Diverse models have been proposed for explaining the electrical performance of memristive devices. In principle, the behavior of internal variables associated to each one could be extracted from experimental results. ...Diverse models have been proposed for explaining the electrical performance of memristive devices. In principle, the behavior of internal variables associated to each one could be extracted from experimental results. In a former work, thermally grown TiOmemristive structures were built and characterized to obtain the constitutive relationship(magnetic flux versus charge). The aim of this work is to continue that analysis by determining the microscopic parameters within the frame of a simple model. We use the already obtained memristance dependence of time and the basic expressions from the non-linear model proposed by Strukov et al. to compute the state-variable,the mobility of the doping species, the speed of the boundary between the doped and the undoped regions, the voltages and the electric fields on the distinct regions. The power dissipation and its time evolution are also presented. Moreover, a quite different window function from those formerly proposed, which was estimated from experimental data, is also determined. This information provides a straightforward picture of the ionic transport during one cycle of a square voltage waveform within the framework of this simple model. Finally, a quality factor is proposed as the key parameter for actual memristors viewed under the same model.展开更多
为提高锂离子电池在复杂工况下的预测能力和建模精度,提出一种基于滑动窗口和长短时记忆(long short term memory,LSTM)神经网络的锂离子电池建模方法。首先建立了基于神经网络的锂离子电池模型,确定了神经网络的基本结构,通过LSTM层、...为提高锂离子电池在复杂工况下的预测能力和建模精度,提出一种基于滑动窗口和长短时记忆(long short term memory,LSTM)神经网络的锂离子电池建模方法。首先建立了基于神经网络的锂离子电池模型,确定了神经网络的基本结构,通过LSTM层、向量拼接层和全连接层分别实现了时序特征提取、特征融合和回归预测。然后提出了滑动窗口的输入向量处理方法,滑动窗口每次向前推进一个时间点,通过限制时间窗口内所能处理的最大信元数对数据量进行限制,为多个LSTM层的并行计算和深隐层的拼接层和全连接层预留了计算量的裕度,实现了对模型中循环网络层深度的优化选择。为解决模型在多工况下运行的泛化问题,提出使用离线数据集的预训练和在线数据的参数修正的训练方法,通过大量离线数据集的反复训练,使模型学习电池的共性部分;再使用部分在线数据,对网络参数进行调整,将其应用于预测中。最后使用恒流/恒压、随机电流脉冲、大功率脉冲等多个工况的数据分别进行测试。结果表明,基于长短时记忆神经网络的建模方法能够准确预测电池输出电压和荷电状态。展开更多
文摘Diverse models have been proposed for explaining the electrical performance of memristive devices. In principle, the behavior of internal variables associated to each one could be extracted from experimental results. In a former work, thermally grown TiOmemristive structures were built and characterized to obtain the constitutive relationship(magnetic flux versus charge). The aim of this work is to continue that analysis by determining the microscopic parameters within the frame of a simple model. We use the already obtained memristance dependence of time and the basic expressions from the non-linear model proposed by Strukov et al. to compute the state-variable,the mobility of the doping species, the speed of the boundary between the doped and the undoped regions, the voltages and the electric fields on the distinct regions. The power dissipation and its time evolution are also presented. Moreover, a quite different window function from those formerly proposed, which was estimated from experimental data, is also determined. This information provides a straightforward picture of the ionic transport during one cycle of a square voltage waveform within the framework of this simple model. Finally, a quality factor is proposed as the key parameter for actual memristors viewed under the same model.
文摘为提高锂离子电池在复杂工况下的预测能力和建模精度,提出一种基于滑动窗口和长短时记忆(long short term memory,LSTM)神经网络的锂离子电池建模方法。首先建立了基于神经网络的锂离子电池模型,确定了神经网络的基本结构,通过LSTM层、向量拼接层和全连接层分别实现了时序特征提取、特征融合和回归预测。然后提出了滑动窗口的输入向量处理方法,滑动窗口每次向前推进一个时间点,通过限制时间窗口内所能处理的最大信元数对数据量进行限制,为多个LSTM层的并行计算和深隐层的拼接层和全连接层预留了计算量的裕度,实现了对模型中循环网络层深度的优化选择。为解决模型在多工况下运行的泛化问题,提出使用离线数据集的预训练和在线数据的参数修正的训练方法,通过大量离线数据集的反复训练,使模型学习电池的共性部分;再使用部分在线数据,对网络参数进行调整,将其应用于预测中。最后使用恒流/恒压、随机电流脉冲、大功率脉冲等多个工况的数据分别进行测试。结果表明,基于长短时记忆神经网络的建模方法能够准确预测电池输出电压和荷电状态。
