边坡变形的LMD-ELM多尺度预测模型研究

• 作 者: 李胜 韩永亮 高宏 罗明坤 胡海永
• 机 构: 辽宁工程技术大学矿业学院,煤科集团沈阳研究院有限公司,
• 出 处: 中国安全科学学报
• 发表日期: 2015年11期
• 关键词: LMD-ELM 尺度 边坡变形 预测模型 周期项 预测模型研究 时间序列 均值 边坡变形监测 随机项

为准确预测边坡变形,有效预防边坡灾害发生,提出构建基于局域均值分解(LMD)和极限学习机(ELM)的边坡变形多尺度预测模型。用LMD方法,将边坡变形时间序列分解为多尺度且相对平稳的随机项、周期项和趋势项。针对各项时间序列,分别构建基于ELM的预测模型。经累加各分项预测值,获得模型最终预测结果。以甘肃某边坡变形为案例,进行实证分析。结果表明:LMD-ELM模型能够充分挖掘数据内部隐含的变形规律,有效诠释多尺度变形与其诱发因素间复杂的响应关系,预测精度、运行速度和拟合泛化能力较其他模型有所提高。

In order to accurately predict the slope deformation and effectively prevent the occurrence of slope disasters, a multiscale slope deformation prediction model based on Local Mean Decomposition (LMD) and Extreme Learning Machine (ELM) is proposed. Using the LMD method, the slope deformation time series is decomposed into multi-scale and relatively stable random items, periodic items and trend items. For each time series, the ELM-based prediction model is constructed. After accumulating the predicted values ​​of each item, the final prediction result of the model is obtained. Deformation of a slope in Gansu Province as a case, an empirical analysis. The results show that the LMD-ELM model can fully exploit the inherent deformation rules in the data and effectively interpret the complex response relationship between multi-scale deformation and its evoked factors. The prediction accuracy, operating speed and generalization ability are improved compared with other models.