Journal of King Saud University: Computer and Information Sciences (Sep 2022)
An approach based on genetic algorithms and neural networks for QoS-aware IoT services composition
Abstract
In the Internet of Things (IoT) environment, the services provided by the connected objects are published as services through the web. This allows to machines to interact between them and, makes the IoT services composition possible. However, the vast proliferation of smart object generates services with the same functionalities but different in terms of quality of services (QoS) proprieties. This makes the satisfaction of the user requirements often complex and a NP-hard problem. Indeed, respecting the QoS constraints (user preferences in terms of QoS) is a challenge, due to the high number of candidate services for the composition. This challenge consists of selecting the most appropriate services so that the composite service must meet both the functional and the non-functional requirements of the user. To deal with this challenge, we propose an approach based on Genetic Algorithms (GA) and Neural Networks (NN) for QoS-aware IoT Services Composition in the context of large-scale environments. The combination between GA and NN allows finding the quasi-optimal IoT service composition. This latter is based on global QoS optimization. To reach this objective, the QoS intervals are decomposed into M QoS-levels to engage them into the theoretical composition. Then, the proposed first GA is used to obtain the ideal theoretical composition with an overall QoS optimization. Afterward, the proposed NN is used to eliminate the inappropriate concrete IoT services, and retain only the services having the same categories of the atomic theoretical services composing the ideal theoretical composition. This allows us to optimize the search space as well as the execution time. Finally, we apply the second GA on the concrete services of the retained categories, in order to obtain the IoT service concrete composition with an overall QoS optimization. The simulation results show that the proposed approach has the best composition time, the best Hypervolume indicator and the best compositional optimality compared to SC-FLA, Improved GA and MGA approaches. On another side, it has almost the same performances compared to TS-QCA and, it finds the near-optimal composition in a very short time compared to PSA, which is an optimal approach. Thus, the obtained results show the effectiveness of our approach.
