基于Attention-Based LSTM算法的文本分类模型

本次研究针对文本数据处理工作中的文本分类项目提出了一套基于Attention-Based LSTM算法的分类模型,根据Attention-Model的基本原理对Attention-Based LSTM算法数据处理方式进行了详细介绍。最后将Attention-Based LSTM算法应用于来自国内外主流门户网站文本数据的分类处理工作。经统计分析发现,Attention-Based LSTM算法相比于常规LSTM算法和Bi-LSTM体现出了更高的分类准确率水平,在文本数据处理方面具有一定的应用价值。

一种改进的Attention-Based LSTM特征选择模型

为了对高校学术活动进行信息提取和文本分类,基于结合注意力机制的长短期记忆网络(Attention-Based Long Short-Term Memory)特征选择模型,构建了高校学术活动分类系统。通过大量分析高校学术活动语料的特点,准确抽取学术活动的相关内容,改善了文本数据质量;提出了一种改进的Attention-Based LSTM特征选择模型,降低了数据维度,有效地突出了重点信息。实验结果表明,该方法提高了分类的准确率,其分类效果明显优于普通LSTM(Long Short-Term Memory)模型和传统模型的处理结果。

A Health State Prediction Model Based on Belief Rule Base and LSTM for Complex Systems

In industrial production and engineering operations,the health state of complex systems is critical,and predicting it can ensure normal operation.Complex systems have many monitoring indicators,complex coupling structures,non-linear and time-varying characteristics,so it is a challenge to establish a reliable prediction model.The belief rule base(BRB)can fuse observed data and expert knowledge to establish a nonlinear relationship between input and output and has well modeling capabilities.Since each indicator of the complex system can reflect the health state to some extent,the BRB is built based on the causal relationship between system indicators and the health state to achieve the prediction.A health state prediction model based on BRB and long short term memory for complex systems is proposed in this paper.Firstly,the LSTMis introduced to predict the trend of the indicators in the system.Secondly,the Density Peak Clustering(DPC)algorithmis used todetermine referential values of indicators for BRB,which effectively offset the lack of expert knowledge.Then,the predicted values and expert knowledge are fused to construct BRB to predict the health state of the systems by inference.Finally,the effectiveness of the model is verified by a case study of a certain vehicle hydraulic pump.

Encoder-Decoder Based LSTM Model to Advance User QoE in 360-Degree Video

The development of multimedia content has resulted in a massiveincrease in network traffic for video streaming. It demands such types ofsolutions that can be addressed to obtain the user’s Quality-of-Experience(QoE). 360-degree videos have already taken up the user’s behavior by storm.However, the users only focus on the part of 360-degree videos, known as aviewport. Despite the immense hype, 360-degree videos convey a loathsomeside effect about viewport prediction, making viewers feel uncomfortablebecause user viewport needs to be pre-fetched in advance. Ideally, we canminimize the bandwidth consumption if we know what the user motionin advance. Looking into the problem definition, we propose an EncoderDecoder based Long-Short Term Memory (LSTM) model to more accuratelycapture the non-linear relationship between past and future viewport positions. This model takes the transforming data instead of taking the direct inputto predict the future user movement. Then, this prediction model is combinedwith a rate adaptation approach that assigns the bitrates to various tiles for360-degree video frames under a given network capacity. Hence, our proposedwork aims to facilitate improved system performance when QoE parametersare jointly optimized. Some experiments were carried out and compared withexisting work to prove the performance of the proposed model. Last but notleast, the experiments implementation of our proposed work provides highuser’s QoE than its competitors.

Night Vision Object Tracking System Using Correlation Aware LSTM-Based Modified Yolo Algorithm

Improved picture quality is critical to the effectiveness of object recog-nition and tracking.The consistency of those photos is impacted by night-video systems because the contrast between high-profile items and different atmospheric conditions,such as mist,fog,dust etc.The pictures then shift in intensity,colour,polarity and consistency.A general challenge for computer vision analyses lies in the horrid appearance of night images in arbitrary illumination and ambient envir-onments.In recent years,target recognition techniques focused on deep learning and machine learning have become standard algorithms for object detection with the exponential growth of computer performance capabilities.However,the iden-tification of objects in the night world also poses further problems because of the distorted backdrop and dim light.The Correlation aware LSTM based YOLO(You Look Only Once)classifier method for exact object recognition and deter-mining its properties under night vision was a major inspiration for this work.In order to create virtual target sets similar to daily environments,we employ night images as inputs;and to obtain high enhanced image using histogram based enhancement and iterative wienerfilter for removing the noise in the image.The process of the feature extraction and feature selection was done for electing the potential features using the Adaptive internal linear embedding(AILE)and uplift linear discriminant analysis(ULDA).The region of interest mask can be segmen-ted using the Recurrent-Phase Level set Segmentation.Finally,we use deep con-volution feature fusion and region of interest pooling to integrate the presently extremely sophisticated quicker Long short term memory based(LSTM)with YOLO method for object tracking system.A range of experimentalfindings demonstrate that our technique achieves high average accuracy with a precision of 99.7%for object detection of SSAN datasets that is considerably more than that of the other standard object detection mechanism.Our approach may therefore satisfy the true demands of night scene target detection applications.We very much believe that our method will help future research.

Mathematical methods for main tenance and operation cost prediction based on transfer learning in State Grid

The electric power enterprise is an important basic energy industry for national development,and it is also the first basic industry of the national economy.With the continuous expansion of State Grid,the progressively complex operating conditions,and the increasing scope and frequency of data collection,how to make reasonable use of electrical big data,improve utilization,and provide a theoretical basis for the reliability of State Grid operation,has become a new research hot spot.Since electrical data has the characteristics of large volume,multiple types,low-value density,and fast processing speed,it is a challenge to mine and analyze it deeply,extract valuable information efficiently,and serve for the actual problem.According to the features of these data,this paper uses artificial intelligence methods such as time series and support vector regression to establish a data mining network model for standard cost prediction through transfer learning.The experimental results show that the model in this paper obtains better prediction results on a small sample data set,which verifies the feasibility of the deep transfer model.Compared with activity-based costing and the traditional prediction method,the average absolute error of the proposed method is reduced by 10%,which is effective and superior.

基于多头自注意力机制的LSTM-TCN基站流量预测算法

基站流量预测对于蜂窝网络的规划、资源分配和用户体验优化至关重要。为提高基站流量预测精度,文中设计一种结合多头自注意机制(MHSA)的LSTM-TCN基站流量预测算法。其中:MHSA能够从多个角度强化基站流量数据的内在关联,增强了模型对流量数据重要特征的表达能力;LSTM-TCN模型中长短期记忆(LSTM)网络捕捉流量数据中的长短时依赖性;时间卷积网络(TCN)进一步捕捉流量数据中的全局特征,使得模型能够提取基站流量数据在不同时间尺度上的变化模式和时间依赖关系,提高基站流量预测模型的拟合能力和预测精度。实验结果表明,该流量预测算法与其他算法相比,在运营商基站流量数据的预测中有效降低了均方根误差和平均绝对误差,提高了决定系数,验证了该流量预测算法的有效性,从而为基站休眠节能提供决策支持。

ISSA优化Attention双向LSTM的短期电力负荷预测

针对短期电力负荷数据的复杂性和多样性,提出一种含Attention的双向LSTM预测方法,简称Bi-LSTM-AT。该方法将电力负荷历史数据作为输入且考虑温度、湿度和日期类型因素的影响。通过建模学习构建网络模型,挖掘网络特征内部变化规律,通过映射加权和学习参数矩阵赋予Bi-LSTM-AT网络隐含状态相应的权重。同时,针对该模型超参数选择困难的问题,提出利用改进麻雀算法实现该模型超参数的优化选择,使得全年最后两天预测值的MAPE为0.42%、RMSE为0.29%和MAE为0.21%,验证了模型线性回归拟合能力的准确性和稳定性。

基于DCNN-LSTM负荷预测算法的5G基站节能系统研究

伴随着5G网络的大规模快速建设,运营商乃至整体通信行业的能耗压力在同步凸显。通过节能降耗实现行业可持续发展成为当前5G网络发展的新研究方向。以小区物理资源块(physical resource block,PRB)利用率为负荷评估指标,对小区指标进行深度特征提取,提出了一套深度卷积神经网络和长短期记忆(DCNN-LSTM)深度学习算法模型实现PRB利用率未来值预测,进一步结合小区瞬时任务中大小包比例,对各种基站设定动态化的节能策略。并引入网络能耗管理网元,对整体5G接入网络的能耗进行动态化统一管理,在保障无线网络服务质量的基础上,实现了5G基站的智能化节能运作。

基于改进蝙蝠算法优化LSTM网络的短时客流预测

准确地预测地铁站短时客流量,对地铁站通风空调系统的节能优化具有重要意义。充分考虑地铁客流量非线性、随机性、周期性等特点,提出一种基于改进蝙蝠算法(IBA)优化长短期记忆(LSTM)神经网络的短时客流量预测模型(IBALSTM)。引入反向学习、动态自适应惯性权重与拉格朗日插值法等方法改进蝙蝠的全局搜索与局部寻优能力,克服标准蝙蝠算法易早熟、易陷入局部最优值的问题;利用改进的蝙蝠算法对LSTM网络的隐含层节点数、迭代次数、初始学习率、学习率下降因子4个参数进行优化;利用西安某地铁站自动检票系统(AFC)采集的客流数据,对模型的有效性进行检验。实验结果表明:该预测模型在均方误差、均方根误差、平均绝对百分比误差等方面均优于标准蝙蝠-LSTM模型、LSTM预测模型、BP预测模型及BP-Adaboost预测模型,所提出的方法可有效应用于短时客流量预测。

基于LSTM网络的Web软件系统实时剩余寿命预测

Web软件系统剩余使用寿命的预测精度是影响Web软件系统抗衰决策的重要方面,为此,提出了一种基于长短期记忆网络的Web软件系统实时剩余寿命预测方法.首先搭建加速寿命测试实验平台,收集反映Web软件系统老化趋势的性能指标,然后根据该指标数据的时序特性,建立了一种基于长短时记忆网络(LSTM)的Web软件系统实时剩余寿命预测模型,并对该模型进行了训练.实验结果表明,该预测模型能够有效对Web软件系统的剩余寿命进行实时预测,具有更好的准确性和适用性.将所提模型应用于Web软件系统寿命预测中,能够有效完成预测,该方法为优化系统抗衰决策提供了技术支撑.

面向天基红外预警的高动态弱小目标LSTM检测方法研究

在天基红外预警任务中,高动态弱小目标具有成像尺寸小、运动规律未知的特点。现有红外弱小目标探测任务主要关注匀速直线运动目标检测问题,对高动态目标的有效检测算法尚需进一步开发。针对天基红外预警任务中高动态非线性运动目标检测问题,提出了一种基于LSTM的红外弱小目标检测算法。首先设计了提取可疑目标位置信息的预处理方法,解决了LSTM网络结构与序列图像不匹配的问题;然后,针对传统算法难以检测非线性运动轨迹的问题,利用LSTM提取目标运动特征,实现序列图像中高动态目标的检测。通过与序列假设检验等算法的对比,在自研的红外序列图像数据集上验证了所提出的算法能够实现不低于0.9347的精确率与不低于0.8633的召回率。

基于LSTM的POI个性化推荐框架

近年来,随着基于位置的社会网络(Location-Based Social Network, LBSN)热度的不断增加,为用户推荐下一个POI (Point-Of-Interests)也显得越来越重要.而对应的各种应用搜集到的用户的行为时间、地理、好友以及标签等信息的增多,使得POI推荐变得更加容易.目前针对POI推荐,已经有部分算法提出,但是他们受限于自身的局限性,还都不能很好的解决这个问题,例如,个性化马尔科夫链(Factorizing Personalized Markov Chain, FPMC)、张量分解(Tensor Factorization, TF)、RNN (Recurrent Neural Networks)等.但是,这些模型由于其本身缺陷,都不能完美的糅合POI场景中的所有信息.在这篇文章中,我们扩展了长短时记忆循环神经网络(Long-ShorT Memory recurrent neural networks, LSTM),提出一种全新的推荐框架POI-LSTM来解决POI推荐问题. POI-LSTM借鉴Embedding的思想,对用户信息、好友关系、POI信息和评论信息进行向量化后,输入到神经网络中,同时利用LSTM捕捉用户的兴趣特征和兴趣的变化趋势,最终能够在不同的输入层拟合社交网络信息和语义信息,同时利用用户历史行为的时间和地理位置信息来为用户推荐下一个兴趣点.

基于自适应VMD-注意力机制LSTM的时间序列预测

真实世界中的时间序列存在非平稳、非周期性波动等特点,直接对其进行预测通常难以达到较好的效果,且极易出现滞后现象。针对上述问题,提出了一种基于自适应变分模态分解的时间序列预测方法。为降低复杂度,首先利用变分模态分解将原始时间序列分解为一系列相对平稳的子序列,并根据最大信息系数确定子序列的输入长度,进而采用带注意力机制的双向长短期记忆网络模型进行预测。为确保该方法能发挥最佳性能,提出一种网格贝叶斯优化方法,对方法参数进行自适应寻优。实验部分将所提方法对时间序列分别进行了单步、多步预测。通过与现有常用方法的对比可以看出,所提方法具有明显精度优势。此外,通过实验结果发现,所提方法可通过更换回归模型来取得预测精度和计算效率的折中,证明其场景适应能力较强。

基于几何特征与LSTM网络结合的动作识别算法

为进一步提高基于人体骨架动作识别的识别率,打破以往大部分基于深度学习的方法的输入都为人体骨架关节坐标的局限性,提出了一种将骨架几何特征与长短期记忆(LSTM)网络结合的动作识别算法。选择基于关节与选定直线之间距离的骨架几何特征代替骨架关节坐标作为网络的输入,并引入了基于LSTM的网络结构,即时序关注LSTM网络。利用时序关注LSTM网络具有重点关注最具识别性的帧的能力,在SBU Interaction数据集和UT Kinect数据集上分别取得了99.25%和98.79%的识别率。实验结果证明:该方法对基于人体骨架动作识别的有效性。

时序分解和CNN-LSTM相融合的月径流预报模型

针对常规模型无法充分提取径流序列复杂非线性特征信息的不足,提出一种基于局部加权回归周期趋势分解算法(STL)与卷积神经网络(CNN)和长短时记忆神经网络(LSTM)相融合的月径流预报模型。该模型首先利用STL将径流序列分解为趋势项、季节项和随机波动的余项,分解后的各分量序列输入CNN进行卷积运算和子采样层重采样,CNN输出的特征序列通过LSTM拟合时序关系后由全连接层输出径流预测值。以黑河流域讨赖河基准站的月径流数据为例,对比分析LSTM、STL-CNN、STL-CNN-LSTM三种模型的预测效果,验证结果表明:STL和CNN-LSTM相融合的模型预报误差最小、精度等级最高。该模型相较于直接对原始径流序列进行分析的常规模型,可以较为显著地提高月径流预测的能力。

基于集成LSTM模型的数据驱动需求预测

电力用户参与电网调度能够有效提升电网灵活性,但其行为的不确定性限制了需求响应的发展。针对此问题,文中首先构建激励型需求响应的实现框架,阐述负荷聚合商(LA)如何整合需求侧资源参与电力市场业务,并将用户随激励政策进行响应的行为转换为需求弹性。然后,基于长短时记忆(LSTM)算法,提出一种集成LSTM的数据驱动的需求弹性预测方法,同时为提升预测模型性能,对源数据进行平滑与缩放处理,并增加损失函数权重系数。算例结果表明,与传统LSTM算法及k近邻预测法相比,文中所提预测方法用于用户需求弹性预测时平均预测误差分别降低了5.33%和28.8%,用于总负荷预测时平均绝对百分比误差(MAPE)分别降低了2.06%和3.09%。同时文中基于集成LSTM分析了平滑、缩放数据预处理对预测精度的影响,结果表明对原始数据进行预处理可有效提升预测精度。

基于GWO⁃LSTM与NKDE的变压器油中溶解气体体积分数点—区间联合预测方法

变压器油中溶解气体体积分数是表征变压器健康状态及故障特性的重要参量。因此,准确预测变压器油中溶解气体的体积分数,有助于及时把握变压器的状态演化与故障发展趋势。现有对气体体积分数预测的研究多集中在点预测方面,难以全面反映气体体积分数的不确定性信息。针对此问题,提出了一种基于灰狼优化长短期记忆网络(long short⁃term memory based on grey wolf optimization,GWO⁃LSTM)与非参数核密度估计(non⁃parametric kernel density estimation,NKDE)的变压器油中溶解气体体积分数点—区间联合预测方法。首先,搭建变压器油中溶解气体体积分数点—区间联合预测模型的整体结构,阐述预测的实现过程;其次,利用自适应噪声完备集合经验模态分解方法将气体体积分数原始序列分解成若干个较为平缓的子序列,再基于GWO⁃LSTM对上述子序列分别进行点预测,并将所有子序列点预测结果叠加合成还原为气体体积分数点预测结果;然后,基于气体体积分数点预测结果及NKDE构造气体体积分数预测误差的概率密度估计函数,进而生成不同置信水平下的区间预测结果;最后,对所提方法进行算例分析,算例结果验证了所提方法的有效性。

基于Bi-LSTM-6Tags的智能中文分词方法

针对当前基于深度学习模型中文分词算法中存在的语义理解不全和词位信息不足的问题,提出了基于双向长短期记忆(Bi-LSTM)神经网络模型的六词位标注集中文分词方法。首先,利用双向长短期记忆神经网络模型自动发现文本特征;然后,通过六词位标注集从文本深层语义上高效准确完成中文分词任务;最后,通过第二国际汉语分词评测(SIGHAN)提供的Backoff2005语料集进行实验验证,在相同实验条件下,该方法与条件随机场(CRF)方法、单向长短期记忆神经网络方法、双向长短期记忆神经网络四词位方法进行比较,分别可以提高分词准确率3%、4%、1%,从而证明该中文分词方法是合理和有效的。

基于改进SSA-LSTM的销量预测研究

为提高销量预测的准确率,提出一种基于改进麻雀搜索算法优化LSTM的销量预测模型。首先在算法迭代过程中通过重心反向学习对个体进行变异,以增强算法跳出局部最优的能力;其次利用改进的算法对LSTM神经网络的超参数进行优化,解决了依靠主观经验选取超参数时存在精度不佳的问题;最后在原始销售数据的基础上加入大量零售数据等多个特征变量进行辅助预测,提高模型预测准确性。实验结果表明,该模型相较于其他模型具有更好的预测效果。