Reinforcement-Learning-Based Variational Quantum Circuits Optimization for Combinatorial Problems