Research In Self-Optimizing Method For Cyber-Physical Systems

With the lively development and common using of Communication Technologies, Software Technologies, Embedded Technologies, Cybernetics and Artificial Intelligence in our daily life，our living condition become more and more comfortable. AI related technologies research and application make people not just satisfy with simple usage of systems,they expect more personalized service. They hope systems could sense their requirements and make better use of the limited source. By such motivation, Cyber-Physical Systems(CPS) get generated. Because of its hybrid subsystems, CPS has huge scale. It mainly contains computing process,network process and physical process. CPS is a kind of robotic which is based on network while composited by hybrid subsystems are the greatest challenge to CPS. It is essential important to make the whole system become more flexible to its running environment by coordinating its subsystems.This paper aims at solving the problem of CPS Self-Optimizing. Based on our points, to have better analysis of CPS, it should be divided into different layers. Firstly we should tackle the problem of self-optimizing of underlying configuration of system, which is called dynamic configuration of systems, then is CPS implementation level which should also have the capability of Self-Optimizing. Based on previous analysis, we should analyze the autonomy of abstract level. In this paper, we have done the following work:(1) Firstly, analyze the importance of system framework to the Self-Optimizing of CPS and give some common examples of CPS framework. Then propose a novel framework which is based on system’s components, only in this way, we can realize the Self-Optimizing of system.(2) Based on underlying level of systems, analyze the dynamic configuration of systems. We propose a extendable dynamic configure algorithm based on the formal language-XML. Because of huge requests of system, we adopt classical interrupt management method of CPU and put it into effect on the systems to have a better management of system requests. Combine CPU’s interrupt method with Perceptron, we propose the passive perceptron algorithm.(3) Based on dynamic configuration,firstly, we can realize the Self-Optimizing of software system. Combine CPS characters with general features of software systems,we think that currently software Self-Optimizing algorithms can not really achieve systems self-adaption. Hidden Markov Model(HMM) has a nice description of statuses of systems while it can not model the statues which are not stored in database of functions.To tackle this problem,we propose a new method which is called constant probability of single system. Based on this idea, we combine the online evolutionary algorithm with HMM algorithm together to solve previous problem, the result shows that it is much better than existed algorithms.(4)Based on previous analysis and algorithms,lastly, we analysis the autonomy of system.We analyzed Reinforcement Learning Algorithm(Q Algorithm) and found out that the point of this algorithm is systems usually interact with their living condition.According to this, propose the CPS autonomy algorithm based on similarity computing. At last,we do the simulation based on the proposed framework.(5)To verify those proposed algorithms, simulation need to be carried out. Therefore, we do three different types of simulation and compare the simulation result with existed algorithms.