In service-oriented environments, abstract business processes can be implemented by concrete services to build complex applications. Given global user constraints, service selection allows to identify the best combination of services with respect to the business constraints. Generally, the selection problem is challenging, but it is particularly complex when dealing with QoS (quality of service) values, which can change during the time associated with temporal constraints. Indeed, these constraints make the selection problem heavily constrained, which can present a barrier for enabling effective service selection. Unlike static QoS values which have been deeply studied in the existing service selection approaches, time-dependent QoS associated with temporal constraints are insufficiently taken into consideration. Moreover, existing approaches cannot handle heavily constrained problems and usually do not provide strategies to detect the source of failure in order to enhance the selection problem in case there is no solution. In this paper, we introduce a new service selection approach, while considering time-dependent QoS values associated with temporal constraints. First, pruning techniques are proposed. The aim of the pruning process is twofold: (1) it allows for reducing the search space and thus, enhancing the efficiency of the selection process; (2) it allows for improving the selection problem by detecting at earlier stages the possible causes of failure, even before the selection process. Second, based on the pruning phase, improvement techniques are proposed to identify possible actions for finding a solution. Third, an exact and an approximate service selection algorithms under several constraints are given. Finally, we demonstrate the effectiveness of our approach through experimental results.