Modeling And Optimizing Cyber Physical Systems Based On Pattern Ontology And Machine Learning

In recent years,the Cyber-Physical Systems have received increasing attention from academia and industry.Their complexity makes their modeling and optimization problem a research hotspot.The Cyber-Physical Systems are hybrid and full of uncertainty which brings some difficulties to the modeling of Cyber-Physical Systems.Meanwhile,Cyber-Physical Systems are domain-specific.The quality of Cyber-Physical Systems modeling heavily depends on the designer's experience.In addition,the controllable and uncontrol-lable factors coexist in the cyber world of Cyber-Physical Systems.It is also a challenge for researchers to perform the optimization of designs from the views of these factors.To address these problems,this thesis proposes methods for modeling and optimizing Cyber-Physical Systems based on pattern ontology and machine learning.The main work includes the following aspects:.A method for modeling Cyber-Physical Systems based on stochastic hybrid au-tomata and pattern ontology is presented.Cyber-Physical Systems are modeled by network of stochastic hybrid automata.By extracting the basic concepts of s-tochastic hybrid automata,the upper pattern ontology is constructed.And the do-main pattern ontology is constructed by instantiating the upper pattern ontology.An approach for modeling Cyber-Physical Systems based on the pattern ontology is presented.And then we achieve the quantitative assessment of system performance by using the statistical model checker-UPPAAL-SMC..For the uncontrolled factors of physical world,a method for choosing the re-source scheduling strategy based on the support vector classification is present-ed.The performance of different scheduling strategies under different user behav-ior is regarded as initial data.And the sample set is obtained by data preprocessing.Then we establish the classification model by training the sample set.We also show how to choose the resource scheduling strategy using the classification model..For the controlled factors of physical world,a method for parameter config-uration multi-objective optimization based on support vector regression and Pareto optimization is presented.First the parameter instances of to be optimized parameters are generated.The quantitative assessment of some selected instances is regarded as the training set.The relation model between system performance and parameter instances is established based on the training set.By predicting the performance of other parameter instances,we get the performance of a large num-ber of parameter instances under multi targets in a short time.At Last,we get the multi-objective parameter instance by Pareto optimization.At last,we use a typical Cyber-Physical System-Smart Building System to validate our proposed approaches.Experiment results show that the proposed pattern ontology based approach can help the system designers establish high-quality system model ef-ficiently.And the proposed optimization methods based on machine learning can help the system designers obtain the optimized system deign which meets the non-functional requirements.