遗传算法优化BP神经网络的岩质边坡稳定性预测

为了解决边坡工程中非线性变化给稳定性预测造成的困难,建立了GA-BP神经网络计算模型预测岩质边坡稳定性。采用定性评价和相互作用矩阵复核的方式,选取边坡坡度、边坡高度、斜坡结构类型、岩体强度、控滑结构面倾角、岩体结构特征、地表变形强度、人类活动强度8个评价因子作为BP神经网络的输入变量;利用遗传算法对神经网络的初始权值和阈值进行优化后训练岩质边坡稳定性预测模型;对比分析GA-BP神经网络和BP神经网络的预测效果。结果表明,优化后的预测结果误差绝对值小于0.15的占85%,未优化的传统神经网络仅占45%,优化后的预测结果更加接近真实值,表明遗传算法对传统BP神经网络的优化是有效的。研究结果对建立岩质边坡稳定性预测模型具有一定的参考价值。

Prediction of rock slope stability based on BP neural network optimized by genetic algorithm

In order to resolve the difficulty of stability prediction caused by nonlinear change in slope engineering, a GA-BP neural network model was established to predict the stability of the rock slope. Firstly, eight evaluation factors, including slope, height, slope structure type, rock mass strength, angle of sliding control structure plane, rock mass structure characteristics, surface deformation strength and human activity intensity, were selected as input variables of BP neural network by qualitative evaluation and interaction matrix review. Secondly, the initial weights and thresholds of neural network were optimized by genetic algorithm to train the prediction model of rock slope stability. Finally, the prediction effects of GA-BP neural network and BP neural network were compared and analyzed. Results show that 85% of the optimized prediction error absolute values are less than 0.15, while non-optimized errors only account for 45% of traditional neural network. The optimized prediction results are more close to the real value which indicates that the genetic algorithm is effective for the traditional BP neural network optimization. The study can be consulted for establishing prediction models of rock slope stability.