so there is nonlinear relationship between mine earthquake and human exploitation to a great extent.This prediction model speeds up the convergence rate of BP network,and improves the training accuracy,and the relative error of predictions decreases by 52.64%.The method provides some reference value for the study of mine earthquake of the hard-rock metal mines.更多还原