With the continuous growth in communication demand, improving the efficiency of resource allocation becomes crucial. Furthermore, flexible resource allocation for meeting the non-uniform and time-varying traffic demand has emerged as an important task in multi-beam satellite systems. To improve power utilization and meet dynamic traffic demand, this paper formulates an optimization objective that minimizes the trade-off between the unmet traffic demand and power consumption. This is realized by optimizing the allocation of channel and their power, while considering the impact of co-channel interference(CCI). We propose the deep reinforcement learning (DRL) technique to optimize resource allocation. Simulation comparisons between our proposed algorithm and benchmark schemes show its effectiveness in achieving a balance between power allocation and traffic demands. Notably, our algorithm outperforms others in terms of power consumption and meeting traffic demand.