【作者】 陈辉；

【导师】 周纯杰；

【作者基本信息】 华中科技大学 ， 控制理论与控制工程， 2011， 博士

【摘要】 网络控制系统是信息时代控制系统的主要方式,控制节点之间通过通信协议栈组成控制网络,控制网络中协议的异构性、环境的动态性和任务的多样性,是影响网络控制系统性能的重要因素。在此背景下,本文提出了一套完整的面向控制网络的节点可重构自组织通信协议栈的设计、实现和优化方案,利用可重构自组织技术对通信协议栈的部分或整体进行更新和替换,使其能适应不同的应用对象和环境。总结了可重构自组织协议栈技术的概念、原理和开发流程。分析了有限状态机和Petri网两种形式化方法在协议设计和评价方面的优势。作为可重构自组织协议栈开发的关键技术,分析了虚拟在解决平台无关的应用层接口设计上的优越性,并指出了代码自动管理和网络服务质量(QoS)自适应调度算法的局限性。针对控制网络的需求和约束,提出了一种面向控制网络的可重构自组织协议栈体系结构,对其中通信协议栈、重构接口和重构管理器三大组件及其接口进行定义。给出了一种模型驱动的闭环重构管理框架,描述了参数重构和代码重构的控制流程,为开展基于形式化技术的可重构自组织协议栈开发工作奠定基础。以规范与描述语言(SDL)为基础,给出一种协议栈代码动态重构的实现框架。建立了与可重构自组织协议栈体系结构对应的SDL模型集。在此基础上,针对状态空间爆炸问题,设计了一种二进制编码规则处理从SDL模型空间到C代码的映射。利用标准的ELF文件,定义了一种协议配置文件及其读写方法,保障协议代码在网络迁移过程中的完整性,并用于支持代码目标节点上的动态加载。给出了SDL的形式化测试方法,在Linux平台下初步评估了动态重构性能,表明基于SDL的动态重构方法能有效避免协议代码实现过程中的二义性,减少功能漏洞。利用确定与随机Petri网(DSPN),提出一种可重构自组织协议栈动态性能评价框架。建立了协议实时性、任务和资源约束关系模型,用于合成综合性能评估模型。定义了DSPN可重构自组织协议栈综合性能模型,并在此基础上,给出了实时性、网络利用率和效率三种性能指标的量化方法。通过静态和动态性能仿真,能方便得到给定环境下协议栈性能的近似真实结果。在协议栈性能优化方面,结合神经网络和遗传算法,提出一种QoS自适应调度算法的改进方案,基于Elman神经网络的QoS预测值,遗传算法能预见性的主动输出新的调度策略,保证网络的稳定性。针对工业无线通信领域的应用需求,以IEEE802.15.4标准为底层硬件基础,设计了一种基于可重构自组织技术的无线通信协议栈(RcWSN)原型节点,在Mesh结构下,该节点能够以一种低功耗强实时高可靠的方式相互通信,通过路由和调度重构实现对环境和应用变化的自适应。最后对全文进行总结,对本文存在的不足进行了阐述,并对控制网络节点可重构自组织协议栈的技术发展进行了展望。更多还原

【Abstract】 Networked control system (NCS) has become the mainstream form of the automation industry. In such a system, nodes are networked and interacted with each other according to a communication protocol stack. The distributed infrastructure complexity, communication environment heterogeneity, and control strategy diversity bring great impact on the system performance. This paper proposes a reconfigurable self-organizing protocol stack (RSPS) for nodes in NCS, including the design, implementation and optimization solutions. Full or partial reconfiguration in the protocol stack itself or from the third part provides the network adaptability for application and environment variations.The principle and development process for reconfigurable self-organizing protocol stack are firstly summarised. The advantages of Finite State Machines (FSM) and Petri nets in the protocol design and evaluation process are analyzed. Then, the virturalization technology is shown as an advanced platform-independent solution for the development of RSPS. Besides, as two other key technology to implement RSPS, limits on the existing methods for the code automated management and QoS aware scheduling are pointed out.To meet the requirements and constraints in NCS, a kind of RSPS architecture is proposed. This architecture defines three functional components and relationship with each other:a protocol stack governing the node interactions, a reconfiguration interface support code updating, and a reconfiguration manager determines the new protocol file. Then, a model driven reconfiguration management framework is given to clarified two reconfiguration levels within the RSPS architecture, called parameter-level and code-level reconfiguration. The management framework is the basis of the following RSPS researches.On basis of the Specification and Description Language (SDL), a framework to implementing the dynamic code-level reconfiguration is given. Firstly, a set of SDL models that are mapping with the functions defined in the RSPS architecture are built. Then, a binary coding rule is proposed cope with the states explosion problem, for exploring and reducing the reconfiguration space and generating the C Code. The standard ELF (Executable and Linkable Format) file is employed to construct the protocol configuration file, which guarantees the completeness during code migration and supports dynamic code loading in the target platform. The method of formal testing is given, and a case on Linux platform shows that the SDL framework enables an unambiguous design process and supports easy testing of real reconfiguration capability.A synthetical performance evaluation model for RSPS is proposed with the help of Deterministic and Stochastic Petri Nets (DSPN). The extended DSPN modeling rule is defined for the protocol modeling, and under this concept, the genetic performance sub-models of timing, task interrelation, resource are given. The synthetically DSPN model provide the convenience to get an approximate real performance result by static and dynamic testing, such as the qualified metrics of real time performance, network utilization and efficiency. On the aspect of performance optimization, the self-adaptive scheduling scheme is improved by a combination of neural network (NN) and genetic algorithm (GA). The network stability can be maintained, because the GA scheduler can reconfigures the scheduling table proactively based on the predicted QoS variations, which are captured by an Elman NN.Focusing on the requirement of industrial wireless communication, a prototype design of RSPS based Wireless Sensor Network (RcWSN) is given, which is compatible with the standard IEEE 802.15.4. Under a mesh topology, RcWSN takes advantages in the aspects of real-time performance, adaptability and energy cost. It can adapt to the changes of enviroment or application with the mechanisms of routing and scheudling reconfiguration.Finally, the full text of this article is summarized, the deficiencies are explained and the future of RPS is prospected.更多还原