Research Of Fast Optimized Scheduling Strategies And Distributed Computing For Two-layer Networked Learning Control System And Extended Application

With the rapid development of network and communication technology,intergrating with the traditional control theory, networked control systems (NCS)emerge as the times require. Networked control system has been gradually applied tothe field of advanced automatic manufacturing, electric power, aerospace and robot,due to various advantages including flexible design, resources sharing, andpracticability. However, network control system, which is supported by commoncommunication network, is subject to many constraints of network capacity andcommunication parameters. To some extent, the performance of the system not onlydepends on control algorithms, but on reasonably effective scheduling of networkcommunication resources as well. In the meanwhile, control algorithms andoptimization scheduling methods trend to be more complex in order to meet thedeepening needs of control. Also, it is really a severe challenge to meet the fastcomputational requirements with limited computing resources, which can further affectthe overall system performance. Therefore, under the limited network communicationbandwidth and computing resources, a two-layer networked learning control system(NLCS) architecture based on fast computing platform is proposed to reach the goal ofsystem real-time requirements. The focus of this paper is on the development of fairoptimization scheduling strategy and the design of fast computing platform. The mainwork is summarized as follows:Firstly, a two-layer NLCS architecture based on fast computing platform ispresented and the differences with NCS are analyzed. A fair multi-objective bandwidthoptimization scheduling model with communication constraints is formulated. For thecomplexity of the algorithms caused by increasing demand of users and the real-timerequirements in networked control system, we introduced a fast computing platforminto the traditional networked learning control system. This new architecture not onlycan guarantee the high reliability of the local control network, but also ensure real-timerequirement of computing resources sharing and calculations control and optimization.Furthermore, considering the bandwidth scheduling problem with communicationconstraints in a two-layer networked learning control system, a fair non-cooperativegame model combined with expert knowledge is proposed. A suitable utility functionis also designed, forming multi-constrained network bandwidth scheduling optimization strategy.Secondly, for solving the optimal network bandwidth scheduling problem, a newoptimization method based on quantum evolution is proposed. Unlike traditionalintelligent optimization methods, by introducing microscopic particles in the theory ofquantum mechanics, quantum state probability amplitude is used in quantum-inspiredoptimization method to encode the current position of the search group for increasingthe diversification of the solution space. Moreover, quantum rotation gates are utilizedfor the variability of group position and the search of the optimal location, which canfurther increase the diversity of search population. The quantum-inspired optimizationmethod can gain the convergence to a global optimal solution to the networkbandwidth scheduling problem, avoiding sticking into local optimal solution.Meanwhile, through setting of proper value for step function of quantum gate, we canachieve algorithm search with arbitrary precision.Thirdly, two new optimization algorithm solutions are proposed for fast solvingthe optimal network bandwidth scheduling problem accelerating factor basedshuffled frog leaping algorithm (SFLA) and two-layer hierarchical market competitionalgorithm (THMCA). Although quantum-inspired optimization method is able toenhance the quality of the optimal solution, its calculation time is instability. Thismethod is only suitable for offline optimization. Consequently, a new meta-heuristicoptimization method with accelerating factor called SFLA is proposed, which isefficient in finding global solutions. The SFLA combines the advantages of the boththe genetic-based memetic algorithm (MA) and the social behavior-based PSOalgorithm. The most distinguished advantage of SFLA is its fast convergence speed.However SFLA suffers from the curse of dimensionality problem, especially in solvingheavy constrains, large scale optimization problem in power system. Inspired bycompetitions among enterprises in economic activities, a novel two-layer hierarchicalmarket competition algorithm (THMCA) is proposed. Market competitions amongthese conglomerates lead to the convergence to a monopoly at the end, resulting in anoptimal solution of the above problem. The algorithm can effectively reduce thedimensions and complexity of the optimization problem, ensuring the quality ofsolutions and computing time, which is suitable for seeking fast solution of large-scaleconstrained optimization problem.Fourthly, in order to further improve the calculation speed of complex algorithmsin two-layer NLCS with limited local computing resources, cooperative game based parallel grid computing and load balancing strategy are established. Integrating gridcomputing into the two-layer NLCS, the computational model of parallel computingtasks is formulated, including selfish grid mathematical models and foreign job costmodel. And cooperative game theory based fair analysis of the model verify that loadbalancing is the global optimal strategy. Furthermore, a novel load balancing scheme,cooperative game based bounded iterative load balancing algorithm, is proposed forheterogeneous load clusters in the multi-cluster selfish grid. Finally, distributedcomputing resources are intergrated as high-performance computing platform.Effective scheduling of computing tasks is used to meet the real-time computingrequirements.Fifthly, we investigate a flexible elastic cloud computing based fast virtualdeployment scheme for a two-layer NLCS. Selfish grid still needs to consume a certainamount of local computing resources. And integrating distributed computing resourceshave high computing costs of the overall system. Therefore, high performance clusterarchitecture with virtual properties that hide the complexity of distributed physicalinfrastructures is presented. The architecture is support for cloud computing todynamically deliver heterogeneous computational environments and partition thecluster capacity, adapting to variable demands in a networked control system. Also, aperformance model employing cloud resources for elastic clusters is developed, whichplans the capacity of the cluster to meet a performance policy and cost request tocomplete a given work load. The performance of model has been evaluated in theexecution of heuristic computing workloads. Finally, the comparison experimentalresults have demonstrated that the virtualization based elastic clusters constitute afeasible and high performing computing platform for a two-layer NLCS.Finally, after the theoretical research and simulation results gain the effectiveverification of the proposed scheme, the best two-layer hierarchical marketcompetition algorithm (THMCA) combined with expert system is applied for solvingthe unit commitment problem in power systems under the fast elastic cloud computingplatform. Expert system are used to produce several expert rules for heavy constraintshandling not only in the prescheduling process and in the THMCA process as well,ensuring that the positions of all companies are feasible and near-optimal solutions tothe UC problem. The algorithm running on fast elastic cloud computing platform isshown to have a fast execution speed for UC application and the comparisonsimulation results on a power system with up to100generating units have demonstrated the effectiveness on cost reduction of the proposed method.