Improving Performance of Feedback-Based Real-Time Networks using Model Checking and Reinforcement Learning