Identifying the parameters of a dynamic model under real-time operating conditions is a challenging task, especially when dealing with uncertain disturbances. This study introduces a reinforcement learning-based approach for robust parameter estimation, enabling the determination of optimal model parameters directly from raw data. The proposed method operates as an online parameter identification algorithm, eliminating the need for labeled training datasets or observation windows. It demonstrates superior performance compared to classical methods, effectively rejecting unknown disturbances in dynamic processes and adapting seamlessly to new environments without requiring historical data. The algorithm’s robustness and effectiveness are validated through simulations focused on acquiring accurate parameter values.