Abstract: In industrial production processes, the inconsistency of data sampling intervals is a major challenge for key variable prediction, and most traditional models are unable to handle data sequences with irregular sampling intervals. To address the problem that conventional models cannot directly model data with inconsistent sampling intervals, this study proposed a prediction method based on an Adaptive Interval-Weighted Network （AIWN）. The AIWN introduced a dynamic weight generator that transformed time intervals into weight values, thereby adaptively adjusting the influence of historical information on prediction accuracy. In addition, the AIWN incorporated a skip connection mechanism to perform a lightweight transformation on information from the previous time step, preserving historical information while conducting a convex combination of the original historical information and the transformed information based on the time-interval-derived weights, thereby achieving adaptive input construction.The results demonstrated that, through validation in the flue gas acid production process, the AIWN exhibited strong predictive performance for key variables, achieving an R²; value above 0.9. The research conclusions provide a reference for process variable prediction under irregular sampling conditions in industrial processes and effectively assist operators in monitoring changes in process variables in a timely manner.