基于自然语言理解的UAV指令理解系统设计

在有人机与无人机协同完成任务时,为了使有人机操作人员更简便的控制无人机,如何对自然语言指令进行理解成为当前的热点问题.文中设计了基于自然语言理解的UAV指令理解系统,提出了该系统的结构组成、构建方案,详细讨论了各功能模块的作用和实现方案.该系统主要功能包括语音识别、自然语言理解和任务规划.考虑到自然语言理解实现的需求,定义了受限的自然语言指令文法规则,并采用自动分词算法对命令词汇进行切分,然后基于语法树的实例匹配算法对指令进行理解.最后,对其中的关键算法进行了仿真实验.实验结果表明,该系统对于无人机指令具有良好的理解结果.     

智能空间下面向动作序列生成的服务机器人指令解析方法

服务机器人在执行家庭任务时存在任务分析困难、环境适应性差、人机交互不友好等问题。为了更加准确地理解用户意图,规划符合当前环境的服务步骤,研究了智能空间下面向序列生成的服务机器人指令智能解析方法,以加强机器人对于家庭环境下服务任务的认知能力。首先,从指令关键词入手,提出了一个基于门控机制的意图识别模型,加强了机器人对用户指令的认知能力。然后,提出了一种以服务策略作为中间状态的序列文本生成机制,用于辅助生成机器人的动作序列。同时,结合机器人与环境的多模态感知,集成了一种基于智能空间本体交互的策略修正机制,以生成最符合当前环境的服务策略。最后,融合了基于领域与问题描述的任务规划模块,为机器人生成可执行的动作序列,从而提升服务任务的执行质量。实验结果表明,本文方法在保证交互友好的同时,能够准确理解复杂的用户指令,并最终生成机器人能执行的可靠动作序列。     

AUV控制系统规划层DEDS建模与仿真

地形勘查使命执行过程中,AUV(自治式水下机器人)控制系统的规划层对于AUV自主智能决策起到了至关重要的作用,对于AUV控制系统规划层进行DEDS(离散事件系统)建模与仿真研究可以使我们深入地了解并验证AUV自主能力.介绍了AUV控制系统逻辑模型以及监督控制理论,采用分层PETRI网对规划层进行了离散事件系统建模,采用了一种改进的RW离散事件监控方法对规划层模型进行了协调,通过对于任务库所的扩展,详细阐述了这种分层结构.采用一种半物理仿真平台,对模型的几种典型状态进行了仿真研究,最后通过一个具体的使命对AUV控制系统的规划层模型进行了仿真验证,仿真结果表明规划层模型能够正确协调各个任务,处理突发事件,指导AUV最大限度完成地形勘查使命.     

基于平行Petri网的制造系统调度与控制一体化方法

为了消除制造系统调度层与控制层之间的隔阂, 实现对生产事件快速灵活响应, 本文提出了一种调度与控制一体化的方法. 首先, 定义了一种新型Petri网模型, 即平行Petri网, 从而集成地描述了传感器、执行器、任务和资源信息, 构建制造系统的信息物理系统模型; 其次, 提出了一种从平行Petri网到赋时Petri网的抽象简化方法, 大规模压缩优化调度所需搜索的状态空间; 再次, 定义了策略Petri网以描述最优调度策略. 最后, 给出了平行Petri网与策略Petri网同步执行算法, 使得平行Petri网与物理系统同步执行.     

融合ERNIE和注意力机制的中文关系抽取模型

关系抽取任务是要在实体识别的基础上确定无结构文本中实体对间的关系类别,即判断实体间的关系.针对目前中文关系抽取精度不足以及静态词向量无法很好地解读文本的问题,本文提出一种融合ERNIE预训练模型和注意力机制的TEXTCNN中文关系抽取模型.ERNIE词向量针对中文的特点以词组为单位做掩盖进行模型训练,实现了对中文文本更好的语义表达,再通过TEXTCNN模型对输入数据进行特征提取,融合注意力机制聚焦于影响最终结果的关键特征,从而实现特征优化提取.本文在百度发布的SKE数据集上进行实验,重点探索ERNIE模型结合注意力机制对中文文本的特征表达效果,结果表明本文模型可以更好地学习中文文本中的特征并用于关系抽取,有效提高关系抽取任务的准确率.     

基于群智能-一致性理论的无人机编队全过程飞行航迹规划方法

针对无人机编队执行任务全过程飞行规划问题,提出一种基于多步粒子群优化的无人机编队航迹规划算法.首先,对无人机和执行任务策略进行建模,将编队执行任务全过程划分为编队成形、执行任务、返航、解散和无人机降落5个阶段,设计不同阶段的飞行策略;其次,针对不同的终端约束条件,设计多类多层优化指标,提出多步粒子群算法,并引入模型预测控制滚动优化航路点,得到适用于不同阶段的能严格满足约束条件的航路规划方法;然后,建立旋转坐标系,将航路点信息转换为编队控制律中的理想航向和高度信息,得到能通过航路点的编队控制算法;最后,利用编队控制算法去执行航路规划方法给出的航路点,生成航迹,得到编队航迹规划算法.仿真结果表明,所提规划方法比传统方法更适用于编队飞行,能为编队规划执行任务全过程的平滑航迹,具有良好的通用性.     

基于混合注意力机制的中文机器阅读理解

预训练语言模型在机器阅读理解领域具有较好表现,但相比于英文机器阅读理解,基于预训练语言模型的阅读理解模型在处理中文文本时表现较差,只能学习文本的浅层语义匹配信息。为了提高模型对中文文本的理解能力,提出一种基于混合注意力机制的阅读理解模型。该模型在编码层使用预训练模型得到序列表示,并经过BiLSTM处理进一步加深上下文交互,再通过由两种变体自注意力组成的混合注意力层处理,旨在学习深层语义表示,以加深对文本语义信息的理解,而融合层结合多重融合机制获取多层次的表示,使得输出的序列携带更加丰富的信息,最终使用双层BiLSTM处理输入输出层得到答案位置。在CMRC2018数据集上的实验结果表明,与复现的基线模型相比,该模型的EM值和F1值分别提升了2.05和0.465个百分点,能够学习到文本的深层语义信息,有效改进预训练语言模型。     

基于类卷积交互式注意力机制的属性抽取研究

在基于深度学习的属性抽取研究中,注意力机制是常用的模型之一.目前,面向属性抽取的注意力机制存在2个局限性：其一,注意力机制多为自注意力机制,这是一种全局式注意力机制,其将不相关的噪音(距离目标词较远且与之不相关的词)带入注意力向量的计算;其二,目前的注意力机制多为单层注意力机制,注意力一次建模后缺少交互性.针对这2个局限性,提出一种面向属性抽取的类卷积交互式注意力机制.该方法先将目标句输入到双向循环神经网络,借以获得每个词的隐式表达,再经过类卷积交互式注意力机制进行表示学习.类卷积交互式注意力机制分为2层注意力计算：第1层按序(从句首到句末)通过滑动窗口控制每个词的上下文宽度,并计算每个词的注意力分布向量;第2层将第1层的注意力分布向量与所有单词进行交互注意力计算,将得到的注意力向量与第1层的注意力向量拼接,最终输入到条件随机场进行属性标记.在2014—2016语义评估(semantic evaluation, SemEval)官方数据集上验证了模型的有效性.相比于基线模型,在4个数据集上的F1值分别提高了2.21,1.35,2.22,2.21个百分点.     

基于AI人工智能的学前教育机器人对话系统研究

为了实现学前教育机器人对话系统的优化，以语音增强为切入点，提出一种基于双阶段注意力机制的儿童语音增强方法，以优化学前教育机器人对话系统，使机器人能够更准确地理解和执行学前儿童语音中的文本指令。首先分析了主流语音增强方法的基本原理以及单通道时域语音增强存在的问题；然后搭建双阶段注意力增强网络的语音增强方法，并在学前教育机器人对话系统中加入双阶段注意力语音增强模块进行测试。结果表明：双阶段注意力语音增强方法的SSNR得分相比于得分较高的通道注意力增强网络提高了3.3 dB,loss值是四种模型中最低的即0.98,且语音增强时域波形图清晰；综上研究得出，提出的语音增强方法具有可行性和有效性。     

一种可信可控的网络体系及协议结构

互联网体系架构正面临着严峻的安全和管理挑战,迫切需要具备可信性和可控性特征的新架构.已有的网络体系结构要么是基于边缘论和面向非连接的设计思想,导致分组传输路径不可控,要么是重新设计现有网络的体系架构,代价巨大.文中提出了一种可信可控的网络体系结构,其特征是在现有网络体系架构的基础上增加一个可信可控4层逻辑结构,从而实现网络组元及用户行为的可预期、可管理,它包括决策层、观测层、资源层和可信接口层4个层次;在明确新网络体系结构中网络控制对象为逻辑流的前提下,给出了一种包含感知与监测、理解与检测、判断与决策和控制及可达4个功能环节的闭环自反馈控制方法,确保网络系统可自诊断、自恢复地回归稳定态.同时指出具备跨层交互能力的可信接口层是新体系实现可信可控的关键,继而给出了位于该层的可信可控协议模型,并定义了确保协议执行可预期的协议基本功能;指出了逻辑流中信任流是协议的主要控制对象,并基于新体系结构下网络控制方法,通过融合信任管理和不可否认服务的方式给出了协议可信的控制方法.为验证协议模型和控制方法的有效性,给出了协议的具体实施方案,包括协议基本要素和协议两阶段执行过程.最后,给出了新体系与现有技术相比较所具备的优势,并对下一步工作进行了展望.     

A cognitive-behavioral model of pathological Internet use

This article introduces a cognitive-behavioral model of Pathological Internet Use (PIU). While previous studies on Internet addiction have described behavioral factors, such as withdrawal and tolerance, the present article focuses on the maladaptive cognitions associated with PIU. The cognitive-behavioral model of PIU distinguishes between specific PIU and generalized PIU. Specific PIU refers to the condition in which an individual pathologically uses the Internet for a particular purpose, such as online sex or online gambling, whereas generalized PIU describes a more global set of behaviors. The model implies a more important role of cognitions in PIU, and describes the means by which PIU is both developed and maintained. Furthermore, it provides a framework for the development of cognitive-behavioral interventions for PIU.     

Stressful life events and problematic Internet use by adolescent females and males: A mediated moderation model

In a sample of 660 adolescents (M age = 14.14 years; 55% females), this study examined a mediated moderation model in which temperamental effortful control and sensation seeking moderated the relationship between stressful life events and problematic Internet use (PIU), and this moderating effect was mediated by maladaptive cognitions. Findings revealed that effortful control buffered the risk of stressful life events for females’ PIU, and this effect was mediated through maladaptive cognitions. The risk-buffering effect of effortful control for males’ PIU was not significant, although it buffered the risk of maladaptive cognitions for PIU. The risk-enhancing effect of sensation seeking was not significant in both female and male adolescents. In addition, males scored higher on risks but lower on protective factors of PIU than females, which explain the gender difference in PIU. The authors discuss the theoretical and practical implications of these findings.     

Family factors in Internet addiction among Chinese youth: A review of English- and Chinese-language studies

Approximately 513 million Chinese citizens used the Internet in 2011, with adolescents reporting comparatively high levels of use. Although numerous studies (reviewed herein) indicate that Internet Addiction/Pathological Internet Use (IA/PIU) is endemic among Chinese youth and trending upward, no prior review has examined family correlates of IA/PIU in Chinese youth. Thus, our principal aim was to evaluate methodological features and substantive findings of all studies examining family correlates of IA/PIU in Chinese youth. Internet, demographic, psychosocial, and psychiatric/behavioral correlates of IA/PIU, and prevalence estimates for adolescent IA/PIU, were also examined using the large set of studies evaluated in association with our principal aim. Comprehensive bibliographic searches identified 42 pertinent investigations. Youth with IA/PIU reported greater global dissatisfaction with their families; less organized, cohesive and adaptable families; greater inter-parental and parent–child conflict; and perceived their parents as more punitive, and less supportive, warm, and involved compared to non-IA youth. IA/PIU youth were significantly more likely to have divorced parents, live with a single parent, and be an only child than non-IA/PIU youth. IA/PIU is prevalent among Chinese youth and associated with diverse family, psychosocial and psychiatric/behavioral impairments, but rarely is the focus of prevention and treatment interventions.     

Problematic internet use among older adolescents: A conceptual framework

Problematic internet use (PIU) is of growing significance to adolescent health, but lacks a strong conceptual framework to guide ongoing work. We used Concept Mapping methodology to construct a theoretical framework describing PIU. This validated approach includes five steps: preparation, generation (brainstorming), structuring (sort and rank), representation (statistical analysis) and interpretation (focus groups). Analysis included hierarchical cluster analysis over the overall square similarity matrix to determine a cluster map. A total of 193 college students participated in one or more phases of the study. Students were from 13 universities across 10 states, had an average age of 19.4 years (SD = 1.8), were 67% female and 77% Caucasian. A total of 20 health professionals participated. The seven clusters depicted on the PIU Concept Map include: psychosocial risk factors, physical impairment, emotional impairment, social/functional impairment, risky internet use, impulsive internet use and internet use dependence. The stress value for the fit of the multidimensional solution to the structuring data was 0.28, indicating adequate fit. The concept map may be used towards development of a future comprehensive model of PIU as well as scale development efforts.     

Social relationship on problematic Internet use (PIU) among adolescents in South Korea: A moderated mediation model of self-esteem and self-control

Despite previous research efforts on identifying the risk and protective factors of problematic Internet use (PIU), the specific mechanism among these factors are largely unknown. Thus, the present study examined the effect of adolescents’ social relationships on their PIU, as well as the effect of self-esteem as a mediator and the effect of self-control as a moderator. Survey data from 750 South Korean middle and high school students in Seoul and Gyeonggi areas were analyzed using structural equation modeling. The results are as follows. First, PIU is significantly associated with relationship with peers and with mother. Second, self-esteem did not mediate the effect of communication with mother on PIU, but was found to partially mediate the effect of peer relationship on PIU. Third, adolescents’ self-control significantly moderated the indirect effect of peer relationship on PIU via self-esteem. Through detailed analyses, this study identified self-esteem as a mediator and self-control as a moderator in the relationship between social relationships and PIU. Implications of these results for understanding the relationship among social relationships, self-esteem, self-control, and PIU are discussed.     

Toward a comprehensive theory of problematic Internet use: Evaluating the role of self-esteem,anxiety, flow,and the self-rated importance of Internet activities

Problematic Internet use (PIU) has long been a matter of public concern. Study 1 examined the correlates of a well-validated measure of Internet addiction [Young, K. S. (1998). Caught in the net. New York: John Wiley and Sons] with 315 students (39% male). Hierarchical regression analyses indicated that low self-esteem, unrealistic optimism, and the importance of 7-positive activities (I7PA) all contributed to the prediction of PIU. The inclusion of controls for aspects of Internet uses and demographic characters showed a clear distinction between PIU and frequent Internet use. Subjects who anticipated that negative outcomes were more likely to happen to them than to peers were none-the-less more likely to engage in PIU. The more important positive activities were, the greater the PIU. SEM model suggested that unrealistic optimism was best treated as part of PIU and that low self-esteem and the I7PA were the powerful predictors of PIU. Study 2 tested SEM models of psychosocial characteristics with 279 students (48% male). Anxiety and flow were added as predictors and the felt safety for social contacts was added to Global PIU following Caplan [Caplan, S. E. (2002). Problematic Internet use and psychosocial well-being: Development of a theory-based cognitive-behavioral measurement instrument. Computers in Human Behavior, 18, 553–575]. The more complex model provided excellent fit and showed that low self-esteem, anxiety, and the I7PA were positively related to PIU. The relationship both of flow and sensation-seeking was mediated by the 17PA. Research on the topic has been relatively atheoretical. We propose a theoretical template to guide future research on PIU.     

Temperament and adolescent problematic Internet use: The mediating role of deviant peer affiliation

Problematic Internet use (PIU) creates psychological, social, school and/or work difficulties in a person's life. This study examined the unique roles of four dimensions of temperament (effortful control, sensation seeking, anger/frustration, and shyness) on adolescent PIU, as well as the mediating role of deviant peer affiliation (DPA) on these pathways. Participants were 2758 Chinese adolescents (46% male; mean age = 13.53 years, SD = 1.06) selected by stratified and random cluster sampling from 10 middle schools in southern China. After covariates were controlled, structural equation modeling (SEM) demonstrated that effortful control negatively predicted PIU, while sensation seeking, anger/frustration and shyness positively predicted PIU. SEM also identified that DPA partially mediated the pathway from all four temperament dimensions to PIU. These findings provide evidence for the relationship between specific temperament dimensions and adolescent PIU, and highlight DPA as one explanation for the prevalence of PIU.     

融合注意力机制的人机交互信息半监督敏感数据抽取算法

为提高敏感数据抽取效果,提出了融合注意力机制的人机交互信息半监督敏感数据抽取方法。融合类卷积以及人机交互注意力机制构建融合交互注意力机制双向长短词记忆（Bi-LSTM-CRF）模型,通过模型的类卷积交互注意力机制将敏感词转化为字符矩阵,采用Bi-LSTM对该矩阵进行编码获得敏感词字符级特点的分布式排列,通过Bi-LSTM对该分布式排列的二次编码获得敏感词上下文信息的隐藏状态,基于该隐藏状态通过类卷积注意力层与交互注意力层进行注意力加权,获得类卷积注意力矩阵与交互注意力矩阵,拼接两个矩阵得到双层注意力矩阵,利用交互注意力层门控循环单元升级双层注意力矩阵成新的注意力矩阵,经全连接降维获取敏感词对应的预测标签,实现人机交互信息半监督敏感数据抽取。实验结果说明：该方法可有效降低敏感数据抽取复杂度,具有较高的敏感数据抽取查全率。     

智能教学系统中基于案例推理的教学策略研究

把基于案例推理(CBR)技术引入智能教学系统中教学策略的推理和控制中,提出了在全局教学策略和局部教学策略控制中CBR应用模型,并提出了在这种模式下,案例库的初始化、案例的学习与修正方法.该模型应用以前学生学习的经验作为以后学生学习的引导,从而实现教学策略的自动组织、自动规划,实现学生的自主选择,自由学习.最后进行了总结,提出了下一步研究的方向.