Power system control is one of the numerous tasks of the transmission system operator, whose purpose is to ensure the safe and reliable operation of the power grid. The complexity of the current and expected future power system is increasing due to the integration of renewable energy sources and new devices into the grid. As the structure of the power system becomes more complex, its operation and management become increasingly challenging. Changes result in increased system complexity, where the number of active control points in the network is too large to be managed manually. Therefore, the application of artificial intelligence technology in the power system emerges as a possible solution to this problem. The focus of this research is on power flow control to prevent overloads in the grid, which can lead to cascading failures and system collapse. The technical method proposed for this application is changing the network topology through bus splitting. To ensure a fast response in decision-making and the ability to learn automatically from data, the proposed method is based on machine learning algorithms. Specifically, reinforcement learning (RL) algorithms have been chosen due to their main characteristic, which is the agent learning the action selection strategy through interaction with the environment. In addition to the final research product, an autonomous RL agent for power flow control using bus splitting, the paper provides a detailed presentation of all research elements that preceded the final outcome. The developed environment for simulating the operation of the power system and RL is shown, the test network and operating scenarios used are described in detail, and the algorithm for selecting bus topologies, which led to a significant improvement in the RL agent's performance, is presented. Through the analysis of results, the significant potential of using reinforcement learning in power system control has been confirmed.