Research On Distributed Heterogeneous Model Composition And Execution

On the analysis of complex decision problems, modelswhich are geographically dispersed, functional singleness and different types in nature have to be logically and intelligently composed and executed, to solve the complex user requests that single model cannot solve. This is the problem of heterogeneous model composition and execution. The difficulty of distributed heterogeneous model composition and execution lies in: Firstly, the heterogeneous nature of models requires a method to compose the quantitative and qualitative models. Secondly, the dynamic nature of distributed environment requires the composite model can be repaired more effectively and intelligently when encountered the execution faults in the execution, while the existing recovery mechanisms cannot ensure the successful execution of the composite model. In response to these circumstances, this dissertation constructs a distributed model management framework, and studiesfrom four aspects which are heterogeneous model representation, the automatic generation of composite model, execution and evaluation. Then, asystematic solution is proposed.A unified and complete model management framework is the foundation and support for the research of model composition and execution. This dissertation constructs a Web service based distributed model management framework that integrates fragmented standards and technologies, and may let the model composition and execution become coherent and smooth. Web service is a standardized set of distributed application development and access methods, it can be used to describe various model resources and software components, to achieve the loose coupling between a variety of related technologies, standards and model functions.Model represents is premise of functional design of model composition. This dissertation unifies and abstracts the heterogeneous model representations and thus forms a general principle. Since there are no methods about the service-oriented representation of qualitative models can be referred to, this dissertation represents it as a set of if-then statements and completes the service-oriented representation using business engines. To ensure the models are composed on functions, this dissertation proposes a model ontology encoding the domain-independent basic modeling concepts and domain-specific modeling related knowledge, and then gives the semantic description of heterogeneous model services by conferring them with preconditions and effects, andreservesthe interfaces for the heterogeneous model composition. Thus, the distributed heterogeneous model resources are represented as the form of Semantic Web services.The key of heterogeneous model composition is to establish the relationship between quantitative and qualitative models, and then design new composition methodbased on this relationship. This dissertation analyzes the concrete form between heterogeneous models, and proposes a new idea that converting the heterogeneous model composition to a special AI planning problem. The inputs and outputs of heterogeneous model can be used to determine the satisfaction of model’s preconditions and effects. This kind of relationship is uncertain and partially observable, and need to be determined at run-time. According to this feature, this dissertationproposes a mapping between the syntax and semantics descriptions to theuncertain, partially observable planning problems, and solves it using the planning as model checking technology.To ensure the successful execution of the composite model, this dissertation improves and unifies the present failure recovery strategies, and makes the composite model can be repaired intelligently and efficiently. The network linking model resources is dynamic, it affects the success or failure of composition plans. To eliminate the information asymmetry between modelers and end-users, this dissertationproposes to build model ontology to share the knowledge about failure recovery. Besides, to realize intelligently repairing the composite model, this dissertationconverts the failure recovery problem to a local model composition problem.The evaluation of the composite model plan is an extension of the above study. The solving process of automatic composition of heterogeneous models and failure recovery process may generate multiple plans, and all of them have the same function but different non-functional performances. To characterize the non-functional properties of composite plans, this dissertation uses the aggregate Qo S scheme to represent the global quality of service. To make the optimal selection between multiple composite plans and represent the explicitand implicitpreference of users, this dissertation uses the ELECTRE-II method for the multi-objective evaluation and merits of sorting, thus ensuring the integrity and systemic of the whole process of heterogeneous model composition and execution.