<p>Plasma fueling serves as a key control technique, enabling precise feedback control of plasma density and its profiles. To achieve effective control, real-time acquisition of fueling beam profiles is necessary to meet the requirements for millisecond-level response times and accurate feedback control algorithms. Fuel injection, such as supersonic molecular beam injection, is significantly influenced by various parameters, particularly rapid variations in fueling beam profiles. These variations are primarily driven by sudden, large-scale gas injection, which induces local transient pressure drops that can significantly impact the profile control in real-time plasma density and profile regulation. Establishing a profile database through experimental testing or calibrating beam simulations based on test results is impractical due to their high computational demands and long processing times, which are unsuitable for real-time applications with rapidly changing parameters. This study presents a neural network-based approach to predict the parameter profiles of the supersonic molecular beam based on gas source pressure variations, enabling real-time plasma profile control. The method’s key advantage lies in its low computational demands and time requirements, making it suitable for feedback control applications. It not only enhances real-time response but also optimizes data processing, allowing for faster adaptation to dynamic environments. Extensive testing across various data scales demonstrates that the model maintains high predictive accuracy and generalization ability even with limited data. This method is particularly advantageous for real-time prediction tasks in complex physical systems with strong regularities and has the potential to provide critical data support for real-time plasma profile control in future fusion devices.</p>