Static uniaxial compression tests were conducted on 16 granite specimens after thermal treatment using a heating device and an electro-hydraulic servo pressure-testing machine. The effects of air cooling and water coo...Static uniaxial compression tests were conducted on 16 granite specimens after thermal treatment using a heating device and an electro-hydraulic servo pressure-testing machine. The effects of air cooling and water cooling on the physical and mechanical properties of the high-temperature granite specimens were studied. Test results showed that the longitudinal wave velocities of the high-temperature specimens gradually decreased after they were cooled by water and air. The peak stress and elastic modulus of the samples decreased gradually with an increase in temperature, whereas their peak strain increased gradually. The effects of peak stress and peak strain were considerably more evident when cooling by water than by air. This result demonstrated that the thermal cracking of the granite specimens, and consequently, their internal micro-fractures, further developed when the specimens were cooled by water.展开更多
为了进一步提高水位预测的准确性,本文提出一种融入改进注意力机制的长短期记忆网络(Long Short Time Memory,LSTM)预测模型。该模型将输入序列拆分为时间序列和特征序列,在LSTM网络模型前引入注意力机制对两个序列分别进行注意力计算,...为了进一步提高水位预测的准确性,本文提出一种融入改进注意力机制的长短期记忆网络(Long Short Time Memory,LSTM)预测模型。该模型将输入序列拆分为时间序列和特征序列,在LSTM网络模型前引入注意力机制对两个序列分别进行注意力计算,然后再进行融合,LSTM网络能够根据重要程度自适应地选择最重要的输入特征,注意力机制层的参数通过竞争随机搜索算法获取,从而进一步增强了模型的鲁棒性。最后在鄱阳湖的水位数据上进行预测实验,结果表明:相对于支持向量回归(SVR)、LSTM等模型,本文提出基于改进注意力机制的LSTM模型具有更好的预测精度,可为水位预测和水资源的精准调度提供技术支持。展开更多
多数传统语音增强算法是基于平稳噪声下分析的,且没有从语音质量及可懂度角度全面衡量增强性能。因此,提出了基于多窗谱估计与归一化最小均方(Normalized Least Mean Square,NLMS)自适应滤波算法的单通道语音增强方案。首先利用多窗谱...多数传统语音增强算法是基于平稳噪声下分析的,且没有从语音质量及可懂度角度全面衡量增强性能。因此,提出了基于多窗谱估计与归一化最小均方(Normalized Least Mean Square,NLMS)自适应滤波算法的单通道语音增强方案。首先利用多窗谱估计谱减法(Multiwindow Spectral Subtraction,MSS)解决谱减法产生的“音乐噪声”问题;其次将估计出的期望信号与纯净参考信号的差值作为误差信号,由自适应滤波的NLMS算法代替传统的最小均方(Least Mean Square,LMS)算法,以降低滤波器成本及运算时间,求取滤波器权系数值,并不断迭代更新修正滤波器;最后分析了所提算法在不同噪声环境下的增强性能,并与传统的各种谱减算法对比,从语音质量及可懂度出发衡量语音增强效果。结果表明,所提算法的增强效果优于各类谱减法。展开更多
基金Supported by"The Training Plan of College Students'Creation"(2017A53449)in Jilin UniversityNew Energy Item of Jilin Province Combining with Universities(SF2017-5-5)
文摘Static uniaxial compression tests were conducted on 16 granite specimens after thermal treatment using a heating device and an electro-hydraulic servo pressure-testing machine. The effects of air cooling and water cooling on the physical and mechanical properties of the high-temperature granite specimens were studied. Test results showed that the longitudinal wave velocities of the high-temperature specimens gradually decreased after they were cooled by water and air. The peak stress and elastic modulus of the samples decreased gradually with an increase in temperature, whereas their peak strain increased gradually. The effects of peak stress and peak strain were considerably more evident when cooling by water than by air. This result demonstrated that the thermal cracking of the granite specimens, and consequently, their internal micro-fractures, further developed when the specimens were cooled by water.
文摘为了进一步提高水位预测的准确性,本文提出一种融入改进注意力机制的长短期记忆网络(Long Short Time Memory,LSTM)预测模型。该模型将输入序列拆分为时间序列和特征序列,在LSTM网络模型前引入注意力机制对两个序列分别进行注意力计算,然后再进行融合,LSTM网络能够根据重要程度自适应地选择最重要的输入特征,注意力机制层的参数通过竞争随机搜索算法获取,从而进一步增强了模型的鲁棒性。最后在鄱阳湖的水位数据上进行预测实验,结果表明:相对于支持向量回归(SVR)、LSTM等模型,本文提出基于改进注意力机制的LSTM模型具有更好的预测精度,可为水位预测和水资源的精准调度提供技术支持。
文摘多数传统语音增强算法是基于平稳噪声下分析的,且没有从语音质量及可懂度角度全面衡量增强性能。因此,提出了基于多窗谱估计与归一化最小均方(Normalized Least Mean Square,NLMS)自适应滤波算法的单通道语音增强方案。首先利用多窗谱估计谱减法(Multiwindow Spectral Subtraction,MSS)解决谱减法产生的“音乐噪声”问题;其次将估计出的期望信号与纯净参考信号的差值作为误差信号,由自适应滤波的NLMS算法代替传统的最小均方(Least Mean Square,LMS)算法,以降低滤波器成本及运算时间,求取滤波器权系数值,并不断迭代更新修正滤波器;最后分析了所提算法在不同噪声环境下的增强性能,并与传统的各种谱减算法对比,从语音质量及可懂度出发衡量语音增强效果。结果表明,所提算法的增强效果优于各类谱减法。
