Heating, ventilating, and air conditioning (HVAC) systems are widely used and can constitute up to 48% of a commercial building's energy consumption. Computational methods of optimizing the energy consumption of these systems require a model of the system to be identified. Modeling the complexities of HVAC systems is well-suited for machine learning algorithms and has been the focus of an increasing amount of research on this topic. Existing works have approached modeling HVAC systems using data driven methods, physics based methods, or a mixture of the two. In this thesis, we propose a data driven modeling method that generalizes well to data outside the bounds of the training data by modeling the ambient temperature of the building and the influence of the HVAC system separately. Such a model can simulate an HVAC system and can be used as part of larger generalization frameworks, such as reinforcement learning, to optimize the HVAC system. We evaluate our model by its ability to accurately predict the temperature of a zone given a window of environmental data and find that CoolWave produces a strong improvement in prediction quality and generalizability when compared the performance to an implementation of Wavenet, an Auto-Regressor, and a Gaussian Process Regressor.