Abstract: The equivalent stress of forgings has a great influence on the quality of forgings. It is difficult to realize on-line measurement and comprehensive evaluation in forging production. In this paper, elastic-plastic finite element method was used to carry out finite element simulation and experimental verification of titanium alloy disc forgings, and the distribution of equivalent stress after loading was analyzed. The experimental results were analyzed by grey correlation analysis. Results show that the initial workblank temperature has the greatest effect on the equivalent stress of the forging, and the room temperature has the least effect. Based on the equivalent stress distribution, principal component analysis (PCA) was used to establish a stress comprehensive evaluation model, with workblank temperature, mould temperature, pressing speed, room temperature and friction factor as the input layer and stress comprehensive evaluation value as the output layer. A back propagation (BP) artificial neural network model was established to predict the synthetic equivalent stress. Results show that the predicted results of BP neural network are consistent with the experimental results. The artificial neural network model is proved to be effective in predicting the synthetic equivalent stress. The neural network model can realize the online prediction of the stress comprehensive evaluation value of forgings according to the technological parameters in forging production, and indirectly evaluate the forging process, which can be combined with the quality information obtained from other online measurement methods to provide a basis for the quality control and process improvement of forging production.