This paper presents a method for transforming raw cloud-based thermostat data for cycling systems into a set of operating modes that is useful for large scale data analysis. Thermostat data typically includes the setpoint temperatures, the actual indoor temperature and the operating mode. These raw thermostat operating modes include `cooling on', `heating on', and `system off'. The transformed operating modes include regulating modes, tracking modes, and free response modes. These new modes, which can be generated during data preprocessing, are used to more clearly show key system performance metrics and identify change points in the time series thermostat data. This paper includes the filtering logic used to label the operating modes and examples of insightful behavior that has been captured using this preprocessing method.