In the oil and gas industry, one of the most important activities is the intervention in wells for maintenance services, which is necessary to ensure the production of oil. These interventions are performed by workover rigs that are available to serve a large number of wells according to a schedule. In this thesis, we proposed three integer linear programming models to efficiently address the problem of intervention in onshore oil wells. The first model determines the schedule of a set of homogeneous rigs, with the objective of minimizing the total production loss. This model is an improvement of the model proposed by Costa and Ferreira Filho (2004). The second model is an extension of the previous one and also considers the sizing of a heterogeneous rig fleet, with the objective of minimizing the production loss cost and the rig rental cost. The third model is a stochastic approach that extends the second model and consists of sizing a rig fleet considering the uncertainty in the intervention time. The uncertainty in the intervention time is represented by the generation of scenarios, using for this purpose the Monte Carlo, Scenario Reduction, and Quasi-Monte Carlo methods. The stability tests proposed by Kaut and Wallace (2003) are applied to evaluate the scenario generation methods and to establish the number of appropriate scenarios to solve the problem. To evaluate the performance of the proposed models, several computational experiments were performed in small, medium and large instances. All instances are based on real cases in Brazil. The results show that the proposed models were able to solve all of the instances considered, including the large instances, proving to be efficient when compared to various metaheuristics, as they produce exact solutions in small computational time. An analysis of the impact on the solutions when there is a change in the oil price and the planning horizon is also carried out. The resolution methodology employed in the third model showed that the Quasi-Monte Carlo method provided the best scenarios to represent the uncertainty and also the potential of the model to solve large-scale problems.