SummaryIn sensor clouds environments, the provisioning process is a crucial task since it is responsible for selecting physical sensors that will be used to create virtual sensors. However, most works consider the allocation of all sensors within the region of interest, causing serious problems such as the wasting of energy consumption. The objective of this paper is to present ACxSIM, an automatic approach to the provisioning of virtual sensors. ACxSIM includes two algorithms: adaptive clustering algorithm based on similarity (ACASIM) and ant colony optimization for sensor selection based on similarity (ACOSIM). ACASIM first clusters the sensor nodes based on the similarity of its measurements (exploiting the temporal and spatial correlations between them), which may create clusters with nodes not physically close to each other. Therefore, in ACASIM, a cluster represents different geographical areas whose nodes have correlated measurements (according to a defined error threshold). Later, ACOSIM, based on ant colony optimization algorithm, creates virtual sensors by selecting only a subset of nodes from each cluster. In this way, the overall energy consumption of sensor nodes is reduced, prolonging the lifetime of the sensor cloud. Results from experiments in Intel Lab dataset show that the ACxSIM reduces energy consumption by 73.97%, providing a solution to be considered in sensor cloud scenarios.