为了提高高速车辆-道岔复杂动力学响应的仿真计算效率和计算精度,构建了基于长短时记忆(Long Short Term Memory,LSTM)网络的动力学响应预测模型。首先,利用刚柔耦合动力学模型,生成不同水平不平顺状态下的轮轨垂向力样本。然后,将水平...为了提高高速车辆-道岔复杂动力学响应的仿真计算效率和计算精度,构建了基于长短时记忆(Long Short Term Memory,LSTM)网络的动力学响应预测模型。首先,利用刚柔耦合动力学模型,生成不同水平不平顺状态下的轮轨垂向力样本。然后,将水平不平顺幅值作为模型输入,轮轨垂向力作为输出,引入dropout参数,训练LSTM网络并采用均方误差和决定系数来评价模型性能。最后,基于计算结果分析轮重减载率和统计超限概率。结果表明:本文搭建的LSTM网络在训练轮次达到5000次后,均方误差基本稳定在0.00267,测试集的决定系数为0.903,模型具有较高的可靠性,可用于预测不平顺状态下的车辆-道岔动力学响应;与传统动力学模型计算相比,LSTM模型计算效率提高了约26倍,大幅提高了计算效率且计算精度达到要求;水平不平顺幅值达到6 mm时,超限概率达到了9.08%,超过了容许阈值。展开更多
The systematical and scalable frameworks were provided for estimating the blocking probabilities under asynchronous traffic in optical burst switching(OBS) nodes with limited wavelength conversion capability(LWCC) . T...The systematical and scalable frameworks were provided for estimating the blocking probabilities under asynchronous traffic in optical burst switching(OBS) nodes with limited wavelength conversion capability(LWCC) . The relevant system architectures of limited range and limited number of wavelength converters(WCs) deployed by a share-per-fiber(SPF) mode were developed,and the novel theoretical analysis of node blocking probability was derived by combining the calculation of discouraged arrival rate in a birth-death process and two-dimensional Markov chain model of SPF. The simulation results on single node performance verify the accuracy and effectiveness of the analysis models. Under most scenarios,it is difficult to distinguish the plots generated by the analysis and simulation. As the conversion degree increases,the accuracy of the analysis model worsens slightly. However,the utmost error on burst loss probability is far less than one order of magnitude and hence,still allows for an accurate estimate. Some results are of actual significance to the construction of next-generation commercial OBS backbones.展开更多
针对量子粒子群算法在处理自变量具有有限定义域的问题时易陷入局部最优解的问题,对算法的量子模型加以改进,提出了基于非对称势的量子粒子群算法(asymmetric potential well based quantum particle swarm optimization,AQPSO)。该算...针对量子粒子群算法在处理自变量具有有限定义域的问题时易陷入局部最优解的问题,对算法的量子模型加以改进,提出了基于非对称势的量子粒子群算法(asymmetric potential well based quantum particle swarm optimization,AQPSO)。该算法认为粒子处于非对称势阱中,势阱的参数由当前的最优位置和自变量的定义域共同决定。而在求解粒子在空间分布的波函数时,又采用了参数消减方法,只需人工指定越限概率,简化算法流程。最后,通过算例验证,该方法的全局搜索能力显著提升,在处理高维、复杂、强干扰性问题时,具有显著优势。展开更多
文摘为了提高高速车辆-道岔复杂动力学响应的仿真计算效率和计算精度,构建了基于长短时记忆(Long Short Term Memory,LSTM)网络的动力学响应预测模型。首先,利用刚柔耦合动力学模型,生成不同水平不平顺状态下的轮轨垂向力样本。然后,将水平不平顺幅值作为模型输入,轮轨垂向力作为输出,引入dropout参数,训练LSTM网络并采用均方误差和决定系数来评价模型性能。最后,基于计算结果分析轮重减载率和统计超限概率。结果表明:本文搭建的LSTM网络在训练轮次达到5000次后,均方误差基本稳定在0.00267,测试集的决定系数为0.903,模型具有较高的可靠性,可用于预测不平顺状态下的车辆-道岔动力学响应;与传统动力学模型计算相比,LSTM模型计算效率提高了约26倍,大幅提高了计算效率且计算精度达到要求;水平不平顺幅值达到6 mm时,超限概率达到了9.08%,超过了容许阈值。
基金Project(60632010) supported by the National Natural Science Foundation of China
文摘The systematical and scalable frameworks were provided for estimating the blocking probabilities under asynchronous traffic in optical burst switching(OBS) nodes with limited wavelength conversion capability(LWCC) . The relevant system architectures of limited range and limited number of wavelength converters(WCs) deployed by a share-per-fiber(SPF) mode were developed,and the novel theoretical analysis of node blocking probability was derived by combining the calculation of discouraged arrival rate in a birth-death process and two-dimensional Markov chain model of SPF. The simulation results on single node performance verify the accuracy and effectiveness of the analysis models. Under most scenarios,it is difficult to distinguish the plots generated by the analysis and simulation. As the conversion degree increases,the accuracy of the analysis model worsens slightly. However,the utmost error on burst loss probability is far less than one order of magnitude and hence,still allows for an accurate estimate. Some results are of actual significance to the construction of next-generation commercial OBS backbones.
文摘针对量子粒子群算法在处理自变量具有有限定义域的问题时易陷入局部最优解的问题,对算法的量子模型加以改进,提出了基于非对称势的量子粒子群算法(asymmetric potential well based quantum particle swarm optimization,AQPSO)。该算法认为粒子处于非对称势阱中,势阱的参数由当前的最优位置和自变量的定义域共同决定。而在求解粒子在空间分布的波函数时,又采用了参数消减方法,只需人工指定越限概率,简化算法流程。最后,通过算例验证,该方法的全局搜索能力显著提升,在处理高维、复杂、强干扰性问题时,具有显著优势。
