【作者】 夏锋；

【导师】 孙优贤；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2006， 博士

【摘要】 随着近年来信息技术的快速发展与应用，实时控制系统(RTCS)资源受限的特征日益突出；同时，RTCS往往需要工作于负载可变的动态不确定环境中，从而导致系统中可获取的计算、通信资源体现出动态不确定性。在这种新的实现环境下，传统以控制与调度分离为特征的控制系统设计与实现方法无法提供必需的控制质量(QoC)保障。从资源调度角度看，当前开环的实时调度算法在应用于实时控制系统中时严重缺乏柔性，通常不能实现可用资源的优化利用。因此，有必要发展新途径、新方法，为动态环境下的实时控制提供使能技术，并在实现平台动态资源约束下优化整体QoC。本文考察了计算、通信与控制等学科不断融合的发展趋势，立足于控制与调度协同设计的研究新思路，从动态资源管理的独特视角深入研究了RTCS中现实存在且尚待解决的一些新课题。将反馈调度作为基本技术架构，为动态资源约束下的RTCS性能优化提供了一系列全新的、有效的解决方案。针对RTCS的实际应用需求和当前的研究现状，系统地阐述了反馈调度的基本科学和技术问题。基于统一的框架平台，重点考虑了CPU、能量和网络带宽这三类典型资源约束，有针对性地开发了一系列适用不同场合的反馈调度方法，提供了闭环动态资源调度的使能技术，从不同角度不同层面解决了反馈调度架构中的一些关键性问题，同时系统地解决了RTCS中瓶颈资源的动态分配问题，提供了柔性的QoC管理机制，实现了动态应用环境下的整体性能优化。在CPU调度方面，本文考虑处理器速率有限的多任务嵌入式控制系统。首先以优化整体控制性能为目标，给出了最优反馈调度问题的数学描述，并探讨了相关的数学求解方法。为了克服数学优化方法运算量大、一般不适宜在线使用的缺点，提出了神经反馈调度方法。该方法在降低调度开销的同时提供了几乎最优的整体控制性能，并且具备较好的适应性、鲁棒性和容错性等。为了在执行时间不可获知且存在测量噪声的情形下实现有效的反馈调度，我们将模糊控制技术引入反馈调度框架体系，提出了模糊反馈调度方法。由于模糊逻辑能够非常有效的处理非线性、非精确性和不确定性，模糊反馈调度方法不依赖控制任务的执行时间参数，通过对负载和可用资源的不确定性进行智能化处理，使得反馈调度系统对于时态参数的测量噪声具有较好的鲁棒性，而且易于实现、调度开销小。在能量管理方面，本文试图在保障嵌入式控制系统QoC的同时尽可能降低处理器能耗。为了有效处理CPU负载动态变化、控制任务执行时间不确定等现实问题，提出了一种集成能量管理与控制质量管理的反馈控制实时调度方法，即能量感知反馈调度。该方法利用动态电压调节技术，通过对CPU处理速率的动态调节来间接改变任务执行时间，以控制CPU利用率。在建立了动态电压调节系统的解析数学模型之后，给出了一种基于控制理论的反馈调度器设计、分析方法，从而获得了具有确定性的反馈调度性能，提供了一种有效的闭环能量管理机制。为了在能量感知反馈调度的基本框架下进一步降低系统能耗，我们通过考察控制系统在可变采样周期下的动态特性，基于适度退化的基本思想提出了一种增强型能量感知反馈调度方法。该方法在利用动态电压调节技术对CPU速率进行调节的基础上，根据各控制回路的实际控制性能对其采样周期进行动态调节，实现了控制任务的柔性时间约束；通过在保障回路QoC的前提下尽可能增大采样周期，获得了更低的能耗，进一步提高了系统能效。在带宽分配方面，本文重点研究优先级驱动的总线网络化控制系统和基于随机介质访问机制的无线控制系统。针对多回路网络化控制系统，以集成控制与网络调度的新观点进行研究，提出了一种集成反馈调度方法，根据可用带宽资源的动态变化，采用一种级联的反馈调度策略来调节采样周期，以将丢包率控制在期望的较低水平，并对可用带宽进行了优化分配；为了进一步提高整体控制性能，基于一种直接反馈调度策略，根据各控制回路的实际性能对回路优先级进行动态配置。该方法在低载情况下可以最大化资源利用，在系统出现过载时可以获取性能的适度退化，因而能够明显改善整体控制性能。针对无线控制系统中可用链路资源的不确定性，我们基于跨层设计思想，提出了一种自适应反馈调度方法。该方法在物理层和应用层之间实现信息交互，以处理信号干扰和链路传输速率变化的影响，并通过控制网络丢包率来优化控制性能。为了提高反馈调度器的响应速度，同时尽可能降低调度开销，提出了面向反馈调度的事件驱动激励方式，从而有效增强了反馈调度方法的应用效果。本文在研究方法上突破了控制与调度分离的传统设计理念，丢弃了固定时间约束的传统控制任务模型，为新技术形势下的RTCS研究开辟了新途径，提供了新思路、新方法。通过本文的研究，初步建立了一套系统化的反馈调度方法体系，从多个角度多个层面丰富了反馈调度这一新兴交叉学科方向的研究成果，从而为集成计算、通信与控制提供了使能技术，必将有效促进控制、计算、通信三大学科领域的进一步融合发展。更多还原

【Abstract】 With rapid evolution and application of information technologies, real-time control systems (RTCS) are becoming ever-increasingly resource limited in recent years. At the same time, real-time control systems often have to operate in dynamic environments that feature workload fluctuation. As a consequence, the available computing and communication resources are typically non-deterministic. In this new implementation environment, traditional control systems design and implementation methodology, which separates control from scheduling, cannot always provide quality of control (QoC) guarantees. From a resource scheduling perspective, existing open-loop real-time scheduling algorithms obviously lack flexibility when applied to RTCS, thereby cannot achieve optimal usage of available resources. Therefore, it is of paramount importance to develop novel frameworks and methods in order for enabling real-time control under dynamic environments and optimizing overall QoC under dynamic resource constraints of implementation platforms.This thesis observes the emerging trend of convergence of control with computing and communication. Following the methodology of control and scheduling codesign, we examine some practical and open problems in RTCS from a unique viewpoint of dynamic resource management. In the context of feedback scheduling, a series of novel and effective solutions are developed for performance optimization of RTCS subject to dynamic resource constraints. The essential technological issues of feedback scheduling are formulated, with real-world application requirements and state of the art in mind. Considering constraints of three representative kinds of resources, i.e. CPU time, energy, and network bandwidth, we present different feedback scheduling methods that exploit a unified framework. They provide enabling technologies for closed-loop dynamic resource scheduling, and tackle a set of essential problems in the emerging field of feedback scheduling. Meanwhile, this thesis holistically addresses the problem of dynamic allocation of bottleneck resources within RTCS, thus providing flexible QoC management mechanisms and achieving overall performance optimization under dynamic environments.As regarding CPU scheduling, we are concerned with multitasking embedded control systems where the processing capacity of the processor is limited. With the goal of optimizing overall control performance, we analytically formulate the problem of optimal feedback scheduling, and then discuss relevant algorithmic solutions. Because algorithmic optimizer is too computationally expensive to be used online, we suggest a new methodology called neural feedback scheduling using neural networks. It could considerably reduce scheduling overhead, while delivering almost-optimal overall control performance. Besides, it is also characterized by good adaptability, robustness, as well as fault-tolerance, etc. Taking into account the unavailability of task execution times and measurement noises, we propose a fuzzy feedback scheduling scheme, which introduces fuzzy control technique into the area of feedback scheduling. Thanksto the powerful capacity of fuzzy logic in handling non-linear, imprecise and uncertain situations, fuzzy feedback scheduling doesn’t rely on task execution times, and makes the feedback scheduling system robust against measurement noises inside temporal parameters by copying with the uncertainty of workload and available resources in an intelligent fashion. Additional merits of this method include, e.g., ease of implementation and low scheduling overhead.In the part of energy management, we aim to reduce the energy consumption of the processor as much as possible while preserving the QoC of embedded control systems. A feedback control real-time scheduling approach, i.e. energy-aware feedback scheduling, which combines energy management with QoC management, is presented to solve such practical problems as CPU workload variations and task execution time uncertainty. This approach exploits dynamic voltage scaling technique and indirectly changes task execution times through dynamically adjusting CPU speed. The objective is to control the CPU utilization at a desired level. After analytically modeling the dynamic voltage scaling system, we present a control theoretic design and analysis method for feedback scheduler. In this way, deterministic performance of feedback scheduling is achieved and closed-loop energy management is realized. With the goal of further reducing energy consumption based on energy-aware feedback scheduling, we analyze the dynamic behavior of control systems with variable sampling periods and suggest an enhanced energy-aware feedback scheduling scheme, which adopts the methodology of graceful performance degradation. It dynamically adapts CPU speed using the dynamic voltage scaling technique, and at the same time, adjusts the sampling period of each control loop according to its current control performance. In this way, it takes advantage of flexible timing constraints of control tasks. Since sampling periods are enlarged as much as possible as long as the QoC is not jeopardized, energy consumption is further reduced, which in turn yields higher energy efficiency.With regard to bandwidth allocation, we focus on networked control systems that use priority-based fieldbuses and wireless control systems employing random medium access protocols. From a viewpoint of feedback control and network scheduling codesign, we present an integrated feedback scheduling scheme for multi-loop networked control systems. Exploiting a cascaded feedback scheduler, it adapts sampling periods with respect to the dynamic changes of available bandwidth. The objective is to maintain the deadline miss ratio at a desired low level and optimize the bandwidth allocation. In order to further improve overall control performance, we suggest a direct feedback scheduling mechanism, which re-assigns priorities to control loops at runtime based on their actual control performance. It is argued that integrated feedback scheduling could maximize resource utilization in underloaded scenarios, and could achieve graceful degradation of control performance under overload conditions. Consequently, the overall control performance could be significantly improved. Recognizing the inherent uncertainty of available link resources in wireless control systems, we present an adaptive feedback scheduling scheme that integrates the methodology of cross-layer design. This scheme deals with disturbing signals andvariations of link transmission rate by exchanging information between physical and application layers. The deadline miss ratio is controlled in a way that the overall control performance is optimized. Furthermore, we suggest an event-based invocation mechanism for feedback schedulers, which is quite effective in that quick response is enhanced while scheduling overhead decreases at the same time. The result is that practical efficiency of feedback schedulers would be considerably improved.In this thesis we go beyond the traditional control systems design methodology that features separation of concerns of control and scheduling, and discard the traditional control task model with fixed timing constraints. Completely fresh methodology and solutions are provided for RTCS research and development in the context of new technological trends. This thesis preliminarily establishes a systematic framework of feedback scheduling with a set of promising results, which enrich this emerging area from various perspectives. With an enabling technology produced, it is believed that our results will contribute to future holistic integration of control with computing and communication.更多还原