Current housing unit design focuses on performance optimization, such as building energy use, daylighting, and occupant comfort. There is significant potential for efficient optimization in the early design stage. This paper proposed a building energy demand optimization framework that considers building design variables for housing unit design in Beijing. First, a high-rise apartment building was designed using geometric and thermal envelope parameters. Second, a thermal load simulation of the housing unit was performed. Third, sensitivity analysis and multi-objective optimization were conducted to evaluate the building’s performance. Finally, a neural network was trained to predict the thermal load. A comparison of the base case and optimal cases revealed that more than 20% of energy demand could be saved. In addition, the effect of the parameters on the thermal load and the lighting schedules on the unit’s energy load were analyzed.