Dynamic optimization of television media parameters based on reinforcement learning and causal reasoning

To address the issue of quality of experience (QoE) degradation caused by incorrect bitrate selection in film and television media, this paper proposes a dynamic optimization method for film and television media parameters based on reinforcement learning (RL) and causal reasoning. First, based on causal reasoning for RL algorithm optimization (CRPPO), by learning the causal structure of the environment, further construct a causal mask acting on the action space to help reduce the decision space of RL. Then, predict the bandwidth in the next period based on network information, enhance the accuracy of bandwidth prediction to assist the CRPPO algorithm in decision-making. According to the cross-layer bandwidth prediction value, buffer status information, and QoE metrics, flexibly generate bitrate selection strategies to achieve global QoE optimization. Experimental outcome indicates that the comprehensive QoE of the proposed method is at least improved by 14.6%, proving the effectiveness of the proposed method.