Hall thrusters represent a promising alternative to chemical propulsion for deep-space missions. They are impacted, however, by plasma instabilities that can degrade their performance and lifetime. Active control is an approach that can be taken to control oscillations in the load by actively compensating for them. While passive control approaches exist, some of which include adjusting the background pressure during ground testing and adjusting the harness inductance, active control better addresses instabilities that occur at the microsecond time scale in real-time. One such form is voltage control, which can vary the injected voltage to the discharge plasma in real time in order to control oscillations. This study first explores pressure control and inductance control as passive control approaches, then seeks to find a way to improve our understanding of the discharge plasma load, through a 0D ionization model along with system identification using open-loop voltage perturbations and real-time active PID control to create a discharge plasma prediction model using neural networks. We use this neural network load model to better inform and understand real-time active machine learning voltage control. This approach could help eliminate challenges with differences in performance between ground and space, as machine learning offers a way for a power processing unit to actively learn from changes in its operating environment.