Quality of Service-Driven Services Composition Approaches in Uncertain Internet of Things Environments

Abstract

The Quality of Service (QoS)-aware Internet of Things (IoT) Service Composition (QIoTSC) problem, subject to global QoS-user constraints, is recognized as one of the NP-hard, challenging, and constrained combinatorial multi-objective optimization issues. In uncertain IoT environments, the QoS parameters of elementary IoT services are frequently non-deterministic, exhibiting vague and ambiguous values due to various environmental factors such as changes in network architectures, communication congestion, and economic policies. Consequently, QoS ambiguity is considered when formulating the QIoTSC problem. Given that fuzzy numbers are robust, versatile, and general models for expressing uncertain values, the QIoTSC problem is formulated as a fuzzy constrained multi-objective optimization (FMOQIoTSC) one. To address the FMOQIoTSC, a novel Fuzzy Multi-Objective Teaching Learning-Based Optimization (FMOTLBO) algorithm is developed. Indeed, the non-dominated ranking method and crowding distance computation from the well-known NSGA-II algorithm are integrated into FMOTLBO, while the deterministic dominance relation and crisp crowding distance formula of NSGA-II are adapted to handle the fuzzy ambiguity of QoS values. Additionally, FMOTLBO utilizes an external archive to store its Pareto-optimal non-dominated solutions. Moreover, rather than employing the continuous equations from the teaching and learning processes of the conventional TLBO algorithm to generate new solution positions, FMOTLBO introduces and applies new discrete methods for positioning learners. Furthermore, a discarded learner substitution process is incorporated into FMOTLBO to enhance diversity and prevent the algorithm from becoming trapped in local optima. Comparative results between FMOTLBO and a recent bio-inspired multi-objective QIoTSC approach, using real and simulated QoS datasets of varying sizes, demonstrate the superior performance of FMOTLBO over the compared approach.