Along with the adaptation of Internet of Things (IoT) to support various industrial applications, the cooperation and coordination of smart things is a promising strategy for satisfying requirements that are beyond the capacity of any single smart thing. To address this challenge, a two-tier IoT service framework is proposed, where the functionalities provided by smart things are encapsulated into IoT services, which are categorized into service classes. The service network is constructed by considering the invocation possibility between service classes, and service class chains are generated using traditional Web service composition techniques, where the functional specification of certain requirements is considered. Considering factors, such as spatial and temporal-constraints, energy efficiency, and the configurability of IoT services, selecting IoT services for the instantiation of service classes contained in chains is reduced to a multiobjective and multiconstrained optimization problem. Heuristic algorithms, such as the genetic algorithm (GA), ant colony optimization (ACO) and particle swarm optimization (PSO), are adopted to search for optimal IoT service compositions. An experimental evaluation shows that PSO performs better than the GA and ACO in searching for approximately optimal IoT service compositions and reduces the energy consumption, thus prolonging the network lifetime