With the acceleration of urbanization, traditional population prediction models have shown limitations in dealing with complex and changeable regional interactions, especially in quantifying the spatial correlation and resource siphoning effect between urban network nodes. To this end, this study introduces the PageRank algorithm into the field of urban population growth prediction. First, cities are abstracted as network nodes, and a directed weighted graph is constructed with population mobility as edge weights. Then, the transition probability matrix of the traditional PageRank algorithm is improved, and coefficients such as urban resource carrying capacity are introduced as damping factors. Finally, the data is integrated to construct a dynamic iterative equation for training and verification. The experimental results show that the average prediction error of the model in the five core cities is 120,000 people; in the trend identification experiment, the PageRank classification accuracy is 90%; in the multi-region generalization test, the STD-MAE (Standard Deviation of Mean Absolute Error) is controlled within 10,000 people, and the trend consistency is as high as 85.0%. The results verify the effectiveness of PageRank in improving prediction accuracy, trend judgment ability and model stability.