基于RS-MPSO-LSSVM的冲击地压危险性等级预测研究

发布时间：2018-05-13 02:16

本文选题：冲击地压 + 粗糙集　；参考：《辽宁工程技术大学》2015年硕士论文

【摘要】：冲击地压已经与矿井瓦斯、矿井火灾、矿井地热等并列成为世界范围内最严重的煤矿动力灾害。冲击地压发生时岩体弹性能大于岩体塑性能的部分以冲击波的形式向周围传播,冲击机理不仅复杂多变,而且影响因素众多,用常规的数学、力学理论对冲击地压预测建模型困难相当大。已经成为国际采矿工程和岩土力学领域急待研究解决的科学难题。为了紧密联系工程实际而建立有说服力的方法,对现场测试技术的研究意义重大并且必须开展,而现场测试技术研究的关键就是采取怎样手段而利用测试数据对冲击地压等级进行预测。本文根据徐州矿务局张集矿近几年观测的动力现象及所经历的冲击事故资料,提出用粗糙集理论和改进粒子群的最小二乘支持向量机(RS-MPSO-LSSVM)相结合的算法对历史数据进行预测研究。利用模糊C均值聚类的连续属性离散化算法得到连续的冲击地压危险性等级原始的属性指标的离散性数据,利用粗糙集理论的属性及对象约减思想对各冲击影响指标进行去冗余化处理,也就是在不改变分类能力的条件下,剔除无关冗余知识,降低空间维度。将约简属性作为LSSVM的输入向量,并选择了合适的支持向量机核函数。由于最小二乘支持向量机参数选择不同会导致不同的模型训练时间、预测误差,利用改进的粒子群算法来优化LSSVM惩罚因子C和核宽度系数σ,得到RS-MPSO-SVM预测模型,并利用测试样本对预测模型进行仿真研究。通过与BP神经网络预测模型及标准粒子群优化的最小二乘支持向量机(RS-PSO-LSSVM)的预测模型比较分析发现,在样本信息完全相同的情况下,本文提出的粗糙集—改进粒子群的最小二乘支持向量机(RS-MPSO-LSSVM)的冲击地压危险性等级预测模型具有更快的收敛速度、更优的预测精度特点。该研究成果可以对矿井冲击地压监测人员提供合理参考,为冲击地压事故预测提供一条新的思路。
[Abstract]:Rock burst has become the most serious coal mine power disaster in the world, along with mine gas, mine fire and mine geothermal. When the rock burst occurs, the elastic energy of rock mass is larger than the plastic property of rock mass. The shock wave propagates around the rock. The impact mechanism is not only complex and changeable, but also has many influencing factors. The mechanics theory is very difficult to build the model of the rock burst prediction. It has become an urgent scientific problem in the field of mining engineering and geotechnical mechanics. In order to establish a persuasive method to closely connect with the engineering practice, it is of great significance to the research of field test technology and must be carried out. The key of field test technology research is how to use the test data to predict the rock burst grade. Based on the dynamic phenomena observed in Zhangji Mine of Xuzhou Mining Bureau in recent years and the impact accident data experienced in recent years, this paper presents a method of combining rough set theory with least square support vector machine (LSSVM) of improved particle swarm to predict historical data. The continuous attribute discretization algorithm based on fuzzy C-means clustering is used to obtain the discrete data of the original attribute index of the continuous rock burst hazard grade. The attributes of rough set theory and the idea of object reduction are used to deredundancy each impact index, that is, eliminating irrelevant redundant knowledge and reducing spatial dimension without changing the classification ability. The reduction attribute is used as the input vector of LSSVM, and the appropriate kernel function of support vector machine is selected. Because different parameters of LS-SVM result in different training time and prediction error, the improved particle swarm optimization algorithm is used to optimize LSSVM penalty factor C and kernel width coefficient 蟽 to obtain RS-MPSO-SVM prediction model. Test samples are used to simulate the prediction model. Comparing with BP neural network prediction model and standard particle swarm optimization least squares support vector machine (LSSVM) prediction model, it is found that, when the sample information is identical, The RS-MPSO-LSSVM-based rough set-improved least squares support vector machine (LSSVM) model proposed in this paper has faster convergence speed and better prediction accuracy. The research results can provide reasonable reference for mine rock burst monitoring personnel and provide a new way of thinking for prediction of impact ground pressure accidents.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TD324

【相似文献】