Acoustic liners are sound-absorbing materials widely used across various industrial applications. In the aerospace sector, they are typically installed in jet engine nacelles to reduce noise. Additionally, they are employed in other industries, such as ducts for industrial machinery and ventilation systems in air conditioning. One of the key limitations of traditional acoustic liners is their static nature—they cannot adapt to changing conditions, making them unsuitable for handling highly stochastic acoustic fields. Furthermore, to absorb lower frequencies, they require a large volume, conflicting with the spatial constraints of liners. As a result, there is a growing need to develop new liner technologies that surpass these limitations. This work focuses specifically on advancing control methods for an innovative technology known as the electroacoustic resonator. An electroacoustic resonator consists of a speaker membrane and a controller. The controller drives the membrane with an electric current and based on sound pressure measurements from a microphone, adjusts the membrane's velocity to target the desired noise absorption across different frequency ranges [1]. In the current version of this actuator, the target impedance can be predefined, but real-time impedance adaptation is not yet possible. Meanwhile, reinforcement learning techniques have gained widespread use across various engineering fields. They differ from supervised learning algorithms by their no reliance on pre-existing datasets, using a reward function alongside environment observations to achieve their control objective seeking a long-term maximum reward. This work seeks to leverage these algorithms to develop an adaptable, reinforcement learning-based impedance control system for electroacoustic resonators. To achieve this goal, this work evaluates the reinforcement learning algorithms Deep Deterministic Policy Gradient and Twin Delayed DDPG (TD3) applied to the control of electroacoustic absorbers. These algorithms will be trained using a digital twin of an impedance tube where one-dimensional acoustic propagation occurs. A test will also be conducted using the Continual Action Contextual Bandit (CATS) [3] to evaluate the effectiveness of sequential decision-making in the algorithm performance. At the end the models will be evaluated in an experimental setup