基于DE-MELM的土质边坡稳定性预测方法

Prediction Method of Soil Slope Stability Based on DE-MELM

为防止滑坡,避免发生事故,将差分进化(DE)算法与改进的极限学习机(ELM)有机组合,提出了一种基于DE-MELM的土质边坡稳定性预测方法。该方法首先在M估计基础上利用加权最小二乘方法计算ELM输出权值,以减少数据粗差对ELM预测的干扰;然后采用DE算法优化该ELM隐含层输入权值和偏差,以降低随机选取参数对预测性能的影响;最后通过所建立的DE-MELM土质边坡稳定性预测模型进行实例仿真验证。仿真验证结果表明:较之于标准ELM方法和基于M估计的ELM方法,所提出的DE-MELM方法仅经过15次迭代即可取得较为理想的预测精度,并对数据粗差具有较强的抗干扰能力,从而验证了其可行性和有效性。

To prevent landslide accidents,integrating Differential Evolution(DE)algorithm with a modified Extreme Learning Machine(ELM),this paper proposes a prediction method for soil slope stability.Firstly,by M estimation based weighted least squares method,the paper calculates the output weight of ELM to reduce interference of data gross error on prediction.Then,through introducing DE algorithm,the paper optimizes the input weight and deviation of ELM hidden layer to reduce influence of random parameter selection on prediction.Finally,using the built model,the paper predicts the stability of soil slopes.The simulation indicates that compared with ELM and ELM with M estimation,the proposed method can get better prediction accuracy through only 15 iterations,with anti-interference ability to data gross error,which verifies its feasibility and effectiveness.