Research On Personalized QoS Prediction And Recommendation For Manufacturing Cloud Service

The rapid development of the Internet-based platform has brought new vitality to the service-oriented manufacturing industry.Cloud manufacturing platform that integrates advanced technologies such as big data,cloud computing,and the Internet of Things has emerged at the moment,and is widely applied among large,medium and small manufacturing enterprises.Manufacturing cloud service,as one of the key components,continues to expand in size and grow rapidly,resulting in more and more functionally equivalent service on the Internet.Thus,how to accurately predict the quality of service,namely the QoS,of manufacturing cloud service in different scenarios,and then make a reliable recommendation is a hot issue in current research.In addition,with the continuous extension of cloud manufacturing,the personalized needs of users are becoming more and more complex,and manufacturing cloud service is beginning to develop toward the trend of network collaboration,which makes service composition optimization gradually becoming another hot issue recently.Furthermore,it is worth mentioning that the successful application of key technologies for service prediction and recommendation in a cloud manufacturing environment not only brings new opportunities for related research on a single service,but also raises higher challenges for in-depth research on service composition.Inspired by this,this dissertation focuses on the existing problems in the current personalized prediction and recommendation approaches and service composition optimization(SCO)algorithms,as well as the deficiencies of the subsequent prediction and recommendation research for service composition.And the main research contents of our systematic research work are as follows:(1)Considering the fact that traditional approaches have a low accuracy and are easily affected by the data sparsity problem,a hybrid QoS prediction approach based on local collaborative filtering and global case-based reasoning is proposed by mining the implicit data and explicit information in the "user-service" complex network.Firstly,in order to achieve higher prediction accuracy,the personalized influence of historical data between similar users and services is fully considered when constructing the similarity model,and then a similarity-enhanced local collaborative filtering approach is proposed.Secondly,in order to overcome the data sparsity problem,the concept of case similarity is introduced,and a global QoS prediction approach based on the case-based reasoning is proposed,which can make full use of both user and service information based on the case library.Finally,an ensemble model is established to combine the results of two approaches.Related experiments are conducted to demonstrate the effectiveness of our approach,particularly when the data is very sparse.(2)In addition to the data sparsity problem,considering that dissimilar task requested by the similar users may lead to quite different prediction results,and thus cannot achieve higher prediction accuracy and recommendation quality.To solve the above problems,the concepts of task similarity and trust are introduced from the context of the "user-task-service" complex network,and a personalized clustering-based and reliable trust-aware QoS prediction approach is proposed for the cloud service recommendation.Firstly,on the basis of in-depth analysis task characteristics of the cloud manufacturing platform,task similarity is calculated by incorporating explicit textual information and rating information as well as implicit contextual information,and then integrate the task similarity into a K-medoids clustering algorithm to identify a group of similar users.Secondly,in order to ensure the reliability of the data source,a trust-aware collaborative filtering approach is designed by fusing local and global trust value in social network,to build a trust network of similar users.Finally,combining our clustering-based algorithm and trust-aware approach to provide personalized QoS prediction and reliable recommendation.Related experiments are conducted to demonstrate that task similarity and trust do improve the effectiveness of cloud service prediction and recommendation.(3)Considering the dynamic optimization problem of service composition under the new situation of complex personalized manufacturing requirements,the concept of service correlation is introduced in "user-service composition" scenario,and a correlation-aware service composition optimization algorithm is proposed.Firstly,in order to consider the complex correlation relationship between services,a comprehensive QoS correlation model is established by analyzing the basic QoS attributes and the characteristics of the service composition structure.Secondly,a parallel max-min ant system optimization algorithm is designed.In this algorithm,another special ant is employed to maintain the diversity of the population,and then a local learning strategy is simultaneously adopted to accelerate the convergence rate.Moreover,the case library,enhanced with an autonomous learning mechanism,is also applied to further improve the searching efficiency.Finally,the comprehensive QoS correlation model is embedded in the algorithm to seek the optimal solution.Related experiments are conducted to demonstrate the effectiveness of our model and algorithm.(4)Based on the above research work,in the cloud manufacturing environment,a new idea for personalized prediction and recommendation of service composition is proposed.Firstly,an overall prediction scheme for service composition is given out by constructing a service composition prediction framework and proposing a diversity recommendation strategy.Then,a prototype of a manufacturing cloud service-oriented hybrid recommendation system is designed,including requirement release module,QoS prediction module,SCO module and service recommendation module.Moreover,in the system,we take our proposed prediction approaches and service composition optimization algorithm and diversity recommendation strategy as well into consideration,respectively.Finally,these system modules and components are verified and visualized in two different application scenario cases of “Drone production and manufacturing” and“Extruder project”,respectively.