Trend Prediction: Construction of Extended Cellular Automata Model

This chapter focuses on the prediction of TSUE and develops the extended CA model. It first identifies the core limitations of the traditional CA models, including excessive reliance on empirical rules and insufficient capacity for geographic feature representation. On this basis, machine learning algorithms (CNN, LSTM, and RF) are incorporated to optimize the traditional CA models, forming a coupled CNN-LSTM-RF-CA framework. The proposed model integrates data-driven mining and knowledge-guided logic, extracts complex spatiotemporal features, and calculates cellular transition probabilities to significantly improve simulation accuracy. Empirical verification based on panel data of 340 Chinese cities demonstrates that the extended CA model outperforms the traditional model in multiple accuracy metrics. This study provides a reliable methodological approach for the dynamic prediction of TSUE.