In this paper, we investigate a resource efficiency

problem for multi-LEO communication systems, addressing the

challenges posed by limited spectrum and energy resources, as

well as severe cross-layer interference. Specifically, we jointly

optimize the transmit power, antenna beamwidth, and both elevation

and azimuth angles, in order to simultaneously minimize

resource inefficiency and average interference of the downlink

transmissions. To solve this optimization problem, we propose

an evolutionary reinforcement learning (ERL) framework, where

the networks are updated using gradient descent in deep reinforcement

learning (DRL), and enhanced via a genetic algorithm.

Statistical results demonstrate the effectiveness of the proposed

method, with a 30.9% increase in resource efficiency and a 39.2%

decrease in interference level.