Intelligent Agents (IA) and Multi-agent Systems (MAS) are utilized in many domains such as traffic control, distributed problem solving, robotics, and so on. In order to make decisions, they can apply two approaches: Reinforcement Learning (RL) and reasoning. In many research studies, Rule-Based (RB) reasoning is the basic solution to guide the IA in acting. In recent years, based on the power of RL, many IA and MAS exploit this approach to act in different environments. Each of these approaches has its own advantages and drawbacks, and very few researchers tried to integrate them in a seamless and systematic manner. This paper aims to strengthen the IA and MAS through the integration of learning and reasoning. To reach this goal, we run an IA with initialized rules based on which the search space for RL is reduced, and the learning is faster. Based on this improved RL, new rules will be discovered and added to the rules set. This cycle will be continued until the problem is solved. Therefore, in the proposed approach, RB improves the RL and vice versa. The proposed method is evaluated in terms of accuracy and confidence in comparison to the Q-learning as a standard RL method and an RB method. The results reveal that the proposed integrated approach outperforms the solely RL and RB methods based on these criteria.