(1. 武汉科技大学 资源与环境工程学院，武汉 430081；
2. 冶金矿产资源高效利用与造块湖北省重点实验室，武汉 430081)
摘 要: 为建立边坡坡态控制参数优化与边坡稳定性系数之间的非线性关系，提出SR-BP神经网络坡态控制参数优化模型，预测不同坡态控制参数优化方案下的边坡稳定性。以黄山某石灰石露天矿高边坡为例，采用强度折减法，计算不同坡态控制参数方案矩阵下的边坡稳定性系数，获得样本数据，提出改进的隐含层节点数求解经验公式，构建SR-BP神经网络坡态控制参数优化模型，并将平均绝对误差(MAE)、均方根误差(RMSE)以及相关系数(R)作为性能评价指标，分析实际样本值与模型预测值的相对误差。结果表明：改进的隐含层节点数求解经验公式充分考虑了输入层和输出层节点数对隐含层节点数的影响；SR-BP神经网络坡态控制参数优化模型表达了坡坡态控制参数优化与边坡稳定性系数之间的非线性关系，其实际样本值与模型预测值相对误差均在6%以内，且MAE为0.013，RMSE为0.026，R接近于1，证明模型拟合较好，预测精度较高。研究成果可为矿山坡态控制参数初步设计及优化提供一定的指导意义及理论基础。
关键字: 安全工程；坡态控制参数；稳定性系数；强度折减法(SR)；BP神经网络
（1. School of Resources and Environmental Engineering, Wuhan University of Science and Technology, Wuhan 430081, China;
2. Hubei Key Laboratory for Efficient Utilization and Agglomeration of Metallurgic Mineral Resources, Wuhan 430081, China）
Abstract:In order to establish the nonlinear relationship between the optimization of slope state control parameters and the slope stability coefficient, a strength reduction(SR)-BP neural network optimization model for slope state control parameters was proposed to predict the slope stability under different slope state control parameters optimization schemes. Taking the high slope of a limestone open-pit mine in Huangshan as an example, the strength reduction method was used to calculate the slope stability coefficient under the scheme matrix of different slope state control parameters, and the sample data are obtained. An improved empirical formula of hidden layer node number was proposed to construct the parameter optimization model of BP neural network for slope state control. And then mean absolute error (MAE), root mean square error (RMSE) and correlation coefficient (R) were used as performance evaluation indexes to analyze the relative error between actual sample value and model prediction value. The results show that, the influence of the number of nodes in input layer and output layer on the number of nodes in hidden layer is fully considered in the improved empirical formula; and the model of SR-BP neural network for slope state control parameter optimization expresses the nonlinear relationship between the optimization of slope state control parameters and the slope stability coefficient. The relative error between the actual sample value and the model prediction value is less than 6%, and MAE is 0.013, RMSE is 0.026, R is close to 1, which proves that the model fits well and the prediction accuracy is high. The research results can provide a certain guiding significance and theoretical basis for the preliminary design and optimization of mine slope control parameters.
Key words: safety engineering; slope control parameters; stability coefficient; strength reduction method(SR); BP neural network
