<p>The flow stress of surface hardened alloy structural steel during high temperature deformation was investigated using a Gleeble thermal machine. The Arrhenius constitutive equation, incorporating strain rate, was constructed to predict the flow stress of the alloy. On this basis, a convergent Monte Carlo method was developed to optimize the value of the hot deformation activation energy and to predict optimized flow stress values. Based on the BP artificial neural network, a sparse detection neural network (SDNN) model was developed to predict flow stress. The results indicate that the Arrhenius model is basically accurate in predicting the flow stress value, and with greater accuracy observed at lower stress levels. As the flow stress value increases, the prediction becomes less accurate. The convergent Monte Carlo method improved the fitting accuracy of flow stress prediction, particularly at higher flow stress levels. The SDNN model results show that the accuracy is significantly improved, with the average absolute relative error reduced to 1.69%. Even at high flow stress levels, the prediction remains accurate, enabling the use of sparse samples for precise prediction across the dataset.</p>