Reconfigurable Intelligent Surface (RIS) technology relies on its reconfigurable electromagnetic properties and offers an efficient solution for enhancing signal quality in coal mine communications. RIS technology significantly enhances signal coverage and transmission quality in complex, confined environments. This paper proposes a channel propagation optimization scheme for coal mine RIS communication systems, using the Deep Deterministic Policy Gradient (DDPG) algorithm. By jointly optimizing base station power allocation and RIS phase shift, this paper comparatively analyzes RIS reflection performance under ideal and non-ideal conditions, focusing on its impact on system propagation rates. A comparison of system stability and convergence rates among the DDPG, A3C, and DQN algorithms reveals that, under the DDPG optimization scheme, the average link rate reaches 6.6 bps/Hz with ideal RIS reflection and 4.6 bps/Hz with non-ideal conditions when the base station transmit power is defined as 38 dBm. Furthermore, increasing the number of RIS units from 8 to 32 results in a system link rate improvement from 5 bps/Hz to 6.8 bps/Hz. The research results provide new design ideas for optimizing coal mine RIS communication systems and open up new solutions for the use of artificial intelligence in complex coal mine tunnel environments.