Research On Key Techniques For Context-Aware Services Coordination Of Internet Of Things

With the development of Internet of Things(IoT),various IoT systems are pervading every corner of our daily lives,such as smart city,intelligent transportation,smart home,etc.An IoT system provides intelligent services by organizing a large number of heterogeneous components deployed in an open environment for task coordination.IoT systems are featured with heterogeneity,large scale,and dynamic context,which brings a great challenge to the effective coordination of IoT.Service-Oriented Architecture(SOA)is widely adopted to realize the coordination of IoT systems due to its inherent support for interoperability.The SOA-based coordination for IoT encapsulates capabilities of software and hardware components as IoT services,focusing on the exploration of key techniques such as service description,service recommendation,and service composition under the SOA infrastructure.However,there are still some problems to be addressed.Firstly,service modeling protocols are verbose and IoT contextual characteristics are largely ignored or not explicitly defined in existing IoT service descriptions.Secondly,the dependency on precise descriptions of user preferences is heavy,and the combination of contextual and functional characteristics is insufficient in IoT service recommendations.Thirdly,little attention is paid to context-awareness and an effective way for composition process design and implementation is absent in IoT service compositions.This dissertation proposes a context-aware IoT services coordination approach from the perspective of SOA,to cope with the IoT characteristics of heterogeneity,large scale,and dynamic context.The aim is to provide a systematic solution for effective IoT services coordination in terms of three key techniques,namely unified service description,active service recommendation,and adaptive service composition.The main contributions are as follows:(1)A lightweight and context-aware IoT service description model is proposed in order to address the problem of unified description of heterogeneous IoT services.For contextual characteristics of IoT,multi-dimensional contexts are explicitly defined and then incorporated into the service description model to improve the quality of services provisioning.In addition,a context-aware IoT service description model is built based on the microservice architecture.It not only provides a unified service model with lightweight protocols,but also comprehensively describes the context,service,and interface characteristics of IoT services.By utilizing the proposed model,the heterogeneity among IoT components can be shielded at the service layer,and thus the complexity of service integration is reduced.A case study is conducted with two typical scenarios of elderly care.The results of the case study show that the proposed model provides a unified way for IoT service description and explicitly incorporates the context into the description model.Compared with existing approaches,the proposed model resorts to a lightweight microservice architecture and explicitly defines the IoT context,which provide a more comprehensive and unified way for IoT service description.(2)A deep collaborative filtering-based context-aware IoT service recommendation approach is proposed in order to address the problem of active recommendation of large-scale IoT services.The probable functional preferences of users are derived by mining historical preferences and potential preferences from usage history to ease the service provisioning.In addition,the deep implications of contextual features are learned using a neural network and then incorporated into the service recommendation for context-awareness.A deep collaborative filtering model is designed by combing deep neural networks and collaborative filtering to capture the interactions between users and services from contextual and functional features and thereby recommend services.It realizes the proactive service provision and reduces the difficulty of service selection.A series of experiments are conducted using a service dataset to evaluate the proposed approach.The experimental results show that the proposed approach is effective in IoT service recommendation,and the incorporation of contextual features can improve the accuracy of service recommendation.(3)A deep reinforcement learning-based context-aware IoT service composition approach is proposed in order to address the problem of adaptive IoT service composition in dynamic contextual environments.To enable the generation of the optimal service composition policy,contextual constraints and pragmatics of services are modeled to characterize the service provision capabilities in a specific context.Then,the pragmatics of services are used to drive the learning process of deep reinforcement learning toward the context-aware service composition policy.For the process design and implementation,the Business Process Modeling Notation(BPMN)and model-driven development techniques are used to automate the transformation of service composition processes.Finally,the composition process is configured with the optimal policy to derive a composition process instance that meets both the contextual and functional requirements.The proposed approach is evaluated with a case study and a simulation dataset.The experimental results show that the proposed approach can realize context-aware and on-demand service compositions,and it is adaptable to the dynamically changing environment.