摘要
针对经典智能算法用于滑坡位移预测时存在的网络结构参数选取复杂、易陷入局部极小等缺陷,提出了基于改进极限学习机ELM(Extreme Learning Machine)的滑坡位移预测模型。在滑坡变形位移状态辨识基础上,根据其位移变化特征,将滑坡位移曲线类型划分减速-匀速型、匀速-增速型、减速-匀速-增速型、复合型4类,将改进的ELM算法分别用于4种不同类型的滑坡位移预测。基于改进ELM算法构建滑坡位移预测模型时,采用二值区间搜索算法选定最佳隐含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、卧龙寺、古树屋、新滩滑坡体为例,对ELM预测的适用性进行讨论,实验结果表明,基于ELM构建不同类型滑坡位移预测模型时,具有较高的预测精度,且在网络学习速度等方面优势明显,适用于复杂状况下滑坡体的位移预测。
Considering complication to the parameter selection of conventional intelligent algorithm and being easy to fall into local minimum in prediction of landslide displacement,this paper proposes a new prediction model of landslide displacement based on improved Extreme Learning Machine.Characteristics of the landslide deformation accumulative displacement curve are studied and according to patterns,it is divided into four types：decelerating-uniform velocity,uniform velocity-accelerating,decelerating-uniform velocity-accelerating and compound type.The optimum number of neurons on hidden layer and excitation function of ELM are searched out according to the 2D range query-based and the technique of rolling modeling adopted in prediction in order to improve the prediction results.Taking landslides in Lianziya,Wolongsi,Gushuwu and Xintan examples,the applicability of improved ELM algorithm to landslide deformation displacement prediction is analyzed and compared.The experimental results show that ELM is valid and feasible in prediction of landslide under complicated conditions with higher precision.
作者
高彩云
高宁
GAO Cai-yun;GAO Ning(School of Geomatics and Urban Information,Henan University of Urban Construction,Pingdingshan 467036,China;Jiangxi Province Key Lab for Digital Land,East China Institute of Technology,Nanchang 330013,China)
出处
《西安科技大学学报》
CAS
北大核心
2018年第4期683-689,共7页
Journal of Xi’an University of Science and Technology
基金
国家自然科学基金(51474217)
江西省数字国土重点实验室开放研究基金(DLLJ201710
DLLJ201508)
河南省高等学校重点科研项目基金(18A420002
16A420001)
河南城建学院青年骨干教师资助项目(YCJQNGGJS201701)
河南城建学院学术技术带头人资助项目(YCJXSJSDTR201704)
2017年度河南省高等学校青年骨干教师培养计划(2017GGJS150)
关键词
滑坡位移
极限学习机
预测
landslide displacement
extreme learning machine
forecasting
