基于PCA和MCMC的贝叶斯方法的海下矿山水害源识别分析

Application of PCA and Bayesian MCMC to discriminate between water sources in seabed gold mines

  • 摘要: 海底金矿矿山水害对矿山生产、人员施工及矿山设备等产生较大威胁,是矿山开采中的自然灾害之一,快速有效的判别出矿山水害水源对于事故的防治有重要意义。三山岛金矿的巷道围岩裂隙普遍并长期存在涌水现象,矿区开采中矿井水害的水源主要有海水、第四系水、基岩裂隙水、地下水等,为了准确快速的判别矿井水水源,有效预防矿井水突水及水害威胁,本研究结合监测点水样的水文地质条件与不同监测点水样的水化学成分分析,选取Mg2+、Na++K+、Ca2+、SO42−、Cl和HCO3 共6项指标作为判别因子,通过主成分分析得出不同水样的矿化程度。在贝叶斯算法分析原理的基础上,将马尔可夫链蒙特卡洛(Markov Chain Monte Carlo, MCMC)引入到贝叶斯方法中,运用统计软件SPSS统计,构建贝叶斯判别分析模型,得出基于水样样本信息的算法估计的后验分布,得出矿山水害水源的分析方法。运用三山岛金矿水害取水点的水样分析数据进行详细的分析验证,建立矿井突水水源模型,进行不同水样的信息分析,得出贝叶斯统计函数并进行水源判别结果分析,验证了贝叶斯矿山水害水源判别模型的准确性和实用性,对现场工作的开展和水害防治有一定的指导意义。

     

    Abstract: Water hazards in submarine gold mines pose a great threat to mine production, construction personnel, and mining equipment, and represent one of the natural disasters that occur in mining. To prevent and control accidents, it is critical to quickly and effectively identify water sources. Cracks in the rocks surrounding the roadway in the Sanshandao Gold Mine are a widespread and long-term water gushing phenomenon. The main sources of mine water hazards in mining areas are seawater, Quaternary water, bedrock fissure water, and groundwater. To accurately and quickly identify mine water sources and effectively prevent inrushes of mine water and water-hazard threats, the hydrogeological conditions and chemical composition of water samples from different monitoring points were analyzed and six indicators, i.e., Mg2+, Na++K+, Ca2+, SO4 2−, Cl, and HCO3 , were selected as discriminant factors. Based on the analysis principle of the Bayesian algorithm, the Markov chain Monte Carlo (MCMC) approach was introduced into the Bayesian method. A Bayesian discriminant analysis model was then constructed using SPSS Statistics and the MCMC Bayesian method. The posterior distribution estimated by the algorithm is based on water-sample information, which enables the analysis of the mine water source. Based on the water-sample data from a water intake point at the Sanshandao Gold Mine, detailed analysis and verification were performed, and a water-source model for the inrush of mine water was established. An analysis of different water samples was then performed. Through the selection of variables, variables with a strong discriminant ability and high degree of correlation were introduced into the discriminant function to obtain the Bayesian statistical function, thus enabling a discriminatory analysis of the water sources. The accuracy and practicability of the proposed Bayesian mine-water-source identification model were verified. This model has certain significance for guiding future field work and water-hazard prevention and control efforts.

     

/

返回文章
返回