Research On Key Technologies Of Resource Scheduling For Cyber-physical System

With the fast development and integration of technologies like MEMS(Micro-Electro-Mechanical Systems), sensors, pervasive computing and communication, the fusion of "3C"(Computing Communication & Control) is becoming a new technical trend. Therefore, Cyber-Physical System(CPS) as a concept was proposed while some other new concepts like cloud computing and network of things emerged. Here, CPS is defined as a sort of complex systems that integrate information processing, network transmission and physical control. CPS combines information technology, communication technology and control technology to achieve a fusion of information processing as well as physical processes in two dimensions(i.e. space and time). CPS is deemed to the core technology, which is likely to facilitate another industrial revolution.Compared to the embedded real-time systems and networked control systems, CPS strives to coordinate effectively and schedule optimally of informational physical resources. In this sense, CPS aims at performing real time sensing and dynamic monitoring for large-scale complex systems, and further providing more flexible, intelligent and efficient information-related as well as control-related services for end-users. In CPS systems, there would be a large number of heterogeneous resources with different functions. The resources in CPS are of high heterogeneity. Also the resources would be of high flexibility to access/exit. The corresponding states vary constantly. In addition, the topology of the CPS changes continuously. It is difficult to forecast accurately the physical environment of the CPS. These aforementioned issues challenge heavily in resource scheduling in CPS. So, there is a need for a flexible and effective resource scheduling techniques to cope with uncertainty changes of the physical environment and user application requests. This sort of technique is likely to be one of the key technologies in CPS research to realize efficiently stable operation of CPS. Therefore, in this paper, the scheduling problems relating to sensing resources, computing resources and mobile resources in CPS are identified, conceptualized and operationalized. As listed as follows, the main contributions of this paper are:(I) The basic concepts and the constitutions of CPS are framed and a service-oriented architecture framework for CPS is proposed. It can shield the heterogeneity of resources, simplify the capability description of resources, package the function of resources and unify the interface of resources with the SOA. The architecture of CPS is mapped into node layer, network layer, resource layer and service layer according to the service-oriented thoughts. Such a categorization of layers could enable researchers and developers to better understand the idea of cyber-physical within the proposed architecture. In allusion to the problems of CPS resources varying dynamically over time and space, we propose a capability-based CPS resource model. This main thought is borrowed from the knowledge of people’s daily management: when using the resources the system we care about their ability to complete the task rather than the concrete status and function of the resource. According to the proposed capability-based model, CPS resources are grouped into sensing resources, computing resources and actuating resources. This sort of grouping provides a foundation to manage and schedule CPS resources.(II) Sensing resource-scheduling problems are studied for CPS. The effectivity and insufficiency of the greedy sequential selection algorithm are analyzed when solving sensor selection problem. Two improving methods are proposed for solving the problem of randomness of the initial solution. Currently, sensing resource scheduling, i.e. so-called SSP(Sensor Selection Problems) is a hotspot in CPS research areas. We study the effectivity and shortcomings of greedy sequential decision algorithm when solving the SSP and prove the greedy sequential decision algorithm is efficient by experiments and theorem-proof. However, as the state of CPS contains multiple components, the initial solution of greedy sequential Selection algorithm has to use a random strategy. This strategy will degrade the algorithm due to its arbitrary initial solution. Against this problem, this paper proposes two solutions, i.e., exhaustive search(EGSS) and backward greedy(BGSS) sequential Selection algorithm. Further, a simulation experiment within a target-tracking scenario is conducted to test the proposed algorithms. The experimental results show that the proposed algorithms can effectively solve the problem of arbitrary initial solution. Therefore, the improved algorithms are better than the original algorithm if the system state is vectorial.(III) The relationship between system state estimation and similarity of sensor observation models is systematically researched. An observation-model clustering based energy-efficient sensor selection strategy is proposed. The objective function of sensor selection model is established with the Fisher Information Matrix. It is proved that the smaller the similarity of observation models is in the selection subset, the higher the estimation accuracy would be. The proposed strategy has two main steps:(1) clustering the sensors base on the observation models; and(2) selecting one sensor in each cluster. Suitable sensor would be selected from each cluster based on two indicators, i.e. sensitivity of the sensor; and the residual energy of the sensor. Finally, simulation experiment is carried out within a target-tracking scenario. The experimental results show that the selected sensors by the proposed strategy can maintain the sensors’ residual energy balance and simultaneously it can gain a higher state estimation accuracy of the system.(IV) The scheduling problems of computing resources in CPS are studied. A distributed heterogeneous computing-resources scheduling algorithm called IHEFT is proposed. As a typical distributed heterogeneous system, every single resource of CPS has limited capability of computing. The system uses distributed parallel computing techniques to obtain high computing performance in dealing with computing tasks such as the system state estimation. In this technique, it is hypothesized that the resources are connected in a high-speed network. The computing tasks of CPS are modeled as Directed Acyclic Graph(DAG). Each node in DAG represents that an atomic task is assigned to the appropriate computing resource. Then the tasks assigned to the same resource will be scheduled according to the starting and ending time of each task to realize the scheduling of computing resources. We provide a better distributed heterogeneous list scheduling algorithm called IHEFT based on the classical algorithm HEFT. The theoretical analysis and experimental results illustrate that IHEFT performs better than HEFT CPOP, LDCP.(V) Mobile resource scheduling problem of CPS is studied, and a state transition-based mobile resource scheduling strategy is provided. Mobile resources are sort of indispensable important resources, which can effectively improve the sensing and actuating capabilities of CPS through changing the location of the resources. In this paper, the decision problems(of determining the task scheduling sequence when the mobile resource waits for processing multiple tasks) are studied. The state transition model of mobile resource task scheduling problem is built according to different types of tasks. The optimal scheduling sequence of each mobile resource is obtained based on the state transition models. The state transition model of CPS mobile resource, in dealing with sensing and actuating tasks with the example of UAV reconnaissance missions and combat missions, is built. The optimal scheduling sequence is reached based on computing the costs and benefits of the task execution order.In conclusion, focusing on the problems of resource-scheduling of CPS, this paper analyzes some problems relating to sensing resource scheduling, computing resource scheduling and mobility resource scheduling, respectively. This paper proposes the corresponding scheduling algorithms and strategies. These algorithms and strategies, to some extent, will contribute theoretically and practically to promoting the overall scheduling performance as well as the related resource management techniques in CPS.