Simulating human behavior in virtual environments is still a challenge. Yet, this feature is crucial to make them more believable. Therefore, many approaches have been proposed in order to find ways to faithfully simulate human behavior. One of the key features to make a virtual agent more believable is the simulation of needs such as hunger or tiredness. Unfortunately, most of the existing approaches in needs simulation do not really address the issue of long-term simulations where some problems may appear such as time drift. Yet, this kind of simulation is useful in many fields like video games or virtual data generation. This is why, in this paper, we focus on the creation of a needs model designed for long-term simulations. According to this model, needs can evolve over several simulated days without interruption. This model is configured to obtain a proper relation between control and autonomy in order to have a coherent behavior during long periods. This paper deals with the key features to set up this needs model and introduces some preliminary results to check the coherence of the agent behavior.