基于NPCA-GA-BP神经网络的采场稳定性预测方法

为提高采场稳定性的预测精度，充分考虑采场稳定性高度非线性和受多因素影响的特点，提出了一种基于NPCA-GA-BP神经网络的采场稳定性预测方法。选择影响采场稳定性的10个指标，运用非线性主成分分析减少指标的维度，提取4个主成分综合指标代替原有的10个指标，简化了神经网络结构，提升了运算速度。利用GA的全局寻优特点优化BP神经网络的权值和阈值，进一步增加了神经网络预测精度。以某矿山实测数据为例，对该预测方法进行验证，对比结果显示：NPCA-GA-BP和GA-BP模型的平均相对误差比BP模型分别降低了10.5%和7.6%，表明通过遗传算法优化BP神经网络可显著提高预测精度；NPCA-GA-BP模型的平均相对误差比GA-BP模型降低了2.9%，表明通过非线性主成分分析减少了变量的维度，提高了预测准确率。研究表明：NPCA-GA-BP预测方法具有更高的采场稳定性预测精度，对实现智慧矿山有一定的指导意义。

Prediction Method of Stope Stability Based on NPCA-GA-BP Neural Network

Stope stability is a geological mechanics problem that cannot be ignored in mining，and its stability directly affects the safety of mine production and engineering decision-making.Therefore，scientific prediction of stope stability plays a crucial role in mining safety.The stability of stope is a typical nonlinear problem.Since BP neural network has the virtue of tackling complex nonlinear systems，it can be applied to stope stability prediction.Nevertheless，the existing prediction methods either only focus on optimizing the weights and thresholds of the neural network or only consider that the stability of the stope is under the influence of multiple factors and the influencing indexes have a strong correlation，but do not consider the two methods in an integrated manner.Hence，the prediction accuracy of stope stability based on neural network is low，which cannot provide valid support for mine management.Due to the highly nonlinear characteristics of the mining stability system，the traditional principal component analysis will lose a large amount of information.Therefore，we propose a stope stability prediction method using nonlinear principal component analysis combined with BP neural network optimized by the genetic algorithm，which effectively improves the prediction accuracy of stope stability.The nonlinear principal component analysis method performs nonlinear dimensionality reduction on the impact indicators of stope stability，replacing the original multiple indicators with a few principal components that retain the original information，simplifying the neural network structure，and improving the operational efficiency.GA aims to optimize the initial weights and thresholds of the BP neural network to overcome the defects of unstable initial weight thresholds and further improve the accuracy of quarry stability prediction.Taking the measured data of a mine as an example，the effectiveness of the proposed method is verified.The comparison results show that the average relative errors of NPCA-GA-BP and GA-BP models are 10.5% and 7.6% lower than those of BP models，respectively，indicating that the BP neural network is optimized by the genetic algorithm can significantly improve the prediction accuracy.The average relative error of the NPCA-GA-BP model is 2.9% lower than that of the GA-BP model，indicating that the dimension of variables is reduced and the prediction accuracy is increased through nonlinear principal component analysis.It can be concluded that the NPCA-GA-BP prediction method has a higher prediction accuracy of stope stability，and has certain guiding significance for realizing intelligent mine.