在多元时间序列预测方法中,传统的模型无法敏锐地捕获时间序列短期突变信号从而导致预测趋势滞后和误差较大。本文提出了一种基于时钟触发长短期记忆(Clockwork Triggered Long Short Term Memory,CWTLSTM)网络的多元时序预测模型,通过...在多元时间序列预测方法中,传统的模型无法敏锐地捕获时间序列短期突变信号从而导致预测趋势滞后和误差较大。本文提出了一种基于时钟触发长短期记忆(Clockwork Triggered Long Short Term Memory,CWTLSTM)网络的多元时序预测模型,通过增强对短期信息的捕获能力提高了预测精度。CWTLSTM将网络中所有的神经元进行分组,对每个分组赋予不同的激活频率,每一组神经元只在时间步长等于周期的整数倍时才被激活。根据周期是否为1将网络分为主干网络链和短期输入增强链,短期输入增强链在靠近输出位置的时间步上激活时,将输入信息的运算结果单向地传递给主干网络链,增强此时的输入权重,使模型在存储长期信息的基础上能快速响应短期突变信息带来的数据波动。在空气污染数据集和水泥篦冷机数据集上的验证结果表明,本文模型在减少预测误差与趋势判断上均有较好的表现。展开更多
Based on the spherical cavity expansion theory in the elastic half space,the ground surface movement characteristics of shallowly buried explosions are analyzed.The results show that the induced seismic wave is a long...Based on the spherical cavity expansion theory in the elastic half space,the ground surface movement characteristics of shallowly buried explosions are analyzed.The results show that the induced seismic wave is a longitudinal wave in the near zone and a Rayleigh wave in the far zone.The maximum displacement(velocity) of the longitudinal wave and the Rayleigh wave are inversely proportional to the scaled distance,and can be described by exponential function with exponents equal to 1.4 and 0.5,respectively.The vibration frequencies of the waves have almost no change.The vibration frequency of the longitudinal wave approximates the natural vibration frequency of the cavity in the broken area,and the vibration frequency of the Rayleigh wave is about half that of the longitudinal wave.On the same reduced buried depth and reduced distance,the particle displacement is directly proportional to the product of the boundary loading and cavity radius,and is inversely proportional to the transversal wave velocity.Meanwhile,the particle velocity is directly proportional to the boundary loading and inversely proportional to the wave velocity ratio.In the far zone,the buried depth of the explosive only has a slight effect on the longitudinal wave,but has a larger effect on the Rayleigh wave.展开更多
文摘在多元时间序列预测方法中,传统的模型无法敏锐地捕获时间序列短期突变信号从而导致预测趋势滞后和误差较大。本文提出了一种基于时钟触发长短期记忆(Clockwork Triggered Long Short Term Memory,CWTLSTM)网络的多元时序预测模型,通过增强对短期信息的捕获能力提高了预测精度。CWTLSTM将网络中所有的神经元进行分组,对每个分组赋予不同的激活频率,每一组神经元只在时间步长等于周期的整数倍时才被激活。根据周期是否为1将网络分为主干网络链和短期输入增强链,短期输入增强链在靠近输出位置的时间步上激活时,将输入信息的运算结果单向地传递给主干网络链,增强此时的输入权重,使模型在存储长期信息的基础上能快速响应短期突变信息带来的数据波动。在空气污染数据集和水泥篦冷机数据集上的验证结果表明,本文模型在减少预测误差与趋势判断上均有较好的表现。
基金Science Fund for Creative Research Group of the National Natural Science Foundation of China under Grant No.51021001China Postdoctoral Science Foundation under Grant No.2013M541675National Natural Science Foundation of China under Grant No.51309233
文摘Based on the spherical cavity expansion theory in the elastic half space,the ground surface movement characteristics of shallowly buried explosions are analyzed.The results show that the induced seismic wave is a longitudinal wave in the near zone and a Rayleigh wave in the far zone.The maximum displacement(velocity) of the longitudinal wave and the Rayleigh wave are inversely proportional to the scaled distance,and can be described by exponential function with exponents equal to 1.4 and 0.5,respectively.The vibration frequencies of the waves have almost no change.The vibration frequency of the longitudinal wave approximates the natural vibration frequency of the cavity in the broken area,and the vibration frequency of the Rayleigh wave is about half that of the longitudinal wave.On the same reduced buried depth and reduced distance,the particle displacement is directly proportional to the product of the boundary loading and cavity radius,and is inversely proportional to the transversal wave velocity.Meanwhile,the particle velocity is directly proportional to the boundary loading and inversely proportional to the wave velocity ratio.In the far zone,the buried depth of the explosive only has a slight effect on the longitudinal wave,but has a larger effect on the Rayleigh wave.
