Improving heating, ventilating, and air conditioning (HVAC) efficiency is crucial for energy savings and carbon emission reduction. In this study, we employed deep reinforcement learning (DRL) to optimize HVAC system control in commercial buildings. Traditional control methods, such as rule-based and model predictive control, often fall short in dynamic and complex environments. In contrast, DRL combines reinforcement learning with deep neural networks to provide a more adaptive and efficient approach. Focusing on a multi-floor commercial building, we used a binary on/off control strategy to streamline decision-making and enhance scalability. The HVAC control problem is modeled as a finite Markov process, with a deep Q-network optimizing operations based on parameters such as indoor/outdoor temperatures, cloud coverage, and occupancy levels. A comparative analysis using simulations and real-world data collected by sensors from a commercial building in South Korea showed that the DRL-based method significantly reduced the HVAC operation frequency and on/off cycles, achieving superior energy savings while maintaining comfortable temperature levels. These results highlight the potential of DRL for effective HVAC management by balancing energy efficiency with occupant comfort.