基于门控循环单元与误差修正的短期负荷预测

针对负荷预测模型迭代训练过程中存在误差积累的问题,提出结合叠式双向门控循环单元（SBiGRU）、完整自适应噪声集成经验模态分解（CEEMDAN）和误差修正的组合预测模型. 建立SBiGRU模型学习在气温、日期类型影响下负荷序列的时序特征,误差特征体现在SBiGRU模型预测产生的误差序列中;使用CEEMDAN算法将误差序列分解为数个本征模态函数（IMF）分量与趋势分量,对每项分量再次建立SBiGRU模型进行学习与预测,并对各分量的预测值进行序列重构,得到误差的预测结果;对预测结果进行求和以修正误差. 模型评估结果表明,组合模型的预测准确精度为98.86%,与SBiGRU、BiRNN、支持向量回归等方法相比,该模型具有更好的精度.     

基于VMD与改进QRGRU的超短期风电功率概率预测

风电功率概率预测是分析未来风电功率不确定性的有效方法之一。为提高风电功率概率预测精度,文中提出基于变分模态分解(VMD)与改进门控循环单元分位数回归(QRGRU)的超短期风电功率概率预测方法。首先,采用VMD将原始风电功率序列分解成不同特征的模态函数;然后,对每个模态函数分别建立基于QRGRU的概率预测模型,并将变量间的网络结构约束作为目标函数的惩罚项,改进QRGRU权重在迭代修正过程中的平稳性;最后,在不同分位数条件下叠加各个模态函数预测值,并采用非参数核密度估计方法得到未来风电功率的概率密度函数。结合某风电场实测数据开展具体算例分析,结果表明所提方法能够兼顾区间覆盖率,减少区间宽度,在不同预测步长中均能表现较好的预测效果。     

应用长短期记忆循环神经网络的弱反射信号增强方法

由于沉积环境的特殊性和复杂性,地下介质中不同反射界面的波阻抗差可能差异巨大。如果储层的有效反射信息较弱,在地震数据中极可能被强反射信息掩盖,不易被识别,影响了储层识别效果,因此亟需一种解释性处理技术突出弱反射信息。常规方法一般是先从地震数据中分离出强反射分量,再将它削弱或删除。但如果地震子波提取不准确,减去法中强反射残留会引入虚假信号。文中提出了一种“升弱降强”的新思路,通过构建幂次反射系数映射模型缩小弱反射信号与强反射信号的相对差异。首先计算测井反射系数的幂次反射系数,将弱反射系数相对增大、强反射系数相对减小,得到拟反射系数序列;再用原始反射系数序列和拟反射系数序列分别与地震子波进行褶积运算,得到合成地震记录和拟合成地震记录,生成训练样本集;然后用该样本集训练长短期记忆(LSTM)循环神经网络,建立合成地震记录与拟合成地震记录的映射关系;最后将该网络应用于地震数据,增强了地震弱反射信号。模型和实际数据应用结果表明,该方法能有效增强地层本身引起的弱反射信号,提高地震数据的储层识别能力。     

Soft sensor modeling method based on hybrid model of nearest neighbor and neural network

Soft-sensing modeling can effectively solve the problems of large measurement lag, high price, and complex maintenance of online analytical instruments in the production process. At present, neural network based on data-driven is one of the main tools of soft sensor. In the process of modeling data collection, the collection of dominant variables is much more difficult than that of auxiliary variables, resulting in a large amount of unlabeled data. However, traditional soft sensor modeling methods ignore these unlabeled data and only use a small amount of labeled data for modeling, which has negative effect on the prediction accuracy of the model. To solve the problem of label missing, the nearest neighbor algorithm is used to pseudo label the unlabeled data. At the same time, a network structure is designed by combining convolution operation and gated recurrent unit neural network (GRU) to further utilize the unlabeled data, extract the dynamic feature from data at different time, and improve the prediction accuracy of the neural network. Finally, the method is applied to the prediction of propane concentration on the top of propylene distillation column. The results show that the model can solve the problem of label missing in the nonlinear dynamic system and has higher prediction accuracy.     

基于深度学习的锂离子电池SOC和SOH联合估算

锂离子电池常被作为储能元件以实现电能的存储和转化,然而其荷电状态(state of charge,SOC)和健康状态(state of health,SOH)无法被直接测量。为了实现锂离子电池SOC和SOH联合估算,该文分析SOC和SOH之间的关联性,并提出一种基于深度学习的锂离子电池SOC和SOH联合估算方法。该方法能够基于门控循环单元循环神经网络(recurrent neural network with gated recurrent unit,GRU-RNN)和卷积神经网络(convolutional neural network,CNN),利用锂离子电池电压、电流、温度,实现锂离子电池全使用周期内SOC和SOH的同时估算,而且由于将锂离子电池的SOH估算值考虑到SOC估算中,能够消除锂离子电池老化因素对锂离子电池SOC估算造成的负面影响,从而提升SOC估算精度。两个锂离子电池测试数据集上的实验结果表明,提出的估算方法能够在不同温度和不同工况下实现锂离子电池全使用周期SOC和SOH联合估算,且获得较高的精度。     

基于改进卷积双向门控循环网络的轴承故障诊断

针对传统深度学习方法没有充分利用轴承信号的时序特点,以及难以处理动态数据的问题,提出一种基于改进卷积双向门控循环神经网络的轴承故障智能诊断方法。采用卷积神经网络从输入信号中提取代表性特征,引入双向门控循环神经网络挖掘故障数据在时间维度上的语义信息,通过注意力机制自适应地对特征图通道赋予不同权值,从而实现高精度的轴承故障诊断。在公开轴承数据集上进行实验,实验结果表明,该方法能够正确地将轴承故障分类,分类精度可达996%。     

基于递归过程神经网络的航空发动机滑油系统状态监测

针对航空发动机滑油系统状态监测问题,提出了递归过程神经网络模型。其隐层和输出层为过程神经元,该网络的输入信号为时变函数或过程,并且含有一个特别的关联层,在建模过程中能储存系统过去更多时刻的状态信息,使得网络结构适于预测时间序列问题。文中给出了相应的学习算法,并且分别利用人工神经网络和递归过程神经网络对航空发动机滑油系统状态进行预测。结果表明,递归过程神经网络预测精度高,优于传统人工神经网络的预测能力。为航空发动机滑油系统状态监测问题提供了一种有效的方法。     

Short-Term Traffic Flow Prediction Based on Road Network Topology

Accurate short-term traffic flow prediction plays a crucial role in intelligent transportation system (ITS), because it can assist both traffic authorities and individual travelers make better decisions. Previous researches mostly focus on shallow traffic prediction models, which performances were unsatisfying since short-term traffic flow exhibits the characteristics of high nonlinearity, complexity and chaos. Taking the spatial and temporal correlations into consideration, a new traffic flow prediction method is proposed with the basis on the road network topology and gated recurrent unit (GRU). This method can help researchers without professional traffic knowledge extracting generic traffic flow features effectively and efficiently. Experiments are conducted by using real traffic flow data collected from the Caltrans Performance Measurement System (PEMS) database in San Diego and Oakland from June 15, 2017 to September 27, 2017. The results demonstrate that our method outperforms other traditional approaches in terms of mean absolute percentage error (MAPE), symmetric mean absolute percentage error (SMAPE) and root mean square error (RMSE).     

Nonuniform sampling theorems for random signals in the linear canonical transform domain

Nonuniform sampling can be encountered in various practical processes because of random events or poor timebase. The analysis and applications of the nonuniform sampling for deterministic signals related to the linear canonical transform (LCT) have been well considered and researched, but up to now no papers have been published regarding the various nonuniform sampling theorems for random signals related to the LCT. The aim of this article is to explore the nonuniform sampling and reconstruction of random signals associated with the LCT. First, some special nonuniform sampling models are briefly introduced. Second, based on these models, some reconstruction theorems for random signals from various nonuniform samples associated with the LCT have been derived. Finally, the simulation results are made to prove the accuracy of the sampling theorems. In addition, the latent real practices of the nonuniform sampling for random signals have been also discussed.     

基于多反馈环结构提高硬件储备池记忆能力

针对单反馈环硬件储备池记忆能力不足的问题,提出一种基于多反馈环结构提高储备池记忆能力的方法.通过增加反馈环将更早输入信号产生的响应重新注入回储备池,使更早期的输入信号也影响储备池的内部动态,从而增强储备池的记忆能力.仿真研究了单个和多个反馈环结构的储备池的记忆能力及其在需要更长记忆能力的NARMA30任务中的预测性能.结果表明,把反馈环增加到10个,虚节点数为50时储备池的记忆能力由单反馈环的18.2提高到40.2.仅用两个反馈环,虚节点数为1000时NARMA30预测的归一化均方根误差可从单反馈环的0.27降到0.09.说明通过合理设置多个反馈环的参数,可以设计出任务需要的特定记忆能力,部分解决了储备池的适应性问题.