Abstract: Objective The accurate prediction of gray-spot area of rail flash butt welded joints is of great significance for the evaluation of rail welding quality. This paper aims to improve prediction accuracy of gray-spot area in welded joints. Methods A prediction model of gray-spot area of welded joints based on principal component analysis (PCA) and back propagation (BP) neural network optimized by genetic algorithm improved particle swarm optimization algorithm (GPSO) was proposed in this paper. PCA was used to reduce the dimension of the characteristic quantities that affected gray-spot area, remove redundant information contained in the original data, and auxiliary variables extracted by PCA were used as the input of the prediction network model. GPSO was used to optimize the initial weights and thresholds of BP neural network, and gray-spot area prediction model based on PCA-GPSO-BP neural network was established. Combined with example data, it was predicted and compared with traditional BP, PCA-BP, PCA-PSO-BP models. Results The results showed that PCA-GPSO-BP model had the highest prediction accuracy, compared with the traditional BP model, the three error indicators of MAX, MAE and RMSE were reduced by 50.97%, 68.51% and 62.43% respectively. The correlation coefficient between the predicted value and the actual value of gray-spot area in the test samples was 0.995 6. Conclusion The PCA-GPSO-BP model can effectively improve prediction accuracy of gray-spot area of rail flash butt welded joints, which has significant engineering application value.