Flying ad-hoc networks (FANETs) encounter considerable challenges, including low data transmission reliability and suboptimal energy efficiency, primarily due to the high-speed mobility and frequent topological variations of UAV nodes. In particular, although network coding can enhance data transmission robustness, the traditional fixed redundancy strategy fails to adapt to the dynamic network environment in real time. This leads to either excessive or insufficient redundancy packets, which severely restricts the global optimization of reliability and energy consumption. To address this issue, this paper proposes an adaptive adjustment algorithm for network coding redundancy based on deep reinforcement learning. The algorithm generates optimal redundancy parameters by constructing a Markov decision process model and integrating multi-dimensional indicators in real time, such as packet delivery rate, node energy state, and network load capacity. As a result, this approach enables the synergistic optimization of reliability enhancement and energy consumption control. Experimental results demonstrate that, in scenarios with time-varying network states and node energy constraints, the proposed method outperforms the fixed redundancy strategy. It improves the end-to-end transmission success rate, reduces unnecessary redundant packet transmissions, and surpasses existing benchmark methods in terms of packet coding and decoding performance, energy consumption, and transmission efficiency. This study presents a new paradigm for the intelligent optimization of network coding parameters in highly dynamic networks.