Machine Learning Algorithms Optimize Prediction Accuracy in Submersible Ballast Tank High-pressure Air Blowing Processes

Abstract The optimization of prediction accuracy in the high-pressure air blowing processes of submersible ballast tanks is crucial for enhancing the safety and efficiency of submersible operations. This study presents a comparative analysis between machine learning algorithms and traditional numerical simulation methods for predicting the dynamics of air blowing in submersible ballast tanks. The investigation focuses on the performance of the Back Propagation Neural Network (BPNN) and the Radial Basis Function Neural Network (RBFNN), employing a series of experimental data to train and validate the models.The results indicate that while traditional numerical simulations exhibit a maximum relative error of 124.805%, the RBFNN, in particular, achieves a remarkable linear correlation coefficient of 1, a mean square error of 5.6758e-23, and a mean absolute percentage error of 7.4888e-11. These findings demonstrate the superior predictive capability of the RBFNN, suggesting that machine learning algorithms can significantly improve the accuracy of predictions concerning the high-pressure air blowing process. The study advocates for the adoption of machine learning techniques in submersible engineering, potentially transforming the reliability and safety of deep-sea exploration endeavors.