<p>A series of isothermal holding experiments were conducted under different experimental conditions to investigate the austenite grain growth behavior of AerMet100 ultra-high-strength steel. Both cellular automaton and phase field methods were employed to simulate the grain growth behavior of AerMet100 steel. The feasibilities of these methods were verified based on the results obtained from experiments and simulations. Higher accuracy in the prediction of dynamic grain growth of AerMet100 steel was observed with the PF method. An Anelli grain growth model was established to incorporate optimized initial grain conditions. To optimize the entire process from input to output data, an advanced Sine chaotic mapping lens opposition-based learning rime optimization algorithm–machine learning model was proposed by utilizing artificial neural network technology. The reliance on the specific number of hidden functions or layers for accurate predictions was eliminated by the proposed model. By expanding the dataset, a Shapley additive explanations interpretable analysis framework was also constructed. According to the analysis results, the austenite grain growth behavior of AerMet100 steel is enhanced by increasing the heating temperature and prolonging the isothermal holding time. The austenite grain size of AerMet100 steel is more significantly affected by the heating temperature compared to the isothermal holding time.</p>