Quality prediction of products on the Surface Mount Technology (SMT) production line is crucial for improving product yield. However, existing methods rely on predicting based on the mapping relationship between process parameters and product quality. In real production environments, product quality is often influenced by unquantifiable factors such as environmental temperature and stencil wear. Therefore, when predicting product quality, it is necessary to consider the actual production conditions on the assembly line and incorporate the time dependency and trends among product qualities to ensure the accuracy of prediction results. This paper proposes a SMT product quality prediction model that combines an enhanced Informer model with change-point detection. The solder paste printing quality of an SMT production line at a research institute is used as the validation object. The ratio of actual to theoretical values of solder paste volume and area serves as the indicator for assessing the quality of solder paste printing. Firstly, change-point detection is utilized to optimize the training data, thereby improving the model’s prediction accuracy. Secondly, the one-dimensional convolution in the encoder of the Informer model is replaced with time convolution to enhance the model’s receptive field and prevent the “leakage” of future-to-past information. Experimental results show that the model exhibits higher prediction accuracy. Compared to existing mainstream prediction algorithms, MAE for the ratio of actual to theoretical solder paste area is reduced by 19% to 29%, and RMSE is reduced by 16% to 23%. The MAE for the ratio of actual to theoretical solder paste volume is reduced by 8% to 18%, and the RMSE is reduced by 10% to 16%. This confirms the higher prediction accuracy of the model proposed in this paper.