As a widely adopted low-power wide-area network technology, LoRa (Long Range) enables long-distance bidirectional communication between devices and has been extensively adopted in the Internet of Things (IoT). However, inefficient utilization of radio resources, such as improper assignment of the spread spectrum factor by LoRa terminal devices, can substantially degrade network performance, reduce device battery life, and impair adaptability to dynamic network conditions. To address this issue, a reinforcement learning-based method for assigning LoRa spread spectrum factors is proposed. First, an energy consumption model and a collision model are established by considering factors such as the spreading factor, transmission load, and other relevant parameters. These models serve as the basis for evaluating the reward function during model training. A reinforcement learning model formulated as a Markov Decision Process (MDP) is designed and trained using appropriate algorithms, resulting in a system capable of adaptively selecting a near-optimal spreading factor for terminal devices. Subsequently, leveraging existing simulation tools, the performance of the proposed mechanism is analyzed in terms of energy efficiency and packet collision rate.