【作者】 王军号；

【导师】 孟祥瑞；

【作者基本信息】 安徽理工大学 ， 采矿工程， 2013， 博士

【摘要】 论文针对煤矿安全监测监控系统存在的技术落后、功能单一、存在监控盲区以及不能联动联控等问题,把智能化的物联网感知技术应用于煤矿安全监测监控领域,以“感知”为突破口,重点研究了感知矿山物联网关键技术：感知煤矿安全状况的分布式信息融合感知算法和感知传感器节点健康状态的故障诊断感知算法。论文取得了以下研究成果：(1)定义了煤矿井下物联网感知域的概念,从感知层的拓扑结构、路由汇聚机制以及中间件等方面进行了设计：在物联网感知域内构建了开放的基于分簇的分布式感知架构——分布式星状无线传感器网络(DSWSN);在LEACH和PEGASIS协议的基础上进行了改进,形成了高效的路由汇聚机制,提高了服务质量(QoS),满足了通信的可靠性和实时性要求；建立了基于分簇的协作型多功能中间件体系结构,把簇层和资源管理层通过移动Agent技术有机结合起来,能够充分支持应用程序的开发,利用应用程序表示形式转换实现多种类型的应用形式的协作统一；给出了用于数据存储和传输的云数据服务平台(PaaS服务)部署方式,建立了高效的煤矿物联网安全监控感知平台。(2)在分析煤矿井下复杂环境的基础上建立了感知煤矿安全状况的信息融合策略。在数据预处理模块中采用置信距离测度与采集数据的时间戳相结合的动态限幅滤波算法对数据进行预处理以消除疏失误差。运用最优加权估计算法进行数据级融合,依据传感器方差的自相关和互相关估计,在总均方误差最小和满足无偏性的最优条件下,根据各个传感器得到的测量值以自适应的方式找到其对应的权数,使融合后的值达到最优,获得更加准确的现场监测信息。在决策级融合算法中建立了基于模糊粗糙-灰色关联分析(FR-GC)的算法模型。在该模型中,不需要预先给定额外信息,而是通过数据的不可分辨关系来提取隐藏在数据中的潜在信息,保证了分析的客观性；同时利用煤矿环境特征向量与标准特征向量的灰色关联度进行系统特征优势分析,从整体上考虑煤矿环境的安全性,最终根据关联度的大小给出系统的安全判决。实验表明本算法具有权值分布合理,绝对误差波动平稳,动态响应特性好,收敛速度快,能有效滤除干扰数据等特征；利用模糊粗糙模型与灰色关联分析之间较强的互补性关系,改善了待决策样本与识别模式的亲和度,突出了定量程度,具有较高的感知区分度,减少了主观因素的影响,提高了决策的客观性。(3)分析了煤矿安全监控系统传感器存在的4种故障模式,在此基础上建立了传感器的故障诊断策略。以瓦斯传感器节点为例,针对常见的常值型、漂移型、偏置型和周期型4种隐性软故障,以小波分析和FRBF神经网络为基础,提出了由加Hamming窗的Shannon为母小波的小波包分解提取特征能量谱与扩展Kalman滤波算法(EKF)优化的FRBF神经网络进行模式分类辨识的传感器节点故障诊断方法。对传感器的输出信号进行小波包分解,运用基于代价函数的局域判别基(LDB)算法进行裁剪,获取最优的特征能量谱,经处理后作为特征向量训练EKF-FRBF神经网络,采用参数增广和统计动力学方法,通过带有整定因子的EKF参数估计,用来辨识传感器节点的故障类型。实验表明,该方法的辨识正确率在95%以上,误报率和漏报率都明显优于其他算法,能够有效用于物联网系统中传感器节点的在线故障诊断。(4)分析了在DSWSN系统中,智能移动Sink节点的功能与特点,分别从仿真设计、硬件设计和软件设计三个方面逐步推进,完成了Sink节点的设计开发。通过实验证明,该Sink节点可以很好地完成对监控数据的处理和传输,实现了对煤矿安全状况和节点健康状态的正确感知,具有电路简单、功能完善和技术性能高的特点,是一种比较可取的物联网汇聚节点的设计方案,从而打造出一张更加密集、更加有效的煤矿安全生产物联网。通过信息融合与故障诊断两种感知算法的密切配合,实现了信息互补与协同感知,大大降低了监控系统的不确定性和不可靠性,减少了由于单一传感器受信息量局限引起的误报错报和冲突,提升了对煤矿安全的快速监测和预警预报能力,为煤矿安全生产提供了强有力的保障。论文的研究能够充分发挥物联网感知技术在煤矿井下应用的优势,为提升煤矿生产效率和加大安全管理提供了一个全新的综合信息化平台。更多还原

【Abstract】 By targeting at issues such as backward technologies, low functionality, monitoring gaps and inability in joint actions and controls which are existing in coal mine’s safety monitoring systems, the paper focuses on research of key technologies for perception of internet of things at coal mines by applying intelligentized perception technology of internet of things into such monitoring systems, with "perception" as the breakthrough point. The key technologies are distributed information fusion perception algorithm for perception of safety conditions at coal mines and fault diagnosis perception algorithm for perception of health status of sensor nodes. The paper has accomplished the following achievements:(1) It has defined concept of perception domain of internet of things and completed designs from many aspects, topology structure of perception layer, routing and aggregation mechanism and intermediate components and so on:An open and cluster-based distributed perception architecture, distributed star wireless sensor network (DSWSN) was built; Improvements based on LEACH and PEGASIS protocols to develop an efficient routing and aggregation mechanism were made, which can improve quality of services (QoS) and satisfy reliability and real-time requirements of communications; A cluster-based coordinating multifunctional structure of intermediate component systems aiming at integrating cluster layers and resource management layer with mobile Agent technology was built, which can sufficiently support development of application programs and utilize application programs to represent formal transformations for realization of coordination and unification of many types of application forms; and a deployment way of cloud data service platform (PaaS service) for data storage and transmission was developed, and an efficient safety monitoring and perception platform of internet of things for coal mines was constructed.(2) The paper has developed an information fusion strategy for perception of safety conditions at coal mines based on analyses of complicated underground mine environmental conditions and has adopted dynamic amplitude limiting filtering algorithm which integrates confidence distance measure and timestamp of data collection in data pre-processing module, to eliminate negligence and errors. Optimal weighting algorithm is used to make data level fusion to optimize post-fusion values and obtain more accurate site monitoring information by relying on estimates of self-related or mutually related sensor variances and finding out corresponding weight number of each sensor in a self-adapting way by utilizing measured values of each sensor under the optimal conditions of minimal total mean square error and satisfying unbiasedness. A fuzzy rough-gray correlation (FR-GC) based algorithmic model has been established in decision level fusion algorithm, in which no additional information needs to be provided in advance and data’s indiscernibility relation is used to extract potential information hidden inside data, which guarantees objectivity of analyses. At the same time, analyses of system features have been made by using gray correlation of coal mine’s environmental feature vectors and standard feature vectors, and coal mine’s environmental safety has been considered in an all-round way, and in the end, judgment about system safety was made according to the correlation. Tests indicate this algorithm is characterized with rationality in weight distribution, stability in absolute errors, soundness in dynamic response characteristics, high speed in convergence speed and ability to effectively remove disturbing data. It can improve affinity between samples to be decided and identification mode with the stronger complementary relation between fuzzy rough model and gray correlation analyses, which highlights quantitative degrees, has higher perception distinction degrees, reduces affect from objective factors and increases decision-making objectivities.(3) Four fault modes of sensors in the monitoring system have been analyzed, based on which, a sensor fault diagnosis strategy has been established. For example, by targeting at the four common latent soft faults of constant value faults, drifting type faults, biased faults and periodic faults from which gas sensor node suffers, the paper proposes a sensor fault diagnosis method, which adopts wavelet analysis and FRBF neural network as the basis and makes mode pattern classification and identification with wavelet packet adopting Hamming window added Shannon as mother wavelet to decompose and extract characteristic energy spectrum, and with FRBF neural network optimized by expanded Kalman filtering algorithm (EKF). Sensor output signals can be decomposed with wavelet packets, and cut with cost function based local discriminant bases (LDB) algorithm to obtain optimum characteristic energy spectrum, which will be used as characteristic vector after processed to train EKF-FRBF neutral network. Then parameter augmentation and statistical dynamics method, and EKF parameter estimation with regulated factors can be used to identify fault type of sensor node. Tests indicate the identification accuracy of this method is over95%, and both its false alarm rate and missing alarm rate are apparently lower than others. So this method can be adopted effectively in on-line fault diagnoses of sensor nodes in the system of things of internet.(4) It analyzed functions and features of intelligent mobile Sink nodes in DSWSN system and completed design and development of Sink nodes by making gradual progress in the three aspects of simulation design, hardware design and software design. Tests demonstrate this Sink node can well perform processing and transmission of monitoring data and achieve correct perception of safety status and node health state at coal mines. With its advantages in simple circuits, complete functions and advanced technical performances, this is satisfactory design for aggregation nodes of things of internet, and can be used to build a more densely-distributed and more efficient things of internet for safe production at coal mines.Close cooperation between information fusion and fault diagnosis perception algorithms has achieved information complementation and coordinated perception, which can significantly reduce uncertainty and unreliability of monitoring systems, and false alarms and conflicts caused by information limits of single sensor, and promote real-time safety monitoring and early safety warning ability of coal mines. So it can provide a strong guarantee for safe production at coal mines. The research done by the author can give full play to advantages in applying perception technology of things of internet in underground sections of coal mines and can provide a brand-new comprehensive informationalized platform for boosting production efficiency and safety management performances at coal mines.更多还原