Thrust prediction for cold gas thrusters using an optimized neural network approach

This paper investigates the use of feedforward neural networks for predicting the thrust of cold gas thrusters, a critical task in space propulsion systems. We implement the proposed model using MATLAB and apply the holdout and K-fold methods to separate the data into training, test, and validation sets. These methods efficiently handle large datasets within time constraints. In addition, we compare two optimization methods, Adam and stochastic gradient descent (SGD), to assess their influence on model performance. Our proposed method integrates SGD for efficient training and uses a correlation matrix-based approach to eliminate redundant parameters, ensuring faster convergence and improved reliability. Simulation results indicate that the mean square error (MSE) for SGD and Adam optimizations after 100 epochs is 0.061 and 0.137, respectively, demonstrating SGD’s robustness and generalizability for cold gas thrusters. These findings underscore the effectiveness of optimization techniques in improving the accuracy of thrust predictions.