A Time Series Based Research On QoS Prediction Of Semantic Web Service Composition

As the main implementation technologies of SOA, web services can realize the interoperability between applications using web technology standards. Semantic web services become the focus of attention, because it can realize semantic reasoning and the automation of web service via increasing the semantic information on the basis of web services. However, the performance of a single service is limited. How to describe services formalized according semantic information and then realize web service composition to get value-added services and this is the problem which semantic web service composition research.OWL-S, a kind of ontology description language is one of the main criteria for composition modeling and description. Application domain knowledge and relevance between all kinds of resources is described, by which OWL-S expresses semantic information. Today, as the current research in semantic ontology language focuses on its formalization and qualitative analysis, the non-functional properties and quantitative analysis is relatively less. Service combination program which does not meet the requirements in non-functional properties cannot meet the needs of users. In order to enable users to judge whether a service meets the requirements in non-functional properties or to choose the most optimal ones from the functionally similar services group, we should predict the Quality of Service(QoS) of composite services.Based on the disadvantages of the method of presetting distribution type and averaging historical value, this paper introduces the network environment factors to the service composition QoS prediction. In this paper, in order to realize semantic expression of atomic process and composite process, we propose an approach employing non-Markovian stochastic Petri net(NMSPN) as the intermediate representation and map OWL-S elements including atomic process and composite process into equivalent NMSPN representations. Time property, reliability and cost are considered and its expression and calculation method are analyzed in the paper. We employ the time series model to predict the future QoS of individual service activities based on historical QoS attribute values. By this way, the influence of the changing network environment factors on service composition QoS properties is considered and the influence of abnormal data to forecast the result is weakened. For composite service, we employ reduction based model to simplify complex logical relations until it is an atomic process. The predicted values are used as the inputs of the QoS reduction model to calculate the QoS values of the whole process.In order to analyze the feasibility and effectiveness of this predicted framework, an instance is introduced. Time series forecasting results to atomic services can match the actual curve of QoS preferably. We conduct a comparative study and accuracy analysis with traditional QoS prediction models including deterministic-time-based model and general-time-based model to evaluate the effect of the prediction QoS of composite services based on time series forecasting results as input.