This paper proposes a centralized AI-based indoor temperature control system for distributed HVAC systems, presenting an efficient control method utilizing AI algorithms and the Modbus TCP protocol. The overall system structure consists of a Sub-part that manages sensors, heaters, and air conditioners, and a 메인파트 that generates AI-based control commands. The AI control model employing LSTM and DQN algorithms is implemented on an NVIDIA Jetson Orin Nano, trained using 8,761 annual temperature data for 2023 in the Cheonan City, Korea. Additionally, a hysteresis control method is introduced to reduce frequent switching of heaters and air conditioners, enhancing overall power efficiency. The Sub-part uses Raspberry Pi 4B to store temperature, humidity, and power usage data in real-time within an InfluxDB database, enabling remote monitoring. Meanwhile, the 메인파트 facilitates remote management of the overall system status and control process through a web server. Through the proposed approach, both energy efficiency and indoor comfort can be achieved by individually optimizing environmental conditions in each space while minimizing total energy consumption.