基于新型智能算法ELM的滑坡变形位移预测  

Displacement prediction of landslide based on new intelligent algorithm of ELM

作　　者：高彩云[1,2] 

机构地区：[1]河南城建学院测绘与城市空间信息学院,河南平顶山467036 [2]东华理工大学江西省数字国土重点实验室,江西南昌330013

出　　处：《人民长江》2017年第7期46-49,69,共5页Yangtze River

基　　金：国家自然科学基金项目(51474217);江西省数字国土重点实验室开放研究基金项目(DLLJ201508);矿山空间信息技术国家测绘地理信息局重点实验室基金项目(KLM201306);河南省高等学校重点科研基金项目(16A420001)

摘　　要：针对采用经典智能算法进行滑坡变形预测时存在学习速度慢、网络结构参数选取复杂等问题,构建了基于新型智能算法ELM(Extreme Learning Machine)的滑坡位移预测模型,采用二值区间搜索算法选定最佳隐层神经元个数和激励函数,并融入数据滚动建模思想,以期提高网络泛化能力和预测精度。以链子崖、古树屋两滑坡体为例,将ELM与经典智能算法LMBP、RBF的预测效果进行对比,算例结果表明:ELM算法具有较高的预测精度,且在网络学习速度等方面优势明显。Considering slow learning speed and complex selection of network structural parameters of conventional intelligent algorithm in landslide displacement prediction, a prediction model for landslide displacement based on Extreme Learning Machine （ELM） is presented in this paper. The number of optimum neurons on hidden layer and excitation function of ELM are deter- mined according to the 2D range search algorithm and the technique of rolling modeling is adopted in prediction in order to improve the network generalization ability and prediction accuracy. Finally, taking Lianziya landslide and Gushuwu landslide as the case, a comparative study was carried out between ELM models with conventional algorithms like LMBP and RBF respectively. The results show that the ELM algorithm has higher accuracy and better network learning speed.

关 键 词：ELM 隐层神经元 激励函数 滑坡变形 

分 类 号：P642[天文地球—工程地质学]